1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Copyright (c) 1994 Christopher G. Demetriou 11 * Copyright (c) 2005 Robert N. M. Watson 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 8.1 (Berkeley) 6/14/93 42 */ 43 44 #include <sys/cdefs.h> 45 __FBSDID("$FreeBSD$"); 46 47 #include "opt_mac.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/acct.h> 52 #include <sys/fcntl.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/lock.h> 56 #include <sys/mac.h> 57 #include <sys/mount.h> 58 #include <sys/mutex.h> 59 #include <sys/namei.h> 60 #include <sys/proc.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sched.h> 63 #include <sys/sx.h> 64 #include <sys/sysctl.h> 65 #include <sys/sysent.h> 66 #include <sys/syslog.h> 67 #include <sys/sysproto.h> 68 #include <sys/tty.h> 69 #include <sys/vnode.h> 70 71 /* 72 * The routines implemented in this file are described in: 73 * Leffler, et al.: The Design and Implementation of the 4.3BSD 74 * UNIX Operating System (Addison Welley, 1989) 75 * on pages 62-63. 76 * 77 * Arguably, to simplify accounting operations, this mechanism should 78 * be replaced by one in which an accounting log file (similar to /dev/klog) 79 * is read by a user process, etc. However, that has its own problems. 80 */ 81 82 /* 83 * Internal accounting functions. 84 * The former's operation is described in Leffler, et al., and the latter 85 * was provided by UCB with the 4.4BSD-Lite release 86 */ 87 static comp_t encode_comp_t(u_long, u_long); 88 static void acctwatch(void); 89 static void acct_thread(void *); 90 static int acct_disable(struct thread *); 91 92 /* 93 * Accounting vnode pointer, saved vnode pointer, and flags for each. 94 * acct_sx protects against changes to the active vnode and credentials 95 * while accounting records are being committed to disk. 96 */ 97 static int acct_configured; 98 static int acct_suspended; 99 static struct vnode *acct_vp; 100 static struct ucred *acct_cred; 101 static int acct_flags; 102 static struct sx acct_sx; 103 104 SX_SYSINIT(acct, &acct_sx, "acct_sx"); 105 106 /* 107 * State of the accounting kthread. 108 */ 109 static int acct_state; 110 111 #define ACCT_RUNNING 1 /* Accounting kthread is running. */ 112 #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */ 113 114 /* 115 * Values associated with enabling and disabling accounting 116 */ 117 static int acctsuspend = 2; /* stop accounting when < 2% free space left */ 118 SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, 119 &acctsuspend, 0, "percentage of free disk space below which accounting stops"); 120 121 static int acctresume = 4; /* resume when free space risen to > 4% */ 122 SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, 123 &acctresume, 0, "percentage of free disk space above which accounting resumes"); 124 125 static int acctchkfreq = 15; /* frequency (in seconds) to check space */ 126 127 static int 128 sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS) 129 { 130 int error, value; 131 132 /* Write out the old value. */ 133 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int)); 134 if (error || req->newptr == NULL) 135 return (error); 136 137 /* Read in and verify the new value. */ 138 error = SYSCTL_IN(req, &value, sizeof(int)); 139 if (error) 140 return (error); 141 if (value <= 0) 142 return (EINVAL); 143 acctchkfreq = value; 144 return (0); 145 } 146 SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW, 147 &acctchkfreq, 0, sysctl_acct_chkfreq, "I", 148 "frequency for checking the free space"); 149 150 SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0, 151 "Accounting configured or not"); 152 153 SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0, 154 "Accounting suspended or not"); 155 156 /* 157 * Accounting system call. Written based on the specification and 158 * previous implementation done by Mark Tinguely. 