2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1999-2005 Apple Inc.
5 * Copyright (c) 2006-2007, 2016-2017 Robert N. M. Watson
8 * Portions of this software were developed by BAE Systems, the University of
9 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
10 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
11 * Computing (TC) research program.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
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. Neither the name of Apple Inc. ("Apple") nor the names of
22 * its contributors may be used to endorse or promote products derived
23 * from this software without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS 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 APPLE OR ITS CONTRIBUTORS BE LIABLE FOR
29 * 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,
33 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
34 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35 * POSSIBILITY OF SUCH DAMAGE.
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/condvar.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/namei.h>
56 #include <sys/queue.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/domain.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
65 #include <sys/ucred.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
71 #include <bsm/audit.h>
72 #include <bsm/audit_internal.h>
73 #include <bsm/audit_kevents.h>
75 #include <netinet/in.h>
76 #include <netinet/in_pcb.h>
78 #include <security/audit/audit.h>
79 #include <security/audit/audit_private.h>
83 FEATURE(audit, "BSM audit support");
85 static uma_zone_t audit_record_zone;
86 static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage");
87 MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
88 MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
89 MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
90 MALLOC_DEFINE(M_AUDITGIDSET, "audit_gidset", "Audit GID set storage");
92 static SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0,
93 "TrustedBSD audit controls");
96 * Audit control settings that are set/read by system calls and are hence
99 * Define the audit control flags.
101 int __read_frequently audit_enabled;
105 * Flags controlling behavior in low storage situations. Should we panic if
106 * a write fails? Should we fail stop if we're out of disk space?
108 int audit_panic_on_write_fail;
114 * Are we currently "failing stop" due to out of disk space?
116 int audit_in_failure;
119 * Global audit statistics.
121 struct audit_fstat audit_fstat;
124 * Preselection mask for non-attributable events.
126 struct au_mask audit_nae_mask;
129 * Mutex to protect global variables shared between various threads and
132 struct mtx audit_mtx;
135 * Queue of audit records ready for delivery to disk. We insert new records
136 * at the tail, and remove records from the head. Also, a count of the
137 * number of records used for checking queue depth. In addition, a counter
138 * of records that we have allocated but are not yet in the queue, which is
139 * needed to estimate the total size of the combined set of records
140 * outstanding in the system.
142 struct kaudit_queue audit_q;
147 * Audit queue control settings (minimum free, low/high water marks, etc.)
149 struct au_qctrl audit_qctrl;
152 * Condition variable to signal to the worker that it has work to do: either
153 * new records are in the queue, or a log replacement is taking place.
155 struct cv audit_worker_cv;
158 * Condition variable to flag when crossing the low watermark, meaning that
159 * threads blocked due to hitting the high watermark can wake up and continue
162 struct cv audit_watermark_cv;
165 * Condition variable for auditing threads wait on when in fail-stop mode.
166 * Threads wait on this CV forever (and ever), never seeing the light of day
169 static struct cv audit_fail_cv;
172 * Optional DTrace audit provider support: function pointers for preselection
176 void *(*dtaudit_hook_preselect)(au_id_t auid, au_event_t event,
178 int (*dtaudit_hook_commit)(struct kaudit_record *kar, au_id_t auid,
179 au_event_t event, au_class_t class, int sorf);
180 void (*dtaudit_hook_bsm)(struct kaudit_record *kar, au_id_t auid,
181 au_event_t event, au_class_t class, int sorf,
182 void *bsm_data, size_t bsm_lenlen);
186 * Kernel audit information. This will store the current audit address
187 * or host information that the kernel will use when it's generating
188 * audit records. This data is modified by the A_GET{SET}KAUDIT auditon(2)
191 static struct auditinfo_addr audit_kinfo;
192 static struct rwlock audit_kinfo_lock;
194 #define KINFO_LOCK_INIT() rw_init(&audit_kinfo_lock, \
196 #define KINFO_RLOCK() rw_rlock(&audit_kinfo_lock)
197 #define KINFO_WLOCK() rw_wlock(&audit_kinfo_lock)
198 #define KINFO_RUNLOCK() rw_runlock(&audit_kinfo_lock)
199 #define KINFO_WUNLOCK() rw_wunlock(&audit_kinfo_lock)
202 audit_set_kinfo(struct auditinfo_addr *ak)
205 KASSERT(ak->ai_termid.at_type == AU_IPv4 ||
206 ak->ai_termid.at_type == AU_IPv6,
207 ("audit_set_kinfo: invalid address type"));
215 audit_get_kinfo(struct auditinfo_addr *ak)
218 KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 ||
219 audit_kinfo.ai_termid.at_type == AU_IPv6,
220 ("audit_set_kinfo: invalid address type"));
228 * Construct an audit record for the passed thread.
