kern_acct.c revision 252422
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 * Copyright (c) 2005 Robert N. M. Watson 6 * All rights reserved. 7 * 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 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * Copyright (c) 1994 Christopher G. Demetriou 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. All advertising materials mentioning features or use of this software 48 * must display the following acknowledgement: 49 * This product includes software developed by the University of 50 * California, Berkeley and its contributors. 51 * 4. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)kern_acct.c 8.1 (Berkeley) 6/14/93 68 */ 69 70#include <sys/cdefs.h> 71__FBSDID("$FreeBSD: head/sys/kern/kern_acct.c 252422 2013-06-30 19:08:06Z mjg $"); 72 73#include <sys/param.h> 74#include <sys/systm.h> 75#include <sys/acct.h> 76#include <sys/fcntl.h> 77#include <sys/kernel.h> 78#include <sys/kthread.h> 79#include <sys/limits.h> 80#include <sys/lock.h> 81#include <sys/mount.h> 82#include <sys/mutex.h> 83#include <sys/namei.h> 84#include <sys/priv.h> 85#include <sys/proc.h> 86#include <sys/resourcevar.h> 87#include <sys/sched.h> 88#include <sys/sx.h> 89#include <sys/sysctl.h> 90#include <sys/sysent.h> 91#include <sys/syslog.h> 92#include <sys/sysproto.h> 93#include <sys/tty.h> 94#include <sys/vnode.h> 95 96#include <security/mac/mac_framework.h> 97 98/* 99 * The routines implemented in this file are described in: 100 * Leffler, et al.: The Design and Implementation of the 4.3BSD 101 * UNIX Operating System (Addison Welley, 1989) 102 * on pages 62-63. 103 * On May 2007 the historic 3 bits base 8 exponent, 13 bit fraction 104 * compt_t representation described in the above reference was replaced 105 * with that of IEEE-754 floats. 106 * 107 * Arguably, to simplify accounting operations, this mechanism should 108 * be replaced by one in which an accounting log file (similar to /dev/klog) 109 * is read by a user process, etc. However, that has its own problems. 110 */ 111 112/* Floating point definitions from <float.h>. */ 113#define FLT_MANT_DIG 24 /* p */ 114#define FLT_MAX_EXP 128 /* emax */ 115 116/* 117 * Internal accounting functions. 118 * The former's operation is described in Leffler, et al., and the latter 119 * was provided by UCB with the 4.4BSD-Lite release 120 */ 121static uint32_t encode_timeval(struct timeval); 122static uint32_t encode_long(long); 123static void acctwatch(void); 124static void acct_thread(void *); 125static int acct_disable(struct thread *, int); 126 127/* 128 * Accounting vnode pointer, saved vnode pointer, and flags for each. 129 * acct_sx protects against changes to the active vnode and credentials 130 * while accounting records are being committed to disk. 131 */ 132static int acct_configured; 133static int acct_suspended; 134static struct vnode *acct_vp; 135static struct ucred *acct_cred; 136static struct plimit *acct_limit; 137static int acct_flags; 138static struct sx acct_sx; 139 140SX_SYSINIT(acct, &acct_sx, "acct_sx"); 141 142/* 143 * State of the accounting kthread. 144 */ 145static int acct_state; 146 147#define ACCT_RUNNING 1 /* Accounting kthread is running. */ 148#define ACCT_EXITREQ 2 /* Accounting kthread should exit. */ 149 150/* 151 * Values associated with enabling and disabling accounting 152 */ 153static int acctsuspend = 2; /* stop accounting when < 2% free space left */ 154SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, 155 &acctsuspend, 0, "percentage of free disk space below which accounting stops"); 156 157static int acctresume = 4; /* resume when free space risen to > 4% */ 158SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, 159 &acctresume, 0, "percentage of free disk space above which accounting resumes"); 160 161static int acctchkfreq = 15; /* frequency (in seconds) to check space */ 162 163static int 164sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS) 165{ 166 int error, value; 167 168 /* Write out the old value. */ 169 error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int)); 170 if (error || req->newptr == NULL) 171 return (error); 172 173 /* Read in and verify the new value. */ 174 error = SYSCTL_IN(req, &value, sizeof(int)); 175 if (error) 176 return (error); 177 if (value <= 0) 178 return (EINVAL); 179 acctchkfreq = value; 180 return (0); 181} 182SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW, 183 &acctchkfreq, 0, sysctl_acct_chkfreq, "I", 184 "frequency for checking the free space"); 185 186SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0, 187 "Accounting configured or not"); 188 189SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0, 190 "Accounting suspended or not"); 191 192/* 193 * Accounting system call. Written based on the specification and previous 194 * implementation done by Mark Tinguely. 