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 #include <libkern/OSAtomic.h>
  71
  72 #include <sys/kdebug.h>
  73
  74 uint32_t system_inshutdown = 0;
  75
  76 /* XXX should be in a header file somewhere, but isn't */
  77 extern void (*unmountroot_pre_hook)(void);
  78
  79 unsigned int proc_shutdown_exitcount = 0;
  80
  81 static int  sd_openlog(vfs_context_t);
  82 static int  sd_closelog(vfs_context_t);
  83 static void sd_log(vfs_context_t, const char *, ...);
  84 static void proc_shutdown(void);
  85 static void kernel_hwm_panic_info(void);
  86 extern void IOSystemShutdownNotification(void);
  87
  88 struct sd_filterargs{
  89         int delayterm;
  90         int shutdownstate;
  91 };
  92
  93
  94 struct sd_iterargs {
  95         int signo;              /* the signal to be posted */
  96         int setsdstate;         /* shutdown state to be set */
  97         int countproc;          /* count processes on action */
  98         int activecount;        /* number of processes on which action was done */
  99 };
 100
 101 static vnode_t sd_logvp = NULLVP;
 102 static off_t sd_log_offset = 0;
 103
 104
 105 static int sd_filt1(proc_t, void *);
 106 static int sd_filt2(proc_t, void *);
 107 static int  sd_callback1(proc_t p, void * arg);
 108 static int  sd_callback2(proc_t p, void * arg);
 109 static int  sd_callback3(proc_t p, void * arg);
 110
 111 extern boolean_t panic_include_zprint;
 112 extern vm_offset_t panic_kext_memory_info;
 113 extern vm_size_t panic_kext_memory_size;
 114
 115 static void
 116 kernel_hwm_panic_info(void)
 117 {
 118         mach_memory_info_t      *memory_info;
 119         unsigned int            num_sites;
 120         kern_return_t           kr;
 121
 122         panic_include_zprint = TRUE;
 123         panic_kext_memory_info = 0;
 124         panic_kext_memory_size = 0;
 125
 126         num_sites = VM_KERN_MEMORY_COUNT + VM_KERN_COUNTER_COUNT;
 127         panic_kext_memory_size = round_page(num_sites * sizeof(mach_zone_info_t));
 128
 129         kr = kmem_alloc(kernel_map, (vm_offset_t *) &panic_kext_memory_info, panic_kext_memory_size, VM_KERN_MEMORY_OSFMK);
 130         if (kr != KERN_SUCCESS) {
 131                 panic_kext_memory_info = 0;
 132                 return;
 133         }
 134         memory_info = (mach_memory_info_t *)panic_kext_memory_info;
 135         vm_page_diagnose(memory_info, num_sites);
 136         return;
 137 }
 138
 139 int
 140 reboot_kernel(int howto, char *message)
 141 {
 142         int hostboot_option=0;
 143
 144         if (!OSCompareAndSwap(0, 1, &system_inshutdown)) {
 145                 if ( (howto&RB_QUICK) == RB_QUICK)
 146                         goto force_reboot;
 147                 return (EBUSY);
 148         }
 149         /*
 150          * Temporary hack to notify the power management root domain
 151          * that the system will shut down.
 152          */
 153         IOSystemShutdownNotification();
 154
 155         if ((howto&RB_QUICK)==RB_QUICK) {
 156                 printf("Quick reboot...\n");
 157                 if ((howto&RB_NOSYNC)==0) {
 158                         sync((proc_t)NULL, (void *)NULL, (int *)NULL);
 159                 }
 160         }
 161         else if ((howto&RB_NOSYNC)==0) {
 162                 int iter, nbusy;
 163
 164                 printf("syncing disks... ");
 165
 166                 /*
 167                  * Release vnodes held by texts before sync.
