audit.c revision 184856
1/*- 2 * Copyright (c) 1999-2005 Apple Inc. 3 * Copyright (c) 2006-2007 Robert N. M. Watson 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. Neither the name of Apple Inc. ("Apple") nor the names of 15 * its contributors may be used to endorse or promote products derived 16 * from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR 22 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 26 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 27 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31#include <sys/cdefs.h> 32__FBSDID("$FreeBSD: head/sys/security/audit/audit.c 184856 2008-11-11 21:57:03Z csjp $"); 33 34#include <sys/param.h> 35#include <sys/condvar.h> 36#include <sys/conf.h> 37#include <sys/file.h> 38#include <sys/filedesc.h> 39#include <sys/fcntl.h> 40#include <sys/ipc.h> 41#include <sys/kernel.h> 42#include <sys/kthread.h> 43#include <sys/malloc.h> 44#include <sys/mount.h> 45#include <sys/namei.h> 46#include <sys/priv.h> 47#include <sys/proc.h> 48#include <sys/queue.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/protosw.h> 52#include <sys/domain.h> 53#include <sys/sysctl.h> 54#include <sys/sysproto.h> 55#include <sys/sysent.h> 56#include <sys/systm.h> 57#include <sys/ucred.h> 58#include <sys/uio.h> 59#include <sys/un.h> 60#include <sys/unistd.h> 61#include <sys/vnode.h> 62 63#include <bsm/audit.h> 64#include <bsm/audit_internal.h> 65#include <bsm/audit_kevents.h> 66 67#include <netinet/in.h> 68#include <netinet/in_pcb.h> 69 70#include <security/audit/audit.h> 71#include <security/audit/audit_private.h> 72 73#include <vm/uma.h> 74 75static uma_zone_t audit_record_zone; 76static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage"); 77MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage"); 78MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage"); 79MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage"); 80 81SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0, 82 "TrustedBSD audit controls"); 83 84/* 85 * Audit control settings that are set/read by system calls and are hence 86 * non-static. 87 * 88 * Define the audit control flags. 89 */ 90int audit_enabled; 91int audit_suspended; 92 93/* 94 * Flags controlling behavior in low storage situations. Should we panic if 95 * a write fails? Should we fail stop if we're out of disk space? 96 */ 97int audit_panic_on_write_fail; 98int audit_fail_stop; 99int audit_argv; 100int audit_arge; 101 102/* 103 * Are we currently "failing stop" due to out of disk space? 104 */ 105int audit_in_failure; 106 107/* 108 * Global audit statistics. 109 */ 110struct audit_fstat audit_fstat; 111 112/* 113 * Preselection mask for non-attributable events. 114 */ 115struct au_mask audit_nae_mask; 116 117/* 118 * Mutex to protect global variables shared between various threads and 119 * processes. 120 */ 121struct mtx audit_mtx; 122 123/* 124 * Queue of audit records ready for delivery to disk. We insert new records 125 * at the tail, and remove records from the head. Also, a count of the 126 * number of records used for checking queue depth. In addition, a counter 127 * of records that we have allocated but are not yet in the queue, which is 128 * needed to estimate the total size of the combined set of records 129 * outstanding in the system. 130 */ 131struct kaudit_queue audit_q; 132int audit_q_len; 133int audit_pre_q_len; 134 135/* 136 * Audit queue control settings (minimum free, low/high water marks, etc.) 137 */ 138struct au_qctrl audit_qctrl; 139 140/* 141 * Condition variable to signal to the worker that it has work to do: either 142 * new records are in the queue, or a log replacement is taking place. 143 */ 144struct cv audit_worker_cv; 145 146/* 147 * Condition variable to flag when crossing the low watermark, meaning that 148 * threads blocked due to hitting the high watermark can wake up and continue 149 * to commit records. 