1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1986, 1988, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ddb.h" 43 #include "opt_ekcd.h" 44 #include "opt_kdb.h" 45 #include "opt_panic.h" 46 #include "opt_printf.h" 47 #include "opt_sched.h" 48 #include "opt_watchdog.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/bio.h> 53 #include <sys/boottrace.h> 54 #include <sys/buf.h> 55 #include <sys/conf.h> 56 #include <sys/compressor.h> 57 #include <sys/cons.h> 58 #include <sys/disk.h> 59 #include <sys/eventhandler.h> 60 #include <sys/filedesc.h> 61 #include <sys/jail.h> 62 #include <sys/kdb.h> 63 #include <sys/kernel.h> 64 #include <sys/kerneldump.h> 65 #include <sys/kthread.h> 66 #include <sys/ktr.h> 67 #include <sys/malloc.h> 68 #include <sys/mbuf.h> 69 #include <sys/mount.h> 70 #include <sys/priv.h> 71 #include <sys/proc.h> 72 #include <sys/reboot.h> 73 #include <sys/resourcevar.h> 74 #include <sys/rwlock.h> 75 #include <sys/sbuf.h> 76 #include <sys/sched.h> 77 #include <sys/smp.h> 78 #include <sys/sysctl.h> 79 #include <sys/sysproto.h> 80 #include <sys/taskqueue.h> 81 #include <sys/vnode.h> 82 #include <sys/watchdog.h> 83 84 #include <crypto/chacha20/chacha.h> 85 #include <crypto/rijndael/rijndael-api-fst.h> 86 #include <crypto/sha2/sha256.h> 87 88 #include <ddb/ddb.h> 89 90 #include <machine/cpu.h> 91 #include <machine/dump.h> 92 #include <machine/pcb.h> 93 #include <machine/smp.h> 94 95 #include <security/mac/mac_framework.h> 96 97 #include <vm/vm.h> 98 #include <vm/vm_object.h> 99 #include <vm/vm_page.h> 100 #include <vm/vm_pager.h> 101 #include <vm/swap_pager.h> 102 103 #include <sys/signalvar.h> 104 105 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer"); 106 107 #ifndef PANIC_REBOOT_WAIT_TIME 108 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ 109 #endif 110 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME; 111 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN, 112 &panic_reboot_wait_time, 0, 113 "Seconds to wait before rebooting after a panic"); 114 static int reboot_wait_time = 0; 115 SYSCTL_INT(_kern, OID_AUTO, reboot_wait_time, CTLFLAG_RWTUN, 116 &reboot_wait_time, 0, 117 "Seconds to wait before rebooting"); 118 119 /* 120 * Note that stdarg.h and the ANSI style va_start macro is used for both 121 * ANSI and traditional C compilers. 122 */ 123 #include <machine/stdarg.h> 124 125 #ifdef KDB 126 #ifdef KDB_UNATTENDED 127 int debugger_on_panic = 0; 128 #else 129 int debugger_on_panic = 1; 130 #endif 131 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, 132 CTLFLAG_RWTUN | CTLFLAG_SECURE, 133 &debugger_on_panic, 0, "Run debugger on kernel panic"); 134 135 static bool debugger_on_recursive_panic = false; 136 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic, 137 CTLFLAG_RWTUN | CTLFLAG_SECURE, 138 &debugger_on_recursive_panic, 0, "Run debugger on recursive kernel panic"); 139 140 int debugger_on_trap = 0; 141 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap, 142 CTLFLAG_RWTUN | CTLFLAG_SECURE, 143 &debugger_on_trap, 0, "Run debugger on kernel trap before panic"); 144 145 #ifdef KDB_TRACE 146 static int trace_on_panic = 1; 147 static bool trace_all_panics = true; 148 #else 149 static int trace_on_panic = 0; 150 static bool trace_all_panics = false; 151 #endif 152 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, 153 CTLFLAG_RWTUN | CTLFLAG_SECURE, 154 &trace_on_panic, 0, "Print stack trace on kernel panic"); 155 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN, 156 &trace_all_panics, 0, "Print stack traces on secondary kernel panics"); 157 #endif /* KDB */ 158 159 static int sync_on_panic = 0; 160 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN, 161 &sync_on_panic, 0, "Do a sync before rebooting from a panic"); 162 163 static bool poweroff_on_panic = 0; 164 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN, 165 &poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic"); 166 167 static bool powercycle_on_panic = 0; 168 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN, 169 &powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic"); 170 171 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 172 "Shutdown environment"); 173 174 #ifndef DIAGNOSTIC 175 static int show_busybufs; 176 #else 177 static int show_busybufs = 1; 178 #endif 179 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW, 180 &show_busybufs, 0, 181 "Show busy buffers during shutdown"); 182 183 int suspend_blocked = 0; 184 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW, 185 &suspend_blocked, 0, "Block suspend due to a pending shutdown"); 186 187 #ifdef EKCD 188 FEATURE(ekcd, "Encrypted kernel crash dumps support"); 189 190 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data"); 191 192 struct kerneldumpcrypto { 193 uint8_t kdc_encryption; 194 uint8_t kdc_iv[KERNELDUMP_IV_MAX_SIZE]; 195 union { 196 struct { 197 keyInstance aes_ki; 198 cipherInstance aes_ci; 199 } u_aes; 200 struct chacha_ctx u_chacha; 201 } u; 202 #define kdc_ki u.u_aes.aes_ki 203 #define kdc_ci u.u_aes.aes_ci 204 #define kdc_chacha u.u_chacha 205 uint32_t kdc_dumpkeysize; 206 struct kerneldumpkey kdc_dumpkey[]; 207 }; 208 #endif 209 210 struct kerneldumpcomp { 211 uint8_t kdc_format; 212 struct compressor *kdc_stream; 213 uint8_t *kdc_buf; 214 size_t kdc_resid; 215 }; 216 217 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di, 218 uint8_t compression); 219 static void kerneldumpcomp_destroy(struct dumperinfo *di); 220 static int kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg); 221 222 static int kerneldump_gzlevel = 6; 223 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN, 224 &kerneldump_gzlevel, 0, 225 "Kernel crash dump compression level"); 226 227 /* 228 * Variable panicstr contains argument to first call to panic; used as flag 229 * to indicate that the kernel has already called panic. 230 */ 231 const char *panicstr; 232 bool __read_frequently panicked; 233 234 int __read_mostly dumping; /* system is dumping */ 235 int rebooting; /* system is rebooting */ 236 /* 237 * Used to serialize between sysctl kern.shutdown.dumpdevname and list 238 * modifications via ioctl. 