 168                  */
 169
 170                 /* handle live procs (deallocate their root and current directories), suspend initproc */
 171                 proc_shutdown();
 172
 173 #if CONFIG_AUDIT
 174                 audit_shutdown();
 175 #endif
 176
 177                 if (unmountroot_pre_hook != NULL)
 178                         unmountroot_pre_hook();
 179
 180                 sync((proc_t)NULL, (void *)NULL, (int *)NULL);
 181
 182                 if (kdebug_enable)
 183                         kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace");
 184
 185                 /*
 186                  * Unmount filesystems
 187                  */
 188                 vfs_unmountall();
 189
 190                 /* Wait for the buffer cache to clean remaining dirty buffers */
 191                 for (iter = 0; iter < 100; iter++) {
 192                         nbusy = count_busy_buffers();
 193                         if (nbusy == 0)
 194                                 break;
 195                         printf("%d ", nbusy);
 196                         delay_for_interval( 1 * nbusy, 1000 * 1000);
 197                 }
 198                 if (nbusy)
 199                         printf("giving up\n");
 200                 else
 201                         printf("done\n");
 202         }
 203 #if NETWORKING
 204         /*
 205          * Can't just use an splnet() here to disable the network
 206          * because that will lock out softints which the disk
 207          * drivers depend on to finish DMAs.
 208          */
 209         if_down_all();
 210 #endif /* NETWORKING */
 211
 212 force_reboot:
 213
 214         if (howto & RB_PANIC) {
 215                 if (strncmp(message, "Kernel memory has exceeded limits", 33) == 0) {
 216                         kernel_hwm_panic_info();
 217                 }
 218                 panic ("userspace panic: %s", message);
 219         }
 220
 221         if (howto & RB_POWERDOWN)
 222                 hostboot_option = HOST_REBOOT_HALT;
 223         if (howto & RB_HALT)
 224                 hostboot_option = HOST_REBOOT_HALT;
 225
 226         if (howto & RB_UPSDELAY) {
 227                 hostboot_option = HOST_REBOOT_UPSDELAY;
 228         }
 229
 230         host_reboot(host_priv_self(), hostboot_option);
 231         /*
 232          * should not be reached
 233          */
 234         return (0);
 235 }
 236
 237 static int
 238 sd_openlog(vfs_context_t ctx)
 239 {
 240         int error = 0;
 241         struct timeval tv;
 242
 243         /* Open shutdown log */
 244         if ((error = vnode_open(PROC_SHUTDOWN_LOG, (O_CREAT | FWRITE | O_NOFOLLOW), 0644, 0, &sd_logvp, ctx))) {
 245                 printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG, error);
 246                 sd_logvp = NULLVP;
 247                 return error;
 248         }
 249
 250         vnode_setsize(sd_logvp, (off_t)0, 0, ctx);
 251
 252         /* Write a little header */
 253         microtime(&tv);
 254         sd_log(ctx, "Process shutdown log.  Current time is %lu (in seconds).\n\n", tv.tv_sec);
 255
 256         return 0;
 257 }
 258
 259 static int
 260 sd_closelog(vfs_context_t ctx)
 261 {
 262         int error = 0;
 263         if (sd_logvp != NULLVP) {
 264                 VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx);
 265                 error = vnode_close(sd_logvp, FWRITE, ctx);
 266         }
 267
 268         return error;
 269 }
 270
 271 static void
 272 sd_log(vfs_context_t ctx, const char *fmt, ...)
 273 {
 274         int resid, log_error, len;
 275         char logbuf[100];
 276         va_list arglist;
 277
 278         /* If the log isn't open yet, open it */
 279         if (sd_logvp == NULLVP) {
 280                 if (sd_openlog(ctx) != 0) {
 281                         /* Couldn't open, we fail out */
 282                         return;
 283                 }
 284         }
 285
 286         va_start(arglist, fmt);
 287         len = vsnprintf(logbuf, sizeof(logbuf), fmt, arglist);
 288         log_error = vn_rdwr(UIO_WRITE, sd_logvp, (caddr_t)logbuf, len, sd_log_offset,
 289                         UIO_SYSSPACE, IO_UNIT | IO_NOAUTH, vfs_context_ucred(ctx), &resid, vfs_context_proc(ctx));
 290         if (log_error == EIO || log_error == 0) {
 291                 sd_log_offset += (len - resid);
 292         }
 293
 294         va_end(arglist);
 295
 296 }
 297
 298 static int
 299 sd_filt1(proc_t p, void * args)
 300 {
 301         proc_t self = current_proc();
 302         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 303         int delayterm = sf-> delayterm;
 304         int shutdownstate = sf->shutdownstate;
 305
 306         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 307                 ||(p == self) || (p->p_stat == SZOMB)
