2 * Copyright (c) 1999-2008 Apple Inc.
3 * Copyright (c) 2006-2008, 2016 Robert N. M. Watson
6 * Portions of this software were developed by BAE Systems, the University of
7 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
8 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
9 * Computing (TC) research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of Apple Inc. ("Apple") nor the names of
20 * its contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR
27 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 * POSSIBILITY OF SUCH DAMAGE.
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/condvar.h>
43 #include <sys/filedesc.h>
44 #include <sys/fcntl.h>
46 #include <sys/kernel.h>
47 #include <sys/kthread.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
52 #include <sys/queue.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/domain.h>
58 #include <sys/sysproto.h>
59 #include <sys/sysent.h>
60 #include <sys/systm.h>
61 #include <sys/ucred.h>
64 #include <sys/unistd.h>
65 #include <sys/vnode.h>
67 #include <bsm/audit.h>
68 #include <bsm/audit_internal.h>
69 #include <bsm/audit_kevents.h>
71 #include <netinet/in.h>
72 #include <netinet/in_pcb.h>
74 #include <security/audit/audit.h>
75 #include <security/audit/audit_private.h>
79 #include <machine/stdarg.h>
82 * Worker thread that will schedule disk I/O, etc.
84 static struct proc *audit_thread;
87 * audit_cred and audit_vp are the stored credential and vnode to use for
88 * active audit trail. They are protected by the audit worker lock, which
89 * will be held across all I/O and all rotation to prevent them from being
90 * replaced (rotated) while in use. The audit_file_rotate_wait flag is set
91 * when the kernel has delivered a trigger to auditd to rotate the trail, and
92 * is cleared when the next rotation takes place. It is also protected by
93 * the audit worker lock.
95 static int audit_file_rotate_wait;
96 static struct ucred *audit_cred;
97 static struct vnode *audit_vp;
98 static off_t audit_size;
99 static struct sx audit_worker_lock;
101 #define AUDIT_WORKER_LOCK_INIT() sx_init(&audit_worker_lock, \
102 "audit_worker_lock");
103 #define AUDIT_WORKER_LOCK_ASSERT() sx_assert(&audit_worker_lock, \
105 #define AUDIT_WORKER_LOCK() sx_xlock(&audit_worker_lock)
106 #define AUDIT_WORKER_UNLOCK() sx_xunlock(&audit_worker_lock)
109 audit_worker_sync_vp(struct vnode *vp, struct mount *mp, const char *fmt, ...)
116 error = vn_start_write(vp, &mp1, 0);
118 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
119 (void)VOP_FSYNC(vp, MNT_WAIT, curthread);
121 vn_finished_write(mp1);
129 * Write an audit record to a file, performed as the last stage after both
130 * preselection and BSM conversion. Both space management and write failures
131 * are handled in this function.
133 * No attempt is made to deal with possible failure to deliver a trigger to
134 * the audit daemon, since the message is asynchronous anyway.
137 audit_record_write(struct vnode *vp, struct ucred *cred, void *data,
140 static struct timeval last_lowspace_trigger;
141 static struct timeval last_fail;
142 static int cur_lowspace_trigger;
143 struct statfs *mnt_stat;
149 AUDIT_WORKER_LOCK_ASSERT();
159 error = vfs_busy(mp, 0);
164 mnt_stat = &mp->mnt_stat;
167 * First, gather statistics on the audit log file and file system so
168 * that we know how we're doing on space. Consider failure of these
169 * operations to indicate a future inability to write to the file.
171 error = VFS_STATFS(mp, mnt_stat);
176 * We handle four different space-related limits:
178 * - A fixed (hard) limit on the minimum free blocks we require on
179 * the file system, and results in record loss, a trigger, and
180 * possible fail stop due to violating invariants.
182 * - An administrative (soft) limit, which when fallen below, results
183 * in the kernel notifying the audit daemon of low space.
185 * - An audit trail size limit, which when gone above, results in the
186 * kernel notifying the audit daemon that rotation is desired.
188 * - The total depth of the kernel audit record exceeding free space,
189 * which can lead to possible fail stop (with drain), in order to
190 * prevent violating invariants. Failure here doesn't halt
191 * immediately, but prevents new records from being generated.
193 * Possibly, the last of these should be handled differently, always
194 * allowing a full queue to be lost, rather than trying to prevent
197 * First, handle the hard limit, which generates a trigger and may
198 * fail stop. This is handled in the same manner as ENOSPC from
199 * VOP_WRITE, and results in record loss.
201 if (mnt_stat->f_bfree < AUDIT_HARD_LIMIT_FREE_BLOCKS) {
207 * Second, handle falling below the soft limit, if defined; we send
208 * the daemon a trigger and continue processing the record. Triggers
209 * are limited to 1/sec.