239 */ 240 static struct mtx dumpconf_list_lk; 241 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF); 242 243 /* Our selected dumper(s). */ 244 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs = 245 TAILQ_HEAD_INITIALIZER(dumper_configs); 246 247 /* Context information for dump-debuggers. */ 248 static struct pcb dumppcb; /* Registers. */ 249 lwpid_t dumptid; /* Thread ID. */ 250 251 static struct cdevsw reroot_cdevsw = { 252 .d_version = D_VERSION, 253 .d_name = "reroot", 254 }; 255 256 static void poweroff_wait(void *, int); 257 static void shutdown_halt(void *junk, int howto); 258 static void shutdown_panic(void *junk, int howto); 259 static void shutdown_reset(void *junk, int howto); 260 static int kern_reroot(void); 261 262 /* register various local shutdown events */ 263 static void 264 shutdown_conf(void *unused) 265 { 266 267 EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, 268 SHUTDOWN_PRI_FIRST); 269 EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, 270 SHUTDOWN_PRI_LAST + 100); 271 EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, 272 SHUTDOWN_PRI_LAST + 100); 273 } 274 275 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL); 276 277 /* 278 * The only reason this exists is to create the /dev/reroot/ directory, 279 * used by reroot code in init(8) as a mountpoint for tmpfs. 280 */ 281 static void 282 reroot_conf(void *unused) 283 { 284 int error; 285 struct cdev *cdev; 286 287 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev, 288 &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot"); 289 if (error != 0) { 290 printf("%s: failed to create device node, error %d", 291 __func__, error); 292 } 293 } 294 295 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL); 296 297 /* 298 * The system call that results in a reboot. 299 */ 300 /* ARGSUSED */ 301 int 302 sys_reboot(struct thread *td, struct reboot_args *uap) 303 { 304 int error; 305 306 error = 0; 307 #ifdef MAC 308 error = mac_system_check_reboot(td->td_ucred, uap->opt); 309 #endif 310 if (error == 0) 311 error = priv_check(td, PRIV_REBOOT); 312 if (error == 0) { 313 if (uap->opt & RB_REROOT) 314 error = kern_reroot(); 315 else 316 kern_reboot(uap->opt); 317 } 318 return (error); 319 } 320 321 static void 322 shutdown_nice_task_fn(void *arg, int pending __unused) 323 { 324 int howto; 325 326 howto = (uintptr_t)arg; 327 /* Send a signal to init(8) and have it shutdown the world. */ 328 PROC_LOCK(initproc); 329 if ((howto & RB_POWEROFF) != 0) { 330 BOOTTRACE("SIGUSR2 to init(8)"); 331 kern_psignal(initproc, SIGUSR2); 332 } else if ((howto & RB_POWERCYCLE) != 0) { 333 BOOTTRACE("SIGWINCH to init(8)"); 334 kern_psignal(initproc, SIGWINCH); 335 } else if ((howto & RB_HALT) != 0) { 336 BOOTTRACE("SIGUSR1 to init(8)"); 337 kern_psignal(initproc, SIGUSR1); 338 } else { 339 BOOTTRACE("SIGINT to init(8)"); 340 kern_psignal(initproc, SIGINT); 341 } 342 PROC_UNLOCK(initproc); 343 } 344 345 static struct task shutdown_nice_task = TASK_INITIALIZER(0, 346 &shutdown_nice_task_fn, NULL); 347 348 /* 349 * Called by events that want to shut down.. e.g <CTL><ALT><DEL> on a PC 350 */ 351 void 352 shutdown_nice(int howto) 353 { 354 355 if (initproc != NULL && !SCHEDULER_STOPPED()) { 356 BOOTTRACE("shutdown initiated"); 357 shutdown_nice_task.ta_context = (void *)(uintptr_t)howto; 358 taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task); 359 } else { 360 /* 361 * No init(8) running, or scheduler would not allow it 362 * to run, so simply reboot. 363 */ 364 kern_reboot(howto | RB_NOSYNC); 365 } 366 } 367 368 static void 369 print_uptime(void) 370 { 371 int f; 372 struct timespec ts; 373 374 getnanouptime(&ts); 375 printf("Uptime: "); 376 f = 0; 377 if (ts.tv_sec >= 86400) { 378 printf("%ldd", (long)ts.tv_sec / 86400); 379 ts.tv_sec %= 86400; 380 f = 1; 381 } 382 if (f || ts.tv_sec >= 3600) { 383 printf("%ldh", (long)ts.tv_sec / 3600); 384 ts.tv_sec %= 3600; 385 f = 1; 386 } 387 if (f || ts.tv_sec >= 60) { 388 printf("%ldm", (long)ts.tv_sec / 60); 389 ts.tv_sec %= 60; 390 f = 1; 391 } 392 printf("%lds\n", (long)ts.tv_sec); 393 } 394 395 /* 396 * Set up a context that can be extracted from the dump. 397 */ 398 void 399 dump_savectx(void) 400 { 401 402 savectx(&dumppcb); 403 dumptid = curthread->td_tid; 404 } 405 406 int 407 doadump(boolean_t textdump) 408 { 409 boolean_t coredump; 410 int error; 411 412 error = 0; 413 if (dumping) 414 return (EBUSY); 415 if (TAILQ_EMPTY(&dumper_configs)) 416 return (ENXIO); 417 418 dump_savectx(); 419 dumping++; 420 421 coredump = TRUE; 422 #ifdef DDB 423 if (textdump && textdump_pending) { 424 coredump = FALSE; 425 textdump_dumpsys(TAILQ_FIRST(&dumper_configs)); 426 } 427 #endif 428 if (coredump) { 429 struct dumperinfo *di; 430 431 TAILQ_FOREACH(di, &dumper_configs, di_next) { 432 error = dumpsys(di); 433 if (error == 0) 434 break; 435 } 436 } 437 438 dumping--; 439 return (error); 440 } 441 442 /* 443 * Trace the shutdown reason. 444 */ 445 static void 446 reboottrace(int howto) 447 { 448 if ((howto & RB_DUMP) != 0) { 449 if ((howto & RB_HALT) != 0) 450 BOOTTRACE("system panic: halting..."); 451 if ((howto & RB_POWEROFF) != 0) 452 BOOTTRACE("system panic: powering off..."); 453 if ((howto & (RB_HALT|RB_POWEROFF)) == 0) 454 BOOTTRACE("system panic: rebooting..."); 455 } else { 456 if ((howto & RB_HALT) != 0) 457 BOOTTRACE("system halting..."); 458 if ((howto & RB_POWEROFF) != 0) 459 BOOTTRACE("system powering off..."); 460 if ((howto & (RB_HALT|RB_POWEROFF)) == 0) 461 BOOTTRACE("system rebooting..."); 462 } 463 } 464 465 /* 466 * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or 467 * power off. 468 */ 469 void 470 kern_reboot(int howto) 471 { 472 static int once = 0; 473 474 if (initproc != NULL && curproc != initproc) 475 BOOTTRACE("kernel shutdown (dirty) started"); 476 else 477 BOOTTRACE("kernel shutdown (clean) started"); 478 479 /* 480 * Normal paths here don't hold Giant, but we can wind up here 481 * unexpectedly with it held. Drop it now so we don't have to 482 * drop and pick it up elsewhere. The paths it is locking will 483 * never be returned to, and it is preferable to preclude 484 * deadlock than to lock against code that won't ever 485 * continue. 