195 */ 196int 197sys_acct(struct thread *td, struct acct_args *uap) 198{ 199 struct nameidata nd; 200 int error, flags, i, replacing; 201 202 error = priv_check(td, PRIV_ACCT); 203 if (error) 204 return (error); 205 206 /* 207 * If accounting is to be started to a file, open that file for 208 * appending and make sure it's a 'normal'. 209 */ 210 if (uap->path != NULL) { 211 NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, 212 UIO_USERSPACE, uap->path, td); 213 flags = FWRITE | O_APPEND; 214 error = vn_open(&nd, &flags, 0, NULL); 215 if (error) 216 return (error); 217 NDFREE(&nd, NDF_ONLY_PNBUF); 218#ifdef MAC 219 error = mac_system_check_acct(td->td_ucred, nd.ni_vp); 220 if (error) { 221 VOP_UNLOCK(nd.ni_vp, 0); 222 vn_close(nd.ni_vp, flags, td->td_ucred, td); 223 return (error); 224 } 225#endif 226 VOP_UNLOCK(nd.ni_vp, 0); 227 if (nd.ni_vp->v_type != VREG) { 228 vn_close(nd.ni_vp, flags, td->td_ucred, td); 229 return (EACCES); 230 } 231#ifdef MAC 232 } else { 233 error = mac_system_check_acct(td->td_ucred, NULL); 234 if (error) 235 return (error); 236#endif 237 } 238 239 /* 240 * Disallow concurrent access to the accounting vnode while we swap 241 * it out, in order to prevent access after close. 242 */ 243 sx_xlock(&acct_sx); 244 245 /* 246 * Don't log spurious disable/enable messages if we are 247 * switching from one accounting file to another due to log 248 * rotation. 249 */ 250 replacing = (acct_vp != NULL && uap->path != NULL); 251 252 /* 253 * If accounting was previously enabled, kill the old space-watcher, 254 * close the file, and (if no new file was specified, leave). Reset 255 * the suspended state regardless of whether accounting remains 256 * enabled. 257 */ 258 acct_suspended = 0; 259 if (acct_vp != NULL) 260 error = acct_disable(td, !replacing); 261 if (uap->path == NULL) { 262 if (acct_state & ACCT_RUNNING) { 263 acct_state |= ACCT_EXITREQ; 264 wakeup(&acct_state); 265 } 266 sx_xunlock(&acct_sx); 267 return (error); 268 } 269 270 /* 271 * Create our own plimit object without limits. It will be assigned 272 * to exiting processes. 273 */ 274 acct_limit = lim_alloc(); 275 for (i = 0; i < RLIM_NLIMITS; i++) 276 acct_limit->pl_rlimit[i].rlim_cur = 277 acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY; 278 279 /* 280 * Save the new accounting file vnode, and schedule the new 281 * free space watcher. 282 */ 283 acct_vp = nd.ni_vp; 284 acct_cred = crhold(td->td_ucred); 285 acct_flags = flags; 286 if (acct_state & ACCT_RUNNING) 287 acct_state &= ~ACCT_EXITREQ; 288 else { 289 /* 290 * Try to start up an accounting kthread. We may start more 291 * than one, but if so the extras will commit suicide as 292 * soon as they start up. 293 */ 294 error = kproc_create(acct_thread, NULL, NULL, 0, 0, 295 "accounting"); 296 if (error) { 297 (void) acct_disable(td, 0); 298 sx_xunlock(&acct_sx); 299 log(LOG_NOTICE, "Unable to start accounting thread\n"); 300 return (error); 301 } 302 } 303 acct_configured = 1; 304 sx_xunlock(&acct_sx); 305 if (!replacing) 306 log(LOG_NOTICE, "Accounting enabled\n"); 307 return (error); 308} 309 310/* 311 * Disable currently in-progress accounting by closing the vnode, dropping 312 * our reference to the credential, and clearing the vnode's flags. 