1 /* 2 * Copyright (c) 1999-2005 Apple Computer, 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 Computer, 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 * $FreeBSD$ 31 */ 32 33 #include <sys/param.h> 34 #include <sys/condvar.h> 35 #include <sys/conf.h> 36 #include <sys/file.h> 37 #include <sys/filedesc.h> 38 #include <sys/fcntl.h> 39 #include <sys/ipc.h> 40 #include <sys/kernel.h> 41 #include <sys/kthread.h> 42 #include <sys/malloc.h> 43 #include <sys/mount.h> 44 #include <sys/namei.h> 45 #include <sys/priv.h> 46 #include <sys/proc.h> 47 #include <sys/queue.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/protosw.h> 51 #include <sys/domain.h> 52 #include <sys/sysproto.h> 53 #include <sys/sysent.h> 54 #include <sys/systm.h> 55 #include <sys/ucred.h> 56 #include <sys/uio.h> 57 #include <sys/un.h> 58 #include <sys/unistd.h> 59 #include <sys/vnode.h> 60 61 #include <bsm/audit.h> 62 #include <bsm/audit_internal.h> 63 #include <bsm/audit_kevents.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_pcb.h> 67 68 #include <security/audit/audit.h> 69 #include <security/audit/audit_private.h> 70 71 #include <vm/uma.h> 72 73 static uma_zone_t audit_record_zone; 74 static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage"); 75 MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage"); 76 MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage"); 77 MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage"); 78 79 /* 80 * Audit control settings that are set/read by system calls and are hence 81 * non-static. 82 * 83 * Define the audit control flags. 84 */ 85 int audit_enabled; 86 int audit_suspended; 87 88 /* 89 * Flags controlling behavior in low storage situations. Should we panic if 90 * a write fails? Should we fail stop if we're out of disk space? 91 */ 92 int audit_panic_on_write_fail; 93 int audit_fail_stop; 94 int audit_argv; 95 int audit_arge; 96 97 /* 98 * Are we currently "failing stop" due to out of disk space? 99 */ 100 int audit_in_failure; 101 102 /* 103 * Global audit statistiscs. 104 */ 105 struct audit_fstat audit_fstat; 106 107 /* 108 * Preselection mask for non-attributable events. 109 */ 110 struct au_mask audit_nae_mask; 111 112 /* 113 * Mutex to protect global variables shared between various threads and 114 * processes. 115 */ 116 struct mtx audit_mtx; 117 118 /* 119 * Queue of audit records ready for delivery to disk. We insert new records 120 * at the tail, and remove records from the head. Also, a count of the 121 * number of records used for checking queue depth. In addition, a counter 122 * of records that we have allocated but are not yet in the queue, which is 123 * needed to estimate the total size of the combined set of records 124 * outstanding in the system. 125 */ 126 struct kaudit_queue audit_q; 127 int audit_q_len; 128 int audit_pre_q_len; 129 130 /* 131 * Audit queue control settings (minimum free, low/high water marks, etc.) 132 */ 133 struct au_qctrl audit_qctrl; 134 135 /* 136 * Condition variable to signal to the worker that it has work to do: either 137 * new records are in the queue, or a log replacement is taking place. 138 */ 139 struct cv audit_worker_cv; 140 141 /* 142 * Condition variable to flag when crossing the low watermark, meaning that 143 * threads blocked due to hitting the high watermark can wake up and continue 144 * to commit records. 145 */ 146 struct cv audit_watermark_cv; 147 148 /* 149 * Condition variable for auditing threads wait on when in fail-stop mode. 150 * Threads wait on this CV forever (and ever), never seeing the light of day 151 * again. 152 */ 153 static struct cv audit_fail_cv; 154 155 /* 156 * Construct an audit record for the passed thread. 157 */ 158 static int 159 audit_record_ctor(void *mem, int size, void *arg, int flags) 160 { 161 struct kaudit_record *ar; 162 struct thread *td; 163 164 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size")); 165 166 td = arg; 167 ar = mem; 168 bzero(ar, sizeof(*ar)); 169 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC; 170 nanotime(&ar->k_ar.ar_starttime); 171 172 /* 173 * Export the subject credential. 