231 audit_record_ctor(void *mem, int size, void *arg, int flags)
233 struct kaudit_record *ar;
238 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
242 bzero(ar, sizeof(*ar));
243 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
244 nanotime(&ar->k_ar.ar_starttime);
247 * Export the subject credential.
250 cru2x(cred, &ar->k_ar.ar_subj_cred);
251 ar->k_ar.ar_subj_ruid = cred->cr_ruid;
252 ar->k_ar.ar_subj_rgid = cred->cr_rgid;
253 ar->k_ar.ar_subj_egid = cred->cr_groups[0];
254 ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
255 ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
256 ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
257 ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
258 ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
260 * If this process is jailed, make sure we capture the name of the
261 * jail so we can use it to generate a zonename token when we covert
262 * this record to BSM.
265 pr = cred->cr_prison;
266 (void) strlcpy(ar->k_ar.ar_jailname, pr->pr_name,
267 sizeof(ar->k_ar.ar_jailname));
269 ar->k_ar.ar_jailname[0] = '\0';
274 audit_record_dtor(void *mem, int size, void *arg)
276 struct kaudit_record *ar;
278 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
281 if (ar->k_ar.ar_arg_upath1 != NULL)
282 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
283 if (ar->k_ar.ar_arg_upath2 != NULL)
284 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
285 if (ar->k_ar.ar_arg_text != NULL)
286 free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
287 if (ar->k_udata != NULL)
288 free(ar->k_udata, M_AUDITDATA);
289 if (ar->k_ar.ar_arg_argv != NULL)
290 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
291 if (ar->k_ar.ar_arg_envv != NULL)
292 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
293 if (ar->k_ar.ar_arg_groups.gidset != NULL)
294 free(ar->k_ar.ar_arg_groups.gidset, M_AUDITGIDSET);
298 * Initialize the Audit subsystem: configuration state, work queue,
299 * synchronization primitives, worker thread, and trigger device node. Also
300 * call into the BSM assembly code to initialize it.
308 audit_panic_on_write_fail = 0;
310 audit_in_failure = 0;
314 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */
315 audit_fstat.af_currsz = 0;
316 audit_nae_mask.am_success = 0;
317 audit_nae_mask.am_failure = 0;
319 TAILQ_INIT(&audit_q);
322 audit_qctrl.aq_hiwater = AQ_HIWATER;
323 audit_qctrl.aq_lowater = AQ_LOWATER;
324 audit_qctrl.aq_bufsz = AQ_BUFSZ;
325 audit_qctrl.aq_minfree = AU_FS_MINFREE;
327 audit_kinfo.ai_termid.at_type = AU_IPv4;
328 audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY;
330 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
332 cv_init(&audit_worker_cv, "audit_worker_cv");
333 cv_init(&audit_watermark_cv, "audit_watermark_cv");
334 cv_init(&audit_fail_cv, "audit_fail_cv");
336 audit_record_zone = uma_zcreate("audit_record",
337 sizeof(struct kaudit_record), audit_record_ctor,
338 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
340 /* Initialize the BSM audit subsystem. */
343 audit_trigger_init();
345 /* Register shutdown handler. */
346 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
349 /* Start audit worker thread. */
353 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
356 * Drain the audit queue and close the log at shutdown. Note that this can
357 * be called both from the system shutdown path and also from audit
358 * configuration syscalls, so 'arg' and 'howto' are ignored.
360 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
361 * drain before returning, which could lead to lost records on shutdown.
364 audit_shutdown(void *arg, int howto)
367 audit_rotate_vnode(NULL, NULL);
371 * Return the current thread's audit record, if any.
373 struct kaudit_record *
377 return (curthread->td_ar);
381 * XXXAUDIT: There are a number of races present in the code below due to
382 * release and re-grab of the mutex. The code should be revised to become
383 * slightly less racy.
385 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
386 * pre_q space, suspending the system call until there is room?
388 struct kaudit_record *
389 audit_new(int event, struct thread *td)
391 struct kaudit_record *ar;
394 mtx_lock(&audit_mtx);
395 no_record = (audit_suspended || !audit_enabled);
396 mtx_unlock(&audit_mtx);
401 * Note: the number of outstanding uncommitted audit records is
402 * limited to the number of concurrent threads servicing system calls
405 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
406 ar->k_ar.ar_event = event;
408 mtx_lock(&audit_mtx);
410 mtx_unlock(&audit_mtx);
416 audit_free(struct kaudit_record *ar)
419 uma_zfree(audit_record_zone, ar);
423 audit_commit(struct kaudit_record *ar, int error, int retval)
429 struct au_mask *aumask;
434 ar->k_ar.ar_errno = error;
435 ar->k_ar.ar_retval = retval;
436 nanotime(&ar->k_ar.ar_endtime);
439 * Decide whether to commit the audit record by checking the error
440 * value from the system call and using the appropriate audit mask.
