X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/55e303ae13a4cf49d70f2294092726f2fffb9ef2..8a3053a07cee346dca737a5670e546fd26a7c9d6:/bsd/kern/kern_shutdown.c diff --git a/bsd/kern/kern_shutdown.c b/bsd/kern/kern_shutdown.c index 19f68507d..4e231826d 100644 --- a/bsd/kern/kern_shutdown.c +++ b/bsd/kern/kern_shutdown.c @@ -1,16 +1,19 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ - * - * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved. + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in - * compliance with the License. Please obtain a copy of the License at - * http://www.opensource.apple.com/apsl/ and read it before using this - * file. + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. + * + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER @@ -20,7 +23,7 @@ * Please see the License for the specific language governing rights and * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * File: bsd/kern/kern_shutdown.c @@ -33,13 +36,12 @@ #include #include #include -#include +#include #include -#include #include #include -#include -#include +#include +#include #include #include #include @@ -49,43 +51,92 @@ #include #include #include -#include -#if NCPUS > 1 -#include -#include -#include -#endif /* NCPUS > 1 */ #include #include #include +#include #include -#include -int waittime = -1; +#include + +#include /* for thread_block() */ +#include /* for host_priv_self() */ +#include /* for if_down_all() */ +#include /* for count_busy_buffers() */ +#include /* for vfs_unmountall() */ +#include /* for task_suspend() */ +#include /* abused for sync() */ +#include /* for delay_for_interval() */ +#include + +#include + +uint32_t system_inshutdown = 0; + +/* XXX should be in a header file somewhere, but isn't */ +extern void md_prepare_for_shutdown(int, int, char *); +extern void (*unmountroot_pre_hook)(void); + +unsigned int proc_shutdown_exitcount = 0; + +static int sd_openlog(vfs_context_t); +static int sd_closelog(vfs_context_t); +static void sd_log(vfs_context_t, const char *, ...); +static void proc_shutdown(void); + +extern void IOSystemShutdownNotification(void); + +struct sd_filterargs{ + int delayterm; + int shutdownstate; +}; + + +struct sd_iterargs { + int signo; /* the signal to be posted */ + int setsdstate; /* shutdown state to be set */ + int countproc; /* count processes on action */ + int activecount; /* number of processes on which action was done */ +}; + +static vnode_t sd_logvp = NULLVP; +static off_t sd_log_offset = 0; + -void -boot(paniced, howto, command) - int paniced, howto; - char *command; +static int sd_filt1(proc_t, void *); +static int sd_filt2(proc_t, void *); +static int sd_callback1(proc_t p, void * arg); +static int sd_callback2(proc_t p, void * arg); +static int sd_callback3(proc_t p, void * arg); + +int +boot(int paniced, int howto, char *command) { - register int i; - int s; struct proc *p = current_proc(); /* XXX */ int hostboot_option=0; - int funnel_state; - - static void proc_shutdown(); - extern void md_prepare_for_shutdown(int paniced, int howto, char * command); - funnel_state = thread_funnel_set(kernel_flock, TRUE); + if (!OSCompareAndSwap(0, 1, &system_inshutdown)) { + if ( (howto&RB_QUICK) == RB_QUICK) + goto force_reboot; + return (EBUSY); + } + /* + * Temporary hack to notify the power management root domain + * that the system will shut down. + */ + IOSystemShutdownNotification(); md_prepare_for_shutdown(paniced, howto, command); - if ((howto&RB_NOSYNC)==0 && waittime < 0) { + if ((howto&RB_QUICK)==RB_QUICK) { + printf("Quick reboot...\n"); + if ((howto&RB_NOSYNC)==0) { + sync(p, (void *)NULL, (int *)NULL); + } + } + else if ((howto&RB_NOSYNC)==0) { int iter, nbusy; - waittime = 0; - printf("syncing disks... "); /* @@ -95,42 +146,54 @@ boot(paniced, howto, command) /* handle live procs (deallocate their root and current directories). */ proc_shutdown(); +#if CONFIG_AUDIT audit_shutdown(); +#endif + + if (unmountroot_pre_hook != NULL) + unmountroot_pre_hook(); sync(p, (void *)NULL, (int *)NULL); - /* Release vnodes from the VM object cache */ - ubc_unmountall(); + /* + * Now that all processes have been terminated and system is + * sync'ed up, suspend init + */ + + if (initproc && p != initproc) + task_suspend(initproc->task); - IOSleep( 