kern_acct.c revision 155431
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: head/sys/kern/kern_acct.c 155431 2006-02-07 16:04:03Z jhb $"); 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 */ 87static comp_t encode_comp_t(u_long, u_long); 88static void acctwatch(void); 89static void acct_thread(void *); 90static 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 */ 97static int acct_suspended; 98static struct vnode *acct_vp; 99static struct ucred *acct_cred; 100static int acct_flags; 101static struct sx acct_sx; 102 103SX_SYSINIT(acct, &acct_sx, "acct_sx"); 104 105/* 106 * State of the accounting kthread. 107 */ 108static int acct_state; 109 110#define ACCT_RUNNING 1 /* Accounting kthread is running. */ 111#define ACCT_EXITREQ 2 /* Accounting kthread should exit. */ 112 113/* 114 * Values associated with enabling and disabling accounting 115 */ 116static int acctsuspend = 2; /* stop accounting when < 2% free space left */ 117SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, 118 &acctsuspend, 0, "percentage of free disk space below which accounting stops"); 119 120static int acctresume = 4; /* resume when free space risen to > 4% */ 121SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, 122 &acctresume, 0, "percentage of free disk space above which accounting resumes"); 123 124static int acctchkfreq = 15; /* frequency (in seconds) to check space */ 125SYSCTL_INT(_kern, OID_AUTO, acct_chkfreq, CTLFLAG_RW, 126 &acctchkfreq, 0, "frequency for checking the free space"); 127 128SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0, 129 "Accounting suspended or not"); 130 131/* 132 * Accounting system call. Written based on the specification and 133 * previous implementation done by Mark Tinguely. 134 * 135 * MPSAFE 136 */ 137int 138acct(struct thread *td, struct acct_args *uap) 139{ 140 struct nameidata nd; 141 int error, flags; 142 143 /* Make sure that the caller is root. */ 144 error = suser(td); 145 if (error) 146 return (error); 147 148 /* 149 * If accounting is to be started to a file, open that file for 150 * appending and make sure it's a 'normal'. While we could 151 * conditionally acquire Giant here, we're actually interacting with 152 * vnodes from possibly two file systems, making the logic a bit 153 * complicated. For now, use Giant unconditionally. 154 */ 155 mtx_lock(&Giant); 156 if (uap->path != NULL) { 157 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->path, td); 158 flags = FWRITE | O_APPEND; 159 error = vn_open(&nd, &flags, 0, -1); 160 if (error) 161 goto done; 162 NDFREE(&nd, NDF_ONLY_PNBUF); 163#ifdef MAC 164 error = mac_check_system_acct(td->td_ucred, nd.ni_vp); 165 if (error) { 166 VOP_UNLOCK(nd.ni_vp, 0, td); 167 vn_close(nd.ni_vp, flags, td->td_ucred, td); 168 goto done; 169 } 170#endif 171 VOP_UNLOCK(nd.ni_vp, 0, td); 172 if (nd.ni_vp->v_type != VREG) { 173 vn_close(nd.ni_vp, flags, td->td_ucred, td); 174 error = EACCES; 175 goto done; 176 } 177#ifdef MAC 178 } else { 179 error = mac_check_system_acct(td->td_ucred, NULL); 180 if (error) 181 goto done; 182#endif 183 } 184 185 /* 186 * Disallow concurrent access to the accounting vnode while we swap 187 * it out, in order to prevent access after close. 188 */ 189 sx_xlock(&acct_sx); 190 191 /* 192 * If accounting was previously enabled, kill the old space-watcher, 193 * close the file, and (if no new file was specified, leave). Reset 194 * the suspended state regardless of whether accounting remains 195 * enabled. 196 */ 197 acct_suspended = 0; 198 if (acct_vp != NULL) 199 error = acct_disable(td); 200 if (uap->path == NULL) { 201 if (acct_state & ACCT_RUNNING) { 202 acct_state |= ACCT_EXITREQ; 203 wakeup(&acct_state); 204 } 205 sx_xunlock(&acct_sx); 206 goto done; 207 } 208 209 /* 210 * Save the new accounting file vnode, and schedule the new 211 * free space watcher. 