2 * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
29 * File: bsd/kern/kern_shutdown.c
31 * Copyright (C) 1989, NeXT, Inc.
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
39 #include <sys/proc_internal.h>
41 #include <sys/reboot.h>
43 #include <sys/vnode_internal.h>
44 #include <sys/file_internal.h>
45 #include <sys/clist.h>
46 #include <sys/callout.h>
48 #include <sys/msgbuf.h>
49 #include <sys/ioctl.h>
50 #include <sys/signal.h>
52 #include <kern/task.h>
53 #include <sys/quota.h>
54 #include <vm/vm_kern.h>
55 #include <mach/vm_param.h>
56 #include <sys/filedesc.h>
57 #include <mach/host_priv.h>
58 #include <mach/host_reboot.h>
60 #include <security/audit/audit.h>
62 #include <kern/sched_prim.h> /* for thread_block() */
63 #include <kern/host.h> /* for host_priv_self() */
64 #include <net/if_var.h> /* for if_down_all() */
65 #include <sys/buf_internal.h> /* for count_busy_buffers() */
66 #include <sys/mount_internal.h> /* for vfs_unmountall() */
67 #include <mach/task.h> /* for task_suspend() */
68 #include <sys/sysproto.h> /* abused for sync() */
69 #include <kern/clock.h> /* for delay_for_interval() */
70 #include <libkern/OSAtomic.h>
71 #include <IOKit/IOPlatformExpert.h>
73 #include <sys/kdebug.h>
75 uint32_t system_inshutdown
= 0;
77 /* XXX should be in a header file somewhere, but isn't */
78 extern void (*unmountroot_pre_hook
)(void);
80 unsigned int proc_shutdown_exitcount
= 0;
82 static int sd_openlog(vfs_context_t
);
83 static int sd_closelog(vfs_context_t
);
84 static void sd_log(vfs_context_t
, const char *, ...);
85 static void proc_shutdown(void);
86 static void kernel_hwm_panic_info(void);
87 extern void halt_log_enter(const char * what
, const void * pc
, uint64_t time
);
89 #if DEVELOPMENT || DEBUG
90 extern boolean_t
kdp_has_polled_corefile(void);
91 #endif /* DEVELOPMENT || DEBUG */
100 int signo
; /* the signal to be posted */
101 int setsdstate
; /* shutdown state to be set */
102 int countproc
; /* count processes on action */
103 int activecount
; /* number of processes on which action was done */
106 static vnode_t sd_logvp
= NULLVP
;
107 static off_t sd_log_offset
= 0;
110 static int sd_filt1(proc_t
, void *);
111 static int sd_filt2(proc_t
, void *);
112 static int sd_callback1(proc_t p
, void * arg
);
113 static int sd_callback2(proc_t p
, void * arg
);
114 static int sd_callback3(proc_t p
, void * arg
);
116 extern boolean_t panic_include_zprint
;
117 extern mach_memory_info_t
*panic_kext_memory_info
;
118 extern vm_size_t panic_kext_memory_size
;
121 kernel_hwm_panic_info(void)
123 unsigned int num_sites
;
126 panic_include_zprint
= TRUE
;
127 panic_kext_memory_info
= NULL
;
128 panic_kext_memory_size
= 0;
130 num_sites
= vm_page_diagnose_estimate();
131 panic_kext_memory_size
= num_sites
* sizeof(panic_kext_memory_info
[0]);
133 kr
= kmem_alloc(kernel_map
, (vm_offset_t
*)&panic_kext_memory_info
, round_page(panic_kext_memory_size
), VM_KERN_MEMORY_OSFMK
);
134 if (kr
!= KERN_SUCCESS
) {
135 panic_kext_memory_info
= NULL
;
139 vm_page_diagnose(panic_kext_memory_info
, num_sites
, 0);
143 get_system_inshutdown()
145 return (system_inshutdown
);
149 reboot_kernel(int howto
, char *message
)
151 int hostboot_option
=0;
154 if (!OSCompareAndSwap(0, 1, &system_inshutdown
)) {
155 if ( (howto
&RB_QUICK
) == RB_QUICK
)
160 * Notify the power management root domain that the system will shut down.
162 IOSystemShutdownNotification(kIOSystemShutdownNotificationStageProcessExit
);
164 if ((howto
&RB_QUICK
)==RB_QUICK
) {
165 printf("Quick reboot...\n");
166 if ((howto
&RB_NOSYNC
)==0) {
167 sync((proc_t
)NULL
, (void *)NULL
, (int *)NULL
);
170 else if ((howto
&RB_NOSYNC
)==0) {
173 printf("syncing disks... ");
176 * Release vnodes held by texts before sync.