442 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
443 aumask = &audit_nae_mask;
445 aumask = &ar->k_ar.ar_subj_amask;
448 sorf = AU_PRS_FAILURE;
450 sorf = AU_PRS_SUCCESS;
453 * syscalls.master sometimes contains a prototype event number, which
454 * we will transform into a more specific event number now that we
455 * have more complete information gathered during the system call.
457 switch(ar->k_ar.ar_event) {
459 ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
460 ar->k_ar.ar_arg_fflags, error);
463 case AUE_OPENAT_RWTC:
464 ar->k_ar.ar_event = audit_flags_and_error_to_openatevent(
465 ar->k_ar.ar_arg_fflags, error);
469 ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
470 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
474 /* Convert the auditon() command to an event. */
475 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
479 if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
481 audit_msgsys_to_event(ar->k_ar.ar_arg_svipc_which);
485 if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
487 audit_semsys_to_event(ar->k_ar.ar_arg_svipc_which);
491 if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
493 audit_shmsys_to_event(ar->k_ar.ar_arg_svipc_which);
497 auid = ar->k_ar.ar_subj_auid;
498 event = ar->k_ar.ar_event;
499 class = au_event_class(event);
501 ar->k_ar_commit |= AR_COMMIT_KERNEL;
502 if (au_preselect(event, class, aumask, sorf) != 0)
503 ar->k_ar_commit |= AR_PRESELECT_TRAIL;
504 if (audit_pipe_preselect(auid, event, class, sorf,
505 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
506 ar->k_ar_commit |= AR_PRESELECT_PIPE;
509 * Expose the audit record to DTrace, both to allow the "commit" probe
510 * to fire if it's desirable, and also to allow a decision to be made
511 * about later firing with BSM in the audit worker.
513 if (dtaudit_hook_commit != NULL) {
514 if (dtaudit_hook_commit(ar, auid, event, class, sorf) != 0)
515 ar->k_ar_commit |= AR_PRESELECT_DTRACE;
519 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
520 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE |
521 AR_PRESELECT_DTRACE)) == 0) {
522 mtx_lock(&audit_mtx);
524 mtx_unlock(&audit_mtx);
530 * Note: it could be that some records initiated while audit was
531 * enabled should still be committed?
533 mtx_lock(&audit_mtx);
534 if (audit_suspended || !audit_enabled) {
536 mtx_unlock(&audit_mtx);
542 * Constrain the number of committed audit records based on the
543 * configurable parameter.
545 while (audit_q_len >= audit_qctrl.aq_hiwater)
546 cv_wait(&audit_watermark_cv, &audit_mtx);
548 TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
551 cv_signal(&audit_worker_cv);
552 mtx_unlock(&audit_mtx);
556 * audit_syscall_enter() is called on entry to each system call. It is
557 * responsible for deciding whether or not to audit the call (preselection),
558 * and if so, allocating a per-thread audit record. audit_new() will fill in
559 * basic thread/credential properties.
562 audit_syscall_enter(unsigned short code, struct thread *td)
564 struct au_mask *aumask;
573 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
574 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
575 ("audit_syscall_enter: TDP_AUDITREC set"));
578 * In FreeBSD, each ABI has its own system call table, and hence
579 * mapping of system call codes to audit events. Convert the code to
580 * an audit event identifier using the process system call table
581 * reference. In Darwin, there's only one, so we use the global
582 * symbol for the system call table. No audit record is generated
583 * for bad system calls, as no operation has been performed.
585 if (code >= td->td_proc->p_sysent->sv_size)
588 event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
589 if (event == AUE_NULL)
593 * Check which audit mask to use; either the kernel non-attributable
594 * event mask or the process audit mask.
596 auid = td->td_ucred->cr_audit.ai_auid;
597 if (auid == AU_DEFAUDITID)
598 aumask = &audit_nae_mask;
600 aumask = &td->td_ucred->cr_audit.ai_mask;
603 * Determine whether trail or pipe preselection would like an audit
604 * record allocated for this system call.
606 class = au_event_class(event);
607 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
609 * If we're out of space and need to suspend unprivileged
610 * processes, do that here rather than trying to allocate
611 * another audit record.
