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");
80 MALLOC_DEFINE(M_AUDITGIDSET, "audit_gidset", "Audit GID set storage");
82 SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0,
83 "TrustedBSD audit controls");
86 * Audit control settings that are set/read by system calls and are hence
89 * Define the audit control flags.
95 * Flags controlling behavior in low storage situations. Should we panic if
96 * a write fails? Should we fail stop if we're out of disk space?
98 int audit_panic_on_write_fail;
104 * Are we currently "failing stop" due to out of disk space?
106 int audit_in_failure;
109 * Global audit statistics.
111 struct audit_fstat audit_fstat;
114 * Preselection mask for non-attributable events.
116 struct au_mask audit_nae_mask;
119 * Mutex to protect global variables shared between various threads and
122 struct mtx audit_mtx;
125 * Queue of audit records ready for delivery to disk. We insert new records
126 * at the tail, and remove records from the head. Also, a count of the
127 * number of records used for checking queue depth. In addition, a counter
128 * of records that we have allocated but are not yet in the queue, which is
129 * needed to estimate the total size of the combined set of records
130 * outstanding in the system.
132 struct kaudit_queue audit_q;
137 * Audit queue control settings (minimum free, low/high water marks, etc.)
139 struct au_qctrl audit_qctrl;
142 * Condition variable to signal to the worker that it has work to do: either
143 * new records are in the queue, or a log replacement is taking place.
145 struct cv audit_worker_cv;
148 * Condition variable to flag when crossing the low watermark, meaning that
149 * threads blocked due to hitting the high watermark can wake up and continue
152 struct cv audit_watermark_cv;
155 * Condition variable for auditing threads wait on when in fail-stop mode.
156 * Threads wait on this CV forever (and ever), never seeing the light of day
159 static struct cv audit_fail_cv;
162 * Kernel audit information. This will store the current audit address
163 * or host information that the kernel will use when it's generating
164 * audit records. This data is modified by the A_GET{SET}KAUDIT auditon(2)
167 static struct auditinfo_addr audit_kinfo;
168 static struct rwlock audit_kinfo_lock;
170 #define KINFO_LOCK_INIT() rw_init(&audit_kinfo_lock, \
172 #define KINFO_RLOCK() rw_rlock(&audit_kinfo_lock)
173 #define KINFO_WLOCK() rw_wlock(&audit_kinfo_lock)
174 #define KINFO_RUNLOCK() rw_runlock(&audit_kinfo_lock)
175 #define KINFO_WUNLOCK() rw_wunlock(&audit_kinfo_lock)
178 audit_set_kinfo(struct auditinfo_addr *ak)
181 KASSERT(ak->ai_termid.at_type == AU_IPv4 ||
182 ak->ai_termid.at_type == AU_IPv6,
183 ("audit_set_kinfo: invalid address type"));
191 audit_get_kinfo(struct auditinfo_addr *ak)
194 KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 ||
195 audit_kinfo.ai_termid.at_type == AU_IPv6,
196 ("audit_set_kinfo: invalid address type"));
204 * Construct an audit record for the passed thread.
207 audit_record_ctor(void *mem, int size, void *arg, int flags)
209 struct kaudit_record *ar;
213 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
217 bzero(ar, sizeof(*ar));
218 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
219 nanotime(&ar->k_ar.ar_starttime);
222 * Export the subject credential.
225 cru2x(cred, &ar->k_ar.ar_subj_cred);
226 ar->k_ar.ar_subj_ruid = cred->cr_ruid;
227 ar->k_ar.ar_subj_rgid = cred->cr_rgid;
228 ar->k_ar.ar_subj_egid = cred->cr_groups[0];
229 ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
230 ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
231 ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
232 ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
233 ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
238 audit_record_dtor(void *mem, int size, void *arg)
240 struct kaudit_record *ar;
242 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
245 if (ar->k_ar.ar_arg_upath1 != NULL)
246 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
247 if (ar->k_ar.ar_arg_upath2 != NULL)
248 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
249 if (ar->k_ar.ar_arg_text != NULL)
250 free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
251 if (ar->k_udata != NULL)
252 free(ar->k_udata, M_AUDITDATA);
253 if (ar->k_ar.ar_arg_argv != NULL)
254 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
255 if (ar->k_ar.ar_arg_envv != NULL)
256 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
257 if (ar->k_ar.ar_arg_groups.gidset != NULL)
258 free(ar->k_ar.ar_arg_groups.gidset, M_AUDITGIDSET);
262 * Initialize the Audit subsystem: configuration state, work queue,
263 * synchronization primitives, worker thread, and trigger device node. Also
264 * call into the BSM assembly code to initialize it.