174 */ 175 cru2x(td->td_ucred, &ar->k_ar.ar_subj_cred); 176 ar->k_ar.ar_subj_ruid = td->td_ucred->cr_ruid; 177 ar->k_ar.ar_subj_rgid = td->td_ucred->cr_rgid; 178 ar->k_ar.ar_subj_egid = td->td_ucred->cr_groups[0]; 179 ar->k_ar.ar_subj_auid = td->td_ucred->cr_audit.ai_auid; 180 ar->k_ar.ar_subj_asid = td->td_ucred->cr_audit.ai_asid; 181 ar->k_ar.ar_subj_pid = td->td_proc->p_pid; 182 ar->k_ar.ar_subj_amask = td->td_ucred->cr_audit.ai_mask; 183 ar->k_ar.ar_subj_term_addr = td->td_ucred->cr_audit.ai_termid; 184 return (0); 185 } 186 187 static void 188 audit_record_dtor(void *mem, int size, void *arg) 189 { 190 struct kaudit_record *ar; 191 192 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size")); 193 194 ar = mem; 195 if (ar->k_ar.ar_arg_upath1 != NULL) 196 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH); 197 if (ar->k_ar.ar_arg_upath2 != NULL) 198 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH); 199 if (ar->k_ar.ar_arg_text != NULL) 200 free(ar->k_ar.ar_arg_text, M_AUDITTEXT); 201 if (ar->k_udata != NULL) 202 free(ar->k_udata, M_AUDITDATA); 203 if (ar->k_ar.ar_arg_argv != NULL) 204 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT); 205 if (ar->k_ar.ar_arg_envv != NULL) 206 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT); 207 } 208 209 /* 210 * Initialize the Audit subsystem: configuration state, work queue, 211 * synchronization primitives, worker thread, and trigger device node. Also 212 * call into the BSM assembly code to initialize it. 213 */ 214 static void 215 audit_init(void) 216 { 217 218 printf("Security auditing service present\n"); 219 audit_enabled = 0; 220 audit_suspended = 0; 221 audit_panic_on_write_fail = 0; 222 audit_fail_stop = 0; 223 audit_in_failure = 0; 224 audit_argv = 0; 225 audit_arge = 0; 226 227 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */ 228 audit_fstat.af_currsz = 0; 229 audit_nae_mask.am_success = AU_NULL; 230 audit_nae_mask.am_failure = AU_NULL; 231 232 TAILQ_INIT(&audit_q); 233 audit_q_len = 0; 234 audit_pre_q_len = 0; 235 audit_qctrl.aq_hiwater = AQ_HIWATER; 236 audit_qctrl.aq_lowater = AQ_LOWATER; 237 audit_qctrl.aq_bufsz = AQ_BUFSZ; 238 audit_qctrl.aq_minfree = AU_FS_MINFREE; 239 240 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF); 241 cv_init(&audit_worker_cv, "audit_worker_cv"); 242 cv_init(&audit_watermark_cv, "audit_watermark_cv"); 243 cv_init(&audit_fail_cv, "audit_fail_cv"); 244 245 audit_record_zone = uma_zcreate("audit_record", 246 sizeof(struct kaudit_record), audit_record_ctor, 247 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0); 248 249 /* Initialize the BSM audit subsystem. */ 250 kau_init(); 251 252 audit_trigger_init(); 253 254 /* Register shutdown handler. */ 255 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL, 256 SHUTDOWN_PRI_FIRST); 257 258 /* Start audit worker thread. */ 259 audit_worker_init(); 260 } 261 262 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL) 263 264 /* 265 * Drain the audit queue and close the log at shutdown. Note that this can 266 * be called both from the system shutdown path and also from audit 267 * configuration syscalls, so 'arg' and 'howto' are ignored. 268 */ 269 void 270 audit_shutdown(void *arg, int howto) 271 { 272 273 audit_rotate_vnode(NULL, NULL); 274 } 275 276 /* 277 * Return the current thread's audit record, if any. 278 */ 279 struct kaudit_record * 280 currecord(void) 281 { 282 283 return (curthread->td_ar); 284 } 285 286 /* 287 * XXXAUDIT: There are a number of races present in the code below due to 288 * release and re-grab of the mutex. The code should be revised to become 289 * slightly less racy. 290 * 291 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available 292 * pre_q space, suspending the system call until there is room? 293 */ 294 struct kaudit_record * 295 audit_new(int event, struct thread *td) 296 { 297 struct kaudit_record *ar; 298 int no_record; 299 300 mtx_lock(&audit_mtx); 301 no_record = (audit_suspended || !audit_enabled); 302 mtx_unlock(&audit_mtx); 303 if (no_record) 304 return (NULL); 305 306 /* 307 * Note: the number of outstanding uncommitted audit records is 308 * limited to the number of concurrent threads servicing system calls 309 * in the kernel. 