bsd/kern/kern_shutdown.c

   1 /*
   2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   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.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  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.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  *      File:   bsd/kern/kern_shutdown.c
  30  *
  31  *      Copyright (C) 1989, NeXT, Inc.
  32  *
  33  */
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/vm.h>
  39 #include <sys/proc_internal.h>
  40 #include <sys/user.h>
  41 #include <sys/reboot.h>
  42 #include <sys/conf.h>
  43 #include <sys/vnode_internal.h>
  44 #include <sys/file_internal.h>
  45 #include <sys/clist.h>
  46 #include <sys/callout.h>
  47 #include <sys/mbuf.h>
  48 #include <sys/msgbuf.h>
  49 #include <sys/ioctl.h>
  50 #include <sys/signal.h>
  51 #include <sys/tty.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>
  59
  60 #include <security/audit/audit.h>
  61
  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
  71 #include <sys/kdebug.h>
  72
  73 int system_inshutdown = 0;
  74
  75 /* XXX should be in a header file somewhere, but isn't */
  76 extern void md_prepare_for_shutdown(int, int, char *);
  77 extern void (*unmountroot_pre_hook)(void);
  78
  79 int     waittime = -1;
  80 unsigned int proc_shutdown_exitcount = 0;
  81
  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
  87 extern void IOSystemShutdownNotification(void);
  88
  89 struct sd_filterargs{
  90         int delayterm;
  91         int shutdownstate;
  92 };
  93
  94
  95 struct sd_iterargs {
  96         int signo;              /* the signal to be posted */
  97         int setsdstate;         /* shutdown state to be set */
  98         int countproc;          /* count processes on action */
  99         int activecount;        /* number of processes on which action was done */
 100 };
 101
 102 static vnode_t sd_logvp = NULLVP;
 103 static off_t sd_log_offset = 0;
 104
 105
 106 static int sd_filt1(proc_t, void *);
 107 static int sd_filt2(proc_t, void *);
 108 static int  sd_callback1(proc_t p, void * arg);
 109 static int  sd_callback2(proc_t p, void * arg);
 110 static int  sd_callback3(proc_t p, void * arg);
 111
 112 void
 113 boot(int paniced, int howto, char *command)
 114 {
 115         struct proc *p = current_proc();        /* XXX */
 116         int hostboot_option=0;
 117         int funnel_state;
 118
 119         system_inshutdown = 1;
 120
 121         funnel_state = thread_funnel_set(kernel_flock, TRUE);
 122
 123        /*
 124         * Temporary hack to notify the power management root domain
 125         * that the system will shut down.
 126         */
 127         IOSystemShutdownNotification();
 128
 129         md_prepare_for_shutdown(paniced, howto, command);
 130
 131         if ((howto&RB_QUICK)==RB_QUICK && waittime < 0) {
 132                 waittime = 0;
 133                 printf("Quick reboot...\n");
 134                 if ((howto&RB_NOSYNC)==0) {
 135                         sync(p, (void *)NULL, (int *)NULL);
 136                 }
 137         }
 138         else if ((howto&RB_NOSYNC)==0 && waittime < 0) {
 139                 int iter, nbusy;
 140
 141                 waittime = 0;
 142
 143                 printf("syncing disks... ");
 144
 145                 /*
 146                  * Release vnodes held by texts before sync.
 147                  */
 148
 149                 /* handle live procs (deallocate their root and current directories). */
 150                 proc_shutdown();
 151
 152 #if CONFIG_AUDIT
 153                 audit_shutdown();
 154 #endif
 155
 156                 if (unmountroot_pre_hook != NULL)
 157                         unmountroot_pre_hook();
 158
 159                 sync(p, (void *)NULL, (int *)NULL);
 160
 161                 /*
 162                  * Now that all processes have been terminated and system is
 163                  * sync'ed up, suspend init
 164                  */
 165
 166                 if (initproc && p != initproc)
 167                         task_suspend(initproc->task);
 168
 169                 if (kdebug_enable)
 170                         kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace");
 171
 172                 /*
 173                  * Unmount filesystems
 174                  */
 175                 vfs_unmountall();
 176
 177                 /* Wait for the buffer cache to clean remaining dirty buffers */
 178                 for (iter = 0; iter < 100; iter++) {
 179                         nbusy = count_busy_buffers();
 180                         if (nbusy == 0)
 181                                 break;
 182                         printf("%d ", nbusy);
 183                         delay_for_interval( 1 * nbusy, 1000 * 1000);
 184                 }
 185                 if (nbusy)
 186                         printf("giving up\n");
 187                 else
 188                         printf("done\n");
 189         }
 190
 191 #if NETWORKING
 192         /*
 193          * Can't just use an splnet() here to disable the network
 194          * because that will lock out softints which the disk
 195          * drivers depend on to finish DMAs.
