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>
42 #include <sys/kernel.h>
43 #include <sys/kthread.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/namei.h>
49 #include <sys/queue.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/protosw.h>
53 #include <sys/domain.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysproto.h>
56 #include <sys/sysent.h>
57 #include <sys/systm.h>
58 #include <sys/ucred.h>
61 #include <sys/unistd.h>
62 #include <sys/vnode.h>
64 #include <bsm/audit.h>
65 #include <bsm/audit_internal.h>
66 #include <bsm/audit_kevents.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
71 #include <security/audit/audit.h>
72 #include <security/audit/audit_private.h>
76 FEATURE(audit, "BSM audit support");
78 static uma_zone_t audit_record_zone;
79 static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage");
80 MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
81 MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
82 MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
83 MALLOC_DEFINE(M_AUDITGIDSET, "audit_gidset", "Audit GID set storage");
85 static SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0,
86 "TrustedBSD audit controls");
89 * Audit control settings that are set/read by system calls and are hence
92 * Define the audit control flags.
98 * Flags controlling behavior in low storage situations. Should we panic if
99 * a write fails? Should we fail stop if we're out of disk space?
101 int audit_panic_on_write_fail;
107 * Are we currently "failing stop" due to out of disk space?
109 int audit_in_failure;
112 * Global audit statistics.
114 struct audit_fstat audit_fstat;
117 * Preselection mask for non-attributable events.
119 struct au_mask audit_nae_mask;
122 * Mutex to protect global variables shared between various threads and
125 struct mtx audit_mtx;
128 * Queue of audit records ready for delivery to disk. We insert new records
129 * at the tail, and remove records from the head. Also, a count of the
130 * number of records used for checking queue depth. In addition, a counter
131 * of records that we have allocated but are not yet in the queue, which is
132 * needed to estimate the total size of the combined set of records
133 * outstanding in the system.
135 struct kaudit_queue audit_q;
140 * Audit queue control settings (minimum free, low/high water marks, etc.)
142 struct au_qctrl audit_qctrl;
145 * Condition variable to signal to the worker that it has work to do: either
146 * new records are in the queue, or a log replacement is taking place.
148 struct cv audit_worker_cv;
151 * Condition variable to flag when crossing the low watermark, meaning that
152 * threads blocked due to hitting the high watermark can wake up and continue
155 struct cv audit_watermark_cv;
158 * Condition variable for auditing threads wait on when in fail-stop mode.
159 * Threads wait on this CV forever (and ever), never seeing the light of day
162 static struct cv audit_fail_cv;
165 * Kernel audit information. This will store the current audit address
166 * or host information that the kernel will use when it's generating
167 * audit records. This data is modified by the A_GET{SET}KAUDIT auditon(2)
170 static struct auditinfo_addr audit_kinfo;
171 static struct rwlock audit_kinfo_lock;
173 #define KINFO_LOCK_INIT() rw_init(&audit_kinfo_lock, \
175 #define KINFO_RLOCK() rw_rlock(&audit_kinfo_lock)
176 #define KINFO_WLOCK() rw_wlock(&audit_kinfo_lock)
177 #define KINFO_RUNLOCK() rw_runlock(&audit_kinfo_lock)
178 #define KINFO_WUNLOCK() rw_wunlock(&audit_kinfo_lock)
181 audit_set_kinfo(struct auditinfo_addr *ak)
184 KASSERT(ak->ai_termid.at_type == AU_IPv4 ||
185 ak->ai_termid.at_type == AU_IPv6,
186 ("audit_set_kinfo: invalid address type"));
194 audit_get_kinfo(struct auditinfo_addr *ak)
197 KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 ||
198 audit_kinfo.ai_termid.at_type == AU_IPv6,
199 ("audit_set_kinfo: invalid address type"));
207 * Construct an audit record for the passed thread.
