1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1999-2008 Apple Inc. 5 * Copyright (c) 2006-2008, 2016, 2018 Robert N. M. Watson 6 * All rights reserved. 7 * 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. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 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. 24 * 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. 36 */ 37 38 #include <sys/param.h> 39 #include <sys/condvar.h> 40 #include <sys/conf.h> 41 #include <sys/file.h> 42 #include <sys/filedesc.h> 43 #include <sys/fcntl.h> 44 #include <sys/ipc.h> 45 #include <sys/kernel.h> 46 #include <sys/kthread.h> 47 #include <sys/malloc.h> 48 #include <sys/mount.h> 49 #include <sys/namei.h> 50 #include <sys/proc.h> 51 #include <sys/queue.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/protosw.h> 55 #include <sys/domain.h> 56 #include <sys/sx.h> 57 #include <sys/sysproto.h> 58 #include <sys/sysent.h> 59 #include <sys/systm.h> 60 #include <sys/ucred.h> 61 #include <sys/uio.h> 62 #include <sys/un.h> 63 #include <sys/unistd.h> 64 #include <sys/vnode.h> 65 66 #include <bsm/audit.h> 67 #include <bsm/audit_internal.h> 68 #include <bsm/audit_kevents.h> 69 70 #include <netinet/in.h> 71 #include <netinet/in_pcb.h> 72 73 #include <security/audit/audit.h> 74 #include <security/audit/audit_private.h> 75 76 #include <vm/uma.h> 77 78 #include <machine/stdarg.h> 79 80 /* 81 * Worker thread that will schedule disk I/O, etc. 82 */ 83 static struct proc *audit_thread; 84 85 /* 86 * audit_cred and audit_vp are the stored credential and vnode to use for 87 * active audit trail. They are protected by the audit worker lock, which 88 * will be held across all I/O and all rotation to prevent them from being 89 * replaced (rotated) while in use. The audit_file_rotate_wait flag is set 90 * when the kernel has delivered a trigger to auditd to rotate the trail, and 91 * is cleared when the next rotation takes place. It is also protected by 92 * the audit worker lock. 93 */ 94 static int audit_file_rotate_wait; 95 static struct ucred *audit_cred; 96 static struct vnode *audit_vp; 97 static off_t audit_size; 98 static struct sx audit_worker_lock; 99 100 #define AUDIT_WORKER_LOCK_INIT() sx_init(&audit_worker_lock, \ 101 "audit_worker_lock"); 102 #define AUDIT_WORKER_LOCK_ASSERT() sx_assert(&audit_worker_lock, \ 103 SA_XLOCKED) 104 #define AUDIT_WORKER_LOCK() sx_xlock(&audit_worker_lock) 105 #define AUDIT_WORKER_UNLOCK() sx_xunlock(&audit_worker_lock) 106 107 static void 108 audit_worker_sync_vp(struct vnode *vp, struct mount *mp, const char *fmt, ...) 109 { 110 struct mount *mp1; 111 int error; 112 va_list va; 113 114 va_start(va, fmt); 115 error = vn_start_write(vp, &mp1, 0); 116 if (error == 0) { 117 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY); 118 (void)VOP_FSYNC(vp, MNT_WAIT, curthread); 119 VOP_UNLOCK(vp); 120 vn_finished_write(mp1); 121 } 122 vfs_unbusy(mp); 123 vpanic(fmt, va); 124 va_end(va); 125 } 126 127 /* 128 * Write an audit record to a file, performed as the last stage after both 129 * preselection and BSM conversion. Both space management and write failures 130 * are handled in this function. 131 * 132 * No attempt is made to deal with possible failure to deliver a trigger to 133 * the audit daemon, since the message is asynchronous anyway. 134 */ 135 static void 136 audit_record_write(struct vnode *vp, struct ucred *cred, void *data, 137 size_t len) 138 { 139 static struct timeval last_lowspace_trigger; 140 static struct timeval last_fail; 141 static int cur_lowspace_trigger; 142 struct statfs *mnt_stat; 143 struct mount *mp; 144 int error; 145 static int cur_fail; 146 long temp; 147 148 AUDIT_WORKER_LOCK_ASSERT(); 149 150 if (vp == NULL) 151 return; 152 153 mp = vp->v_mount; 154 if (mp == NULL) { 155 error = EINVAL; 156 goto fail; 157 } 158 error = vfs_busy(mp, 0); 159 if (error != 0) { 160 mp = NULL; 161 goto fail; 162 } 163 mnt_stat = &mp->mnt_stat; 164 165 /* 166 * First, gather statistics on the audit log file and file system so 167 * that we know how we're doing on space. Consider failure of these 168 * operations to indicate a future inability to write to the file. 169 */ 170 error = VFS_STATFS(mp, mnt_stat); 171 if (error != 0) 172 goto fail; 173 174 /* 175 * We handle four different space-related limits: 176 * 177 * - A fixed (hard) limit on the minimum free blocks we require on 178 * the file system, and results in record loss, a trigger, and 179 * possible fail stop due to violating invariants. 