272 audit_panic_on_write_fail = 0;
274 audit_in_failure = 0;
278 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */
279 audit_fstat.af_currsz = 0;
280 audit_nae_mask.am_success = 0;
281 audit_nae_mask.am_failure = 0;
283 TAILQ_INIT(&audit_q);
286 audit_qctrl.aq_hiwater = AQ_HIWATER;
287 audit_qctrl.aq_lowater = AQ_LOWATER;
288 audit_qctrl.aq_bufsz = AQ_BUFSZ;
289 audit_qctrl.aq_minfree = AU_FS_MINFREE;
291 audit_kinfo.ai_termid.at_type = AU_IPv4;
292 audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY;
294 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
296 cv_init(&audit_worker_cv, "audit_worker_cv");
297 cv_init(&audit_watermark_cv, "audit_watermark_cv");
298 cv_init(&audit_fail_cv, "audit_fail_cv");
300 audit_record_zone = uma_zcreate("audit_record",
301 sizeof(struct kaudit_record), audit_record_ctor,
302 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
304 /* Initialize the BSM audit subsystem. */
307 audit_trigger_init();
309 /* Register shutdown handler. */
310 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
313 /* Start audit worker thread. */
317 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
320 * Drain the audit queue and close the log at shutdown. Note that this can
321 * be called both from the system shutdown path and also from audit
322 * configuration syscalls, so 'arg' and 'howto' are ignored.
324 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
325 * drain before returning, which could lead to lost records on shutdown.
328 audit_shutdown(void *arg, int howto)
331 audit_rotate_vnode(NULL, NULL);
335 * Return the current thread's audit record, if any.
337 struct kaudit_record *
341 return (curthread->td_ar);
345 * XXXAUDIT: There are a number of races present in the code below due to
346 * release and re-grab of the mutex. The code should be revised to become
347 * slightly less racy.
349 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
350 * pre_q space, suspending the system call until there is room?
352 struct kaudit_record *
353 audit_new(int event, struct thread *td)
355 struct kaudit_record *ar;
358 mtx_lock(&audit_mtx);
359 no_record = (audit_suspended || !audit_enabled);
360 mtx_unlock(&audit_mtx);
365 * Note: the number of outstanding uncommitted audit records is
366 * limited to the number of concurrent threads servicing system calls
369 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
370 ar->k_ar.ar_event = event;
372 mtx_lock(&audit_mtx);
374 mtx_unlock(&audit_mtx);
380 audit_free(struct kaudit_record *ar)
383 uma_zfree(audit_record_zone, ar);
387 audit_commit(struct kaudit_record *ar, int error, int retval)
393 struct au_mask *aumask;
399 * Decide whether to commit the audit record by checking the error
400 * value from the system call and using the appropriate audit mask.
402 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
403 aumask = &audit_nae_mask;
405 aumask = &ar->k_ar.ar_subj_amask;
408 sorf = AU_PRS_FAILURE;
410 sorf = AU_PRS_SUCCESS;
413 * syscalls.master sometimes contains a prototype event number, which
414 * we will transform into a more specific event number now that we
415 * have more complete information gathered during the system call.
417 switch(ar->k_ar.ar_event) {
419 ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
420 ar->k_ar.ar_arg_fflags, error);
423 case AUE_OPENAT_RWTC:
424 ar->k_ar.ar_event = audit_flags_and_error_to_openatevent(
425 ar->k_ar.ar_arg_fflags, error);
429 ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
430 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
434 /* Convert the auditon() command to an event. */
435 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
439 auid = ar->k_ar.ar_subj_auid;
440 event = ar->k_ar.ar_event;
441 class = au_event_class(event);
443 ar->k_ar_commit |= AR_COMMIT_KERNEL;
444 if (au_preselect(event, class, aumask, sorf) != 0)
445 ar->k_ar_commit |= AR_PRESELECT_TRAIL;
446 if (audit_pipe_preselect(auid, event, class, sorf,
447 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
448 ar->k_ar_commit |= AR_PRESELECT_PIPE;
449 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
450 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE)) == 0) {
451 mtx_lock(&audit_mtx);
453 mtx_unlock(&audit_mtx);
458 ar->k_ar.ar_errno = error;
459 ar->k_ar.ar_retval = retval;
460 nanotime(&ar->k_ar.ar_endtime);
463 * Note: it could be that some records initiated while audit was
464 * enabled should still be committed?
466 mtx_lock(&audit_mtx);
467 if (audit_suspended || !audit_enabled) {
469 mtx_unlock(&audit_mtx);
475 * Constrain the number of committed audit records based on the
476 * configurable parameter.
478 while (audit_q_len >= audit_qctrl.aq_hiwater)
479 cv_wait(&audit_watermark_cv, &audit_mtx);
481 TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
484 cv_signal(&audit_worker_cv);
485 mtx_unlock(&audit_mtx);
489 * audit_syscall_enter() is called on entry to each system call. It is
490 * responsible for deciding whether or not to audit the call (preselection),
491 * and if so, allocating a per-thread audit record. audit_new() will fill in
492 * basic thread/credential properties.