486 */ 487 while (mtx_owned(&Giant)) 488 mtx_unlock(&Giant); 489 490 #if defined(SMP) 491 /* 492 * Bind us to the first CPU so that all shutdown code runs there. Some 493 * systems don't shutdown properly (i.e., ACPI power off) if we 494 * run on another processor. 495 */ 496 if (!SCHEDULER_STOPPED()) { 497 thread_lock(curthread); 498 sched_bind(curthread, CPU_FIRST()); 499 thread_unlock(curthread); 500 KASSERT(PCPU_GET(cpuid) == CPU_FIRST(), 501 ("%s: not running on cpu 0", __func__)); 502 } 503 #endif 504 /* We're in the process of rebooting. */ 505 rebooting = 1; 506 reboottrace(howto); 507 508 /* We are out of the debugger now. */ 509 kdb_active = 0; 510 511 /* 512 * Do any callouts that should be done BEFORE syncing the filesystems. 513 */ 514 EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); 515 BOOTTRACE("shutdown pre sync complete"); 516 517 /* 518 * Now sync filesystems 519 */ 520 if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) { 521 once = 1; 522 BOOTTRACE("bufshutdown begin"); 523 bufshutdown(show_busybufs); 524 BOOTTRACE("bufshutdown end"); 525 } 526 527 print_uptime(); 528 529 cngrab(); 530 531 /* 532 * Ok, now do things that assume all filesystem activity has 533 * been completed. 534 */ 535 EVENTHANDLER_INVOKE(shutdown_post_sync, howto); 536 BOOTTRACE("shutdown post sync complete"); 537 538 if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping) 539 doadump(TRUE); 540 541 /* Now that we're going to really halt the system... */ 542 BOOTTRACE("shutdown final begin"); 543 544 if (shutdown_trace) 545 boottrace_dump_console(); 546 547 EVENTHANDLER_INVOKE(shutdown_final, howto); 548 549 /* 550 * Call this directly so that reset is attempted even if shutdown 551 * handlers are not yet registered. 552 */ 553 shutdown_reset(NULL, howto); 554 555 for(;;) ; /* safety against shutdown_reset not working */ 556 /* NOTREACHED */ 557 } 558 559 /* 560 * The system call that results in changing the rootfs. 561 */ 562 static int 563 kern_reroot(void) 564 { 565 struct vnode *oldrootvnode, *vp; 566 struct mount *mp, *devmp; 567 int error; 568 569 if (curproc != initproc) 570 return (EPERM); 571 572 /* 573 * Mark the filesystem containing currently-running executable 574 * (the temporary copy of init(8)) busy. 575 */ 576 vp = curproc->p_textvp; 577 error = vn_lock(vp, LK_SHARED); 578 if (error != 0) 579 return (error); 580 mp = vp->v_mount; 581 error = vfs_busy(mp, MBF_NOWAIT); 582 if (error != 0) { 583 vfs_ref(mp); 584 VOP_UNLOCK(vp); 585 error = vfs_busy(mp, 0); 586 vn_lock(vp, LK_SHARED | LK_RETRY); 587 vfs_rel(mp); 588 if (error != 0) { 589 VOP_UNLOCK(vp); 590 return (ENOENT); 591 } 592 if (VN_IS_DOOMED(vp)) { 593 VOP_UNLOCK(vp); 594 vfs_unbusy(mp); 595 return (ENOENT); 596 } 597 } 598 VOP_UNLOCK(vp); 599 600 /* 601 * Remove the filesystem containing currently-running executable 602 * from the mount list, to prevent it from being unmounted 603 * by vfs_unmountall(), and to avoid confusing vfs_mountroot(). 604 * 605 * Also preserve /dev - forcibly unmounting it could cause driver 606 * reinitialization. 607 */ 608 609 vfs_ref(rootdevmp); 610 devmp = rootdevmp; 611 rootdevmp = NULL; 612 613 mtx_lock(&mountlist_mtx); 614 TAILQ_REMOVE(&mountlist, mp, mnt_list); 615 TAILQ_REMOVE(&mountlist, devmp, mnt_list); 616 mtx_unlock(&mountlist_mtx); 617 618 oldrootvnode = rootvnode; 619 620 /* 621 * Unmount everything except for the two filesystems preserved above. 622 */ 623 vfs_unmountall(); 624 625 /* 626 * Add /dev back; vfs_mountroot() will move it into its new place. 627 */ 628 mtx_lock(&mountlist_mtx); 629 TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list); 630 mtx_unlock(&mountlist_mtx); 631 rootdevmp = devmp; 632 vfs_rel(rootdevmp); 633 634 /* 635 * Mount the new rootfs. 636 */ 637 vfs_mountroot(); 638 639 /* 640 * Update all references to the old rootvnode. 641 */ 642 mountcheckdirs(oldrootvnode, rootvnode); 643 644 /* 645 * Add the temporary filesystem back and unbusy it. 646 */ 647 mtx_lock(&mountlist_mtx); 648 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 649 mtx_unlock(&mountlist_mtx); 650 vfs_unbusy(mp); 651 652 return (0); 653 } 654 655 /* 656 * If the shutdown was a clean halt, behave accordingly. 657 */ 658 static void 659 shutdown_halt(void *junk, int howto) 660 { 661 662 if (howto & RB_HALT) { 663 printf("\n"); 664 printf("The operating system has halted.\n"); 665 printf("Please press any key to reboot.\n\n"); 666 667 wdog_kern_pat(WD_TO_NEVER); 668 669 switch (cngetc()) { 670 case -1: /* No console, just die */ 671 cpu_halt(); 672 /* NOTREACHED */ 673 default: 674 break; 675 } 676 } 677 } 678 679 /* 680 * Check to see if the system panicked, pause and then reboot 681 * according to the specified delay. 682 */ 683 static void 684 shutdown_panic(void *junk, int howto) 685 { 686 int loop; 687 688 if (howto & RB_DUMP) { 689 if (panic_reboot_wait_time != 0) { 690 if (panic_reboot_wait_time != -1) { 691 printf("Automatic reboot in %d seconds - " 692 "press a key on the console to abort\n", 693 panic_reboot_wait_time); 694 for (loop = panic_reboot_wait_time * 10; 695 loop > 0; --loop) { 696 DELAY(1000 * 100); /* 1/10th second */ 697 /* Did user type a key? */ 698 if (cncheckc() != -1) 699 break; 700 } 701 if (!loop) 702 return; 703 } 704 } else { /* zero time specified - reboot NOW */ 705 return; 706 } 707 printf("--> Press a key on the console to reboot,\n"); 708 printf("--> or switch off the system now.\n"); 709 cngetc(); 710 } 711 } 712 713 /* 714 * Everything done, now reset 715 */ 716 static void 717 shutdown_reset(void *junk, int howto) 718 { 719 720 printf("Rebooting...\n"); 721 DELAY(reboot_wait_time * 1000000); 722 723 /* 724 * Acquiring smp_ipi_mtx here has a double effect: 725 * - it disables interrupts avoiding CPU0 preemption 726 * by fast handlers (thus deadlocking against other CPUs) 727 * - it avoids deadlocks against smp_rendezvous() or, more 728 * generally, threads busy-waiting, with this spinlock held, 729 * and waiting for responses by threads on other CPUs 730 * (ie. smp_tlb_shootdown()). 731 * 732 * For the !SMP case it just needs to handle the former problem. 