2 * Copyright (c) 1999-2005 Apple Inc.
3 * Copyright (c) 2006-2007 Robert N. M. Watson
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/condvar.h>
38 #include <sys/filedesc.h>
39 #include <sys/fcntl.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>
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>
60 #include <sys/unistd.h>
61 #include <sys/vnode.h>
63 #include <bsm/audit.h>
64 #include <bsm/audit_internal.h>
65 #include <bsm/audit_kevents.h>
67 #include <netinet/in.h>
68 #include <netinet/in_pcb.h>
70 #include <security/audit/audit.h>
71 #include <security/audit/audit_private.h>
75 static uma_zone_t audit_record_zone;
76 static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage");
77 MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
78 MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
79 MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
81 SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0,
82 "TrustedBSD audit controls");
85 * Audit control settings that are set/read by system calls and are hence
88 * Define the audit control flags.
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?
97 int audit_panic_on_write_fail;
103 * Are we currently "failing stop" due to out of disk space?
105 int audit_in_failure;
108 * Global audit statistics.
110 struct audit_fstat audit_fstat;
113 * Preselection mask for non-attributable events.
115 struct au_mask audit_nae_mask;
118 * Mutex to protect global variables shared between various threads and
121 struct mtx audit_mtx;
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.
131 struct kaudit_queue audit_q;
133 size_t audit_pre_q_len;
136 * Audit queue control settings (minimum free, low/high water marks, etc.)
138 struct au_qctrl audit_qctrl;
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.
144 struct cv audit_worker_cv;
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
151 struct cv audit_watermark_cv;
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
158 static struct cv audit_fail_cv;
161 * Construct an audit record for the passed thread.
164 audit_record_ctor(void *mem, int size, void *arg, int flags)
166 struct kaudit_record *ar;
170 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
174 bzero(ar, sizeof(*ar));
175 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
176 nanotime(&ar->k_ar.ar_starttime);
179 * Export the subject credential.
182 cru2x(cred, &ar->k_ar.ar_subj_cred);
183 ar->k_ar.ar_subj_ruid = cred->cr_ruid;
184 ar->k_ar.ar_subj_rgid = cred->cr_rgid;
185 ar->k_ar.ar_subj_egid = cred->cr_groups[0];
186 ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
187 ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
188 ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
189 ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
190 ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
195 audit_record_dtor(void *mem, int size, void *arg)
197 struct kaudit_record *ar;
199 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
202 if (ar->k_ar.ar_arg_upath1 != NULL)
203 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
204 if (ar->k_ar.ar_arg_upath2 != NULL)
205 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
206 if (ar->k_ar.ar_arg_text != NULL)
207 free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
208 if (ar->k_udata != NULL)
209 free(ar->k_udata, M_AUDITDATA);
210 if (ar->k_ar.ar_arg_argv != NULL)
211 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
212 if (ar->k_ar.ar_arg_envv != NULL)
213 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
217 * Initialize the Audit subsystem: configuration state, work queue,
218 * synchronization primitives, worker thread, and trigger device node. Also
219 * call into the BSM assembly code to initialize it.
227 audit_panic_on_write_fail = 0;
229 audit_in_failure = 0;
233 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */
234 audit_fstat.af_currsz = 0;
235 audit_nae_mask.am_success = 0;
236 audit_nae_mask.am_failure = 0;
238 TAILQ_INIT(&audit_q);
241 audit_qctrl.aq_hiwater = AQ_HIWATER;
242 audit_qctrl.aq_lowater = AQ_LOWATER;
243 audit_qctrl.aq_bufsz = AQ_BUFSZ;
244 audit_qctrl.aq_minfree = AU_FS_MINFREE;
246 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
247 cv_init(&audit_worker_cv, "audit_worker_cv");
248 cv_init(&audit_watermark_cv, "audit_watermark_cv");
249 cv_init(&audit_fail_cv, "audit_fail_cv");
251 audit_record_zone = uma_zcreate("audit_record",
252 sizeof(struct kaudit_record), audit_record_ctor,
253 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
255 /* Initialize the BSM audit subsystem. */
258 audit_trigger_init();
260 /* Register shutdown handler. */
261 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
264 /* Start audit worker thread. */
268 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
271 * Drain the audit queue and close the log at shutdown. Note that this can
272 * be called both from the system shutdown path and also from audit
273 * configuration syscalls, so 'arg' and 'howto' are ignored.
