1 /* $NetBSD: kern_acct.c,v 1.52 2003/04/20 03:20:18 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1994 Christopher G. Demetriou 5 * Copyright (c) 1982, 1986, 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)kern_acct.c 8.8 (Berkeley) 5/14/95 42 */ 43 44 #include <sys/cdefs.h> 45 __KERNEL_RCSID(0, "$NetBSD: kern_acct.c,v 1.52 2003/04/20 03:20:18 christos Exp $"); 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/proc.h> 50 #include <sys/mount.h> 51 #include <sys/vnode.h> 52 #include <sys/file.h> 53 #include <sys/syslog.h> 54 #include <sys/kernel.h> 55 #include <sys/kthread.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/namei.h> 59 #include <sys/errno.h> 60 #include <sys/acct.h> 61 #include <sys/resourcevar.h> 62 #include <sys/ioctl.h> 63 #include <sys/tty.h> 64 65 #include <sys/sa.h> 66 #include <sys/syscallargs.h> 67 68 /* 69 * The routines implemented in this file are described in: 70 * Leffler, et al.: The Design and Implementation of the 4.3BSD 71 * UNIX Operating System (Addison Welley, 1989) 72 * on pages 62-63. 73 * 74 * Arguably, to simplify accounting operations, this mechanism should 75 * be replaced by one in which an accounting log file (similar to /dev/klog) 76 * is read by a user process, etc. However, that has its own problems. 77 */ 78 79 /* 80 * The global accounting state and related data. Gain the lock before 81 * accessing these variables. 82 */ 83 enum { 84 ACCT_STOP, 85 ACCT_ACTIVE, 86 ACCT_SUSPENDED 87 } acct_state; /* The current accounting state. */ 88 struct vnode *acct_vp; /* Accounting vnode pointer. */ 89 struct ucred *acct_ucred; /* Credential of accounting file 90 owner (i.e root). Used when 91 accounting file i/o. */ 92 struct proc *acct_dkwatcher; /* Free disk space checker. */ 93 94 /* 95 * Lock to serialize system calls and kernel threads. 96 */ 97 struct lock acct_lock; 98 #define ACCT_LOCK() \ 99 do { \ 100 (void) lockmgr(&acct_lock, LK_EXCLUSIVE, NULL); \ 101 } while (/* CONSTCOND */0) 102 #define ACCT_UNLOCK() \ 103 do { \ 104 (void) lockmgr(&acct_lock, LK_RELEASE, NULL); \ 105 } while (/* CONSTCOND */0) 106 107 /* 108 * Internal accounting functions. 109 * The former's operation is described in Leffler, et al., and the latter 110 * was provided by UCB with the 4.4BSD-Lite release 111 */ 112 comp_t encode_comp_t __P((u_long, u_long)); 113 void acctwatch __P((void *)); 114 void acct_stop __P((void)); 115 int acct_chkfree __P((void)); 116 117 /* 118 * Values associated with enabling and disabling accounting 119 */ 120 int acctsuspend = 2; /* stop accounting when < 2% free space left */ 121 int acctresume = 4; /* resume when free space risen to > 4% */ 122 int acctchkfreq = 15; /* frequency (in seconds) to check space */ 123 124 void 125 acct_init() 126 { 127 128 acct_state = ACCT_STOP; 129 acct_vp = NULLVP; 130 acct_ucred = NULL; 131 lockinit(&acct_lock, PWAIT, "acctlk", 0, 0); 132 } 133 134 void 135 acct_stop() 136 { 137 int error; 138 139 if (acct_vp != NULLVP && acct_vp->v_type != VBAD) { 140 error = vn_close(acct_vp, FWRITE, acct_ucred, NULL); 141 #ifdef DIAGNOSTIC 142 if (error != 0) 143 printf("acct_stop: failed to close, errno = %d\n", 144 error); 145 #endif 146 acct_vp = NULLVP; 147 } 148 if (acct_ucred != NULL) { 149 crfree(acct_ucred); 150 acct_ucred = NULL; 151 } 152 acct_state = ACCT_STOP; 153 } 154 155 int 156 acct_chkfree() 157 { 158 int error; 159 struct statfs sb; 160 161 error = VFS_STATFS(acct_vp->v_mount, &sb, NULL); 162 if (error != 0) 163 return (error); 164 165 switch (acct_state) { 166 case ACCT_SUSPENDED: 167 if (sb.f_bavail > acctresume * sb.f_blocks / 100) { 168 acct_state = ACCT_ACTIVE; 169 log(LOG_NOTICE, "Accounting resumed\n"); 170 } 171 break; 172 case ACCT_ACTIVE: 173 if (sb.f_bavail <= acctsuspend * sb.f_blocks / 100) { 174 acct_state = ACCT_SUSPENDED; 175 log(LOG_NOTICE, "Accounting suspended\n"); 176 } 177 break; 178 case ACCT_STOP: 179 break; 180 } 181 return (0); 182 } 183 184 /* 185 * Accounting system call. Written based on the specification and 186 * previous implementation done by Mark Tinguely. 