733 */ 734 #ifdef SMP 735 mtx_lock_spin(&smp_ipi_mtx); 736 #else 737 spinlock_enter(); 738 #endif 739 740 cpu_reset(); 741 /* NOTREACHED */ /* assuming reset worked */ 742 } 743 744 #if defined(WITNESS) || defined(INVARIANT_SUPPORT) 745 static int kassert_warn_only = 0; 746 #ifdef KDB 747 static int kassert_do_kdb = 0; 748 #endif 749 #ifdef KTR 750 static int kassert_do_ktr = 0; 751 #endif 752 static int kassert_do_log = 1; 753 static int kassert_log_pps_limit = 4; 754 static int kassert_log_mute_at = 0; 755 static int kassert_log_panic_at = 0; 756 static int kassert_suppress_in_panic = 0; 757 static int kassert_warnings = 0; 758 759 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 760 "kassert options"); 761 762 #ifdef KASSERT_PANIC_OPTIONAL 763 #define KASSERT_RWTUN CTLFLAG_RWTUN 764 #else 765 #define KASSERT_RWTUN CTLFLAG_RDTUN 766 #endif 767 768 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN, 769 &kassert_warn_only, 0, 770 "KASSERT triggers a panic (0) or just a warning (1)"); 771 772 #ifdef KDB 773 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN, 774 &kassert_do_kdb, 0, "KASSERT will enter the debugger"); 775 #endif 776 777 #ifdef KTR 778 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN, 779 &kassert_do_ktr, 0, 780 "KASSERT does a KTR, set this to the KTRMASK you want"); 781 #endif 782 783 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN, 784 &kassert_do_log, 0, 785 "If warn_only is enabled, log (1) or do not log (0) assertion violations"); 786 787 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS, 788 &kassert_warnings, 0, "number of KASSERTs that have been triggered"); 789 790 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN, 791 &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic"); 792 793 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN, 794 &kassert_log_pps_limit, 0, "limit number of log messages per second"); 795 796 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN, 797 &kassert_log_mute_at, 0, "max number of KASSERTS to log"); 798 799 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN, 800 &kassert_suppress_in_panic, 0, 801 "KASSERTs will be suppressed while handling a panic"); 802 #undef KASSERT_RWTUN 803 804 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS); 805 806 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert, 807 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0, 808 kassert_sysctl_kassert, "I", 809 "set to trigger a test kassert"); 810 811 static int 812 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS) 813 { 814 int error, i; 815 816 error = sysctl_wire_old_buffer(req, sizeof(int)); 817 if (error == 0) { 818 i = 0; 819 error = sysctl_handle_int(oidp, &i, 0, req); 820 } 821 if (error != 0 || req->newptr == NULL) 822 return (error); 823 KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i)); 824 return (0); 825 } 826 827 #ifdef KASSERT_PANIC_OPTIONAL 828 /* 829 * Called by KASSERT, this decides if we will panic 830 * or if we will log via printf and/or ktr. 831 */ 832 void 833 kassert_panic(const char *fmt, ...) 834 { 835 static char buf[256]; 836 va_list ap; 837 838 va_start(ap, fmt); 839 (void)vsnprintf(buf, sizeof(buf), fmt, ap); 840 va_end(ap); 841 842 /* 843 * If we are suppressing secondary panics, log the warning but do not 844 * re-enter panic/kdb. 845 */ 846 if (KERNEL_PANICKED() && kassert_suppress_in_panic) { 847 if (kassert_do_log) { 848 printf("KASSERT failed: %s\n", buf); 849 #ifdef KDB 850 if (trace_all_panics && trace_on_panic) 851 kdb_backtrace(); 852 #endif 853 } 854 return; 855 } 856 857 /* 858 * panic if we're not just warning, or if we've exceeded 859 * kassert_log_panic_at warnings. 860 */ 861 if (!kassert_warn_only || 862 (kassert_log_panic_at > 0 && 863 kassert_warnings >= kassert_log_panic_at)) { 864 va_start(ap, fmt); 865 vpanic(fmt, ap); 866 /* NORETURN */ 867 } 868 #ifdef KTR 869 if (kassert_do_ktr) 870 CTR0(ktr_mask, buf); 871 #endif /* KTR */ 872 /* 873 * log if we've not yet met the mute limit. 874 */ 875 if (kassert_do_log && 876 (kassert_log_mute_at == 0 || 877 kassert_warnings < kassert_log_mute_at)) { 878 static struct timeval lasterr; 879 static int curerr; 880 881 if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) { 882 printf("KASSERT failed: %s\n", buf); 883 kdb_backtrace(); 884 } 885 } 886 #ifdef KDB 887 if (kassert_do_kdb) { 888 kdb_enter(KDB_WHY_KASSERT, buf); 889 } 890 #endif 891 atomic_add_int(&kassert_warnings, 1); 892 } 893 #endif /* KASSERT_PANIC_OPTIONAL */ 894 #endif 895 896 /* 897 * Panic is called on unresolvable fatal errors. It prints "panic: mesg", 898 * and then reboots. If we are called twice, then we avoid trying to sync 899 * the disks as this often leads to recursive panics. 900 */ 901 void 902 panic(const char *fmt, ...) 903 { 904 va_list ap; 905 906 va_start(ap, fmt); 907 vpanic(fmt, ap); 908 } 909 910 void 911 vpanic(const char *fmt, va_list ap) 912 { 913 #ifdef SMP 914 cpuset_t other_cpus; 915 #endif 916 struct thread *td = curthread; 917 int bootopt, newpanic; 918 static char buf[256]; 919 920 spinlock_enter(); 921 922 #ifdef SMP 923 /* 924 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from 925 * concurrently entering panic. Only the winner will proceed 926 * further. 927 */ 928 if (panicstr == NULL && !kdb_active) { 929 other_cpus = all_cpus; 930 CPU_CLR(PCPU_GET(cpuid), &other_cpus); 931 stop_cpus_hard(other_cpus); 932 } 933 #endif 934 935 /* 936 * Ensure that the scheduler is stopped while panicking, even if panic 937 * has been entered from kdb. 938 */ 939 td->td_stopsched = 1; 940 941 bootopt = RB_AUTOBOOT; 942 newpanic = 0; 943 if (KERNEL_PANICKED()) 944 bootopt |= RB_NOSYNC; 945 else { 946 bootopt |= RB_DUMP; 947 panicstr = fmt; 948 panicked = true; 949 newpanic = 1; 950 } 951 952 if (newpanic) { 953 (void)vsnprintf(buf, sizeof(buf), fmt, ap); 954 panicstr = buf; 955 cngrab(); 956 printf("panic: %s\n", buf); 957 } else { 958 printf("panic: "); 959 vprintf(fmt, ap); 960 printf("\n"); 961 } 962 #ifdef SMP 963 printf("cpuid = %d\n", PCPU_GET(cpuid)); 964 #endif 965 printf("time = %jd\n", (intmax_t )time_second); 966 #ifdef KDB 967 if ((newpanic || trace_all_panics) && trace_on_panic) 968 kdb_backtrace(); 969 if (debugger_on_panic) 970 kdb_enter(KDB_WHY_PANIC, "panic"); 971 else if (!newpanic && debugger_on_recursive_panic) 972 kdb_enter(KDB_WHY_PANIC, "re-panic"); 973 #endif 974 /*thread_lock(td); */ 975 td->td_flags |= TDF_INPANIC; 976 /* thread_unlock(td); */ 977 if (!sync_on_panic) 978 bootopt |= RB_NOSYNC; 979 if (poweroff_on_panic) 980 bootopt |= RB_POWEROFF; 981 if (powercycle_on_panic) 982 bootopt |= RB_POWERCYCLE; 983 kern_reboot(bootopt); 984 } 985 986 /* 987 * Support for poweroff delay. 