150 */ 151struct cv audit_watermark_cv; 152 153/* 154 * Condition variable for auditing threads wait on when in fail-stop mode. 155 * Threads wait on this CV forever (and ever), never seeing the light of day 156 * again. 157 */ 158static struct cv audit_fail_cv; 159 160/* 161 * Kernel audit information. This will store the current audit address 162 * or host information that the kernel will use when it's generating 163 * audit records. This data is modified by the A_GET{SET}KAUDIT auditon(2) 164 * command. 165 */ 166static struct auditinfo_addr audit_kinfo; 167static struct rwlock audit_kinfo_lock; 168 169#define KINFO_LOCK_INIT() rw_init(&audit_kinfo_lock, "kernel audit info lock") 170#define KINFO_RLOCK() rw_rlock(&audit_kinfo_lock) 171#define KINFO_WLOCK() rw_wlock(&audit_kinfo_lock) 172#define KINFO_RUNLOCK() rw_runlock(&audit_kinfo_lock) 173#define KINFO_WUNLOCK() rw_wunlock(&audit_kinfo_lock) 174 175void 176audit_set_kinfo(struct auditinfo_addr *ak) 177{ 178 179 KASSERT(ak->ai_termid.at_type == AU_IPv4 || 180 ak->ai_termid.at_type == AU_IPv6, 181 ("audit_set_kinfo: invalid address type")); 182 KINFO_WLOCK(); 183 audit_kinfo = *ak; 184 KINFO_WUNLOCK(); 185} 186 187void 188audit_get_kinfo(struct auditinfo_addr *ak) 189{ 190 191 KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 || 192 audit_kinfo.ai_termid.at_type == AU_IPv6, 193 ("audit_set_kinfo: invalid address type")); 194 KINFO_RLOCK(); 195 *ak = audit_kinfo; 196 KINFO_RUNLOCK(); 197} 198 199/* 200 * Construct an audit record for the passed thread. 201 */ 202static int 203audit_record_ctor(void *mem, int size, void *arg, int flags) 204{ 205 struct kaudit_record *ar; 206 struct thread *td; 207 208 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size")); 209 210 td = arg; 211 ar = mem; 212 bzero(ar, sizeof(*ar)); 213 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC; 214 nanotime(&ar->k_ar.ar_starttime); 215 216 /* 217 * Export the subject credential. 218 */ 219 cru2x(td->td_ucred, &ar->k_ar.ar_subj_cred); 220 ar->k_ar.ar_subj_ruid = td->td_ucred->cr_ruid; 221 ar->k_ar.ar_subj_rgid = td->td_ucred->cr_rgid; 222 ar->k_ar.ar_subj_egid = td->td_ucred->cr_groups[0]; 223 ar->k_ar.ar_subj_auid = td->td_ucred->cr_audit.ai_auid; 224 ar->k_ar.ar_subj_asid = td->td_ucred->cr_audit.ai_asid; 225 ar->k_ar.ar_subj_pid = td->td_proc->p_pid; 226 ar->k_ar.ar_subj_amask = td->td_ucred->cr_audit.ai_mask; 227 ar->k_ar.ar_subj_term_addr = td->td_ucred->cr_audit.ai_termid; 228 return (0); 229} 230 231static void 232audit_record_dtor(void *mem, int size, void *arg) 233{ 234 struct kaudit_record *ar; 235 236 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size")); 237 238 ar = mem; 239 if (ar->k_ar.ar_arg_upath1 != NULL) 240 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH); 241 if (ar->k_ar.ar_arg_upath2 != NULL) 242 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH); 243 if (ar->k_ar.ar_arg_text != NULL) 244 free(ar->k_ar.ar_arg_text, M_AUDITTEXT); 245 if (ar->k_udata != NULL) 246 free(ar->k_udata, M_AUDITDATA); 247 if (ar->k_ar.ar_arg_argv != NULL) 248 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT); 249 if (ar->k_ar.ar_arg_envv != NULL) 250 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT); 251} 252 253/* 254 * Initialize the Audit subsystem: configuration state, work queue, 255 * synchronization primitives, worker thread, and trigger device node. Also 256 * call into the BSM assembly code to initialize it. 257 */ 258static void 259audit_init(void) 260{ 261 262 audit_enabled = 0; 263 audit_suspended = 0; 264 audit_panic_on_write_fail = 0; 265 audit_fail_stop = 0; 266 audit_in_failure = 0; 267 audit_argv = 0; 268 audit_arge = 0; 269 270 