1 /* $NetBSD: kern_acct.c,v 1.61 2005/06/23 23:15:12 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_acct.c 8.8 (Berkeley) 5/14/95 37 */ 38 39 /*- 40 * Copyright (c) 1994 Christopher G. Demetriou 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)kern_acct.c 8.8 (Berkeley) 5/14/95 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: kern_acct.c,v 1.61 2005/06/23 23:15:12 thorpej Exp $"); 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/proc.h> 79 #include <sys/mount.h> 80 #include <sys/vnode.h> 81 #include <sys/file.h> 82 #include <sys/syslog.h> 83 #include <sys/kernel.h> 84 #include <sys/kthread.h> 85 #include <sys/lock.h> 86 #include <sys/malloc.h> 87 #include <sys/namei.h> 88 #include <sys/errno.h> 89 #include <sys/acct.h> 90 #include <sys/resourcevar.h> 91 #include <sys/ioctl.h> 92 #include <sys/tty.h> 93 94 #include <sys/sa.h> 95 #include <sys/syscallargs.h> 96 97 /* 98 * The routines implemented in this file are described in: 99 * Leffler, et al.: The Design and Implementation of the 4.3BSD 100 * UNIX Operating System (Addison Welley, 1989) 101 * on pages 62-63. 102 * 103 * Arguably, to simplify accounting operations, this mechanism should 104 * be replaced by one in which an accounting log file (similar to /dev/klog) 105 * is read by a user process, etc. However, that has its own problems. 106 */ 107 108 /* 109 * Lock to serialize system calls and kernel threads. 110 */ 111 static struct lock acct_lock; 112 #define ACCT_LOCK() \ 113 do { \ 114 (void) lockmgr(&acct_lock, LK_EXCLUSIVE, NULL); \ 115 } while (/* CONSTCOND */0) 116 #define ACCT_UNLOCK() \ 117 do { \ 118 (void) lockmgr(&acct_lock, LK_RELEASE, NULL); \ 119 } while (/* CONSTCOND */0) 120 121 /* 122 * The global accounting state and related data. Gain the lock before 123 * accessing these variables. 124 */ 125 static enum { 126 ACCT_STOP, 127 ACCT_ACTIVE, 128 ACCT_SUSPENDED 129 } acct_state; /* The current accounting state. */ 130 static struct vnode *acct_vp; /* Accounting vnode pointer. */ 131 static struct ucred *acct_ucred; /* Credential of accounting file 132 owner (i.e root). Used when 133 accounting file i/o. */ 134 static struct proc *acct_dkwatcher; /* Free disk space checker. */ 135 136 /* 137 * Values associated with enabling and disabling accounting 138 */ 139 int acctsuspend = 2; /* stop accounting when < 2% free space left */ 140 int acctresume = 4; /* resume when free space risen to > 4% */ 141 int acctchkfreq = 15; /* frequency (in seconds) to check space */ 142 143 /* 144 * Encode_comp_t converts from ticks in seconds and microseconds 145 * to ticks in 1/AHZ seconds. The encoding is described in 146 * Leffler, et al., on page 63. 147 */ 148 149 #define MANTSIZE 13 /* 13 bit mantissa. */ 150 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 151 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 152 153 static comp_t 154 encode_comp_t(u_long s, u_long us) 155 { 156 int exp, rnd; 157 158 exp = 0; 159 rnd = 0; 160 s *= AHZ; 161 s += us / (1000000 / AHZ); /* Maximize precision. */ 162 163 while (s > MAXFRACT) { 164 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 165 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 166 exp++; 167 } 168 169 /* If we need to round up, do it (and handle overflow correctly). */ 170 if (rnd && (++s > MAXFRACT)) { 171 s >>= EXPSIZE; 172 exp++; 173 } 174 175 /* Clean it up and polish it off. */ 176 exp <<= MANTSIZE; /* Shift the exponent into place */ 177 exp += s; /* and add on the mantissa. */ 178 return (exp); 179 } 180 181 static int 182 acct_chkfree(void) 183 { 184 int error; 185 struct statvfs sb; 186 int64_t bavail; 187 188 error = VFS_STATVFS(acct_vp->v_mount, &sb, NULL); 189 if (error != 0) 190 return (error); 