 196          */
 197         if_down_all();
 198 #endif /* NETWORKING */
 199
 200         if (howto & RB_POWERDOWN)
 201                 hostboot_option = HOST_REBOOT_HALT;
 202         if (howto & RB_HALT)
 203                 hostboot_option = HOST_REBOOT_HALT;
 204         if (paniced == RB_PANIC)
 205                 hostboot_option = HOST_REBOOT_HALT;
 206
 207     if (howto & RB_UPSDELAY) {
 208         hostboot_option = HOST_REBOOT_UPSDELAY;
 209     }
 210
 211         host_reboot(host_priv_self(), hostboot_option);
 212
 213         thread_funnel_set(kernel_flock, FALSE);
 214 }
 215
 216 static int
 217 sd_openlog(vfs_context_t ctx)
 218 {
 219         int error = 0;
 220         struct timeval tv;
 221
 222         /* Open shutdown log */
 223         if ((error = vnode_open(PROC_SHUTDOWN_LOG, (O_CREAT | FWRITE | O_NOFOLLOW), 0644, 0, &sd_logvp, ctx))) {
 224                 printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG, error);
 225                 sd_logvp = NULLVP;
 226                 return error;
 227         }
 228
 229         vnode_setsize(sd_logvp, (off_t)0, 0, ctx);
 230
 231         /* Write a little header */
 232         microtime(&tv);
 233         sd_log(ctx, "Process shutdown log.  Current time is %lu (in seconds).\n\n", tv.tv_sec);
 234
 235         return 0;
 236 }
 237
 238 static int
 239 sd_closelog(vfs_context_t ctx)
 240 {
 241         int error = 0;
 242         if (sd_logvp != NULLVP) {
 243                 VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx);
 244                 error = vnode_close(sd_logvp, FWRITE, ctx);
 245         }
 246
 247         return error;
 248 }
 249
 250 static void
 251 sd_log(vfs_context_t ctx, const char *fmt, ...)
 252 {
 253         int resid, log_error, len;
 254         char logbuf[100];
 255         va_list arglist;
 256
 257         /* If the log isn't open yet, open it */
 258         if (sd_logvp == NULLVP) {
 259                 if (sd_openlog(ctx) != 0) {
 260                         /* Couldn't open, we fail out */
 261                         return;
 262                 }
 263         }
 264
 265         va_start(arglist, fmt);
 266         len = vsnprintf(logbuf, sizeof(logbuf), fmt, arglist);
 267         log_error = vn_rdwr(UIO_WRITE, sd_logvp, (caddr_t)logbuf, len, sd_log_offset,
 268                         UIO_SYSSPACE, IO_UNIT | IO_NOAUTH, vfs_context_ucred(ctx), &resid, vfs_context_proc(ctx));
 269         if (log_error == EIO || log_error == 0) {
 270                 sd_log_offset += (len - resid);
 271         }
 272
 273         va_end(arglist);
 274
 275 }
 276
 277 static int
 278 sd_filt1(proc_t p, void * args)
 279 {
 280         proc_t self = current_proc();
 281         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 282         int delayterm = sf-> delayterm;
 283         int shutdownstate = sf->shutdownstate;
 284
 285         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 286                 ||(p == self) || (p->p_stat == SZOMB)
 287                 || (p->p_shutdownstate != shutdownstate)
 288                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))
 289                 || ((p->p_sigcatch & sigmask(SIGTERM))== 0)) {
 290                         return(0);
 291                 }
 292         else
 293                 return(1);
 294 }
 295
 296
 297 static int
 298 sd_callback1(proc_t p, void * args)
 299 {
 300         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 301         int signo = sd->signo;
 302         int setsdstate = sd->setsdstate;
 303         int countproc = sd->countproc;
 304
 305         proc_lock(p);
 306         p->p_shutdownstate = setsdstate;
 307         if (p->p_stat != SZOMB) {
 308                 proc_unlock(p);
 309                 if (countproc != 0) {
 310                         proc_list_lock();
 311                         p->p_listflag |= P_LIST_EXITCOUNT;
 312                         proc_shutdown_exitcount++;
 313                         proc_list_unlock();
 314                 }
 315
 316                 psignal(p, signo);
 317                 if (countproc !=  0)
 318                         sd->activecount++;