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */ 271 audit_fstat.af_currsz = 0; 272 audit_nae_mask.am_success = 0; 273 audit_nae_mask.am_failure = 0; 274 275 TAILQ_INIT(&audit_q); 276 audit_q_len = 0; 277 audit_pre_q_len = 0; 278 audit_qctrl.aq_hiwater = AQ_HIWATER; 279 audit_qctrl.aq_lowater = AQ_LOWATER; 280 audit_qctrl.aq_bufsz = AQ_BUFSZ; 281 audit_qctrl.aq_minfree = AU_FS_MINFREE; 282 283 audit_kinfo.ai_termid.at_type = AU_IPv4; 284 audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY; 285 286 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF); 287 KINFO_LOCK_INIT(); 288 cv_init(&audit_worker_cv, "audit_worker_cv"); 289 cv_init(&audit_watermark_cv, "audit_watermark_cv"); 290 cv_init(&audit_fail_cv, "audit_fail_cv"); 291 292 audit_record_zone = uma_zcreate("audit_record", 293 sizeof(struct kaudit_record), audit_record_ctor, 294 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0); 295 296 /* Initialize the BSM audit subsystem. */ 297 kau_init(); 298 299 audit_trigger_init(); 300 301 /* Register shutdown handler. */ 302 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL, 303 SHUTDOWN_PRI_FIRST); 304 305 /* Start audit worker thread. */ 306 audit_worker_init(); 307} 308 309SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL); 310 311/* 312 * Drain the audit queue and close the log at shutdown. Note that this can 313 * be called both from the system shutdown path and also from audit 314 * configuration syscalls, so 'arg' and 'howto' are ignored. 315 * 316 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to 317 * drain before returning, which could lead to lost records on shutdown. 318 */ 319void 320audit_shutdown(void *arg, int howto) 321{ 322 323 audit_rotate_vnode(NULL, NULL); 324} 325 326/* 327 * Return the current thread's audit record, if any. 328 */ 329struct kaudit_record * 330currecord(void) 331{ 332 333 return (curthread->td_ar); 334} 335 336/* 337 * XXXAUDIT: There are a number of races present in the code below due to 338 * release and re-grab of the mutex. The code should be revised to become 339 * slightly less racy. 340 * 341 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available 342 * pre_q space, suspending the system call until there is room? 343 */ 344struct kaudit_record * 345audit_new(int event, struct thread *td) 346{ 347 struct kaudit_record *ar; 348 int no_record; 349 350 mtx_lock(&audit_mtx); 351 no_record = (audit_suspended || !audit_enabled); 352 mtx_unlock(&audit_mtx); 353 if (no_record) 354 return (NULL); 355 356 /* 357 * Note: the number of outstanding uncommitted audit records is 358 * limited to the number of concurrent threads servicing system calls 359 * in the kernel. 360 */ 361 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK); 362 ar->k_ar.ar_event = event; 363 364 mtx_lock(&audit_mtx); 365 audit_pre_q_len++; 366 mtx_unlock(&audit_mtx); 367 368 return (ar); 369} 370 371void 372audit_free(struct kaudit_record *ar) 373{ 374 375 uma_zfree(audit_record_zone, ar); 376} 377 378void 379audit_commit(struct kaudit_record *ar, int error, int retval) 380{ 381 au_event_t event; 382 au_class_t class; 383 au_id_t auid; 384 int sorf; 385 struct au_mask *aumask; 386 387 if (ar == NULL) 388 return; 389 390 /* 391 * Decide whether to commit the audit record by checking the error 392 * value from the system call and using the appropriate audit mask. 393 */ 394 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID) 395 aumask = &audit_nae_mask; 396 else 397 aumask = &ar->k_ar.ar_subj_amask; 398 399 if (error) 400 sorf = AU_PRS_FAILURE; 401 else 402 sorf = AU_PRS_SUCCESS; 403 404 switch(ar->k_ar.ar_event) { 405 case AUE_OPEN_RWTC: 406 /* 407 * The open syscall always writes a AUE_OPEN_RWTC event; 408 * change it to the proper type of event based on the flags 409 * and the error value. 