988 * 989 * Please note that setting this delay too short might power off your machine 990 * before the write cache on your hard disk has been flushed, leading to 991 * soft-updates inconsistencies. 992 */ 993 #ifndef POWEROFF_DELAY 994 # define POWEROFF_DELAY 5000 995 #endif 996 static int poweroff_delay = POWEROFF_DELAY; 997 998 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, 999 &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)"); 1000 1001 static void 1002 poweroff_wait(void *junk, int howto) 1003 { 1004 1005 if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0) 1006 return; 1007 DELAY(poweroff_delay * 1000); 1008 } 1009 1010 /* 1011 * Some system processes (e.g. syncer) need to be stopped at appropriate 1012 * points in their main loops prior to a system shutdown, so that they 1013 * won't interfere with the shutdown process (e.g. by holding a disk buf 1014 * to cause sync to fail). For each of these system processes, register 1015 * shutdown_kproc() as a handler for one of shutdown events. 1016 */ 1017 static int kproc_shutdown_wait = 60; 1018 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, 1019 &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process"); 1020 1021 void 1022 kproc_shutdown(void *arg, int howto) 1023 { 1024 struct proc *p; 1025 int error; 1026 1027 if (KERNEL_PANICKED()) 1028 return; 1029 1030 p = (struct proc *)arg; 1031 printf("Waiting (max %d seconds) for system process `%s' to stop... ", 1032 kproc_shutdown_wait, p->p_comm); 1033 error = kproc_suspend(p, kproc_shutdown_wait * hz); 1034 1035 if (error == EWOULDBLOCK) 1036 printf("timed out\n"); 1037 else 1038 printf("done\n"); 1039 } 1040 1041 void 1042 kthread_shutdown(void *arg, int howto) 1043 { 1044 struct thread *td; 1045 int error; 1046 1047 if (KERNEL_PANICKED()) 1048 return; 1049 1050 td = (struct thread *)arg; 1051 printf("Waiting (max %d seconds) for system thread `%s' to stop... ", 1052 kproc_shutdown_wait, td->td_name); 1053 error = kthread_suspend(td, kproc_shutdown_wait * hz); 1054 1055 if (error == EWOULDBLOCK) 1056 printf("timed out\n"); 1057 else 1058 printf("done\n"); 1059 } 1060 1061 static int 1062 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS) 1063 { 1064 char buf[256]; 1065 struct dumperinfo *di; 1066 struct sbuf sb; 1067 int error; 1068 1069 error = sysctl_wire_old_buffer(req, 0); 1070 if (error != 0) 1071 return (error); 1072 1073 sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req); 1074 1075 mtx_lock(&dumpconf_list_lk); 1076 TAILQ_FOREACH(di, &dumper_configs, di_next) { 1077 if (di != TAILQ_FIRST(&dumper_configs)) 1078 sbuf_putc(&sb, ','); 1079 sbuf_cat(&sb, di->di_devname); 1080 } 1081 mtx_unlock(&dumpconf_list_lk); 1082 1083 error = sbuf_finish(&sb); 1084 sbuf_delete(&sb); 1085 return (error); 1086 } 1087 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname, 1088 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0, 1089 dumpdevname_sysctl_handler, "A", 1090 "Device(s) for kernel dumps"); 1091 1092 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length); 1093 1094 #ifdef EKCD 1095 static struct kerneldumpcrypto * 1096 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption, 1097 const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey) 1098 { 1099 struct kerneldumpcrypto *kdc; 1100 struct kerneldumpkey *kdk; 1101 uint32_t dumpkeysize; 1102 1103 dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize); 1104 kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO); 1105 1106 arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0); 1107 1108 kdc->kdc_encryption = encryption; 1109 switch (kdc->kdc_encryption) { 1110 case KERNELDUMP_ENC_AES_256_CBC: 1111 if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0) 1112 goto failed; 1113 break; 1114 case KERNELDUMP_ENC_CHACHA20: 1115 chacha_keysetup(&kdc->kdc_chacha, key, 256); 1116 break; 1117 default: 1118 goto failed; 1119 } 1120 1121 kdc->kdc_dumpkeysize = dumpkeysize; 1122 kdk = kdc->kdc_dumpkey; 1123 kdk->kdk_encryption = kdc->kdc_encryption; 1124 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv)); 1125 kdk->kdk_encryptedkeysize = htod32(encryptedkeysize); 1126 memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize); 1127 1128 return (kdc); 1129 failed: 1130 zfree(kdc, M_EKCD); 1131 return (NULL); 1132 } 1133 1134 static int 1135 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc) 1136 { 1137 uint8_t hash[SHA256_DIGEST_LENGTH]; 1138 SHA256_CTX ctx; 1139 struct kerneldumpkey *kdk; 1140 int error; 1141 1142 error = 0; 1143 1144 if (kdc == NULL) 1145 return (0); 1146 1147 /* 1148 * When a user enters ddb it can write a crash dump multiple times. 1149 * Each time it should be encrypted using a different IV. 1150 */ 1151 SHA256_Init(&ctx); 1152 SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv)); 1153 SHA256_Final(hash, &ctx); 1154 bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv)); 1155 1156 switch (kdc->kdc_encryption) { 1157 case KERNELDUMP_ENC_AES_256_CBC: 1158 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC, 1159 kdc->kdc_iv) <= 0) { 1160 error = EINVAL; 1161 goto out; 1162 } 1163 break; 1164 case KERNELDUMP_ENC_CHACHA20: 1165 chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL); 1166 break; 1167 default: 1168 error = EINVAL; 1169 goto out; 1170 } 1171 1172 kdk = kdc->kdc_dumpkey; 1173 memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv)); 1174 out: 1175 explicit_bzero(hash, sizeof(hash)); 1176 return (error); 1177 } 1178 1179 static uint32_t 1180 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc) 1181 { 1182 1183 if (kdc == NULL) 1184 return (0); 1185 return (kdc->kdc_dumpkeysize); 1186 } 1187 #endif /* EKCD */ 1188 1189 static struct kerneldumpcomp * 1190 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression) 1191 { 1192 struct kerneldumpcomp *kdcomp; 1193 int format; 1194 1195 switch (compression) { 1196 case KERNELDUMP_COMP_GZIP: 1197 format = COMPRESS_GZIP; 1198 break; 1199 case KERNELDUMP_COMP_ZSTD: 1200 format = COMPRESS_ZSTD; 1201 break; 1202 default: 1203 return (NULL); 1204 } 1205 1206 kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO); 1207 kdcomp->kdc_format = compression; 1208 kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb, 1209 format, di->maxiosize, kerneldump_gzlevel, di); 1210 if (kdcomp->kdc_stream == NULL) { 1211 free(kdcomp, M_DUMPER); 1212 return (NULL); 1213 } 1214 kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP); 1215 return (kdcomp); 1216 } 1217 1218 static void 1219 kerneldumpcomp_destroy(struct dumperinfo *di) 1220 { 1221 struct kerneldumpcomp *kdcomp; 1222 1223 kdcomp = di->kdcomp; 1224 if (kdcomp == NULL) 1225 return; 1226 compressor_fini(kdcomp->kdc_stream); 1227 zfree(kdcomp->kdc_buf, M_DUMPER); 1228 free(kdcomp, M_DUMPER); 1229 } 1230 1231 /* 1232 * Free a dumper. Must not be present on global list. 1233 */ 1234 void 1235 dumper_destroy(struct dumperinfo *di) 1236 { 1237 1238 if (di == NULL) 1239 return; 1240 1241 zfree(di->blockbuf, M_DUMPER); 1242 kerneldumpcomp_destroy(di); 1243 #ifdef EKCD 1244 zfree(di->kdcrypto, M_EKCD); 1245 #endif 1246 zfree(di, M_DUMPER); 1247 } 1248 1249 /* 1250 * Allocate and set up a new dumper from the provided template. 1251 */ 1252 int 1253 dumper_create(const struct dumperinfo *di_template, const char *devname, 1254 const struct diocskerneldump_arg *kda, struct dumperinfo **dip) 1255 { 1256 struct dumperinfo *newdi; 1257 int error = 0; 1258 1259 if (dip == NULL) 1260 return (EINVAL); 1261 1262 /* Allocate a new dumper */ 1263 newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER, 1264 M_WAITOK | M_ZERO); 1265 memcpy(newdi, di_template, sizeof(*newdi)); 1266 newdi->blockbuf = NULL; 1267 newdi->kdcrypto = NULL; 1268 newdi->kdcomp = NULL; 1269 strcpy(newdi->di_devname, devname); 1270 1271 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) { 1272 #ifdef EKCD 1273 newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize, 1274 kda->kda_encryption, kda->kda_key, 1275 kda->kda_encryptedkeysize, kda->kda_encryptedkey); 1276 if (newdi->kdcrypto == NULL) { 1277 error = EINVAL; 1278 goto cleanup; 1279 } 1280 #else 1281 error = EOPNOTSUPP; 1282 goto cleanup; 1283 #endif 1284 } 1285 if (kda->kda_compression != KERNELDUMP_COMP_NONE) { 1286 #ifdef EKCD 1287 /* 1288 * We can't support simultaneous unpadded block cipher 1289 * encryption and compression because there is no guarantee the 1290 * length of the compressed result is exactly a multiple of the 1291 * cipher block size. 1292 */ 1293 if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) { 1294 error = EOPNOTSUPP; 1295 goto cleanup; 1296 } 1297 #endif 1298 newdi->kdcomp = kerneldumpcomp_create(newdi, 1299 kda->kda_compression); 1300 if (newdi->kdcomp == NULL) { 1301 error = EINVAL; 1302 goto cleanup; 1303 } 1304 } 1305 newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO); 1306 1307 *dip = newdi; 1308 return (0); 1309 cleanup: 1310 dumper_destroy(newdi); 1311 return (error); 1312 } 1313 1314 /* 1315 * Create a new dumper and register it in the global list. 1316 */ 1317 int 1318 dumper_insert(const struct dumperinfo *di_template, const char *devname, 1319 const struct diocskerneldump_arg *kda) 1320 { 1321 struct dumperinfo *newdi, *listdi; 1322 bool inserted; 1323 uint8_t index; 1324 int error; 1325 1326 index = kda->kda_index; 1327 MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV && 1328 index != KDA_REMOVE_ALL); 1329 1330 error = priv_check(curthread, PRIV_SETDUMPER); 1331 if (error != 0) 1332 return (error); 1333 1334 error = dumper_create(di_template, devname, kda, &newdi); 1335 if (error != 0) 1336 return (error); 1337 1338 /* Add the new configuration to the queue */ 1339 mtx_lock(&dumpconf_list_lk); 1340 inserted = false; 1341 TAILQ_FOREACH(listdi, &dumper_configs, di_next) { 1342 if (index == 0) { 1343 TAILQ_INSERT_BEFORE(listdi, newdi, di_next); 1344 inserted = true; 1345 break; 1346 } 1347 index--; 1348 } 1349 if (!inserted) 1350 TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next); 1351 mtx_unlock(&dumpconf_list_lk); 1352 1353 return (0); 1354 } 1355 1356 #ifdef DDB 1357 void 1358 dumper_ddb_insert(struct dumperinfo *newdi) 1359 { 1360 TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next); 1361 } 1362 1363 void 1364 dumper_ddb_remove(struct dumperinfo *di) 1365 { 1366 TAILQ_REMOVE(&dumper_configs, di, di_next); 1367 } 1368 #endif 1369 1370 static bool 1371 dumper_config_match(const struct dumperinfo *di, const char *devname, 1372 const struct diocskerneldump_arg *kda) 1373 { 1374 if (kda->kda_index == KDA_REMOVE_ALL) 1375 return (true); 1376 1377 if (strcmp(di->di_devname, devname) != 0) 1378 return (false); 1379 1380 /* 1381 * Allow wildcard removal of configs matching a device on g_dev_orphan. 1382 */ 1383 if (kda->kda_index == KDA_REMOVE_DEV) 1384 return (true); 1385 1386 if (di->kdcomp != NULL) { 1387 if (di->kdcomp->kdc_format != kda->kda_compression) 1388 return (false); 1389 } else if (kda->kda_compression != KERNELDUMP_COMP_NONE) 1390 return (false); 1391 #ifdef EKCD 1392 if (di->kdcrypto != NULL) { 1393 if (di->kdcrypto->kdc_encryption != kda->kda_encryption) 1394 return (false); 1395 /* 1396 * Do we care to verify keys match to delete? It seems weird 1397 * to expect multiple fallback dump configurations on the same 1398 * device that only differ in crypto key. 1399 */ 1400 } else 1401 #endif 1402 if (kda->kda_encryption != KERNELDUMP_ENC_NONE) 1403 return (false); 1404 1405 return (true); 1406 } 1407 1408 /* 1409 * Remove and free the requested dumper(s) from the global list. 1410 */ 1411 int 1412 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda) 1413 { 1414 struct dumperinfo *di, *sdi; 1415 bool found; 1416 int error; 1417 1418 error = priv_check(curthread, PRIV_SETDUMPER); 1419 if (error != 0) 1420 return (error); 1421 1422 /* 1423 * Try to find a matching configuration, and kill it. 1424 * 1425 * NULL 'kda' indicates remove any configuration matching 'devname', 1426 * which may remove multiple configurations in atypical configurations. 