211 if (audit_qctrl.aq_minfree != 0) {
212 temp = mnt_stat->f_blocks / (100 / audit_qctrl.aq_minfree);
213 if (mnt_stat->f_bfree < temp) {
214 if (ppsratecheck(&last_lowspace_trigger,
215 &cur_lowspace_trigger, 1)) {
216 (void)audit_send_trigger(
217 AUDIT_TRIGGER_LOW_SPACE);
218 printf("Warning: disk space low (< %d%% free) "
219 "on audit log file-system\n",
220 audit_qctrl.aq_minfree);
226 * If the current file is getting full, generate a rotation trigger
227 * to the daemon. This is only approximate, which is fine as more
228 * records may be generated before the daemon rotates the file.
230 if (audit_fstat.af_filesz != 0 &&
231 audit_size >= audit_fstat.af_filesz * (audit_file_rotate_wait + 1)) {
232 AUDIT_WORKER_LOCK_ASSERT();
234 audit_file_rotate_wait++;
235 (void)audit_send_trigger(AUDIT_TRIGGER_ROTATE_KERNEL);
239 * If the estimated amount of audit data in the audit event queue
240 * (plus records allocated but not yet queued) has reached the amount
241 * of free space on the disk, then we need to go into an audit fail
242 * stop state, in which we do not permit the allocation/committing of
243 * any new audit records. We continue to process records but don't
244 * allow any activities that might generate new records. In the
245 * future, we might want to detect when space is available again and
246 * allow operation to continue, but this behavior is sufficient to
247 * meet fail stop requirements in CAPP.
249 if (audit_fail_stop) {
250 if ((unsigned long)((audit_q_len + audit_pre_q_len + 1) *
251 MAX_AUDIT_RECORD_SIZE) / mnt_stat->f_bsize >=
252 (unsigned long)(mnt_stat->f_bfree)) {
253 if (ppsratecheck(&last_fail, &cur_fail, 1))
254 printf("audit_record_write: free space "
255 "below size of audit queue, failing "
257 audit_in_failure = 1;
258 } else if (audit_in_failure) {
260 * Note: if we want to handle recovery, this is the
261 * spot to do it: unset audit_in_failure, and issue a
267 error = vn_rdwr(UIO_WRITE, vp, data, len, (off_t)0, UIO_SYSSPACE,
268 IO_APPEND|IO_UNIT, cred, NULL, NULL, curthread);
273 AUDIT_WORKER_LOCK_ASSERT();
277 * Catch completion of a queue drain here; if we're draining and the
278 * queue is now empty, fail stop. That audit_fail_stop is implicitly
279 * true, since audit_in_failure can only be set of audit_fail_stop is
282 * Note: if we handle recovery from audit_in_failure, then we need to
283 * make panic here conditional.
285 if (audit_in_failure) {
286 if (audit_q_len == 0 && audit_pre_q_len == 0) {
287 audit_worker_sync_vp(vp, mp,
288 "Audit store overflow; record queue drained.");
297 * ENOSPC is considered a special case with respect to failures, as
298 * this can reflect either our preemptive detection of insufficient
299 * space, or ENOSPC returned by the vnode write call.
301 if (audit_fail_stop) {
302 audit_worker_sync_vp(vp, mp,
303 "Audit log space exhausted and fail-stop set.");
305 (void)audit_send_trigger(AUDIT_TRIGGER_NO_SPACE);
311 * We have failed to write to the file, so the current record is
312 * lost, which may require an immediate system halt.
314 if (audit_panic_on_write_fail) {
315 audit_worker_sync_vp(vp, mp,
316 "audit_worker: write error %d\n", error);
317 } else if (ppsratecheck(&last_fail, &cur_fail, 1))
318 printf("audit_worker: write error %d\n", error);
324 * Given a kernel audit record, process as required. Kernel audit records
325 * are converted to one, or possibly two, BSM records, depending on whether
326 * there is a user audit record present also. Kernel records need be
327 * converted to BSM before they can be written out. Both types will be
328 * written to disk, and audit pipes.
331 audit_worker_process_record(struct kaudit_record *ar)
333 struct au_record *bsm;
341 * We hold the audit worker lock over both writes, if there are two,
342 * so that the two records won't be split across a rotation and end
343 * up in two different trail files.
345 if (((ar->k_ar_commit & AR_COMMIT_USER) &&
346 (ar->k_ar_commit & AR_PRESELECT_USER_TRAIL)) ||
347 (ar->k_ar_commit & AR_PRESELECT_TRAIL)) {
354 * First, handle the user record, if any: commit to the system trail
355 * and audit pipes as selected.