410 */ 411 ar->k_ar.ar_event = audit_flags_and_error_to_openevent( 412 ar->k_ar.ar_arg_fflags, error); 413 break; 414 415 case AUE_SYSCTL: 416 ar->k_ar.ar_event = audit_ctlname_to_sysctlevent( 417 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg); 418 break; 419 420 case AUE_AUDITON: 421 /* Convert the auditon() command to an event. */ 422 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd); 423 break; 424 } 425 426 auid = ar->k_ar.ar_subj_auid; 427 event = ar->k_ar.ar_event; 428 class = au_event_class(event); 429 430 ar->k_ar_commit |= AR_COMMIT_KERNEL; 431 if (au_preselect(event, class, aumask, sorf) != 0) 432 ar->k_ar_commit |= AR_PRESELECT_TRAIL; 433 if (audit_pipe_preselect(auid, event, class, sorf, 434 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0) 435 ar->k_ar_commit |= AR_PRESELECT_PIPE; 436 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE | 437 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE)) == 0) { 438 mtx_lock(&audit_mtx); 439 audit_pre_q_len--; 440 mtx_unlock(&audit_mtx); 441 audit_free(ar); 442 return; 443 } 444 445 ar->k_ar.ar_errno = error; 446 ar->k_ar.ar_retval = retval; 447 nanotime(&ar->k_ar.ar_endtime); 448 449 /* 450 * Note: it could be that some records initiated while audit was 451 * enabled should still be committed? 452 */ 453 mtx_lock(&audit_mtx); 454 if (audit_suspended || !audit_enabled) { 455 audit_pre_q_len--; 456 mtx_unlock(&audit_mtx); 457 audit_free(ar); 458 return; 459 } 460 461 /* 462 * Constrain the number of committed audit records based on the 463 * configurable parameter. 464 */ 465 while (audit_q_len >= audit_qctrl.aq_hiwater) 466 cv_wait(&audit_watermark_cv, &audit_mtx); 467 468 TAILQ_INSERT_TAIL(&audit_q, ar, k_q); 469 audit_q_len++; 470 audit_pre_q_len--; 471 cv_signal(&audit_worker_cv); 472 mtx_unlock(&audit_mtx); 473} 474 475/* 476 * audit_syscall_enter() is called on entry to each system call. It is 477 * responsible for deciding whether or not to audit the call (preselection), 478 * and if so, allocating a per-thread audit record. audit_new() will fill in 479 * basic thread/credential properties. 480 */ 481void 482audit_syscall_enter(unsigned short code, struct thread *td) 483{ 484 struct au_mask *aumask; 485 au_class_t class; 486 au_event_t event; 487 au_id_t auid; 488 489 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL")); 490 491 /* 492 * In FreeBSD, each ABI has its own system call table, and hence 493 * mapping of system call codes to audit events. Convert the code to 494 * an audit event identifier using the process system call table 495 * reference. In Darwin, there's only one, so we use the global 496 * symbol for the system call table. No audit record is generated 497 * for bad system calls, as no operation has been performed. 498 */ 499 if (code >= td->td_proc->p_sysent->sv_size) 500 return; 501 502 event = td->td_proc->p_sysent->sv_table[code].sy_auevent; 503 if (event == AUE_NULL) 504 return; 505 506 /* 507 * Check which audit mask to use; either the kernel non-attributable 508 * event mask or the process audit mask. 509 */ 510 auid = td->td_ucred->cr_audit.ai_auid; 511 if (auid == AU_DEFAUDITID) 512 aumask = &audit_nae_mask; 513 else 514 aumask = &td->td_ucred->cr_audit.ai_mask; 515 516 /* 517 * Allocate an audit record, if preselection allows it, and store in 518 * the thread for later use. 519 */ 520 class = au_event_class(event); 521 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) { 522 /* 523 * If we're out of space and need to suspend unprivileged 524 * processes, do that here rather than trying to allocate 525 * another audit record. 526 * 527 * Note: we might wish to be able to continue here in the 528 * future, if the system recovers. That should be possible 529 * by means of checking the condition in a loop around 530 * cv_wait(). It might be desirable to reevaluate whether an 531 * audit record is still required for this event by 532 * re-calling au_preselect(). 