310 */ 311 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK); 312 ar->k_ar.ar_event = event; 313 314 mtx_lock(&audit_mtx); 315 audit_pre_q_len++; 316 mtx_unlock(&audit_mtx); 317 318 return (ar); 319 } 320 321 void 322 audit_free(struct kaudit_record *ar) 323 { 324 325 uma_zfree(audit_record_zone, ar); 326 } 327 328 void 329 audit_commit(struct kaudit_record *ar, int error, int retval) 330 { 331 au_event_t event; 332 au_class_t class; 333 au_id_t auid; 334 int sorf; 335 struct au_mask *aumask; 336 337 if (ar == NULL) 338 return; 339 340 /* 341 * Decide whether to commit the audit record by checking the error 342 * value from the system call and using the appropriate audit mask. 343 * 344 * XXXAUDIT: Synchronize access to audit_nae_mask? 345 */ 346 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID) 347 aumask = &audit_nae_mask; 348 else 349 aumask = &ar->k_ar.ar_subj_amask; 350 351 if (error) 352 sorf = AU_PRS_FAILURE; 353 else 354 sorf = AU_PRS_SUCCESS; 355 356 switch(ar->k_ar.ar_event) { 357 case AUE_OPEN_RWTC: 358 /* 359 * The open syscall always writes a AUE_OPEN_RWTC event; 360 * change it to the proper type of event based on the flags 361 * and the error value. 362 */ 363 ar->k_ar.ar_event = flags_and_error_to_openevent( 364 ar->k_ar.ar_arg_fflags, error); 365 break; 366 367 case AUE_SYSCTL: 368 ar->k_ar.ar_event = ctlname_to_sysctlevent( 369 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg); 370 break; 371 372 case AUE_AUDITON: 373 /* Convert the auditon() command to an event */ 374 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd); 375 break; 376 } 377 378 auid = ar->k_ar.ar_subj_auid; 379 event = ar->k_ar.ar_event; 380 class = au_event_class(event); 381 382 ar->k_ar_commit |= AR_COMMIT_KERNEL; 383 if (au_preselect(event, class, aumask, sorf) != 0) 384 ar->k_ar_commit |= AR_PRESELECT_TRAIL; 385 if (audit_pipe_preselect(auid, event, class, sorf, 386 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0) 387 ar->k_ar_commit |= AR_PRESELECT_PIPE; 388 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE | 389 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE)) == 0) { 390 mtx_lock(&audit_mtx); 391 audit_pre_q_len--; 392 mtx_unlock(&audit_mtx); 393 audit_free(ar); 394 return; 395 } 396 397 ar->k_ar.ar_errno = error; 398 ar->k_ar.ar_retval = retval; 399 400 /* 401 * We might want to do some system-wide post-filtering here at some 402 * point. 403 */ 404 405 /* 406 * Timestamp system call end. 407 */ 408 nanotime(&ar->k_ar.ar_endtime); 409 410 /* 411 * Note: it could be that some records initiated while audit was 412 * enabled should still be committed? 413 */ 414 mtx_lock(&audit_mtx); 415 if (audit_suspended || !audit_enabled) { 416 audit_pre_q_len--; 417 mtx_unlock(&audit_mtx); 418 audit_free(ar); 419 return; 420 } 421 422 /* 423 * Constrain the number of committed audit records based on the 424 * configurable parameter. 425 */ 426 while (audit_q_len >= audit_qctrl.aq_hiwater) 427 cv_wait(&audit_watermark_cv, &audit_mtx); 428 429 TAILQ_INSERT_TAIL(&audit_q, ar, k_q); 430 audit_q_len++; 431 audit_pre_q_len--; 432 cv_signal(&audit_worker_cv); 433 mtx_unlock(&audit_mtx); 434 } 435 436 /* 437 * audit_syscall_enter() is called on entry to each system call. It is 438 * responsible for deciding whether or not to audit the call (preselection), 439 * and if so, allocating a per-thread audit record. audit_new() will fill in 440 * basic thread/credential properties. 