159 * 160 * MPSAFE 161 */ 162 int 163 acct(struct thread *td, struct acct_args *uap) 164 { 165 struct nameidata nd; 166 int error, flags, vfslocked; 167 168 /* Make sure that the caller is root. */ 169 error = suser(td); 170 if (error) 171 return (error); 172 173 /* 174 * If accounting is to be started to a file, open that file for 175 * appending and make sure it's a 'normal'. 176 */ 177 if (uap->path != NULL) { 178 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE | AUDITVNODE1, 179 UIO_USERSPACE, uap->path, td); 180 flags = FWRITE | O_APPEND; 181 error = vn_open(&nd, &flags, 0, -1); 182 if (error) 183 return (error); 184 vfslocked = NDHASGIANT(&nd); 185 NDFREE(&nd, NDF_ONLY_PNBUF); 186 #ifdef MAC 187 error = mac_check_system_acct(td->td_ucred, nd.ni_vp); 188 if (error) { 189 VOP_UNLOCK(nd.ni_vp, 0, td); 190 vn_close(nd.ni_vp, flags, td->td_ucred, td); 191 VFS_UNLOCK_GIANT(vfslocked); 192 return (error); 193 } 194 #endif 195 VOP_UNLOCK(nd.ni_vp, 0, td); 196 if (nd.ni_vp->v_type != VREG) { 197 vn_close(nd.ni_vp, flags, td->td_ucred, td); 198 VFS_UNLOCK_GIANT(vfslocked); 199 return (EACCES); 200 } 201 VFS_UNLOCK_GIANT(vfslocked); 202 #ifdef MAC 203 } else { 204 error = mac_check_system_acct(td->td_ucred, NULL); 205 if (error) 206 return (error); 207 #endif 208 } 209 210 /* 211 * Disallow concurrent access to the accounting vnode while we swap 212 * it out, in order to prevent access after close. 213 */ 214 sx_xlock(&acct_sx); 215 216 /* 217 * If accounting was previously enabled, kill the old space-watcher, 218 * close the file, and (if no new file was specified, leave). Reset 219 * the suspended state regardless of whether accounting remains 220 * enabled. 221 */ 222 acct_suspended = 0; 223 if (acct_vp != NULL) { 224 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 225 error = acct_disable(td); 226 VFS_UNLOCK_GIANT(vfslocked); 227 } 228 if (uap->path == NULL) { 229 if (acct_state & ACCT_RUNNING) { 230 acct_state |= ACCT_EXITREQ; 231 wakeup(&acct_state); 232 } 233 sx_xunlock(&acct_sx); 234 return (error); 235 } 236 237 /* 238 * Save the new accounting file vnode, and schedule the new 239 * free space watcher. 240 */ 241 acct_vp = nd.ni_vp; 242 acct_cred = crhold(td->td_ucred); 243 acct_flags = flags; 244 if (acct_state & ACCT_RUNNING) 245 acct_state &= ~ACCT_EXITREQ; 246 else { 247 /* 248 * Try to start up an accounting kthread. We may start more 249 * than one, but if so the extras will commit suicide as 250 * soon as they start up. 251 */ 252 error = kthread_create(acct_thread, NULL, NULL, 0, 0, 253 "accounting"); 254 if (error) { 255 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 256 (void) vn_close(acct_vp, acct_flags, acct_cred, td); 257 VFS_UNLOCK_GIANT(vfslocked); 258 crfree(acct_cred); 259 acct_configured = 0; 260 acct_vp = NULL; 261 acct_cred = NULL; 262 acct_flags = 0; 263 sx_xunlock(&acct_sx); 264 log(LOG_NOTICE, "Unable to start accounting thread\n"); 265 return (error); 266 } 267 } 268 acct_configured = 1; 269 sx_xunlock(&acct_sx); 270 log(LOG_NOTICE, "Accounting enabled\n"); 271 return (error); 272 } 273 274 /* 275 * Disable currently in-progress accounting by closing the vnode, dropping 276 * our reference to the credential, and clearing the vnode's flags. 