533 */ 534 if (audit_in_failure && 535 priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) { 536 cv_wait(&audit_fail_cv, &audit_mtx); 537 panic("audit_failing_stop: thread continued"); 538 } 539 td->td_ar = audit_new(event, td); 540 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) 541 td->td_ar = audit_new(event, td); 542 else 543 td->td_ar = NULL; 544} 545 546/* 547 * audit_syscall_exit() is called from the return of every system call, or in 548 * the event of exit1(), during the execution of exit1(). It is responsible 549 * for committing the audit record, if any, along with return condition. 550 */ 551void 552audit_syscall_exit(int error, struct thread *td) 553{ 554 int retval; 555 556 /* 557 * Commit the audit record as desired; once we pass the record into 558 * audit_commit(), the memory is owned by the audit subsystem. The 559 * return value from the system call is stored on the user thread. 560 * If there was an error, the return value is set to -1, imitating 561 * the behavior of the cerror routine. 562 */ 563 if (error) 564 retval = -1; 565 else 566 retval = td->td_retval[0]; 567 568 audit_commit(td->td_ar, error, retval); 569 td->td_ar = NULL; 570} 571 572void 573audit_cred_copy(struct ucred *src, struct ucred *dest) 574{ 575 576 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit)); 577} 578 579void 580audit_cred_destroy(struct ucred *cred) 581{ 582 583} 584 585void 586audit_cred_init(struct ucred *cred) 587{ 588 589 bzero(&cred->cr_audit, sizeof(cred->cr_audit)); 590} 591 592/* 593 * Initialize audit information for the first kernel process (proc 0) and for 594 * the first user process (init). 595 */ 596void 597audit_cred_kproc0(struct ucred *cred) 598{ 599 600 cred->cr_audit.ai_auid = AU_DEFAUDITID; 601 cred->cr_audit.ai_termid.at_type = AU_IPv4; 602} 603 604void 605audit_cred_proc1(struct ucred *cred) 606{ 607 608 cred->cr_audit.ai_auid = AU_DEFAUDITID; 609 cred->cr_audit.ai_termid.at_type = AU_IPv4; 610} 611 612void 613audit_thread_alloc(struct thread *td) 614{ 615 616 td->td_ar = NULL; 617} 618 619void 620audit_thread_free(struct thread *td) 621{ 622 623 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL")); 624} 625 626void 627audit_proc_coredump(struct thread *td, char *path, int errcode) 628{ 629 struct kaudit_record *ar; 630 struct au_mask *aumask; 631 au_class_t class; 632 int ret, sorf; 633 char **pathp; 634 au_id_t auid; 635 636 ret = 0; 637 638 /* 639 * Make sure we are using the correct preselection mask. 640 */ 641 auid = td->td_ucred->cr_audit.ai_auid; 642 if (auid == AU_DEFAUDITID) 643 aumask = &audit_nae_mask; 644 else 645 aumask = &td->td_ucred->cr_audit.ai_mask; 646 /* 647 * It's possible for coredump(9) generation to fail. Make sure that 648 * we handle this case correctly for preselection. 649 */ 650 if (errcode != 0) 651 sorf = AU_PRS_FAILURE; 652 else 653 sorf = AU_PRS_SUCCESS; 654 class = au_event_class(AUE_CORE); 655 if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 && 656 audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0) 657 return; 658 /* 659 * If we are interested in seeing this audit record, allocate it. 660 * Where possible coredump records should contain a pathname and arg32 661 * (signal) tokens. 662 */ 663 ar = audit_new(AUE_CORE, td); 664 if (path != NULL) { 665 pathp = &ar->k_ar.ar_arg_upath1; 666 *pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK); 667 audit_canon_path(td, path, *pathp); 668 ARG_SET_VALID(ar, ARG_UPATH1); 669 } 670 ar->k_ar.ar_arg_signum = td->td_proc->p_sig; 671 ARG_SET_VALID(ar, ARG_SIGNUM); 672 if (errcode != 0) 673 ret = 1; 674 audit_commit(ar, errcode, ret); 675} 676