495 audit_syscall_enter(unsigned short code, struct thread *td)
497 struct au_mask *aumask;
502 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
503 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
504 ("audit_syscall_enter: TDP_AUDITREC set"));
507 * In FreeBSD, each ABI has its own system call table, and hence
508 * mapping of system call codes to audit events. Convert the code to
509 * an audit event identifier using the process system call table
510 * reference. In Darwin, there's only one, so we use the global
511 * symbol for the system call table. No audit record is generated
512 * for bad system calls, as no operation has been performed.
514 if (code >= td->td_proc->p_sysent->sv_size)
517 event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
518 if (event == AUE_NULL)
522 * Check which audit mask to use; either the kernel non-attributable
523 * event mask or the process audit mask.
525 auid = td->td_ucred->cr_audit.ai_auid;
526 if (auid == AU_DEFAUDITID)
527 aumask = &audit_nae_mask;
529 aumask = &td->td_ucred->cr_audit.ai_mask;
532 * Allocate an audit record, if preselection allows it, and store in
533 * the thread for later use.
535 class = au_event_class(event);
536 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
538 * If we're out of space and need to suspend unprivileged
539 * processes, do that here rather than trying to allocate
540 * another audit record.
542 * Note: we might wish to be able to continue here in the
543 * future, if the system recovers. That should be possible
544 * by means of checking the condition in a loop around
545 * cv_wait(). It might be desirable to reevaluate whether an
546 * audit record is still required for this event by
547 * re-calling au_preselect().
549 if (audit_in_failure &&
550 priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) {
551 cv_wait(&audit_fail_cv, &audit_mtx);
552 panic("audit_failing_stop: thread continued");
554 td->td_ar = audit_new(event, td);
555 if (td->td_ar != NULL)
556 td->td_pflags |= TDP_AUDITREC;
557 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
558 td->td_ar = audit_new(event, td);
559 if (td->td_ar != NULL)
560 td->td_pflags |= TDP_AUDITREC;
566 * audit_syscall_exit() is called from the return of every system call, or in
567 * the event of exit1(), during the execution of exit1(). It is responsible
568 * for committing the audit record, if any, along with return condition.
571 audit_syscall_exit(int error, struct thread *td)
576 * Commit the audit record as desired; once we pass the record into
577 * audit_commit(), the memory is owned by the audit subsystem. The
578 * return value from the system call is stored on the user thread.
579 * If there was an error, the return value is set to -1, imitating
580 * the behavior of the cerror routine.
585 retval = td->td_retval[0];
587 audit_commit(td->td_ar, error, retval);
589 td->td_pflags &= ~TDP_AUDITREC;
593 audit_cred_copy(struct ucred *src, struct ucred *dest)
596 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
600 audit_cred_destroy(struct ucred *cred)
606 audit_cred_init(struct ucred *cred)
609 bzero(&cred->cr_audit, sizeof(cred->cr_audit));
613 * Initialize audit information for the first kernel process (proc 0) and for
614 * the first user process (init).
617 audit_cred_kproc0(struct ucred *cred)
620 cred->cr_audit.ai_auid = AU_DEFAUDITID;
621 cred->cr_audit.ai_termid.at_type = AU_IPv4;
625 audit_cred_proc1(struct ucred *cred)
628 cred->cr_audit.ai_auid = AU_DEFAUDITID;
629 cred->cr_audit.ai_termid.at_type = AU_IPv4;
633 audit_thread_alloc(struct thread *td)
640 audit_thread_free(struct thread *td)
643 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
644 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
645 ("audit_thread_free: TDP_AUDITREC set"));
649 audit_proc_coredump(struct thread *td, char *path, int errcode)
651 struct kaudit_record *ar;
652 struct au_mask *aumask;
662 * Make sure we are using the correct preselection mask.
665 auid = cred->cr_audit.ai_auid;
666 if (auid == AU_DEFAUDITID)
667 aumask = &audit_nae_mask;
669 aumask = &cred->cr_audit.ai_mask;
671 * It's possible for coredump(9) generation to fail. Make sure that
672 * we handle this case correctly for preselection.
675 sorf = AU_PRS_FAILURE;
677 sorf = AU_PRS_SUCCESS;
678 class = au_event_class(AUE_CORE);
679 if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
680 audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
684 * If we are interested in seeing this audit record, allocate it.
685 * Where possible coredump records should contain a pathname and arg32
688 ar = audit_new(AUE_CORE, td);
690 pathp = &ar->k_ar.ar_arg_upath1;
691 *pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
692 audit_canon_path(td, path, *pathp);
693 ARG_SET_VALID(ar, ARG_UPATH1);
695 ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
696 ARG_SET_VALID(ar, ARG_SIGNUM);
699 audit_commit(ar, errcode, ret);