277 */ 278 static int 279 acct_disable(struct thread *td) 280 { 281 int error; 282 283 sx_assert(&acct_sx, SX_XLOCKED); 284 error = vn_close(acct_vp, acct_flags, acct_cred, td); 285 crfree(acct_cred); 286 acct_configured = 0; 287 acct_vp = NULL; 288 acct_cred = NULL; 289 acct_flags = 0; 290 log(LOG_NOTICE, "Accounting disabled\n"); 291 return (error); 292 } 293 294 /* 295 * Write out process accounting information, on process exit. 296 * Data to be written out is specified in Leffler, et al. 297 * and are enumerated below. (They're also noted in the system 298 * "acct.h" header file.) 299 */ 300 int 301 acct_process(struct thread *td) 302 { 303 struct acct acct; 304 struct timeval ut, st, tmp; 305 struct plimit *newlim, *oldlim; 306 struct proc *p; 307 struct rusage *r; 308 int t, ret, vfslocked; 309 310 /* 311 * Lockless check of accounting condition before doing the hard 312 * work. 313 */ 314 if (acct_vp == NULL || acct_suspended) 315 return (0); 316 317 sx_slock(&acct_sx); 318 319 /* 320 * If accounting isn't enabled, don't bother. Have to check again 321 * once we own the lock in case we raced with disabling of accounting 322 * by another thread. 323 */ 324 if (acct_vp == NULL || acct_suspended) { 325 sx_sunlock(&acct_sx); 326 return (0); 327 } 328 329 p = td->td_proc; 330 331 /* 332 * Get process accounting information. 333 */ 334 335 PROC_LOCK(p); 336 /* (1) The name of the command that ran */ 337 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); 338 339 /* (2) The amount of user and system time that was used */ 340 calcru(p, &ut, &st); 341 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 342 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 343 344 /* (3) The elapsed time the command ran (and its starting time) */ 345 tmp = boottime; 346 timevaladd(&tmp, &p->p_stats->p_start); 347 acct.ac_btime = tmp.tv_sec; 348 microuptime(&tmp); 349 timevalsub(&tmp, &p->p_stats->p_start); 350 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 351 352 /* (4) The average amount of memory used */ 353 r = &p->p_stats->p_ru; 354 tmp = ut; 355 timevaladd(&tmp, &st); 356 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 357 if (t) 358 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 359 else 360 acct.ac_mem = 0; 361 362 /* (5) The number of disk I/O operations done */ 363 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 364 365 /* (6) The UID and GID of the process */ 366 acct.ac_uid = p->p_ucred->cr_ruid; 367 acct.ac_gid = p->p_ucred->cr_rgid; 368 369 /* (7) The terminal from which the process was started */ 370 SESS_LOCK(p->p_session); 371 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 372 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev); 373 else 374 acct.ac_tty = NODEV; 375 SESS_UNLOCK(p->p_session); 376 377 /* (8) The boolean flags that tell how the process terminated, etc. */ 378 acct.ac_flag = p->p_acflag; 379 PROC_UNLOCK(p); 380 381 /* 382 * Eliminate any file size rlimit. 383 */ 384 newlim = lim_alloc(); 385 PROC_LOCK(p); 386 oldlim = p->p_limit; 387 lim_copy(newlim, oldlim); 388 newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 389 p->p_limit = newlim; 390 PROC_UNLOCK(p); 391 lim_free(oldlim); 392 393 /* 394 * Write the accounting information to the file. 395 */ 396 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 397 VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE); 398 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct), 399 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED, 400 (int *)0, td); 401 VFS_UNLOCK_GIANT(vfslocked); 402 sx_sunlock(&acct_sx); 403 return (ret); 404 } 405 406 /* 407 * Encode_comp_t converts from ticks in seconds and microseconds 408 * to ticks in 1/AHZ seconds. The encoding is described in 409 * Leffler, et al., on page 63. 