 308                 || (p->p_shutdownstate != shutdownstate)
 309                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))
 310                 || ((p->p_sigcatch & sigmask(SIGTERM))== 0)) {
 311                         return(0);
 312                 }
 313         else
 314                 return(1);
 315 }
 316
 317
 318 static int
 319 sd_callback1(proc_t p, void * args)
 320 {
 321         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 322         int signo = sd->signo;
 323         int setsdstate = sd->setsdstate;
 324         int countproc = sd->countproc;
 325
 326         proc_lock(p);
 327         p->p_shutdownstate = setsdstate;
 328         if (p->p_stat != SZOMB) {
 329                 proc_unlock(p);
 330                 if (countproc != 0) {
 331                         proc_list_lock();
 332                         p->p_listflag |= P_LIST_EXITCOUNT;
 333                         proc_shutdown_exitcount++;
 334                         proc_list_unlock();
 335                 }
 336
 337                 psignal(p, signo);
 338                 if (countproc !=  0)
 339                         sd->activecount++;
 340         } else
 341                 proc_unlock(p);
 342         return(PROC_RETURNED);
 343 }
 344
 345 static int
 346 sd_filt2(proc_t p, void * args)
 347 {
 348         proc_t self = current_proc();
 349         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 350         int delayterm = sf-> delayterm;
 351         int shutdownstate = sf->shutdownstate;
 352
 353         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 354                 ||(p == self) || (p->p_stat == SZOMB)
 355                 || (p->p_shutdownstate == shutdownstate)
 356                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))) {
 357                         return(0);
 358                 }
 359         else
 360                 return(1);
 361 }
 362
 363 static int
 364 sd_callback2(proc_t p, void * args)
 365 {
 366         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 367         int signo = sd->signo;
 368         int setsdstate = sd->setsdstate;
 369         int countproc = sd->countproc;
 370
 371         proc_lock(p);
 372         p->p_shutdownstate = setsdstate;
 373         if (p->p_stat != SZOMB) {
 374                 proc_unlock(p);
 375                 if (countproc !=  0) {
 376                         proc_list_lock();
 377                         p->p_listflag |= P_LIST_EXITCOUNT;
 378                         proc_shutdown_exitcount++;
 379                         proc_list_unlock();
 380                 }
 381                 psignal(p, signo);
 382                 if (countproc !=  0)
 383                         sd->activecount++;
 384         } else
 385                 proc_unlock(p);
 386
 387         return(PROC_RETURNED);
 388
 389 }
 390
 391 static int
 392 sd_callback3(proc_t p, void * args)
 393 {
 394         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 395         vfs_context_t ctx = vfs_context_current();
 396
 397         int setsdstate = sd->setsdstate;
 398
 399         proc_lock(p);
 400         p->p_shutdownstate = setsdstate;
 401         if (p->p_stat != SZOMB) {
 402                /*
 403                 * NOTE: following code ignores sig_lock and plays
 404                 * with exit_thread correctly.  This is OK unless we
 405                 * are a multiprocessor, in which case I do not
 406                 * understand the sig_lock.  This needs to be fixed.
 407                 * XXX
 408                 */
 409                 if (p->exit_thread) {   /* someone already doing it */
 410                         proc_unlock(p);
 411                         /* give him a chance */
 412                         thread_block(THREAD_CONTINUE_NULL);
 413                 } else {
 414                         p->exit_thread = current_thread();
 415                         printf(".");
 416
 417                         sd_log(ctx, "%s[%d] had to be forced closed with exit1().\n", p->p_comm, p->p_pid);
 418
 419                         proc_unlock(p);
 420                         KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_FRCEXIT) | DBG_FUNC_NONE,
 421                                               p->p_pid, 0, 1, 0, 0);
 422                         sd->activecount++;
 423                         exit1(p, 1, (int *)NULL);
 424                 }
 425         } else
 426                 proc_unlock(p);
 427
 428         return(PROC_RETURNED);
 429 }
 430
 431
 432 /*
 433  * proc_shutdown()
 434  *
 435  *      Shutdown down proc system (release references to current and root
 436  *      dirs for each process).
 437  *
 438  * POSIX modifications:
 439  *
 440  *      For POSIX fcntl() file locking call vno_lockrelease() on
 441  *      the file to release all of its record locks, if any.