191 192 bavail = sb.f_bfree - sb.f_bresvd; 193 194 switch (acct_state) { 195 case ACCT_SUSPENDED: 196 if (bavail > acctresume * sb.f_blocks / 100) { 197 acct_state = ACCT_ACTIVE; 198 log(LOG_NOTICE, "Accounting resumed\n"); 199 } 200 break; 201 case ACCT_ACTIVE: 202 if (bavail <= acctsuspend * sb.f_blocks / 100) { 203 acct_state = ACCT_SUSPENDED; 204 log(LOG_NOTICE, "Accounting suspended\n"); 205 } 206 break; 207 case ACCT_STOP: 208 break; 209 } 210 return (0); 211 } 212 213 static void 214 acct_stop(void) 215 { 216 int error; 217 218 if (acct_vp != NULLVP && acct_vp->v_type != VBAD) { 219 error = vn_close(acct_vp, FWRITE, acct_ucred, NULL); 220 #ifdef DIAGNOSTIC 221 if (error != 0) 222 printf("acct_stop: failed to close, errno = %d\n", 223 error); 224 #endif 225 acct_vp = NULLVP; 226 } 227 if (acct_ucred != NULL) { 228 crfree(acct_ucred); 229 acct_ucred = NULL; 230 } 231 acct_state = ACCT_STOP; 232 } 233 234 /* 235 * Periodically check the file system to see if accounting 236 * should be turned on or off. Beware the case where the vnode 237 * has been vgone()'d out from underneath us, e.g. when the file 238 * system containing the accounting file has been forcibly unmounted. 239 */ 240 static void 241 acctwatch(void *arg) 242 { 243 int error; 244 245 log(LOG_NOTICE, "Accounting started\n"); 246 ACCT_LOCK(); 247 while (acct_state != ACCT_STOP) { 248 if (acct_vp->v_type == VBAD) { 249 log(LOG_NOTICE, "Accounting terminated\n"); 250 acct_stop(); 251 continue; 252 } 253 254 error = acct_chkfree(); 255 #ifdef DIAGNOSTIC 256 if (error != 0) 257 printf("acctwatch: failed to statvfs, error = %d\n", 258 error); 259 #endif 260 261 ACCT_UNLOCK(); 262 error = tsleep(acctwatch, PSWP, "actwat", acctchkfreq * hz); 263 ACCT_LOCK(); 264 #ifdef DIAGNOSTIC 265 if (error != 0 && error != EWOULDBLOCK) 266 printf("acctwatch: sleep error %d\n", error); 267 #endif 268 } 269 acct_dkwatcher = NULL; 270 ACCT_UNLOCK(); 271 272 kthread_exit(0); 273 } 274 275 void 276 acct_init(void) 277 { 278 279 acct_state = ACCT_STOP; 280 acct_vp = NULLVP; 281 acct_ucred = NULL; 282 lockinit(&acct_lock, PWAIT, "acctlk", 0, 0); 283 } 284 285 /* 286 * Accounting system call. Written based on the specification and 287 * previous implementation done by Mark Tinguely. 288 */ 289 int 290 sys_acct(struct lwp *l, void *v, register_t *retval) 291 { 292 struct sys_acct_args /* { 293 syscallarg(const char *) path; 294 } */ *uap = v; 295 struct nameidata nd; 296 int error; 297 struct proc *p = l->l_proc; 298 299 /* Make sure that the caller is root. */ 300 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 301 return (error); 302 303 /* 304 * If accounting is to be started to a file, open that file for 305 * writing and make sure it's a 'normal'. 306 */ 307 if (SCARG(uap, path) != NULL) { 308 struct vattr va; 309 size_t pad; 310 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), 311 p); 312 if ((error = vn_open(&nd, FWRITE|O_APPEND, 0)) != 0) 313 return (error); 314 if (nd.ni_vp->v_type != VREG) { 315 VOP_UNLOCK(nd.ni_vp, 0); 316 error = EACCES; 317 goto bad; 318 } 319 if ((error = VOP_GETATTR(nd.ni_vp, &va, p->p_ucred, p)) != 0) { 320 VOP_UNLOCK(nd.ni_vp, 0); 321 goto bad; 322 } 323 324 if ((pad = (va.va_size % sizeof(struct acct))) != 0) { 325 u_quad_t size = va.va_size - pad; 326 #ifdef DIAGNOSTIC 327 printf("Size of accounting file not a multiple of " 328 "%lu - incomplete record truncated\n", 329 (unsigned long)sizeof(struct acct)); 330 #endif 331 VATTR_NULL(&va); 332 va.va_size = size; 333 error = VOP_SETATTR(nd.ni_vp, &va, p->p_ucred, p); 334 if (error != 0) { 335 VOP_UNLOCK(nd.ni_vp, 0); 336 goto bad; 337 } 338 } 339 VOP_UNLOCK(nd.ni_vp, 0); 340 } 341 342 ACCT_LOCK(); 343 344 /* 345 * If accounting was previously enabled, kill the old space-watcher, 346 * free credential for accounting file i/o, 347 * ... (and, if no new file was specified, leave). 