410 */ 411 412 #define MANTSIZE 13 /* 13 bit mantissa. */ 413 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 414 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 415 416 static comp_t 417 encode_comp_t(u_long s, u_long us) 418 { 419 int exp, rnd; 420 421 exp = 0; 422 rnd = 0; 423 s *= AHZ; 424 s += us / (1000000 / AHZ); /* Maximize precision. */ 425 426 while (s > MAXFRACT) { 427 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 428 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 429 exp++; 430 } 431 432 /* If we need to round up, do it (and handle overflow correctly). */ 433 if (rnd && (++s > MAXFRACT)) { 434 s >>= EXPSIZE; 435 exp++; 436 } 437 438 /* Clean it up and polish it off. */ 439 exp <<= MANTSIZE; /* Shift the exponent into place */ 440 exp += s; /* and add on the mantissa. */ 441 return (exp); 442 } 443 444 /* 445 * Periodically check the filesystem to see if accounting 446 * should be turned on or off. Beware the case where the vnode 447 * has been vgone()'d out from underneath us, e.g. when the file 448 * system containing the accounting file has been forcibly unmounted. 449 */ 450 /* ARGSUSED */ 451 static void 452 acctwatch(void) 453 { 454 struct statfs sb; 455 int vfslocked; 456 457 sx_assert(&acct_sx, SX_XLOCKED); 458 459 /* 460 * If accounting was disabled before our kthread was scheduled, 461 * then acct_vp might be NULL. If so, just ask our kthread to 462 * exit and return. 463 */ 464 if (acct_vp == NULL) { 465 acct_state |= ACCT_EXITREQ; 466 return; 467 } 468 469 /* 470 * If our vnode is no longer valid, tear it down and signal the 471 * accounting thread to die. 472 */ 473 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 474 if (acct_vp->v_type == VBAD) { 475 (void) acct_disable(NULL); 476 VFS_UNLOCK_GIANT(vfslocked); 477 acct_state |= ACCT_EXITREQ; 478 return; 479 } 480 481 /* 482 * Stopping here is better than continuing, maybe it will be VBAD 483 * next time around. 484 */ 485 if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) { 486 VFS_UNLOCK_GIANT(vfslocked); 487 return; 488 } 489 VFS_UNLOCK_GIANT(vfslocked); 490 if (acct_suspended) { 491 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks / 492 100)) { 493 acct_suspended = 0; 494 log(LOG_NOTICE, "Accounting resumed\n"); 495 } 496 } else { 497 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks / 498 100)) { 499 acct_suspended = 1; 500 log(LOG_NOTICE, "Accounting suspended\n"); 501 } 502 } 503 } 504 505 /* 506 * The main loop for the dedicated kernel thread that periodically calls 507 * acctwatch(). 508 */ 509 static void 510 acct_thread(void *dummy) 511 { 512 u_char pri; 513 514 /* This is a low-priority kernel thread. */ 515 pri = PRI_MAX_KERN; 516 mtx_lock_spin(&sched_lock); 517 sched_prio(curthread, pri); 518 mtx_unlock_spin(&sched_lock); 519 520 /* If another accounting kthread is already running, just die. */ 521 sx_xlock(&acct_sx); 522 if (acct_state & ACCT_RUNNING) { 523 sx_xunlock(&acct_sx); 524 kthread_exit(0); 525 } 526 acct_state |= ACCT_RUNNING; 527 528 /* Loop until we are asked to exit. */ 529 while (!(acct_state & ACCT_EXITREQ)) { 530 531 /* Perform our periodic checks. */ 532 acctwatch(); 533 534 /* 535 * We check this flag again before sleeping since the 536 * acctwatch() might have shut down accounting and asked us 537 * to exit. 538 */ 539 if (!(acct_state & ACCT_EXITREQ)) { 540 sx_xunlock(&acct_sx); 541 tsleep(&acct_state, pri, "-", acctchkfreq * hz); 542 sx_xlock(&acct_sx); 543 } 544 } 545 546 /* 547 * Acknowledge the exit request and shutdown. We clear both the 548 * exit request and running flags. 549 */ 550 acct_state = 0; 551 sx_xunlock(&acct_sx); 552 kthread_exit(0); 553 } 554