 319         } else
 320                 proc_unlock(p);
 321         return(PROC_RETURNED);
 322 }
 323
 324 static int
 325 sd_filt2(proc_t p, void * args)
 326 {
 327         proc_t self = current_proc();
 328         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 329         int delayterm = sf-> delayterm;
 330         int shutdownstate = sf->shutdownstate;
 331
 332         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 333                 ||(p == self) || (p->p_stat == SZOMB)
 334                 || (p->p_shutdownstate == shutdownstate)
 335                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))) {
 336                         return(0);
 337                 }
 338         else
 339                 return(1);
 340 }
 341
 342 static int
 343 sd_callback2(proc_t p, void * args)
 344 {
 345         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 346         int signo = sd->signo;
 347         int setsdstate = sd->setsdstate;
 348         int countproc = sd->countproc;
 349
 350         proc_lock(p);
 351         p->p_shutdownstate = setsdstate;
 352         if (p->p_stat != SZOMB) {
 353                 proc_unlock(p);
 354                 if (countproc !=  0) {
 355                         proc_list_lock();
 356                         p->p_listflag |= P_LIST_EXITCOUNT;
 357                         proc_shutdown_exitcount++;
 358                         proc_list_unlock();
 359                 }
 360                 psignal(p, signo);
 361                 if (countproc !=  0)
 362                         sd->activecount++;
 363         } else
 364                 proc_unlock(p);
 365
 366         return(PROC_RETURNED);
 367
 368 }
 369
 370 static int
 371 sd_callback3(proc_t p, void * args)
 372 {
 373         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 374         vfs_context_t ctx = vfs_context_current();
 375
 376         int setsdstate = sd->setsdstate;
 377
 378         proc_lock(p);
 379         p->p_shutdownstate = setsdstate;
 380         if (p->p_stat != SZOMB) {
 381                /*
 382                 * NOTE: following code ignores sig_lock and plays
 383                 * with exit_thread correctly.  This is OK unless we
 384                 * are a multiprocessor, in which case I do not
 385                 * understand the sig_lock.  This needs to be fixed.
 386                 * XXX
 387                 */
 388                 if (p->exit_thread) {   /* someone already doing it */
 389                         proc_unlock(p);
 390                         /* give him a chance */
 391                         thread_block(THREAD_CONTINUE_NULL);
 392                 } else {
 393                         p->exit_thread = current_thread();
 394                         printf(".");
 395
 396                         sd_log(ctx, "%s[%d] had to be forced closed with exit1().\n", p->p_comm, p->p_pid);
 397
 398                         proc_unlock(p);
 399                         KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_FRCEXIT) | DBG_FUNC_NONE,
 400                                               p->p_pid, 0, 1, 0, 0);
 401                         sd->activecount++;
 402                         exit1(p, 1, (int *)NULL);
 403                 }
 404         } else
 405                 proc_unlock(p);
 406
 407         return(PROC_RETURNED);
 408 }
 409
 410
 411 /*
 412  * proc_shutdown()
 413  *
 414  *      Shutdown down proc system (release references to current and root
 415  *      dirs for each process).
 416  *
 417  * POSIX modifications:
 418  *
 419  *      For POSIX fcntl() file locking call vno_lockrelease() on
 420  *      the file to release all of its record locks, if any.
 421  */
 422
 423 static void
 424 proc_shutdown(void)
 425 {
 426         vfs_context_t ctx = vfs_context_current();
 427         struct proc *p, *self;
 428         int delayterm = 0;
 429         struct sd_filterargs sfargs;
 430         struct sd_iterargs sdargs;
 431         int error = 0;
 432         struct timespec ts;
 433
 434         /*
 435          *      Kill as many procs as we can.  (Except ourself...)