357 if ((ar->k_ar_commit & AR_COMMIT_USER) &&
358 (ar->k_ar_commit & AR_PRESELECT_USER_TRAIL)) {
359 AUDIT_WORKER_LOCK_ASSERT();
360 audit_record_write(audit_vp, audit_cred, ar->k_udata,
364 if ((ar->k_ar_commit & AR_COMMIT_USER) &&
365 (ar->k_ar_commit & AR_PRESELECT_USER_PIPE))
366 audit_pipe_submit_user(ar->k_udata, ar->k_ulen);
368 if (!(ar->k_ar_commit & AR_COMMIT_KERNEL) ||
369 ((ar->k_ar_commit & AR_PRESELECT_PIPE) == 0 &&
370 (ar->k_ar_commit & AR_PRESELECT_TRAIL) == 0 &&
371 (ar->k_ar_commit & AR_PRESELECT_DTRACE) == 0))
374 auid = ar->k_ar.ar_subj_auid;
375 event = ar->k_ar.ar_event;
376 class = au_event_class(event);
377 if (ar->k_ar.ar_errno == 0)
378 sorf = AU_PRS_SUCCESS;
380 sorf = AU_PRS_FAILURE;
382 error = kaudit_to_bsm(ar, &bsm);
388 printf("audit_worker_process_record: BSM_FAILURE\n");
395 panic("kaudit_to_bsm returned %d", error);
398 if (ar->k_ar_commit & AR_PRESELECT_TRAIL) {
399 AUDIT_WORKER_LOCK_ASSERT();
400 audit_record_write(audit_vp, audit_cred, bsm->data, bsm->len);
403 if (ar->k_ar_commit & AR_PRESELECT_PIPE)
404 audit_pipe_submit(auid, event, class, sorf,
405 ar->k_ar_commit & AR_PRESELECT_TRAIL, bsm->data,
410 * Version of the dtaudit commit hook that accepts BSM.
412 if (ar->k_ar_commit & AR_PRESELECT_DTRACE) {
413 if (dtaudit_hook_bsm != NULL)
414 dtaudit_hook_bsm(ar, auid, event, class, sorf,
415 bsm->data, bsm->len);
422 AUDIT_WORKER_UNLOCK();
426 * The audit_worker thread is responsible for watching the event queue,
427 * dequeueing records, converting them to BSM format, and committing them to
428 * disk. In order to minimize lock thrashing, records are dequeued in sets
429 * to a thread-local work queue.
431 * Note: this means that the effect bound on the size of the pending record
432 * queue is 2x the length of the global queue.
435 audit_worker(void *arg)
437 struct kaudit_queue ar_worklist;
438 struct kaudit_record *ar;
441 TAILQ_INIT(&ar_worklist);
442 mtx_lock(&audit_mtx);
444 mtx_assert(&audit_mtx, MA_OWNED);
449 while (TAILQ_EMPTY(&audit_q))
450 cv_wait(&audit_worker_cv, &audit_mtx);
453 * If there are records in the global audit record queue,
454 * transfer them to a thread-local queue and process them
455 * one by one. If we cross the low watermark threshold,
456 * signal any waiting processes that they may wake up and
457 * continue generating records.
460 while ((ar = TAILQ_FIRST(&audit_q))) {
461 TAILQ_REMOVE(&audit_q, ar, k_q);
463 if (audit_q_len == audit_qctrl.aq_lowater)
465 TAILQ_INSERT_TAIL(&ar_worklist, ar, k_q);
468 cv_broadcast(&audit_watermark_cv);
470 mtx_unlock(&audit_mtx);
471 while ((ar = TAILQ_FIRST(&ar_worklist))) {
472 TAILQ_REMOVE(&ar_worklist, ar, k_q);
473 audit_worker_process_record(ar);
476 mtx_lock(&audit_mtx);
481 * audit_rotate_vnode() is called by a user or kernel thread to configure or
482 * de-configure auditing on a vnode. The arguments are the replacement
483 * credential (referenced) and vnode (referenced and opened) to substitute
484 * for the current credential and vnode, if any. If either is set to NULL,
485 * both should be NULL, and this is used to indicate that audit is being
486 * disabled. Any previous cred/vnode will be closed and freed. We re-enable
487 * generating rotation requests to auditd.
490 audit_rotate_vnode(struct ucred *cred, struct vnode *vp)
492 struct ucred *old_audit_cred;
493 struct vnode *old_audit_vp;
496 KASSERT((cred != NULL && vp != NULL) || (cred == NULL && vp == NULL),
497 ("audit_rotate_vnode: cred %p vp %p", cred, vp));
500 vn_lock(vp, LK_SHARED | LK_RETRY);
501 if (VOP_GETATTR(vp, &vattr, cred) != 0)
509 * Rotate the vnode/cred, and clear the rotate flag so that we will
510 * send a rotate trigger if the new file fills.
513 old_audit_cred = audit_cred;
514 old_audit_vp = audit_vp;
517 audit_size = vattr.va_size;
518 audit_file_rotate_wait = 0;
519 audit_enabled = (audit_vp != NULL);
520 AUDIT_WORKER_UNLOCK();
523 * If there was an old vnode/credential, close and free.
525 if (old_audit_vp != NULL) {
526 vn_close(old_audit_vp, AUDIT_CLOSE_FLAGS, old_audit_cred,
528 crfree(old_audit_cred);
533 audit_worker_init(void)
537 AUDIT_WORKER_LOCK_INIT();
538 error = kproc_create(audit_worker, NULL, &audit_thread, RFHIGHPID,
541 panic("audit_worker_init: kproc_create returned %d", error);