187 */ 188 int 189 sys_acct(l, v, retval) 190 struct lwp *l; 191 void *v; 192 register_t *retval; 193 { 194 struct sys_acct_args /* { 195 syscallarg(const char *) path; 196 } */ *uap = v; 197 struct nameidata nd; 198 int error; 199 struct proc *p = l->l_proc; 200 201 /* Make sure that the caller is root. */ 202 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 203 return (error); 204 205 /* 206 * If accounting is to be started to a file, open that file for 207 * writing and make sure it's a 'normal'. 208 */ 209 if (SCARG(uap, path) != NULL) { 210 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), 211 p); 212 if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0) 213 return (error); 214 VOP_UNLOCK(nd.ni_vp, 0); 215 if (nd.ni_vp->v_type != VREG) { 216 vn_close(nd.ni_vp, FWRITE, p->p_ucred, p); 217 return (EACCES); 218 } 219 } 220 221 ACCT_LOCK(); 222 223 /* 224 * If accounting was previously enabled, kill the old space-watcher, 225 * free credential for accounting file i/o, 226 * ... (and, if no new file was specified, leave). 227 */ 228 acct_stop(); 229 if (SCARG(uap, path) == NULL) 230 goto out; 231 232 /* 233 * Save the new accounting file vnode and credential, 234 * and schedule the new free space watcher. 235 */ 236 acct_state = ACCT_ACTIVE; 237 acct_vp = nd.ni_vp; 238 acct_ucred = p->p_ucred; 239 crhold(acct_ucred); 240 241 error = acct_chkfree(); /* Initial guess. */ 242 if (error != 0) { 243 acct_stop(); 244 goto out; 245 } 246 247 if (acct_dkwatcher == NULL) { 248 error = kthread_create1(acctwatch, NULL, &acct_dkwatcher, 249 "acctwatch"); 250 if (error != 0) 251 acct_stop(); 252 } 253 254 out: 255 ACCT_UNLOCK(); 256 return (error); 257 } 258 259 /* 260 * Write out process accounting information, on process exit. 261 * Data to be written out is specified in Leffler, et al. 262 * and are enumerated below. (They're also noted in the system 263 * "acct.h" header file.) 264 */ 265 int 266 acct_process(p) 267 struct proc *p; 268 { 269 struct acct acct; 270 struct rusage *r; 271 struct timeval ut, st, tmp; 272 int s, t, error = 0; 273 struct plimit *oplim = NULL; 274 275 ACCT_LOCK(); 276 277 /* If accounting isn't enabled, don't bother */ 278 if (acct_state != ACCT_ACTIVE) 279 goto out; 280 281 /* 282 * Raise the file limit so that accounting can't be stopped by 283 * the user. 284 * 285 * XXX We should think about the CPU limit, too. 286 */ 287 if (p->p_limit->p_refcnt > 1) { 288 oplim = p->p_limit; 289 p->p_limit = limcopy(p->p_limit); 290 } 291 p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 292 293 /* 294 * Get process accounting information. 