 436          */
 437         self = (struct proc *)current_proc();
 438
 439         /*
 440          * Signal the init with SIGTERM so that he does not launch
 441          * new processes
 442          */
 443         p = proc_find(1);
 444         if (p && p != self) {
 445                 psignal(p, SIGTERM);
 446         }
 447         proc_rele(p);
 448
 449         printf("Killing all processes ");
 450
 451 sigterm_loop:
 452         /*
 453          * send SIGTERM to those procs interested in catching one
 454          */
 455         sfargs.delayterm = delayterm;
 456         sfargs.shutdownstate = 0;
 457         sdargs.signo = SIGTERM;
 458         sdargs.setsdstate = 1;
 459         sdargs.countproc = 1;
 460         sdargs.activecount = 0;
 461
 462         error = 0;
 463         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 464         proc_rebootscan(sd_callback1, (void *)&sdargs, sd_filt1, (void *)&sfargs);
 465
 466         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 467                 proc_list_lock();
 468                 if (proc_shutdown_exitcount != 0) {
 469                         /*
 470                         * now wait for up to 30 seconds to allow those procs catching SIGTERM
 471                         * to digest it
 472                         * as soon as these procs have exited, we'll continue on to the next step
 473                         */
 474                         ts.tv_sec = 30;
 475                         ts.tv_nsec = 0;
 476                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 477                         if (error != 0) {
 478                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 479                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 480                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 481                                 }
 482                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 483                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 484                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 485                                 }
 486                         }
 487
 488                 }
 489                 proc_list_unlock();
 490         }
 491         if (error == ETIMEDOUT) {
 492                 /*
 493                  * log the names of the unresponsive tasks
 494                  */
 495
 496
 497                 proc_list_lock();
 498
 499                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 500                         if (p->p_shutdownstate == 1) {
 501                                 printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 502                                 sd_log(ctx, "%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 503                         }
 504                 }
 505
 506                 proc_list_unlock();
 507
 508                 delay_for_interval(1000 * 5, 1000 * 1000);
 509         }
 510
 511         /*
 512          * send a SIGKILL to all the procs still hanging around
 513          */
 514         sfargs.delayterm = delayterm;
 515         sfargs.shutdownstate = 2;
 516         sdargs.signo = SIGKILL;
 517         sdargs.setsdstate = 2;
 518         sdargs.countproc = 1;
 519         sdargs.activecount = 0;
 520
 521         /* post a SIGKILL to all that catch SIGTERM and not marked for delay */
 522         proc_rebootscan(sd_callback2, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 523
 524         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 525                 proc_list_lock();
 526                 if (proc_shutdown_exitcount != 0) {
 527                         /*
 528                         * wait for up to 60 seconds to allow these procs to exit normally
 529                         *
 530                         * History:      The delay interval was changed from 100 to 200
 531                         *               for NFS requests in particular.
 532                         */
 533                         ts.tv_sec = 60;
 534                         ts.tv_nsec = 0;
 535                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 536                         if (error != 0) {
 537                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 538                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 539                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 540                                 }
 541                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 542                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 543                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 544                                 }
 545                         }
 546                 }
 547                 proc_list_unlock();
 548         }
 549
 550         /*
 551          * if we still have procs that haven't exited, then brute force 'em
 552          */
 553         sfargs.delayterm = delayterm;
 554         sfargs.shutdownstate = 3;
 555         sdargs.signo = 0;
 556         sdargs.setsdstate = 3;
 557         sdargs.countproc = 0;
 558         sdargs.activecount = 0;
 559
 560         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 561         proc_rebootscan(sd_callback3, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 562         printf("\n");
 563
 564         /* Now start the termination of processes that are marked for delayed termn */
 565         if (delayterm == 0) {
 566                 delayterm = 1;
 567                 goto  sigterm_loop;
 568         }
 569
 570         sd_closelog(ctx);
 571
 572         /* drop the ref on initproc */
 573         proc_rele(initproc);
 574         printf("continuing\n");
 575 }
 576