180 * 181 * - An administrative (soft) limit, which when fallen below, results 182 * in the kernel notifying the audit daemon of low space. 183 * 184 * - An audit trail size limit, which when gone above, results in the 185 * kernel notifying the audit daemon that rotation is desired. 186 * 187 * - The total depth of the kernel audit record exceeding free space, 188 * which can lead to possible fail stop (with drain), in order to 189 * prevent violating invariants. Failure here doesn't halt 190 * immediately, but prevents new records from being generated. 191 * 192 * Possibly, the last of these should be handled differently, always 193 * allowing a full queue to be lost, rather than trying to prevent 194 * loss. 195 * 196 * First, handle the hard limit, which generates a trigger and may 197 * fail stop. This is handled in the same manner as ENOSPC from 198 * VOP_WRITE, and results in record loss. 199 */ 200 if (mnt_stat->f_bfree < AUDIT_HARD_LIMIT_FREE_BLOCKS) { 201 error = ENOSPC; 202 goto fail_enospc; 203 } 204 205 /* 206 * Second, handle falling below the soft limit, if defined; we send 207 * the daemon a trigger and continue processing the record. Triggers 208 * are limited to 1/sec. 209 */ 210 if (audit_qctrl.aq_minfree != 0) { 211 temp = mnt_stat->f_blocks / (100 / audit_qctrl.aq_minfree); 212 if (mnt_stat->f_bfree < temp) { 213 if (ppsratecheck(&last_lowspace_trigger, 214 &cur_lowspace_trigger, 1)) { 215 (void)audit_send_trigger( 216 AUDIT_TRIGGER_LOW_SPACE); 217 printf("Warning: disk space low (< %d%% free) " 218 "on audit log file-system\n", 219 audit_qctrl.aq_minfree); 220 } 221 } 222 } 223 224 /* 225 * If the current file is getting full, generate a rotation trigger 226 * to the daemon. This is only approximate, which is fine as more 227 * records may be generated before the daemon rotates the file. 228 */ 229 if (audit_fstat.af_filesz != 0 && 230 audit_size >= audit_fstat.af_filesz * (audit_file_rotate_wait + 1)) { 231 AUDIT_WORKER_LOCK_ASSERT(); 232 233 audit_file_rotate_wait++; 234 (void)audit_send_trigger(AUDIT_TRIGGER_ROTATE_KERNEL); 235 } 236 237 /* 238 * If the estimated amount of audit data in the audit event queue 239 * (plus records allocated but not yet queued) has reached the amount 240 * of free space on the disk, then we need to go into an audit fail 241 * stop state, in which we do not permit the allocation/committing of 242 * any new audit records. We continue to process records but don't 243 * allow any activities that might generate new records. In the 244 * future, we might want to detect when space is available again and 245 * allow operation to continue, but this behavior is sufficient to 246 * meet fail stop requirements in CAPP. 247 */ 248 if (audit_fail_stop) { 249 if ((unsigned long)((audit_q_len + audit_pre_q_len + 1) * 250 MAX_AUDIT_RECORD_SIZE) / mnt_stat->f_bsize >= 251 (unsigned long)(mnt_stat->f_bfree)) { 252 if (ppsratecheck(&last_fail, &cur_fail, 1)) 253 printf("audit_record_write: free space " 254 "below size of audit queue, failing " 255 "stop\n"); 256 audit_in_failure = 1; 257 } else if (audit_in_failure) { 258 /* 259 * Note: if we want to handle recovery, this is the 260 * spot to do it: unset audit_in_failure, and issue a 261 * wakeup on the cv. 262 */ 263 } 264 } 265 266 error = vn_rdwr(UIO_WRITE, vp, data, len, (off_t)0, UIO_SYSSPACE, 267 IO_APPEND|IO_UNIT, cred, NULL, NULL, curthread); 268 if (error == ENOSPC) 269 goto fail_enospc; 270 else if (error) 271 goto fail; 272 AUDIT_WORKER_LOCK_ASSERT(); 273 audit_size += len; 274 275 /* 276 * Catch completion of a queue drain here; if we're draining and the 277 * queue is now empty, fail stop. That audit_fail_stop is implicitly 278 * true, since audit_in_failure can only be set of audit_fail_stop is 279 * set. 280 * 281 * Note: if we handle recovery from audit_in_failure, then we need to 282 * make panic here conditional. 283 */ 284 if (audit_in_failure) { 285 if (audit_q_len == 0 && audit_pre_q_len == 0) { 286 audit_worker_sync_vp(vp, mp, 287 "Audit store overflow; record queue drained."); 288 } 289 } 290 291 vfs_unbusy(mp); 292 return; 293 294 fail_enospc: 295 /* 296 * ENOSPC is considered a special case with respect to failures, as 297 * this can reflect either our preemptive detection of insufficient 298 * space, or ENOSPC returned by the vnode write call. 299 */ 300 if (audit_fail_stop) { 301 audit_worker_sync_vp(vp, mp, 302 "Audit log space exhausted and fail-stop set."); 303 } 304 (void)audit_send_trigger(AUDIT_TRIGGER_NO_SPACE); 305 audit_trail_suspended = 1; 306 audit_syscalls_enabled_update(); 307 308 /* FALLTHROUGH */ 309 fail: 310 /* 311 * We have failed to write to the file, so the current record is 312 * lost, which may require an immediate system halt. 313 */ 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); 319 if (mp != NULL) 320 vfs_unbusy(mp); 321 } 322 323 /* 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. 