313 */ 314static int 315acct_disable(struct thread *td, int logging) 316{ 317 int error; 318 319 sx_assert(&acct_sx, SX_XLOCKED); 320 error = vn_close(acct_vp, acct_flags, acct_cred, td); 321 crfree(acct_cred); 322 lim_free(acct_limit); 323 acct_configured = 0; 324 acct_vp = NULL; 325 acct_cred = NULL; 326 acct_flags = 0; 327 if (logging) 328 log(LOG_NOTICE, "Accounting disabled\n"); 329 return (error); 330} 331 332/* 333 * Write out process accounting information, on process exit. 334 * Data to be written out is specified in Leffler, et al. 335 * and are enumerated below. (They're also noted in the system 336 * "acct.h" header file.) 337 */ 338int 339acct_process(struct thread *td) 340{ 341 struct acctv2 acct; 342 struct timeval ut, st, tmp; 343 struct plimit *oldlim; 344 struct proc *p; 345 struct rusage ru; 346 int t, ret; 347 348 /* 349 * Lockless check of accounting condition before doing the hard 350 * work. 351 */ 352 if (acct_vp == NULL || acct_suspended) 353 return (0); 354 355 sx_slock(&acct_sx); 356 357 /* 358 * If accounting isn't enabled, don't bother. Have to check again 359 * once we own the lock in case we raced with disabling of accounting 360 * by another thread. 361 */ 362 if (acct_vp == NULL || acct_suspended) { 363 sx_sunlock(&acct_sx); 364 return (0); 365 } 366 367 p = td->td_proc; 368 369 /* 370 * Get process accounting information. 371 */ 372 373 sx_slock(&proctree_lock); 374 PROC_LOCK(p); 375 376 /* (1) The terminal from which the process was started */ 377 if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) 378 acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp); 379 else 380 acct.ac_tty = NODEV; 381 sx_sunlock(&proctree_lock); 382 383 /* (2) The name of the command that ran */ 384 bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); 385 386 /* (3) The amount of user and system time that was used */ 387 rufetchcalc(p, &ru, &ut, &st); 388 acct.ac_utime = encode_timeval(ut); 389 acct.ac_stime = encode_timeval(st); 390 391 /* (4) The elapsed time the command ran (and its starting time) */ 392 tmp = boottime; 393 timevaladd(&tmp, &p->p_stats->p_start); 394 acct.ac_btime = tmp.tv_sec; 395 microuptime(&tmp); 396 timevalsub(&tmp, &p->p_stats->p_start); 397 acct.ac_etime = encode_timeval(tmp); 398 399 /* (5) The average amount of memory used */ 400 tmp = ut; 401 timevaladd(&tmp, &st); 402 /* Convert tmp (i.e. u + s) into hz units to match ru_i*. */ 403 t = tmp.tv_sec * hz + tmp.tv_usec / tick; 404 if (t) 405 acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss + 406 + ru.ru_isrss) / t); 407 else 408 acct.ac_mem = 0; 409 410 /* (6) The number of disk I/O operations done */ 411 acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock); 412 413 /* (7) The UID and GID of the process */ 414 acct.ac_uid = p->p_ucred->cr_ruid; 415 acct.ac_gid = p->p_ucred->cr_rgid; 416 417 /* (8) The boolean flags that tell how the process terminated, etc. */ 418 acct.ac_flagx = p->p_acflag; 419 420 /* Setup ancillary structure fields. */ 421 acct.ac_flagx |= ANVER; 422 acct.ac_zero = 0; 423 acct.ac_version = 2; 424 acct.ac_len = acct.ac_len2 = sizeof(acct); 425 426 /* 427 * Eliminate rlimits (file size limit in particular). 428 */ 429 oldlim = p->p_limit; 430 p->p_limit = lim_hold(acct_limit); 431 PROC_UNLOCK(p); 432 lim_free(oldlim); 433 434 /* 435 * Write the accounting information to the file. 436 */ 437 ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct), 438 (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED, 439 NULL, td); 440 sx_sunlock(&acct_sx); 441 return (ret); 442} 443 444/* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */ 445 446/* Convert timevals and longs into IEEE-754 bit patterns. */ 447 448/* Mantissa mask (MSB is implied, so subtract 1). */ 449#define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1) 450 451/* 452 * We calculate integer values to a precision of approximately 453 * 28 bits. 454 * This is high-enough precision to fill the 24 float bits 455 * and low-enough to avoid overflowing the 32 int bits. 456 */ 457#define CALC_BITS 28 458 459/* log_2(1000000). */ 460#define LOG2_1M 20 461 462/* 463 * Convert the elements of a timeval into a 32-bit word holding 464 * the bits of a IEEE-754 float. 465 * The float value represents the timeval's value in microsecond units. 466 */ 467static uint32_t 468encode_timeval(struct timeval tv) 469{ 470 int log2_s; 471 int val, exp; /* Unnormalized value and exponent */ 472 int norm_exp; /* Normalized exponent */ 473 int shift; 474 475 /* 476 * First calculate value and exponent to about CALC_BITS precision. 477 * Note that the following conditionals have been ordered so that 478 * the most common cases appear first. 