210 audit_record_ctor(void *mem, int size, void *arg, int flags)
212 struct kaudit_record *ar;
217 KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
221 bzero(ar, sizeof(*ar));
222 ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
223 nanotime(&ar->k_ar.ar_starttime);
226 * Export the subject credential.
229 cru2x(cred, &ar->k_ar.ar_subj_cred);
230 ar->k_ar.ar_subj_ruid = cred->cr_ruid;
231 ar->k_ar.ar_subj_rgid = cred->cr_rgid;
232 ar->k_ar.ar_subj_egid = cred->cr_groups[0];
233 ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
234 ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
235 ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
236 ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
237 ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
239 * If this process is jailed, make sure we capture the name of the
240 * jail so we can use it to generate a zonename token when we covert
241 * this record to BSM.
244 pr = cred->cr_prison;
245 (void) strlcpy(ar->k_ar.ar_jailname, pr->pr_name,
246 sizeof(ar->k_ar.ar_jailname));
248 ar->k_ar.ar_jailname[0] = '\0';
253 audit_record_dtor(void *mem, int size, void *arg)
255 struct kaudit_record *ar;
257 KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
260 if (ar->k_ar.ar_arg_upath1 != NULL)
261 free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
262 if (ar->k_ar.ar_arg_upath2 != NULL)
263 free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
264 if (ar->k_ar.ar_arg_text != NULL)
265 free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
266 if (ar->k_udata != NULL)
267 free(ar->k_udata, M_AUDITDATA);
268 if (ar->k_ar.ar_arg_argv != NULL)
269 free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
270 if (ar->k_ar.ar_arg_envv != NULL)
271 free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
272 if (ar->k_ar.ar_arg_groups.gidset != NULL)
273 free(ar->k_ar.ar_arg_groups.gidset, M_AUDITGIDSET);
277 * Initialize the Audit subsystem: configuration state, work queue,
278 * synchronization primitives, worker thread, and trigger device node. Also
279 * call into the BSM assembly code to initialize it.
287 audit_panic_on_write_fail = 0;
289 audit_in_failure = 0;
293 audit_fstat.af_filesz = 0; /* '0' means unset, unbounded. */
294 audit_fstat.af_currsz = 0;
295 audit_nae_mask.am_success = 0;
296 audit_nae_mask.am_failure = 0;
298 TAILQ_INIT(&audit_q);
301 audit_qctrl.aq_hiwater = AQ_HIWATER;
302 audit_qctrl.aq_lowater = AQ_LOWATER;
303 audit_qctrl.aq_bufsz = AQ_BUFSZ;
304 audit_qctrl.aq_minfree = AU_FS_MINFREE;
306 audit_kinfo.ai_termid.at_type = AU_IPv4;
307 audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY;
309 mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
311 cv_init(&audit_worker_cv, "audit_worker_cv");
312 cv_init(&audit_watermark_cv, "audit_watermark_cv");
313 cv_init(&audit_fail_cv, "audit_fail_cv");
315 audit_record_zone = uma_zcreate("audit_record",
316 sizeof(struct kaudit_record), audit_record_ctor,
317 audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
319 /* Initialize the BSM audit subsystem. */
322 audit_trigger_init();
324 /* Register shutdown handler. */
325 EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
328 /* Start audit worker thread. */
332 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
335 * Drain the audit queue and close the log at shutdown. Note that this can
336 * be called both from the system shutdown path and also from audit
337 * configuration syscalls, so 'arg' and 'howto' are ignored.
339 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
340 * drain before returning, which could lead to lost records on shutdown.
343 audit_shutdown(void *arg, int howto)
346 audit_rotate_vnode(NULL, NULL);
350 * Return the current thread's audit record, if any.
352 struct kaudit_record *
356 return (curthread->td_ar);
360 * XXXAUDIT: There are a number of races present in the code below due to
361 * release and re-grab of the mutex. The code should be revised to become
362 * slightly less racy.
364 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
365 * pre_q space, suspending the system call until there is room?