179 /* handle live procs (deallocate their root and current directories), suspend initproc */
181 startTime
= mach_absolute_time();
183 halt_log_enter("proc_shutdown", 0, mach_absolute_time() - startTime
);
186 startTime
= mach_absolute_time();
188 halt_log_enter("audit_shutdown", 0, mach_absolute_time() - startTime
);
191 if (unmountroot_pre_hook
!= NULL
)
192 unmountroot_pre_hook();
194 startTime
= mach_absolute_time();
195 sync((proc_t
)NULL
, (void *)NULL
, (int *)NULL
);
198 startTime
= mach_absolute_time();
199 kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace");
200 halt_log_enter("shutdown.trace", 0, mach_absolute_time() - startTime
);
203 IOSystemShutdownNotification(kIOSystemShutdownNotificationStageRootUnmount
);
206 * Unmount filesystems
209 #if DEVELOPMENT || DEBUG
210 if (!(howto
& RB_PANIC
) || !kdp_has_polled_corefile())
211 #endif /* DEVELOPMENT || DEBUG */
213 startTime
= mach_absolute_time();
215 halt_log_enter("vfs_unmountall", 0, mach_absolute_time() - startTime
);
218 /* Wait for the buffer cache to clean remaining dirty buffers */
219 startTime
= mach_absolute_time();
220 for (iter
= 0; iter
< 100; iter
++) {
221 nbusy
= count_busy_buffers();
224 printf("%d ", nbusy
);
225 delay_for_interval( 1 * nbusy
, 1000 * 1000);
228 printf("giving up\n");
231 halt_log_enter("bufferclean", 0, mach_absolute_time() - startTime
);
235 * Can't just use an splnet() here to disable the network
236 * because that will lock out softints which the disk
237 * drivers depend on to finish DMAs.
239 startTime
= mach_absolute_time();
241 halt_log_enter("if_down_all", 0, mach_absolute_time() - startTime
);
242 #endif /* NETWORKING */
246 if (howto
& RB_PANIC
) {
247 if (strncmp(message
, "Kernel memory has exceeded limits", 33) == 0) {
248 kernel_hwm_panic_info();
250 panic ("userspace panic: %s", message
);
253 if (howto
& RB_POWERDOWN
)
254 hostboot_option
= HOST_REBOOT_HALT
;
256 hostboot_option
= HOST_REBOOT_HALT
;
258 if (howto
& RB_UPSDELAY
) {
259 hostboot_option
= HOST_REBOOT_UPSDELAY
;
262 host_reboot(host_priv_self(), hostboot_option
);
264 * should not be reached
270 sd_openlog(vfs_context_t ctx
)
275 /* Open shutdown log */
276 if ((error
= vnode_open(PROC_SHUTDOWN_LOG
, (O_CREAT
| FWRITE
| O_NOFOLLOW
), 0644, 0, &sd_logvp
, ctx
))) {
277 printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG
, error
);
282 vnode_setsize(sd_logvp
, (off_t
)0, 0, ctx
);
284 /* Write a little header */
286 sd_log(ctx
, "Process shutdown log. Current time is %lu (in seconds).\n\n", tv
.tv_sec
);
292 sd_closelog(vfs_context_t ctx
)
295 if (sd_logvp
!= NULLVP
) {
296 VNOP_FSYNC(sd_logvp
, MNT_WAIT
, ctx
);
297 error
= vnode_close(sd_logvp
, FWRITE
, ctx
);
304 sd_log(vfs_context_t ctx
, const char *fmt
, ...)