275 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
276 * drain before returning, which could lead to lost records on shutdown.
279 audit_shutdown(void *arg, int howto)
282 audit_rotate_vnode(NULL, NULL);
286 * Return the current thread's audit record, if any.
288 struct kaudit_record *
292 return (curthread->td_ar);
296 * XXXAUDIT: There are a number of races present in the code below due to
297 * release and re-grab of the mutex. The code should be revised to become
298 * slightly less racy.
300 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
301 * pre_q space, suspending the system call until there is room?
303 struct kaudit_record *
304 audit_new(int event, struct thread *td)
306 struct kaudit_record *ar;
309 mtx_lock(&audit_mtx);
310 no_record = (audit_suspended || !audit_enabled);
311 mtx_unlock(&audit_mtx);
316 * Note: the number of outstanding uncommitted audit records is
317 * limited to the number of concurrent threads servicing system calls
320 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
321 ar->k_ar.ar_event = event;
323 mtx_lock(&audit_mtx);
325 mtx_unlock(&audit_mtx);
331 audit_free(struct kaudit_record *ar)
334 uma_zfree(audit_record_zone, ar);
338 audit_commit(struct kaudit_record *ar, int error, int retval)
344 struct au_mask *aumask;
350 * Decide whether to commit the audit record by checking the error
351 * value from the system call and using the appropriate audit mask.
353 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
354 aumask = &audit_nae_mask;
356 aumask = &ar->k_ar.ar_subj_amask;
359 sorf = AU_PRS_FAILURE;
361 sorf = AU_PRS_SUCCESS;
363 switch(ar->k_ar.ar_event) {
366 * The open syscall always writes a AUE_OPEN_RWTC event;
367 * change it to the proper type of event based on the flags
368 * and the error value.
370 ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
371 ar->k_ar.ar_arg_fflags, error);
375 ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
376 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
380 /* Convert the auditon() command to an event. */
381 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
385 auid = ar->k_ar.ar_subj_auid;
386 event = ar->k_ar.ar_event;
387 class = au_event_class(event);
389 ar->k_ar_commit |= AR_COMMIT_KERNEL;
390 if (au_preselect(event, class, aumask, sorf) != 0)
391 ar->k_ar_commit |= AR_PRESELECT_TRAIL;
392 if (audit_pipe_preselect(auid, event, class, sorf,
393 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
394 ar->k_ar_commit |= AR_PRESELECT_PIPE;
395 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
396 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE)) == 0) {
397 mtx_lock(&audit_mtx);
399 mtx_unlock(&audit_mtx);
404 ar->k_ar.ar_errno = error;
405 ar->k_ar.ar_retval = retval;
406 nanotime(&ar->k_ar.ar_endtime);
409 * Note: it could be that some records initiated while audit was
410 * enabled should still be committed?
412 mtx_lock(&audit_mtx);
413 if (audit_suspended || !audit_enabled) {
415 mtx_unlock(&audit_mtx);
421 * Constrain the number of committed audit records based on the
422 * configurable parameter.
424 while (audit_q_len >= audit_qctrl.aq_hiwater)
425 cv_wait(&audit_watermark_cv, &audit_mtx);
427 TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
430 cv_signal(&audit_worker_cv);
431 mtx_unlock(&audit_mtx);
435 * audit_syscall_enter() is called on entry to each system call. It is
436 * responsible for deciding whether or not to audit the call (preselection),
437 * and if so, allocating a per-thread audit record. audit_new() will fill in
438 * basic thread/credential properties.