1 * 1000 ); + if (kdebug_enable) + kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace"); /* * Unmount filesystems */ - if (panicstr == 0) - vfs_unmountall(); + vfs_unmountall(); /* Wait for the buffer cache to clean remaining dirty buffers */ - for (iter = 0; iter < 20; iter++) { + for (iter = 0; iter < 100; iter++) { nbusy = count_busy_buffers(); if (nbusy == 0) break; printf("%d ", nbusy); - IOSleep( 4 * nbusy ); + delay_for_interval( 1 * nbusy, 1000 * 1000); } if (nbusy) printf("giving up\n"); else printf("done\n"); } - +#if NETWORKING /* * Can't just use an splnet() here to disable the network * because that will lock out softints which the disk * drivers depend on to finish DMAs. */ if_down_all(); +#endif /* NETWORKING */ +force_reboot: if (howto & RB_POWERDOWN) hostboot_option = HOST_REBOOT_HALT; if (howto & RB_HALT) @@ -138,14 +201,212 @@ boot(paniced, howto, command) if (paniced == RB_PANIC) hostboot_option = HOST_REBOOT_HALT; - if (hostboot_option == HOST_REBOOT_HALT) - IOSleep( 1 * 1000 ); + if (howto & RB_UPSDELAY) { + hostboot_option = HOST_REBOOT_UPSDELAY; + } host_reboot(host_priv_self(), hostboot_option); + /* + * should not be reached + */ + return (0); +} + +static int +sd_openlog(vfs_context_t ctx) +{ + int error = 0; + struct timeval tv; + + /* Open shutdown log */ + if ((error = vnode_open(PROC_SHUTDOWN_LOG, (O_CREAT | FWRITE | O_NOFOLLOW), 0644, 0, &sd_logvp, ctx))) { + printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG, error); + sd_logvp = NULLVP; + return error; + } + + vnode_setsize(sd_logvp, (off_t)0, 0, ctx); + + /* Write a little header */ + microtime(&tv); + sd_log(ctx, "Process shutdown log. Current time is %lu (in seconds).\n\n", tv.tv_sec); + + return 0; +} + +static int +sd_closelog(vfs_context_t ctx) +{ + int error = 0; + if (sd_logvp != NULLVP) { + VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx); + error = vnode_close(sd_logvp, FWRITE, ctx); + } + + return error; +} + +static void +sd_log(vfs_context_t ctx, const char *fmt, ...) +{ + int resid, log_error, len; + char logbuf[100]; + va_list arglist; + + /* If the log isn't open yet, open it */ + if (sd_logvp == NULLVP) { + if (sd_openlog(ctx) != 0) { + /* Couldn't open, we fail out */ + return; + } + } + + va_start(arglist, fmt); + len = vsnprintf(logbuf, sizeof(logbuf), fmt, arglist); + log_error = vn_rdwr(UIO_WRITE, sd_logvp, (caddr_t)logbuf, len, sd_log_offset, + UIO_SYSSPACE, IO_UNIT | IO_NOAUTH, vfs_context_ucred(ctx), &resid, vfs_context_proc(ctx)); + if (log_error == EIO || log_error == 0) { + sd_log_offset += (len - resid); + } + + va_end(arglist); + +} + +static int +sd_filt1(proc_t p, void * args) +{ + proc_t self = current_proc(); + struct sd_filterargs * sf = (struct sd_filterargs *)args; + int delayterm = sf-> delayterm; + int shutdownstate = sf->shutdownstate; + + if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0) + ||(p == self) || (p->p_stat == SZOMB) + || (p->p_shutdownstate != shutdownstate) + ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM)) + || ((p->p_sigcatch & sigmask(SIGTERM))== 0)) { + return(0); + } + else + return(1); +} + + +static int +sd_callback1(proc_t p, void * args) +{ + struct sd_iterargs * sd = (struct sd_iterargs *)args; + int signo = sd->signo; + int setsdstate = sd->setsdstate; + int countproc = sd->countproc; + + proc_lock(p); + p->p_shutdownstate = setsdstate; + if (p->p_stat != SZOMB) { + proc_unlock(p); + if (countproc != 0) { + proc_list_lock(); + p->p_listflag |= P_LIST_EXITCOUNT; + proc_shutdown_exitcount++; + proc_list_unlock(); + } - thread_funnel_set(kernel_flock, FALSE); + psignal(p, signo); + if (countproc != 0) + sd->activecount++; + } else + proc_unlock(p); + return(PROC_RETURNED); } +static int +sd_filt2(proc_t p, void * args) +{ + proc_t self = current_proc(); + struct sd_filterargs * sf = (struct sd_filterargs *)args; + int delayterm = sf-> delayterm; + int shutdownstate = sf->shutdownstate; + + if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0) + ||(p == self) || (p->p_stat == SZOMB) + || (p->p_shutdownstate == shutdownstate) + ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))) { + return(0); + } + else + return(1); +} + +static int +sd_callback2(proc_t p, void * args) +{ + struct sd_iterargs * sd = (struct sd_iterargs *)args; + int signo = sd->signo; + int setsdstate = sd->setsdstate; + int countproc = sd->countproc; + + proc_lock(p); + p->p_shutdownstate = setsdstate; + if (p->p_stat != SZOMB) { + proc_unlock(p); + if (countproc != 0) { + proc_list_lock(); + p->p_listflag |= P_LIST_EXITCOUNT; + proc_shutdown_exitcount++; + proc_list_unlock(); + } + psignal(p, signo); + if (countproc != 0) + sd->activecount++; + } else + proc_unlock(p); + + return(PROC_RETURNED); + +} + +static int +sd_callback3(proc_t p, void * args) +{ + struct sd_iterargs * sd = (struct sd_iterargs *)args; + vfs_context_t ctx = vfs_context_current(); + + int setsdstate = sd->setsdstate; + + proc_lock(p); + p->p_shutdownstate = setsdstate; + if (p->p_stat != SZOMB) { + /* + * NOTE: following code ignores sig_lock and plays + * with exit_thread correctly. This is OK unless we + * are a multiprocessor, in which case I do not + * understand the sig_lock. This needs to be fixed. + * XXX + */ + if (p->exit_thread) { /* someone already doing it */ + proc_unlock(p); + /* give him a chance */ + thread_block(THREAD_CONTINUE_NULL); + } else { + p->exit_thread = current_thread(); + printf("."); + + sd_log(ctx, "%s[%d] had to be forced closed with exit1().\n", p->p_comm, p->p_pid); + + proc_unlock(p); + KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_FRCEXIT) | DBG_FUNC_NONE, + p->p_pid, 0, 1, 0, 0); + sd->activecount++; + exit1(p, 1, (int *)NULL); + } + } else + proc_unlock(p); + + return(PROC_RETURNED); +} + + /* * proc_shutdown() * @@ -159,11 +420,15 @@ boot(paniced, howto, command) */ static void -proc_shutdown() +proc_shutdown(void) { - struct proc *p, *self; - struct vnode **cdirp, **rdirp, *vp; - int restart, i, TERM_catch; + vfs_context_t ctx = vfs_context_current(); + struct proc *p, *self; + int delayterm = 0; + struct sd_filterargs sfargs; + struct sd_iterargs sdargs; + int error = 0; + struct timespec ts; /* * Kill as many procs as we can. (Except ourself...) @@ -171,131 +436,140 @@ proc_shutdown() self = (struct proc *)current_proc(); /* - * Suspend /etc/init + * Signal the init with SIGTERM so that he does not launch + * new processes */ - p = pfind(1); - if (p && p != self) - task_suspend(p->task); /* stop init */ + p = proc_find(1); + if (p && p != self) { + psignal(p, SIGTERM); + } + proc_rele(p); printf("Killing all processes "); +sigterm_loop: /* * send SIGTERM to those procs interested in catching one */ - for (p = allproc.lh_first; p; p = p->p_list.le_next) { - if (((p->p_flag&P_SYSTEM) == 0) && (p->p_pptr->p_pid != 0) && (p != self)) { - if (p->p_sigcatch & sigmask(SIGTERM)) - psignal(p, SIGTERM); - } - } - /* - * now wait for up to 30 seconds to allow those procs catching SIGTERM - * to digest it - * as soon as these procs have exited, we'll continue on to the next step - */ - for (i = 0; i < 300; i++) { - /* - * sleep for a tenth of a second - * and then check to see if the tasks that were sent a - * SIGTERM have exited - */ - IOSleep(100); - TERM_catch = 0; - - for (p = allproc.lh_first; p; p = p->p_list.le_next) { - if (((p->p_flag&P_SYSTEM) == 0) && (p->p_pptr->p_pid != 0) && (p != self)) { - if (p->p_sigcatch & sigmask(SIGTERM)) - TERM_catch++; + sfargs.delayterm = delayterm; + sfargs.shutdownstate = 0; + sdargs.signo = SIGTERM; + sdargs.setsdstate = 1; + sdargs.countproc = 1; + sdargs.activecount = 0; + + error = 0; + /* post a SIGTERM to all that catch SIGTERM and not marked for delay */ + proc_rebootscan(sd_callback1, (void *)&sdargs, sd_filt1, (void *)&sfargs); + + if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) { + proc_list_lock(); + if (proc_shutdown_exitcount != 0) { + /* + * now wait for up to 30 seconds to allow those procs catching SIGTERM + * to digest it + * as soon as these procs have exited, we'll continue on to the next step + */ + ts.tv_sec = 30; + ts.tv_nsec = 0; + error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts); + if (error != 0) { + for (p = allproc.lh_first; p; p = p->p_list.le_next) { + if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT) + p->p_listflag &= ~P_LIST_EXITCOUNT; + } + for (p = zombproc.lh_first; p; p = p->p_list.le_next) { + if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT) + p->p_listflag &= ~P_LIST_EXITCOUNT; + } } + } - if (TERM_catch == 0) - break; + proc_list_unlock(); } - if (TERM_catch) { - /* + if (error == ETIMEDOUT) { + /* * log the names of the unresponsive tasks */ - for (p = allproc.lh_first; p; p = p->p_list.le_next) { - if (((p->p_flag&P_SYSTEM) == 0) && (p->p_pptr->p_pid != 0) && (p != self)) { - if (p->p_sigcatch & sigmask(SIGTERM)) - printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid); + + proc_list_lock(); + + for (p = allproc.lh_first; p; p = p->p_list.le_next) { + if (p->p_shutdownstate == 1) { + printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid); + sd_log(ctx, "%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid); } } - IOSleep(1000 * 5); + + proc_list_unlock(); + + delay_for_interval(1000 * 5, 1000 * 1000); } /* * send a SIGKILL to all the procs still hanging around */ - for (p = allproc.lh_first; p; p = p->p_list.le_next) { - if (((p->p_flag&P_SYSTEM) == 0) && (p->p_pptr->p_pid != 0) && (p != self)) - psignal(p, SIGKILL); - } - /* - * wait for up to 60 seconds to allow these procs to exit normally - */ - for (i = 0; i < 300; i++) { - IOSleep(200); /* double the time from 100 to 200 for NFS requests in particular */ - - for (p = allproc.lh_first; p; p = p->p_list.le_next) { - if (((p->p_flag&P_SYSTEM) == 0) && (p->p_pptr->p_pid != 0) && (p != self)) - break; + sfargs.delayterm = delayterm; + sfargs.shutdownstate = 2; + sdargs.signo = SIGKILL; + sdargs.setsdstate = 2; + sdargs.countproc = 1; + sdargs.activecount = 0; + + /* post a SIGKILL to all that catch SIGTERM and not marked for delay */ + proc_rebootscan(sd_callback2, (void *)&sdargs, sd_filt2, (void *)&sfargs); + + if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) { + proc_list_lock(); + if (proc_shutdown_exitcount != 0) { + /* + * wait for up to 60 seconds to allow these procs to exit normally + * + * History: The delay interval was changed from 100 to 200 + * for NFS requests in particular. + */ + ts.tv_sec = 60; + ts.tv_nsec = 0; + error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts); + if (error != 0) { + for (p = allproc.lh_first; p; p = p->p_list.le_next) { + if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT) + p->p_listflag &= ~P_LIST_EXITCOUNT; + } + for (p = zombproc.lh_first; p; p = p->p_list.le_next) { + if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT) + p->p_listflag &= ~P_LIST_EXITCOUNT; + } + } } - if (!p) - break; + proc_list_unlock(); } /* * if we still have procs that haven't exited, then brute force 'em */ - p = allproc.lh_first; - while (p) { - if ((p->p_flag&P_SYSTEM) || (p->p_pptr->p_pid == 0) || (p == self)) { - p = p->p_list.le_next; - } - else { - /* - * NOTE: following code ignores sig_lock and plays - * with exit_thread correctly. This is OK unless we - * are a multiprocessor, in which case I do not - * understand the sig_lock. This needs to be fixed. - * XXX - */ - if (p->exit_thread) { /* someone already doing it */ - /* give him a chance */ - thread_block(THREAD_CONTINUE_NULL); - } - else { - p->exit_thread = current_act(); - printf("."); - exit1(p, 1, (int *)NULL); - } - p = allproc.lh_first; - } - } + sfargs.delayterm = delayterm; + sfargs.shutdownstate = 3; + sdargs.signo = 0; + sdargs.setsdstate = 3; + sdargs.countproc = 0; + sdargs.activecount = 0; + + /* post a SIGTERM to all that catch SIGTERM and not marked for delay */ + proc_rebootscan(sd_callback3, (void *)&sdargs, sd_filt2, (void *)&sfargs); printf("\n"); - /* - * Forcibly free resources of what's left. - */ - p = allproc.lh_first; - while (p) { - /* - * Close open files and release open-file table. - * This may block! - */ -#ifdef notyet - /* panics on reboot due to "zfree: non-allocated memory in collectable zone" message */ - fdfree(p); -#endif /* notyet */ - p = p->p_list.le_next; + + /* Now start the termination of processes that are marked for delayed termn */ + if (delayterm == 0) { + delayterm = 1; + goto sigterm_loop; } - /* Wait for the reaper thread to run, and clean up what we have done - * before we proceed with the hardcore shutdown. This reduces the race - * between kill_tasks and the reaper thread. - */ - /* thread_wakeup(&reaper_queue); */ - /* IOSleep( 1 * 1000); */ + + sd_closelog(ctx); + + /* drop the ref on initproc */ + proc_rele(initproc); printf("continuing\n"); }