306 int resid
, log_error
, len
;
310 /* If the log isn't open yet, open it */
311 if (sd_logvp
== NULLVP
) {
312 if (sd_openlog(ctx
) != 0) {
313 /* Couldn't open, we fail out */
318 va_start(arglist
, fmt
);
319 len
= vsnprintf(logbuf
, sizeof(logbuf
), fmt
, arglist
);
320 log_error
= vn_rdwr(UIO_WRITE
, sd_logvp
, (caddr_t
)logbuf
, len
, sd_log_offset
,
321 UIO_SYSSPACE
, IO_UNIT
| IO_NOAUTH
, vfs_context_ucred(ctx
), &resid
, vfs_context_proc(ctx
));
322 if (log_error
== EIO
|| log_error
== 0) {
323 sd_log_offset
+= (len
- resid
);
331 sd_filt1(proc_t p
, void * args
)
333 proc_t self
= current_proc();
334 struct sd_filterargs
* sf
= (struct sd_filterargs
*)args
;
335 int delayterm
= sf
-> delayterm
;
336 int shutdownstate
= sf
->shutdownstate
;
338 if (((p
->p_flag
&P_SYSTEM
) != 0) || (p
->p_ppid
== 0)
339 ||(p
== self
) || (p
->p_stat
== SZOMB
)
340 || (p
->p_shutdownstate
!= shutdownstate
)
341 ||((delayterm
== 0) && ((p
->p_lflag
& P_LDELAYTERM
) == P_LDELAYTERM
))
342 || ((p
->p_sigcatch
& sigmask(SIGTERM
))== 0)) {
351 sd_callback1(proc_t p
, void * args
)
353 struct sd_iterargs
* sd
= (struct sd_iterargs
*)args
;
354 int signo
= sd
->signo
;
355 int setsdstate
= sd
->setsdstate
;
356 int countproc
= sd
->countproc
;
359 p
->p_shutdownstate
= setsdstate
;
360 if (p
->p_stat
!= SZOMB
) {
362 if (countproc
!= 0) {
364 p
->p_listflag
|= P_LIST_EXITCOUNT
;
365 proc_shutdown_exitcount
++;
376 return PROC_RETURNED
;
380 sd_filt2(proc_t p
, void * args
)
382 proc_t self
= current_proc();
383 struct sd_filterargs
* sf
= (struct sd_filterargs
*)args
;
384 int delayterm
= sf
-> delayterm
;
385 int shutdownstate
= sf
->shutdownstate
;
387 if (((p
->p_flag
&P_SYSTEM
) != 0) || (p
->p_ppid
== 0)
388 ||(p
== self
) || (p
->p_stat
== SZOMB
)
389 || (p
->p_shutdownstate
== shutdownstate
)
390 ||((delayterm
== 0) && ((p
->p_lflag
& P_LDELAYTERM
) == P_LDELAYTERM
))) {
398 sd_callback2(proc_t p
, void * args
)
400 struct sd_iterargs
* sd
= (struct sd_iterargs
*)args
;
401 int signo
= sd
->signo
;
402 int setsdstate
= sd
->setsdstate
;
403 int countproc
= sd
->countproc
;
406 p
->p_shutdownstate
= setsdstate
;
407 if (p
->p_stat
!= SZOMB
) {
409 if (countproc
!= 0) {
411 p
->p_listflag
|= P_LIST_EXITCOUNT
;
412 proc_shutdown_exitcount
++;
422 return PROC_RETURNED
;
426 sd_callback3(proc_t p
, void * args
)
428 struct sd_iterargs
* sd
= (struct sd_iterargs
*)args
;
429 vfs_context_t ctx
= vfs_context_current();
431 int setsdstate
= sd
->setsdstate
;
434 p
->p_shutdownstate
= setsdstate
;
435 if (p
->p_stat
!= SZOMB
) {
437 * NOTE: following code ignores sig_lock and plays
438 * with exit_thread correctly. This is OK unless we
439 * are a multiprocessor, in which case I do not
440 * understand the sig_lock. This needs to be fixed.
443 if (p
->exit_thread
) { /* someone already doing it */
445 /* give him a chance */
446 thread_block(THREAD_CONTINUE_NULL
);
448 p
->exit_thread
= current_thread();
451 sd_log(ctx
, "%s[%d] had to be forced closed with exit1().\n", p
->p_comm
, p
->p_pid
);
454 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC
, BSD_PROC_FRCEXIT
) | DBG_FUNC_NONE
,
455 p
->p_pid
, 0, 1, 0, 0);
457 exit1(p
, 1, (int *)NULL
);
463 return PROC_RETURNED
;
470 * Shutdown down proc system (release references to current and root
471 * dirs for each process).
473 * POSIX modifications:
475 * For POSIX fcntl() file locking call vno_lockrelease() on
476 * the file to release all of its record locks, if any.
482 vfs_context_t ctx
= vfs_context_current();
483 struct proc
*p
, *self
;
485 struct sd_filterargs sfargs
;
486 struct sd_iterargs sdargs
;
491 * Kill as many procs as we can. (Except ourself...)