212 */ 213 acct_vp = nd.ni_vp; 214 acct_cred = crhold(td->td_ucred); 215 acct_flags = flags; 216 if (acct_state & ACCT_RUNNING) 217 acct_state &= ~ACCT_EXITREQ; 218 else { 219 /* 220 * Try to start up an accounting kthread. We may start more 221 * than one, but if so the extras will commit suicide as 222 * soon as they start up. 223 */ 224 error = kthread_create(acct_thread, NULL, NULL, 0, 0, 225 "accounting"); 226 if (error) { 227 (void) vn_close(acct_vp, acct_flags, acct_cred, td); 228 crfree(acct_cred); 229 acct_vp = NULL; 230 acct_cred = NULL; 231 acct_flags = 0; 232 sx_xunlock(&acct_sx); 233 log(LOG_NOTICE, "Unable to start accounting thread\n"); 234 goto done; 235 } 236 } 237 sx_xunlock(&acct_sx); 238 log(LOG_NOTICE, "Accounting enabled\n"); 239done: 240 mtx_unlock(&Giant); 241 return (error); 242} 243 244/* 245 * Disable currently in-progress accounting by closing the vnode, dropping 246 * our reference to the credential, and clearing the vnode's flags. 247 */ 248static int 249acct_disable(struct thread *td) 250{ 251 int error; 252 253 sx_assert(&acct_sx, SX_XLOCKED); 254 error = vn_close(acct_vp, acct_flags, acct_cred, td); 255 crfree(acct_cred); 256 acct_vp = NULL; 257 acct_cred = NULL; 258 acct_flags = 0; 259 log(LOG_NOTICE, "Accounting disabled\n"); 260 return (error); 261} 262 263/* 264 * Write out process accounting information, on process exit. 265 * Data to be written out is specified in Leffler, et al. 266 * and are enumerated below. (They're also noted in the system 267 * "acct.h" header file.) 268 */ 269int 270acct_process(struct thread *td) 271{ 272 struct acct acct; 273 struct timeval ut, st, tmp; 274 struct plimit *newlim, *oldlim; 275 struct proc *p; 276 struct rusage *r; 277 int t, ret, vfslocked; 278 279 /* 280 * Lockless check of accounting condition before doing the hard 281 * work. 282 */ 283 if (acct_vp == NULL || acct_suspended) 284 return (0); 285 286 sx_slock(&acct_sx); 287 288 /* 289 * If accounting isn't enabled, don't bother. Have to check again 290 * once we own the lock in case we raced with disabling of accounting 291 * by another thread. 292 */ 293 if (acct_vp == NULL || acct_suspended) { 294 sx_sunlock(&acct_sx); 295 return (0); 296 } 297 298 p = td->td_proc; 299 300 /* 301 * Get process accounting information. 302 */ 303 304 PROC_LOCK(p); 305 /* (1) The name of the command that ran */ 306 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); 307 308 /* (2) The amount of user and system time that was used */ 309 calcru(p, &ut, &st); 310 acct.ac_utime = encode_comp_t(ut.tv_sec, ut.tv_usec); 311 acct.ac_stime = encode_comp_t(st.tv_sec, st.tv_usec); 312 313 /* (3) The elapsed time the command ran (and its starting time) */ 314 tmp = boottime; 315 timevaladd(&tmp, &p->p_stats->p_start); 316 acct.ac_btime = tmp.tv_sec; 317 microuptime(&tmp); 318 timevalsub(&tmp, &p->p_stats->p_start); 319 acct.ac_etime = encode_comp_t(tmp.tv_sec, tmp.tv_usec); 320 321 /* (4) The average amount of memory used */ 322 r = &p->p_stats->p_ru; 323 tmp = ut; 324 timevaladd(&tmp, &st); 325 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 326 if (t) 327 acct.ac_mem = (r->ru_ixrss + r->ru_idrss + r->ru_isrss) / t; 328 else 329 acct.ac_mem = 0; 330 331 /* (5) The number of disk I/O operations done */ 332 acct.ac_io = encode_comp_t(r->ru_inblock + r->ru_oublock, 0); 333 334 /* (6) The UID and GID of the process */ 335 acct.ac_uid = p->p_ucred->cr_ruid; 336 acct.ac_gid = p->p_ucred->cr_rgid; 337 338 /* (7) The terminal from which the process was started */ 339 SESS_LOCK(p->p_session); 340 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 341 acct.ac_tty = dev2udev(p->p_pgrp->pg_session->s_ttyp->t_dev); 342 else 343 acct.ac_tty = NODEV; 344 SESS_UNLOCK(p->p_session); 345 346 /* (8) The boolean flags that tell how the process terminated, etc. */ 347 acct.ac_flag = p->p_acflag; 348 PROC_UNLOCK(p); 349 350 /* 351 * Eliminate any file size rlimit. 