441 */ 442 void 443 audit_syscall_enter(unsigned short code, struct thread *td) 444 { 445 struct au_mask *aumask; 446 au_class_t class; 447 au_event_t event; 448 au_id_t auid; 449 450 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL")); 451 452 /* 453 * In FreeBSD, each ABI has its own system call table, and hence 454 * mapping of system call codes to audit events. Convert the code to 455 * an audit event identifier using the process system call table 456 * reference. In Darwin, there's only one, so we use the global 457 * symbol for the system call table. No audit record is generated 458 * for bad system calls, as no operation has been performed. 459 */ 460 if (code >= td->td_proc->p_sysent->sv_size) 461 return; 462 463 event = td->td_proc->p_sysent->sv_table[code].sy_auevent; 464 if (event == AUE_NULL) 465 return; 466 467 /* 468 * Check which audit mask to use; either the kernel non-attributable 469 * event mask or the process audit mask. 470 */ 471 auid = td->td_ucred->cr_audit.ai_auid; 472 if (auid == AU_DEFAUDITID) 473 aumask = &audit_nae_mask; 474 else 475 aumask = &td->td_ucred->cr_audit.ai_mask; 476 477 /* 478 * Allocate an audit record, if preselection allows it, and store in 479 * the thread for later use. 480 */ 481 class = au_event_class(event); 482 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) { 483 /* 484 * If we're out of space and need to suspend unprivileged 485 * processes, do that here rather than trying to allocate 486 * another audit record. 487 * 488 * Note: we might wish to be able to continue here in the 489 * future, if the system recovers. That should be possible 490 * by means of checking the condition in a loop around 491 * cv_wait(). It might be desirable to reevaluate whether an 492 * audit record is still required for this event by 493 * re-calling au_preselect(). 494 */ 495 if (audit_in_failure && 496 priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) { 497 cv_wait(&audit_fail_cv, &audit_mtx); 498 panic("audit_failing_stop: thread continued"); 499 } 500 td->td_ar = audit_new(event, td); 501 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) 502 td->td_ar = audit_new(event, td); 503 else 504 td->td_ar = NULL; 505 } 506 507 /* 508 * audit_syscall_exit() is called from the return of every system call, or in 509 * the event of exit1(), during the execution of exit1(). It is responsible 510 * for committing the audit record, if any, along with return condition. 511 */ 512 void 513 audit_syscall_exit(int error, struct thread *td) 514 { 515 int retval; 516 517 /* 518 * Commit the audit record as desired; once we pass the record into 519 * audit_commit(), the memory is owned by the audit subsystem. The 520 * return value from the system call is stored on the user thread. 521 * If there was an error, the return value is set to -1, imitating 522 * the behavior of the cerror routine. 523 */ 524 if (error) 525 retval = -1; 526 else 527 retval = td->td_retval[0]; 528 529 audit_commit(td->td_ar, error, retval); 530 td->td_ar = NULL; 531 } 532 533 /* 534 * Copy audit state from an existing credential to a new credential. 535 */ 536 void 537 audit_cred_copy(struct ucred *src, struct ucred *dest) 538 { 539 540 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit)); 541 } 542 543 /* 544 * Free audit state from a credential when the credential is freed. 545 */ 546 void 547 audit_cred_destroy(struct ucred *cred) 548 { 549 550 bzero(&cred->cr_audit, sizeof(cred->cr_audit)); 551 } 552 553 /* 554 * Allocate audit state for a new credential. 555 */ 556 void 557 audit_cred_init(struct ucred *cred) 558 { 559 560 bzero(&cred->cr_audit, sizeof(cred->cr_audit)); 561 } 562 563 /* 564 * Initialize audit information for the first kernel process (proc 0) and for 565 * the first user process (init). 566 */ 567 void 568 audit_cred_kproc0(struct ucred *cred) 569 { 570 571 } 572 573 void 574 audit_cred_proc1(struct ucred *cred) 575 { 576 577 cred->cr_audit.ai_auid = AU_DEFAUDITID; 578 } 579 580 /* 581 * Allocate storage for a new thread. 582 */ 583 void 584 audit_thread_alloc(struct thread *td) 585 { 586 587 td->td_ar = NULL; 588 } 589 590 /* 591 * Thread destruction. 592 */ 593 void 594 audit_thread_free(struct thread *td) 595 { 596 597 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL")); 598 } 599