493 self
= (struct proc
*)current_proc();
496 * Signal the init with SIGTERM so that he does not launch
500 if (p
&& p
!= self
) {
505 printf("Killing all processes ");
509 * send SIGTERM to those procs interested in catching one
511 sfargs
.delayterm
= delayterm
;
512 sfargs
.shutdownstate
= 0;
513 sdargs
.signo
= SIGTERM
;
514 sdargs
.setsdstate
= 1;
515 sdargs
.countproc
= 1;
516 sdargs
.activecount
= 0;
519 /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
520 proc_rebootscan(sd_callback1
, (void *)&sdargs
, sd_filt1
, (void *)&sfargs
);
522 if (sdargs
.activecount
!= 0 && proc_shutdown_exitcount
!= 0) {
524 if (proc_shutdown_exitcount
!= 0) {
526 * now wait for up to 3 seconds to allow those procs catching SIGTERM
528 * as soon as these procs have exited, we'll continue on to the next step
532 error
= msleep(&proc_shutdown_exitcount
, proc_list_mlock
, PWAIT
, "shutdownwait", &ts
);
534 for (p
= allproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
535 if ((p
->p_listflag
& P_LIST_EXITCOUNT
) == P_LIST_EXITCOUNT
)
536 p
->p_listflag
&= ~P_LIST_EXITCOUNT
;
538 for (p
= zombproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
539 if ((p
->p_listflag
& P_LIST_EXITCOUNT
) == P_LIST_EXITCOUNT
)
540 p
->p_listflag
&= ~P_LIST_EXITCOUNT
;
546 if (error
== ETIMEDOUT
) {
548 * log the names of the unresponsive tasks
553 for (p
= allproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
554 if (p
->p_shutdownstate
== 1) {
555 printf("%s[%d]: didn't act on SIGTERM\n", p
->p_comm
, p
->p_pid
);
556 sd_log(ctx
, "%s[%d]: didn't act on SIGTERM\n", p
->p_comm
, p
->p_pid
);
564 * send a SIGKILL to all the procs still hanging around
566 sfargs
.delayterm
= delayterm
;
567 sfargs
.shutdownstate
= 2;
568 sdargs
.signo
= SIGKILL
;
569 sdargs
.setsdstate
= 2;
570 sdargs
.countproc
= 1;
571 sdargs
.activecount
= 0;
573 /* post a SIGKILL to all that catch SIGTERM and not marked for delay */
574 proc_rebootscan(sd_callback2
, (void *)&sdargs
, sd_filt2
, (void *)&sfargs
);
578 if (sdargs
.activecount
!= 0 && proc_shutdown_exitcount
!= 0) {
580 if (proc_shutdown_exitcount
!= 0) {
582 * wait for up to 60 seconds to allow these procs to exit normally
584 * History: The delay interval was changed from 100 to 200
585 * for NFS requests in particular.
589 error
= msleep(&proc_shutdown_exitcount
, proc_list_mlock
, PWAIT
, "shutdownwait", &ts
);
591 for (p
= allproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
592 if ((p
->p_listflag
& P_LIST_EXITCOUNT
) == P_LIST_EXITCOUNT
)
593 p
->p_listflag
&= ~P_LIST_EXITCOUNT
;
595 for (p
= zombproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
596 if ((p
->p_listflag
& P_LIST_EXITCOUNT
) == P_LIST_EXITCOUNT
)
597 p
->p_listflag
&= ~P_LIST_EXITCOUNT
;
604 if (error
== ETIMEDOUT
) {
606 * log the names of the unresponsive tasks
611 for (p
= allproc
.lh_first
; p
; p
= p
->p_list
.le_next
) {
612 if (p
->p_shutdownstate
== 2) {
613 printf("%s[%d]: didn't act on SIGKILL\n", p
->p_comm
, p
->p_pid
);
614 sd_log(ctx
, "%s[%d]: didn't act on SIGKILL\n", p
->p_comm
, p
->p_pid
);
622 * if we still have procs that haven't exited, then brute force 'em
624 sfargs
.delayterm
= delayterm
;
625 sfargs
.shutdownstate
= 3;
627 sdargs
.setsdstate
= 3;
628 sdargs
.countproc
= 0;
629 sdargs
.activecount
= 0;
633 /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
634 proc_rebootscan(sd_callback3
, (void *)&sdargs
, sd_filt2
, (void *)&sfargs
);
637 /* Now start the termination of processes that are marked for delayed termn */
638 if (delayterm
== 0) {
646 * Now that all other processes have been terminated, suspend init
648 task_suspend_internal(initproc
->task
);
650 /* drop the ref on initproc */
652 printf("continuing\n");