367 struct kaudit_record *
368 audit_new(int event, struct thread *td)
370 struct kaudit_record *ar;
373 mtx_lock(&audit_mtx);
374 no_record = (audit_suspended || !audit_enabled);
375 mtx_unlock(&audit_mtx);
380 * Note: the number of outstanding uncommitted audit records is
381 * limited to the number of concurrent threads servicing system calls
384 ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
385 ar->k_ar.ar_event = event;
387 mtx_lock(&audit_mtx);
389 mtx_unlock(&audit_mtx);
395 audit_free(struct kaudit_record *ar)
398 uma_zfree(audit_record_zone, ar);
402 audit_commit(struct kaudit_record *ar, int error, int retval)
408 struct au_mask *aumask;
414 * Decide whether to commit the audit record by checking the error
415 * value from the system call and using the appropriate audit mask.
417 if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
418 aumask = &audit_nae_mask;
420 aumask = &ar->k_ar.ar_subj_amask;
423 sorf = AU_PRS_FAILURE;
425 sorf = AU_PRS_SUCCESS;
428 * syscalls.master sometimes contains a prototype event number, which
429 * we will transform into a more specific event number now that we
430 * have more complete information gathered during the system call.
432 switch(ar->k_ar.ar_event) {
434 ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
435 ar->k_ar.ar_arg_fflags, error);
438 case AUE_OPENAT_RWTC:
439 ar->k_ar.ar_event = audit_flags_and_error_to_openatevent(
440 ar->k_ar.ar_arg_fflags, error);
444 ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
445 ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
449 /* Convert the auditon() command to an event. */
450 ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
454 auid = ar->k_ar.ar_subj_auid;
455 event = ar->k_ar.ar_event;
456 class = au_event_class(event);
458 ar->k_ar_commit |= AR_COMMIT_KERNEL;
459 if (au_preselect(event, class, aumask, sorf) != 0)
460 ar->k_ar_commit |= AR_PRESELECT_TRAIL;
461 if (audit_pipe_preselect(auid, event, class, sorf,
462 ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
463 ar->k_ar_commit |= AR_PRESELECT_PIPE;
464 if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
465 AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE)) == 0) {
466 mtx_lock(&audit_mtx);
468 mtx_unlock(&audit_mtx);
473 ar->k_ar.ar_errno = error;
474 ar->k_ar.ar_retval = retval;
475 nanotime(&ar->k_ar.ar_endtime);
478 * Note: it could be that some records initiated while audit was
479 * enabled should still be committed?
481 mtx_lock(&audit_mtx);
482 if (audit_suspended || !audit_enabled) {
484 mtx_unlock(&audit_mtx);
490 * Constrain the number of committed audit records based on the
491 * configurable parameter.
493 while (audit_q_len >= audit_qctrl.aq_hiwater)
494 cv_wait(&audit_watermark_cv, &audit_mtx);
496 TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
499 cv_signal(&audit_worker_cv);
500 mtx_unlock(&audit_mtx);
504 * audit_syscall_enter() is called on entry to each system call. It is
505 * responsible for deciding whether or not to audit the call (preselection),
506 * and if so, allocating a per-thread audit record. audit_new() will fill in
507 * basic thread/credential properties.
510 audit_syscall_enter(unsigned short code, struct thread *td)
512 struct au_mask *aumask;
517 KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
518 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
519 ("audit_syscall_enter: TDP_AUDITREC set"));
522 * In FreeBSD, each ABI has its own system call table, and hence
523 * mapping of system call codes to audit events. Convert the code to
524 * an audit event identifier using the process system call table
525 * reference. In Darwin, there's only one, so we use the global
526 * symbol for the system call table. No audit record is generated
527 * for bad system calls, as no operation has been performed.
529 if (code >= td->td_proc->p_sysent->sv_size)
532 event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
533 if (event == AUE_NULL)
537 * Check which audit mask to use; either the kernel non-attributable
538 * event mask or the process audit mask.