295 */ 296 297 /* (1) The name of the command that ran */ 298 memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm)); 299 300 /* (2) The amount of user and system time that was used */ 301 calcru(p, &ut, &st, NULL); 302 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 303 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 304 305 /* (3) The elapsed time the commmand ran (and its starting time) */ 306 acct.ac_btime = p->p_stats->p_start.tv_sec; 307 s = splclock(); 308 timersub(&time, &p->p_stats->p_start, &tmp); 309 splx(s); 310 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 311 312 /* (4) The average amount of memory used */ 313 r = &p->p_stats->p_ru; 314 timeradd(&ut, &st, &tmp); 315 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 316 if (t) 317 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 318 else 319 acct.ac_mem = 0; 320 321 /* (5) The number of disk I/O operations done */ 322 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 323 324 /* (6) The UID and GID of the process */ 325 acct.ac_uid = p->p_cred->p_ruid; 326 acct.ac_gid = p->p_cred->p_rgid; 327 328 /* (7) The terminal from which the process was started */ 329 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 330 acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev; 331 else 332 acct.ac_tty = NODEV; 333 334 /* (8) The boolean flags that tell how the process terminated, etc. */ 335 acct.ac_flag = p->p_acflag; 336 337 /* 338 * Now, just write the accounting information to the file. 339 */ 340 VOP_LEASE(acct_vp, p, p->p_ucred, LEASE_WRITE); 341 error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, 342 sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, 343 acct_ucred, NULL, p); 344 if (error != 0) 345 log(LOG_ERR, "Accounting: write failed %d\n", error); 346 347 if (oplim) { 348 limfree(p->p_limit); 349 p->p_limit = oplim; 350 } 351 352 out: 353 ACCT_UNLOCK(); 354 return (error); 355 } 356 357 /* 358 * Encode_comp_t converts from ticks in seconds and microseconds 359 * to ticks in 1/AHZ seconds. The encoding is described in 360 * Leffler, et al., on page 63. 361 */ 362 363 #define MANTSIZE 13 /* 13 bit mantissa. */ 364 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 365 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 366 367 comp_t 368 encode_comp_t(s, us) 369 u_long s, us; 370 { 371 int exp, rnd; 372 373 exp = 0; 374 rnd = 0; 375 s *= AHZ; 376 s += us / (1000000 / AHZ); /* Maximize precision. */ 377 378 while (s > MAXFRACT) { 379 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 380 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 381 exp++; 382 } 383 384 /* If we need to round up, do it (and handle overflow correctly). */ 385 if (rnd && (++s > MAXFRACT)) { 386 s >>= EXPSIZE; 387 exp++; 388 } 389 390 /* Clean it up and polish it off. */ 391 exp <<= MANTSIZE; /* Shift the exponent into place */ 392 exp += s; /* and add on the mantissa. */ 393 return (exp); 394 } 395 396 /* 397 * Periodically check the file system to see if accounting 398 * should be turned on or off. Beware the case where the vnode 399 * has been vgone()'d out from underneath us, e.g. when the file 400 * system containing the accounting file has been forcibly unmounted. 401 */ 402 void 403 acctwatch(arg) 404 void *arg; 405 { 406 int error; 407 408 log(LOG_NOTICE, "Accounting started\n"); 409 ACCT_LOCK(); 410 while (acct_state != ACCT_STOP) { 411 if (acct_vp->v_type == VBAD) { 412 log(LOG_NOTICE, "Accounting terminated\n"); 413 acct_stop(); 414 continue; 415 } 416 417 error = acct_chkfree(); 418 #ifdef DIAGNOSTIC 419 if (error != 0) 420 printf("acctwatch: failed to statfs, error = %d\n", 421 error); 422 #endif 423 424 ACCT_UNLOCK(); 425 error = tsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz); 426 ACCT_LOCK(); 427 #ifdef DIAGNOSTIC 428 if (error != 0 && error != EWOULDBLOCK) 429 printf("acctwatch: sleep error %d\n", error); 430 #endif 431 } 432 acct_dkwatcher = NULL; 433 ACCT_UNLOCK(); 434 435 kthread_exit(0); 436 } 437