329 */ 330 static void 331 audit_worker_process_record(struct kaudit_record *ar) 332 { 333 struct au_record *bsm; 334 au_class_t class; 335 au_event_t event; 336 au_id_t auid; 337 int error, sorf; 338 int locked; 339 340 /* 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. 344 */ 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)) { 348 AUDIT_WORKER_LOCK(); 349 locked = 1; 350 } else 351 locked = 0; 352 353 /* 354 * First, handle the user record, if any: commit to the system trail 355 * and audit pipes as selected. 356 */ 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, 361 ar->k_ulen); 362 } 363 364 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); 367 368 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)) 372 goto out; 373 374 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; 379 else 380 sorf = AU_PRS_FAILURE; 381 382 error = kaudit_to_bsm(ar, &bsm); 383 switch (error) { 384 case BSM_NOAUDIT: 385 goto out; 386 387 case BSM_FAILURE: 388 printf("audit_worker_process_record: BSM_FAILURE\n"); 389 goto out; 390 391 case BSM_SUCCESS: 392 break; 393 394 default: 395 panic("kaudit_to_bsm returned %d", error); 396 } 397 398 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); 401 } 402 403 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, 406 bsm->len); 407 408 #ifdef KDTRACE_HOOKS 409 /* 410 * Version of the dtaudit commit hook that accepts BSM. 411 */ 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); 416 } 417 #endif 418 419 kau_free(bsm); 420 out: 421 if (locked) 422 AUDIT_WORKER_UNLOCK(); 423 } 424 425 /* 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. 430 * 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. 433 */ 434 static void 435 audit_worker(void *arg) 436 { 437 struct kaudit_queue ar_worklist; 438 struct kaudit_record *ar; 439 int lowater_signal; 440 441 TAILQ_INIT(&ar_worklist); 442 mtx_lock(&audit_mtx); 443 while (1) { 444 mtx_assert(&audit_mtx, MA_OWNED); 445 446 /* 447 * Wait for a record. 448 */ 449 while (TAILQ_EMPTY(&audit_q)) 450 cv_wait(&audit_worker_cv, &audit_mtx); 451 452 /* 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. 458 */ 459 lowater_signal = 0; 460 while ((ar = TAILQ_FIRST(&audit_q))) { 461 TAILQ_REMOVE(&audit_q, ar, k_q); 462 audit_q_len--; 463 if (audit_q_len == audit_qctrl.aq_lowater) 464 lowater_signal++; 465 TAILQ_INSERT_TAIL(&ar_worklist, ar, k_q); 466 } 467 if (lowater_signal) 468 cv_broadcast(&audit_watermark_cv); 469 470 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); 474 audit_free(ar); 475 } 476 mtx_lock(&audit_mtx); 477 } 478 } 479 480 /* 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. 488 */ 489 void 490 audit_rotate_vnode(struct ucred *cred, struct vnode *vp) 491 { 492 struct ucred *old_audit_cred; 493 struct vnode *old_audit_vp; 494 struct vattr vattr; 495 496 KASSERT((cred != NULL && vp != NULL) || (cred == NULL && vp == NULL), 497 ("audit_rotate_vnode: cred %p vp %p", cred, vp)); 498 499 if (vp != NULL) { 500 vn_lock(vp, LK_SHARED | LK_RETRY); 501 if (VOP_GETATTR(vp, &vattr, cred) != 0) 502 vattr.va_size = 0; 503 VOP_UNLOCK(vp); 504 } else { 505 vattr.va_size = 0; 506 } 507 508 /* 509 * Rotate the vnode/cred, and clear the rotate flag so that we will 510 * send a rotate trigger if the new file fills. 511 */ 512 AUDIT_WORKER_LOCK(); 513 old_audit_cred = audit_cred; 514 old_audit_vp = audit_vp; 515 audit_cred = cred; 516 audit_vp = vp; 517 audit_size = vattr.va_size; 518 audit_file_rotate_wait = 0; 519 audit_trail_enabled = (audit_vp != NULL); 520 audit_syscalls_enabled_update(); 521 AUDIT_WORKER_UNLOCK(); 522 523 /* 524 * If there was an old vnode/credential, close and free. 525 */ 526 if (old_audit_vp != NULL) { 527 vn_close(old_audit_vp, AUDIT_CLOSE_FLAGS, old_audit_cred, 528 curthread); 529 crfree(old_audit_cred); 530 } 531 } 532 533 void 534 audit_worker_init(void) 535 { 536 int error; 537 538 AUDIT_WORKER_LOCK_INIT(); 539 error = kproc_create(audit_worker, NULL, &audit_thread, RFHIGHPID, 540 0, "audit"); 541 if (error) 542 panic("audit_worker_init: kproc_create returned %d", error); 543 } 544