540 auid = td->td_ucred->cr_audit.ai_auid;
541 if (auid == AU_DEFAUDITID)
542 aumask = &audit_nae_mask;
544 aumask = &td->td_ucred->cr_audit.ai_mask;
547 * Allocate an audit record, if preselection allows it, and store in
548 * the thread for later use.
550 class = au_event_class(event);
551 if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
553 * If we're out of space and need to suspend unprivileged
554 * processes, do that here rather than trying to allocate
555 * another audit record.
557 * Note: we might wish to be able to continue here in the
558 * future, if the system recovers. That should be possible
559 * by means of checking the condition in a loop around
560 * cv_wait(). It might be desirable to reevaluate whether an
561 * audit record is still required for this event by
562 * re-calling au_preselect().
564 if (audit_in_failure &&
565 priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) {
566 cv_wait(&audit_fail_cv, &audit_mtx);
567 panic("audit_failing_stop: thread continued");
569 td->td_ar = audit_new(event, td);
570 if (td->td_ar != NULL)
571 td->td_pflags |= TDP_AUDITREC;
572 } else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
573 td->td_ar = audit_new(event, td);
574 if (td->td_ar != NULL)
575 td->td_pflags |= TDP_AUDITREC;
581 * audit_syscall_exit() is called from the return of every system call, or in
582 * the event of exit1(), during the execution of exit1(). It is responsible
583 * for committing the audit record, if any, along with return condition.
586 audit_syscall_exit(int error, struct thread *td)
591 * Commit the audit record as desired; once we pass the record into
592 * audit_commit(), the memory is owned by the audit subsystem. The
593 * return value from the system call is stored on the user thread.
594 * If there was an error, the return value is set to -1, imitating
595 * the behavior of the cerror routine.
600 retval = td->td_retval[0];
602 audit_commit(td->td_ar, error, retval);
604 td->td_pflags &= ~TDP_AUDITREC;
608 audit_cred_copy(struct ucred *src, struct ucred *dest)
611 bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
615 audit_cred_destroy(struct ucred *cred)
621 audit_cred_init(struct ucred *cred)
624 bzero(&cred->cr_audit, sizeof(cred->cr_audit));
628 * Initialize audit information for the first kernel process (proc 0) and for
629 * the first user process (init).
632 audit_cred_kproc0(struct ucred *cred)
635 cred->cr_audit.ai_auid = AU_DEFAUDITID;
636 cred->cr_audit.ai_termid.at_type = AU_IPv4;
640 audit_cred_proc1(struct ucred *cred)
643 cred->cr_audit.ai_auid = AU_DEFAUDITID;
644 cred->cr_audit.ai_termid.at_type = AU_IPv4;
648 audit_thread_alloc(struct thread *td)
655 audit_thread_free(struct thread *td)
658 KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
659 KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
660 ("audit_thread_free: TDP_AUDITREC set"));
664 audit_proc_coredump(struct thread *td, char *path, int errcode)
666 struct kaudit_record *ar;
667 struct au_mask *aumask;
677 * Make sure we are using the correct preselection mask.
680 auid = cred->cr_audit.ai_auid;
681 if (auid == AU_DEFAUDITID)
682 aumask = &audit_nae_mask;
684 aumask = &cred->cr_audit.ai_mask;
686 * It's possible for coredump(9) generation to fail. Make sure that
687 * we handle this case correctly for preselection.
690 sorf = AU_PRS_FAILURE;
692 sorf = AU_PRS_SUCCESS;
693 class = au_event_class(AUE_CORE);
694 if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
695 audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
699 * If we are interested in seeing this audit record, allocate it.
700 * Where possible coredump records should contain a pathname and arg32
703 ar = audit_new(AUE_CORE, td);
707 pathp = &ar->k_ar.ar_arg_upath1;
708 *pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
709 audit_canon_path(td, AT_FDCWD, path, *pathp);
710 ARG_SET_VALID(ar, ARG_UPATH1);
712 ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
713 ARG_SET_VALID(ar, ARG_SIGNUM);
716 audit_commit(ar, errcode, ret);