1427 */ 1428 found = false; 1429 mtx_lock(&dumpconf_list_lk); 1430 TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) { 1431 if (dumper_config_match(di, devname, kda)) { 1432 found = true; 1433 TAILQ_REMOVE(&dumper_configs, di, di_next); 1434 dumper_destroy(di); 1435 } 1436 } 1437 mtx_unlock(&dumpconf_list_lk); 1438 1439 /* Only produce ENOENT if a more targeted match didn't match. */ 1440 if (!found && kda->kda_index == KDA_REMOVE) 1441 return (ENOENT); 1442 return (0); 1443 } 1444 1445 static int 1446 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length) 1447 { 1448 1449 if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset || 1450 offset - di->mediaoffset + length > di->mediasize)) { 1451 if (di->kdcomp != NULL && offset >= di->mediaoffset) { 1452 printf( 1453 "Compressed dump failed to fit in device boundaries.\n"); 1454 return (E2BIG); 1455 } 1456 1457 printf("Attempt to write outside dump device boundaries.\n" 1458 "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n", 1459 (intmax_t)offset, (intmax_t)di->mediaoffset, 1460 (uintmax_t)length, (intmax_t)di->mediasize); 1461 return (ENOSPC); 1462 } 1463 if (length % di->blocksize != 0) { 1464 printf("Attempt to write partial block of length %ju.\n", 1465 (uintmax_t)length); 1466 return (EINVAL); 1467 } 1468 if (offset % di->blocksize != 0) { 1469 printf("Attempt to write at unaligned offset %jd.\n", 1470 (intmax_t)offset); 1471 return (EINVAL); 1472 } 1473 1474 return (0); 1475 } 1476 1477 #ifdef EKCD 1478 static int 1479 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size) 1480 { 1481 1482 switch (kdc->kdc_encryption) { 1483 case KERNELDUMP_ENC_AES_256_CBC: 1484 if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf, 1485 8 * size, buf) <= 0) { 1486 return (EIO); 1487 } 1488 if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC, 1489 buf + size - 16 /* IV size for AES-256-CBC */) <= 0) { 1490 return (EIO); 1491 } 1492 break; 1493 case KERNELDUMP_ENC_CHACHA20: 1494 chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size); 1495 break; 1496 default: 1497 return (EINVAL); 1498 } 1499 1500 return (0); 1501 } 1502 1503 /* Encrypt data and call dumper. */ 1504 static int 1505 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset, 1506 size_t length) 1507 { 1508 static uint8_t buf[KERNELDUMP_BUFFER_SIZE]; 1509 struct kerneldumpcrypto *kdc; 1510 int error; 1511 size_t nbytes; 1512 1513 kdc = di->kdcrypto; 1514 1515 while (length > 0) { 1516 nbytes = MIN(length, sizeof(buf)); 1517 bcopy(virtual, buf, nbytes); 1518 1519 if (dump_encrypt(kdc, buf, nbytes) != 0) 1520 return (EIO); 1521 1522 error = dump_write(di, buf, offset, nbytes); 1523 if (error != 0) 1524 return (error); 1525 1526 offset += nbytes; 1527 virtual = (void *)((uint8_t *)virtual + nbytes); 1528 length -= nbytes; 1529 } 1530 1531 return (0); 1532 } 1533 #endif /* EKCD */ 1534 1535 static int 1536 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg) 1537 { 1538 struct dumperinfo *di; 1539 size_t resid, rlength; 1540 int error; 1541 1542 di = arg; 1543 1544 if (length % di->blocksize != 0) { 1545 /* 1546 * This must be the final write after flushing the compression 1547 * stream. Write as many full blocks as possible and stash the 1548 * residual data in the dumper's block buffer. It will be 1549 * padded and written in dump_finish(). 1550 */ 1551 rlength = rounddown(length, di->blocksize); 1552 if (rlength != 0) { 1553 error = _dump_append(di, base, rlength); 1554 if (error != 0) 1555 return (error); 1556 } 1557 resid = length - rlength; 1558 memmove(di->blockbuf, (uint8_t *)base + rlength, resid); 1559 bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid); 1560 di->kdcomp->kdc_resid = resid; 1561 return (EAGAIN); 1562 } 1563 return (_dump_append(di, base, length)); 1564 } 1565 1566 /* 1567 * Write kernel dump headers at the beginning and end of the dump extent. 1568 * Write the kernel dump encryption key after the leading header if we were 1569 * configured to do so. 1570 */ 1571 static int 1572 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh) 1573 { 1574 #ifdef EKCD 1575 struct kerneldumpcrypto *kdc; 1576 #endif 1577 void *buf; 1578 size_t hdrsz; 1579 uint64_t extent; 1580 uint32_t keysize; 1581 int error; 1582 1583 hdrsz = sizeof(*kdh); 1584 if (hdrsz > di->blocksize) 1585 return (ENOMEM); 1586 1587 #ifdef EKCD 1588 kdc = di->kdcrypto; 1589 keysize = kerneldumpcrypto_dumpkeysize(kdc); 1590 #else 1591 keysize = 0; 1592 #endif 1593 1594 /* 1595 * If the dump device has special handling for headers, let it take care 1596 * of writing them out. 1597 */ 1598 if (di->dumper_hdr != NULL) 1599 return (di->dumper_hdr(di, kdh)); 1600 1601 if (hdrsz == di->blocksize) 1602 buf = kdh; 1603 else { 1604 buf = di->blockbuf; 1605 memset(buf, 0, di->blocksize); 1606 memcpy(buf, kdh, hdrsz); 1607 } 1608 1609 extent = dtoh64(kdh->dumpextent); 1610 #ifdef EKCD 1611 if (kdc != NULL) { 1612 error = dump_write(di, kdc->kdc_dumpkey, 1613 di->mediaoffset + di->mediasize - di->blocksize - extent - 1614 keysize, keysize); 1615 if (error != 0) 1616 return (error); 1617 } 1618 #endif 1619 1620 error = dump_write(di, buf, 1621 di->mediaoffset + di->mediasize - 2 * di->blocksize - extent - 1622 keysize, di->blocksize); 1623 if (error == 0) 1624 error = dump_write(di, buf, di->mediaoffset + di->mediasize - 1625 di->blocksize, di->blocksize); 1626 return (error); 1627 } 1628 1629 /* 1630 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This is to 1631 * protect us from metadata and metadata from us. 1632 */ 1633 #define SIZEOF_METADATA (64 * 1024) 1634 1635 /* 1636 * Do some preliminary setup for a kernel dump: initialize state for encryption, 1637 * if requested, and make sure that we have enough space on the dump device. 1638 * 1639 * We set things up so that the dump ends before the last sector of the dump 1640 * device, at which the trailing header is written. 1641 * 1642 * +-----------+------+-----+----------------------------+------+ 1643 * | | lhdr | key | ... kernel dump ... | thdr | 1644 * +-----------+------+-----+----------------------------+------+ 1645 * 1 blk opt <------- dump extent --------> 1 blk 1646 * 1647 * Dumps written using dump_append() start at the beginning of the extent. 