441 audit_syscall_enter(unsigned short code, struct thread *td)
443 struct au_mask *aumask;
448 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
449 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
450 ("audit_syscall_enter: TDP_AUDITREC set"));
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.
460 if (code >= td->td_proc->p_sysent->sv_size)
463 event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
464 if (event == AUE_NULL)
468 * Check which audit mask to use; either the kernel non-attributable
469 * event mask or the process audit mask.
471 auid = td->td_ucred->cr_audit.ai_auid;
472 if (auid == AU_DEFAUDITID)
473 aumask = &audit_nae_mask;
475 aumask = &td->td_ucred->cr_audit.ai_mask;
478 * Allocate an audit record, if preselection allows it, and store in
479 * the thread for later use.
481 class = au_event_class(event);
482 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
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.
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().
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");
500 td->td_ar = audit_new(event, td);
501 if (td->td_ar != NULL)
502 td->td_pflags |= TDP_AUDITREC;
503 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
504 td->td_ar = audit_new(event, td);
505 if (td->td_ar != NULL)
506 td->td_pflags |= TDP_AUDITREC;
512 * audit_syscall_exit() is called from the return of every system call, or in
513 * the event of exit1(), during the execution of exit1(). It is responsible
514 * for committing the audit record, if any, along with return condition.
517 audit_syscall_exit(int error, struct thread *td)
522 * Commit the audit record as desired; once we pass the record into
523 * audit_commit(), the memory is owned by the audit subsystem. The
524 * return value from the system call is stored on the user thread.
525 * If there was an error, the return value is set to -1, imitating
526 * the behavior of the cerror routine.
531 retval = td->td_retval[0];
533 audit_commit(td->td_ar, error, retval);
535 td->td_pflags &= ~TDP_AUDITREC;
539 audit_cred_copy(struct ucred *src, struct ucred *dest)
542 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
546 audit_cred_destroy(struct ucred *cred)
552 audit_cred_init(struct ucred *cred)
555 bzero(&cred->cr_audit, sizeof(cred->cr_audit));
559 * Initialize audit information for the first kernel process (proc 0) and for
560 * the first user process (init).
563 audit_cred_kproc0(struct ucred *cred)
566 cred->cr_audit.ai_auid = AU_DEFAUDITID;
567 cred->cr_audit.ai_termid.at_type = AU_IPv4;
571 audit_cred_proc1(struct ucred *cred)
574 cred->cr_audit.ai_auid = AU_DEFAUDITID;
575 cred->cr_audit.ai_termid.at_type = AU_IPv4;
579 audit_thread_alloc(struct thread *td)
586 audit_thread_free(struct thread *td)
589 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
590 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
591 ("audit_thread_free: TDP_AUDITREC set"));
595 audit_proc_coredump(struct thread *td, char *path, int errcode)
597 struct kaudit_record *ar;
598 struct au_mask *aumask;
608 * Make sure we are using the correct preselection mask.
611 auid = cred->cr_audit.ai_auid;
612 if (auid == AU_DEFAUDITID)
613 aumask = &audit_nae_mask;
615 aumask = &cred->cr_audit.ai_mask;
617 * It's possible for coredump(9) generation to fail. Make sure that
618 * we handle this case correctly for preselection.
621 sorf = AU_PRS_FAILURE;
623 sorf = AU_PRS_SUCCESS;
624 class = au_event_class(AUE_CORE);
625 if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
626 audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
630 * If we are interested in seeing this audit record, allocate it.
631 * Where possible coredump records should contain a pathname and arg32
634 ar = audit_new(AUE_CORE, td);
636 pathp = &ar->k_ar.ar_arg_upath1;
637 *pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
638 audit_canon_path(td, path, *pathp);
639 ARG_SET_VALID(ar, ARG_UPATH1);
641 ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
642 ARG_SET_VALID(ar, ARG_SIGNUM);
645 audit_commit(ar, errcode, ret);