352 */ 353 newlim = lim_alloc(); 354 PROC_LOCK(p); 355 oldlim = p->p_limit; 356 lim_copy(newlim, oldlim); 357 newlim->pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 358 p->p_limit = newlim; 359 PROC_UNLOCK(p); 360 lim_free(oldlim); 361 362 /* 363 * Write the accounting information to the file. 364 */ 365 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 366 VOP_LEASE(acct_vp, td, acct_cred, LEASE_WRITE); 367 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct), 368 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED, 369 (int *)0, td); 370 VFS_UNLOCK_GIANT(vfslocked); 371 sx_sunlock(&acct_sx); 372 return (ret); 373} 374 375/* 376 * Encode_comp_t converts from ticks in seconds and microseconds 377 * to ticks in 1/AHZ seconds. The encoding is described in 378 * Leffler, et al., on page 63. 379 */ 380 381#define MANTSIZE 13 /* 13 bit mantissa. */ 382#define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ 383#define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ 384 385static comp_t 386encode_comp_t(u_long s, u_long us) 387{ 388 int exp, rnd; 389 390 exp = 0; 391 rnd = 0; 392 s *= AHZ; 393 s += us / (1000000 / AHZ); /* Maximize precision. */ 394 395 while (s > MAXFRACT) { 396 rnd = s & (1 << (EXPSIZE - 1)); /* Round up? */ 397 s >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ 398 exp++; 399 } 400 401 /* If we need to round up, do it (and handle overflow correctly). */ 402 if (rnd && (++s > MAXFRACT)) { 403 s >>= EXPSIZE; 404 exp++; 405 } 406 407 /* Clean it up and polish it off. */ 408 exp <<= MANTSIZE; /* Shift the exponent into place */ 409 exp += s; /* and add on the mantissa. */ 410 return (exp); 411} 412 413/* 414 * Periodically check the filesystem to see if accounting 415 * should be turned on or off. Beware the case where the vnode 416 * has been vgone()'d out from underneath us, e.g. when the file 417 * system containing the accounting file has been forcibly unmounted. 418 */ 419/* ARGSUSED */ 420static void 421acctwatch(void) 422{ 423 struct statfs sb; 424 int vfslocked; 425 426 sx_assert(&acct_sx, SX_XLOCKED); 427 428 /* 429 * If accounting was disabled before our kthread was scheduled, 430 * then acct_vp might be NULL. If so, just ask our kthread to 431 * exit and return. 432 */ 433 if (acct_vp == NULL) { 434 acct_state |= ACCT_EXITREQ; 435 return; 436 } 437 438 /* 439 * If our vnode is no longer valid, tear it down and signal the 440 * accounting thread to die. 441 */ 442 vfslocked = VFS_LOCK_GIANT(acct_vp->v_mount); 443 if (acct_vp->v_type == VBAD) { 444 (void) acct_disable(NULL); 445 VFS_UNLOCK_GIANT(vfslocked); 446 acct_state |= ACCT_EXITREQ; 447 return; 448 } 449 450 /* 451 * Stopping here is better than continuing, maybe it will be VBAD 452 * next time around. 453 */ 454 if (VFS_STATFS(acct_vp->v_mount, &sb, curthread) < 0) { 455 VFS_UNLOCK_GIANT(vfslocked); 456 return; 457 } 458 VFS_UNLOCK_GIANT(vfslocked); 459 if (acct_suspended) { 460 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks / 461 100)) { 462 acct_suspended = 0; 463 log(LOG_NOTICE, "Accounting resumed\n"); 464 } 465 } else { 466 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks / 467 100)) { 468 acct_suspended = 1; 469 log(LOG_NOTICE, "Accounting suspended\n"); 470 } 471 } 472} 473 474/* 475 * The main loop for the dedicated kernel thread that periodically calls 476 * acctwatch(). 477 */ 478static void 479acct_thread(void *dummy) 480{ 481 u_char pri; 482 483 /* This is a low-priority kernel thread. */ 484 pri = PRI_MAX_KERN; 485 mtx_lock_spin(&sched_lock); 486 sched_prio(curthread, pri); 487 mtx_unlock_spin(&sched_lock); 488 489 /* If another accounting kthread is already running, just die. */ 490 sx_xlock(&acct_sx); 491 if (acct_state & ACCT_RUNNING) { 492 sx_xunlock(&acct_sx); 493 kthread_exit(0); 494 } 495 acct_state |= ACCT_RUNNING; 496 497 /* Loop until we are asked to exit. */ 498 while (!(acct_state & ACCT_EXITREQ)) { 499 500 /* Perform our periodic checks. */ 501 acctwatch(); 502 503 /* 504 * We check this flag again before sleeping since the 505 * acctwatch() might have shut down accounting and asked us 506 * to exit. 507 */ 508 if (!(acct_state & ACCT_EXITREQ)) { 509 sx_xunlock(&acct_sx); 510 tsleep(&acct_state, pri, "-", acctchkfreq * hz); 511 sx_xlock(&acct_sx); 512 } 513 } 514 515 /* 516 * Acknowledge the exit request and shutdown. We clear both the 517 * exit request and running flags. 518 */ 519 acct_state = 0; 520 sx_xunlock(&acct_sx); 521 kthread_exit(0); 522} 523