348 */ 349 acct_stop(); 350 if (SCARG(uap, path) == NULL) 351 goto out; 352 353 /* 354 * Save the new accounting file vnode and credential, 355 * and schedule the new free space watcher. 356 */ 357 acct_state = ACCT_ACTIVE; 358 acct_vp = nd.ni_vp; 359 acct_ucred = p->p_ucred; 360 crhold(acct_ucred); 361 362 error = acct_chkfree(); /* Initial guess. */ 363 if (error != 0) { 364 acct_stop(); 365 goto out; 366 } 367 368 if (acct_dkwatcher == NULL) { 369 error = kthread_create1(acctwatch, NULL, &acct_dkwatcher, 370 "acctwatch"); 371 if (error != 0) 372 acct_stop(); 373 } 374 375 out: 376 ACCT_UNLOCK(); 377 return (error); 378 bad: 379 vn_close(nd.ni_vp, FWRITE, p->p_ucred, p); 380 return error; 381 } 382 383 /* 384 * Write out process accounting information, on process exit. 385 * Data to be written out is specified in Leffler, et al. 386 * and are enumerated below. (They're also noted in the system 387 * "acct.h" header file.) 388 */ 389 int 390 acct_process(struct proc *p) 391 { 392 struct acct acct; 393 struct rusage *r; 394 struct timeval ut, st, tmp; 395 int s, t, error = 0; 396 struct plimit *oplim = NULL; 397 398 ACCT_LOCK(); 399 400 /* If accounting isn't enabled, don't bother */ 401 if (acct_state != ACCT_ACTIVE) 402 goto out; 403 404 /* 405 * Raise the file limit so that accounting can't be stopped by 406 * the user. 407 * 408 * XXX We should think about the CPU limit, too. 409 */ 410 if (p->p_limit->p_refcnt > 1) { 411 oplim = p->p_limit; 412 p->p_limit = limcopy(p->p_limit); 413 } 414 p->p_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 415 416 /* 417 * Get process accounting information. 418 */ 419 420 /* (1) The name of the command that ran */ 421 memcpy(acct.ac_comm, p->p_comm, sizeof(acct.ac_comm)); 422 423 /* (2) The amount of user and system time that was used */ 424 calcru(p, &ut, &st, NULL); 425 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 426 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 427 428 /* (3) The elapsed time the commmand ran (and its starting time) */ 429 acct.ac_btime = p->p_stats->p_start.tv_sec; 430 s = splclock(); 431 timersub(&time, &p->p_stats->p_start, &tmp); 432 splx(s); 433 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 434 435 /* (4) The average amount of memory used */ 436 r = &p->p_stats->p_ru; 437 timeradd(&ut, &st, &tmp); 438 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 439 if (t) 440 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 441 else 442 acct.ac_mem = 0; 443 444 /* (5) The number of disk I/O operations done */ 445 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 446 447 /* (6) The UID and GID of the process */ 448 acct.ac_uid = p->p_cred->p_ruid; 449 acct.ac_gid = p->p_cred->p_rgid; 450 451 /* (7) The terminal from which the process was started */ 452 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 453 acct.ac_tty = p->p_pgrp->pg_session->s_ttyp->t_dev; 454 else 455 acct.ac_tty = NODEV; 456 457 /* (8) The boolean flags that tell how the process terminated, etc. */ 458 acct.ac_flag = p->p_acflag; 459 460 /* 461 * Now, just write the accounting information to the file. 462 */ 463 VOP_LEASE(acct_vp, p, p->p_ucred, LEASE_WRITE); 464 error = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, 465 sizeof(acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, 466 acct_ucred, NULL, NULL); 467 if (error != 0) 468 log(LOG_ERR, "Accounting: write failed %d\n", error); 469 470 if (oplim) { 471 limfree(p->p_limit); 472 p->p_limit = oplim; 473 } 474 475 out: 476 ACCT_UNLOCK(); 477 return (error); 478 } 479