479 */ 480 if (tv.tv_sec == 0) { 481 if (tv.tv_usec == 0) 482 return (0); 483 exp = 0; 484 val = tv.tv_usec; 485 } else { 486 /* 487 * Calculate the value to a precision of approximately 488 * CALC_BITS. 489 */ 490 log2_s = fls(tv.tv_sec) - 1; 491 if (log2_s + LOG2_1M < CALC_BITS) { 492 exp = 0; 493 val = 1000000 * tv.tv_sec + tv.tv_usec; 494 } else { 495 exp = log2_s + LOG2_1M - CALC_BITS; 496 val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec + 497 tv.tv_usec) >> exp); 498 } 499 } 500 /* Now normalize and pack the value into an IEEE-754 float. */ 501 norm_exp = fls(val) - 1; 502 shift = FLT_MANT_DIG - norm_exp - 1; 503#ifdef ACCT_DEBUG 504 printf("val=%d exp=%d shift=%d log2(val)=%d\n", 505 val, exp, shift, norm_exp); 506 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp, 507 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK)); 508#endif 509 return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) | 510 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK)); 511} 512 513/* 514 * Convert a non-negative long value into the bit pattern of 515 * an IEEE-754 float value. 516 */ 517static uint32_t 518encode_long(long val) 519{ 520 int norm_exp; /* Normalized exponent */ 521 int shift; 522 523 if (val == 0) 524 return (0); 525 if (val < 0) { 526 log(LOG_NOTICE, 527 "encode_long: negative value %ld in accounting record\n", 528 val); 529 val = LONG_MAX; 530 } 531 norm_exp = fls(val) - 1; 532 shift = FLT_MANT_DIG - norm_exp - 1; 533#ifdef ACCT_DEBUG 534 printf("val=%d shift=%d log2(val)=%d\n", 535 val, shift, norm_exp); 536 printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp, 537 ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK)); 538#endif 539 return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) | 540 ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK)); 541} 542 543/* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */ 544 545/* 546 * Periodically check the filesystem to see if accounting 547 * should be turned on or off. Beware the case where the vnode 548 * has been vgone()'d out from underneath us, e.g. when the file 549 * system containing the accounting file has been forcibly unmounted. 550 */ 551/* ARGSUSED */ 552static void 553acctwatch(void) 554{ 555 struct statfs sb; 556 557 sx_assert(&acct_sx, SX_XLOCKED); 558 559 /* 560 * If accounting was disabled before our kthread was scheduled, 561 * then acct_vp might be NULL. If so, just ask our kthread to 562 * exit and return. 563 */ 564 if (acct_vp == NULL) { 565 acct_state |= ACCT_EXITREQ; 566 return; 567 } 568 569 /* 570 * If our vnode is no longer valid, tear it down and signal the 571 * accounting thread to die. 572 */ 573 if (acct_vp->v_type == VBAD) { 574 (void) acct_disable(NULL, 1); 575 acct_state |= ACCT_EXITREQ; 576 return; 577 } 578 579 /* 580 * Stopping here is better than continuing, maybe it will be VBAD 581 * next time around. 582 */ 583 if (VFS_STATFS(acct_vp->v_mount, &sb) < 0) 584 return; 585 if (acct_suspended) { 586 if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks / 587 100)) { 588 acct_suspended = 0; 589 log(LOG_NOTICE, "Accounting resumed\n"); 590 } 591 } else { 592 if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks / 593 100)) { 594 acct_suspended = 1; 595 log(LOG_NOTICE, "Accounting suspended\n"); 596 } 597 } 598} 599 600/* 601 * The main loop for the dedicated kernel thread that periodically calls 602 * acctwatch(). 603 */ 604static void 605acct_thread(void *dummy) 606{ 607 u_char pri; 608 609 /* This is a low-priority kernel thread. */ 610 pri = PRI_MAX_KERN; 611 thread_lock(curthread); 612 sched_prio(curthread, pri); 613 thread_unlock(curthread); 614 615 /* If another accounting kthread is already running, just die. */ 616 sx_xlock(&acct_sx); 617 if (acct_state & ACCT_RUNNING) { 618 sx_xunlock(&acct_sx); 619 kproc_exit(0); 620 } 621 acct_state |= ACCT_RUNNING; 622 623 /* Loop until we are asked to exit. */ 624 while (!(acct_state & ACCT_EXITREQ)) { 625 626 /* Perform our periodic checks. */ 627 acctwatch(); 628 629 /* 630 * We check this flag again before sleeping since the 631 * acctwatch() might have shut down accounting and asked us 632 * to exit. 633 */ 634 if (!(acct_state & ACCT_EXITREQ)) { 635 sx_sleep(&acct_state, &acct_sx, 0, "-", 636 acctchkfreq * hz); 637 } 638 } 639 640 /* 641 * Acknowledge the exit request and shutdown. We clear both the 642 * exit request and running flags. 643 */ 644 acct_state = 0; 645 sx_xunlock(&acct_sx); 646 kproc_exit(0); 647} 648