613 * Note: we might wish to be able to continue here in the
614 * future, if the system recovers. That should be possible
615 * by means of checking the condition in a loop around
616 * cv_wait(). It might be desirable to reevaluate whether an
617 * audit record is still required for this event by
618 * re-calling au_preselect().
620 if (audit_in_failure &&
621 priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) {
622 cv_wait(&audit_fail_cv, &audit_mtx);
623 panic("audit_failing_stop: thread continued");
626 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
633 * After audit trails and pipes have made their policy choices, DTrace
634 * may request that records be generated as well. This is a slightly
635 * complex affair, as the DTrace audit provider needs the audit
636 * framework to maintain some state on the audit record, which has not
637 * been allocated at the point where the decision has to be made.
638 * This hook must run even if we are not changing the decision, as
639 * DTrace may want to stick event state onto a record we were going to
640 * produce due to the trail or pipes. The event state returned by the
641 * DTrace provider must be safe without locks held between here and
642 * below -- i.e., dtaudit_state must must refer to stable memory.
645 dtaudit_state = NULL;
646 if (dtaudit_hook_preselect != NULL) {
647 dtaudit_state = dtaudit_hook_preselect(auid, event, class);
648 if (dtaudit_state != NULL)
654 * If a record is required, allocate it and attach it to the thread
655 * for use throughout the system call. Also attach DTrace state if
658 * XXXRW: If we decide to reference count the evname_elem underlying
659 * dtaudit_state, we will need to free here if no record is allocated
663 td->td_ar = audit_new(event, td);
664 if (td->td_ar != NULL) {
665 td->td_pflags |= TDP_AUDITREC;
667 td->td_ar->k_dtaudit_state = dtaudit_state;
675 * audit_syscall_exit() is called from the return of every system call, or in
676 * the event of exit1(), during the execution of exit1(). It is responsible
677 * for committing the audit record, if any, along with return condition.
680 audit_syscall_exit(int error, struct thread *td)
685 * Commit the audit record as desired; once we pass the record into
686 * audit_commit(), the memory is owned by the audit subsystem. The
687 * return value from the system call is stored on the user thread.
688 * If there was an error, the return value is set to -1, imitating
689 * the behavior of the cerror routine.
694 retval = td->td_retval[0];
696 audit_commit(td->td_ar, error, retval);
698 td->td_pflags &= ~TDP_AUDITREC;
702 audit_cred_copy(struct ucred *src, struct ucred *dest)
705 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
709 audit_cred_destroy(struct ucred *cred)
715 audit_cred_init(struct ucred *cred)
718 bzero(&cred->cr_audit, sizeof(cred->cr_audit));
722 * Initialize audit information for the first kernel process (proc 0) and for
723 * the first user process (init).
726 audit_cred_kproc0(struct ucred *cred)
729 cred->cr_audit.ai_auid = AU_DEFAUDITID;
730 cred->cr_audit.ai_termid.at_type = AU_IPv4;
734 audit_cred_proc1(struct ucred *cred)
737 cred->cr_audit.ai_auid = AU_DEFAUDITID;
738 cred->cr_audit.ai_termid.at_type = AU_IPv4;
742 audit_thread_alloc(struct thread *td)
749 audit_thread_free(struct thread *td)
752 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
753 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
754 ("audit_thread_free: TDP_AUDITREC set"));
758 audit_proc_coredump(struct thread *td, char *path, int errcode)
760 struct kaudit_record *ar;
761 struct au_mask *aumask;
771 * Make sure we are using the correct preselection mask.
774 auid = cred->cr_audit.ai_auid;
775 if (auid == AU_DEFAUDITID)
776 aumask = &audit_nae_mask;
778 aumask = &cred->cr_audit.ai_mask;
780 * It's possible for coredump(9) generation to fail. Make sure that
781 * we handle this case correctly for preselection.
784 sorf = AU_PRS_FAILURE;
786 sorf = AU_PRS_SUCCESS;
787 class = au_event_class(AUE_CORE);
788 if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
789 audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
793 * If we are interested in seeing this audit record, allocate it.
794 * Where possible coredump records should contain a pathname and arg32
797 ar = audit_new(AUE_CORE, td);
801 pathp = &ar->k_ar.ar_arg_upath1;
802 *pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
803 audit_canon_path(td, AT_FDCWD, path, *pathp);
804 ARG_SET_VALID(ar, ARG_UPATH1);
806 ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
807 ARG_SET_VALID(ar, ARG_SIGNUM);
810 audit_commit(ar, errcode, ret);