1648 * Uncompressed dumps will use the entire extent, but compressed dumps typically 1649 * will not. The true length of the dump is recorded in the leading and trailing 1650 * headers once the dump has been completed. 1651 * 1652 * The dump device may provide a callback, in which case it will initialize 1653 * dumpoff and take care of laying out the headers. 1654 */ 1655 int 1656 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh) 1657 { 1658 #ifdef EKCD 1659 struct kerneldumpcrypto *kdc; 1660 #endif 1661 void *key; 1662 uint64_t dumpextent, span; 1663 uint32_t keysize; 1664 int error; 1665 1666 #ifdef EKCD 1667 /* Send the key before the dump so a partial dump is still usable. */ 1668 kdc = di->kdcrypto; 1669 error = kerneldumpcrypto_init(kdc); 1670 if (error != 0) 1671 return (error); 1672 keysize = kerneldumpcrypto_dumpkeysize(kdc); 1673 key = keysize > 0 ? kdc->kdc_dumpkey : NULL; 1674 #else 1675 error = 0; 1676 keysize = 0; 1677 key = NULL; 1678 #endif 1679 1680 if (di->dumper_start != NULL) { 1681 error = di->dumper_start(di, key, keysize); 1682 } else { 1683 dumpextent = dtoh64(kdh->dumpextent); 1684 span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize + 1685 keysize; 1686 if (di->mediasize < span) { 1687 if (di->kdcomp == NULL) 1688 return (E2BIG); 1689 1690 /* 1691 * We don't yet know how much space the compressed dump 1692 * will occupy, so try to use the whole swap partition 1693 * (minus the first 64KB) in the hope that the 1694 * compressed dump will fit. If that doesn't turn out to 1695 * be enough, the bounds checking in dump_write() 1696 * will catch us and cause the dump to fail. 1697 */ 1698 dumpextent = di->mediasize - span + dumpextent; 1699 kdh->dumpextent = htod64(dumpextent); 1700 } 1701 1702 /* 1703 * The offset at which to begin writing the dump. 1704 */ 1705 di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize - 1706 dumpextent; 1707 } 1708 di->origdumpoff = di->dumpoff; 1709 return (error); 1710 } 1711 1712 static int 1713 _dump_append(struct dumperinfo *di, void *virtual, size_t length) 1714 { 1715 int error; 1716 1717 #ifdef EKCD 1718 if (di->kdcrypto != NULL) 1719 error = dump_encrypted_write(di, virtual, di->dumpoff, length); 1720 else 1721 #endif 1722 error = dump_write(di, virtual, di->dumpoff, length); 1723 if (error == 0) 1724 di->dumpoff += length; 1725 return (error); 1726 } 1727 1728 /* 1729 * Write to the dump device starting at dumpoff. When compression is enabled, 1730 * writes to the device will be performed using a callback that gets invoked 1731 * when the compression stream's output buffer is full. 1732 */ 1733 int 1734 dump_append(struct dumperinfo *di, void *virtual, size_t length) 1735 { 1736 void *buf; 1737 1738 if (di->kdcomp != NULL) { 1739 /* Bounce through a buffer to avoid CRC errors. */ 1740 if (length > di->maxiosize) 1741 return (EINVAL); 1742 buf = di->kdcomp->kdc_buf; 1743 memmove(buf, virtual, length); 1744 return (compressor_write(di->kdcomp->kdc_stream, buf, length)); 1745 } 1746 return (_dump_append(di, virtual, length)); 1747 } 1748 1749 /* 1750 * Write to the dump device at the specified offset. 1751 */ 1752 int 1753 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length) 1754 { 1755 int error; 1756 1757 error = dump_check_bounds(di, offset, length); 1758 if (error != 0) 1759 return (error); 1760 return (di->dumper(di->priv, virtual, offset, length)); 1761 } 1762 1763 /* 1764 * Perform kernel dump finalization: flush the compression stream, if necessary, 1765 * write the leading and trailing kernel dump headers now that we know the true 1766 * length of the dump, and optionally write the encryption key following the 1767 * leading header. 1768 */ 1769 int 1770 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh) 1771 { 1772 int error; 1773 1774 if (di->kdcomp != NULL) { 1775 error = compressor_flush(di->kdcomp->kdc_stream); 1776 if (error == EAGAIN) { 1777 /* We have residual data in di->blockbuf. */ 1778 error = _dump_append(di, di->blockbuf, di->blocksize); 1779 if (error == 0) 1780 /* Compensate for _dump_append()'s adjustment. */ 1781 di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid; 1782 di->kdcomp->kdc_resid = 0; 1783 } 1784 if (error != 0) 1785 return (error); 1786 1787 /* 1788 * We now know the size of the compressed dump, so update the 1789 * header accordingly and recompute parity. 1790 */ 1791 kdh->dumplength = htod64(di->dumpoff - di->origdumpoff); 1792 kdh->parity = 0; 1793 kdh->parity = kerneldump_parity(kdh); 1794 1795 compressor_reset(di->kdcomp->kdc_stream); 1796 } 1797 1798 error = dump_write_headers(di, kdh); 1799 if (error != 0) 1800 return (error); 1801 1802 (void)dump_write(di, NULL, 0, 0); 1803 return (0); 1804 } 1805 1806 void 1807 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh, 1808 const char *magic, uint32_t archver, uint64_t dumplen) 1809 { 1810 size_t dstsize; 1811 1812 bzero(kdh, sizeof(*kdh)); 1813 strlcpy(kdh->magic, magic, sizeof(kdh->magic)); 1814 strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); 1815 kdh->version = htod32(KERNELDUMPVERSION); 1816 kdh->architectureversion = htod32(archver); 1817 kdh->dumplength = htod64(dumplen); 1818 kdh->dumpextent = kdh->dumplength; 1819 kdh->dumptime = htod64(time_second); 1820 #ifdef EKCD 1821 kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto)); 1822 #else 1823 kdh->dumpkeysize = 0; 1824 #endif 1825 kdh->blocksize = htod32(di->blocksize); 1826 strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname)); 1827 dstsize = sizeof(kdh->versionstring); 1828 if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize) 1829 kdh->versionstring[dstsize - 2] = '\n'; 1830 if (panicstr != NULL) 1831 strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); 1832 if (di->kdcomp != NULL) 1833 kdh->compression = di->kdcomp->kdc_format; 1834 kdh->parity = kerneldump_parity(kdh); 1835 } 1836 1837 #ifdef DDB 1838 DB_SHOW_COMMAND_FLAGS(panic, db_show_panic, DB_CMD_MEMSAFE) 1839 { 1840 1841 if (panicstr == NULL) 1842 db_printf("panicstr not set\n"); 1843 else 1844 db_printf("panic: %s\n", panicstr); 1845 } 1846 #endif 1847