 442  */
 443
 444 static void
 445 proc_shutdown(void)
 446 {
 447         vfs_context_t ctx = vfs_context_current();
 448         struct proc *p, *self;
 449         int delayterm = 0;
 450         struct sd_filterargs sfargs;
 451         struct sd_iterargs sdargs;
 452         int error = 0;
 453         struct timespec ts;
 454
 455         /*
 456          *      Kill as many procs as we can.  (Except ourself...)
 457          */
 458         self = (struct proc *)current_proc();
 459
 460         /*
 461          * Signal the init with SIGTERM so that he does not launch
 462          * new processes
 463          */
 464         p = proc_find(1);
 465         if (p && p != self) {
 466                 psignal(p, SIGTERM);
 467         }
 468         proc_rele(p);
 469
 470         printf("Killing all processes ");
 471
 472 sigterm_loop:
 473         /*
 474          * send SIGTERM to those procs interested in catching one
 475          */
 476         sfargs.delayterm = delayterm;
 477         sfargs.shutdownstate = 0;
 478         sdargs.signo = SIGTERM;
 479         sdargs.setsdstate = 1;
 480         sdargs.countproc = 1;
 481         sdargs.activecount = 0;
 482
 483         error = 0;
 484         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 485         proc_rebootscan(sd_callback1, (void *)&sdargs, sd_filt1, (void *)&sfargs);
 486
 487         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 488                 proc_list_lock();
 489                 if (proc_shutdown_exitcount != 0) {
 490                         /*
 491                         * now wait for up to 30 seconds to allow those procs catching SIGTERM
 492                         * to digest it
 493                         * as soon as these procs have exited, we'll continue on to the next step
 494                         */
 495                         ts.tv_sec = 30;
 496                         ts.tv_nsec = 0;
 497                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 498                         if (error != 0) {
 499                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 500                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 501                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 502                                 }
 503                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 504                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 505                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 506                                 }
 507                         }
 508
 509                 }
 510                 proc_list_unlock();
 511         }
 512         if (error == ETIMEDOUT) {
 513                 /*
 514                  * log the names of the unresponsive tasks
 515                  */
 516
 517
 518                 proc_list_lock();
 519
 520                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 521                         if (p->p_shutdownstate == 1) {
 522                                 printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 523                                 sd_log(ctx, "%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 524                         }
 525                 }
 526
 527                 proc_list_unlock();
 528
 529                 delay_for_interval(1000 * 5, 1000 * 1000);
 530         }
 531
 532         /*
 533          * send a SIGKILL to all the procs still hanging around
 534          */
 535         sfargs.delayterm = delayterm;
 536         sfargs.shutdownstate = 2;
 537         sdargs.signo = SIGKILL;
 538         sdargs.setsdstate = 2;
 539         sdargs.countproc = 1;
 540         sdargs.activecount = 0;
 541
 542         /* post a SIGKILL to all that catch SIGTERM and not marked for delay */
 543         proc_rebootscan(sd_callback2, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 544
 545         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 546                 proc_list_lock();
 547                 if (proc_shutdown_exitcount != 0) {
 548                         /*
 549                         * wait for up to 60 seconds to allow these procs to exit normally
 550                         *
 551                         * History:      The delay interval was changed from 100 to 200
 552                         *               for NFS requests in particular.
 553                         */
 554                         ts.tv_sec = 60;
 555                         ts.tv_nsec = 0;
 556                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 557                         if (error != 0) {
 558                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 559                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 560                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 561                                 }
 562                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 563                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 564                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 565                                 }
 566                         }
 567                 }
 568                 proc_list_unlock();
 569         }
 570
 571         /*
 572          * if we still have procs that haven't exited, then brute force 'em
 573          */
 574         sfargs.delayterm = delayterm;
 575         sfargs.shutdownstate = 3;
 576         sdargs.signo = 0;
 577         sdargs.setsdstate = 3;
 578         sdargs.countproc = 0;
 579         sdargs.activecount = 0;
 580
 581         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 582         proc_rebootscan(sd_callback3, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 583         printf("\n");
 584
 585         /* Now start the termination of processes that are marked for delayed termn */
 586         if (delayterm == 0) {
 587                 delayterm = 1;
 588                 goto  sigterm_loop;
 589         }
 590
 591         sd_closelog(ctx);
 592
 593         /*
 594          * Now that all other processes have been terminated, suspend init
 595          */
 596         task_suspend_internal(initproc->task);
 597
 598         /* drop the ref on initproc */
 599         proc_rele(initproc);
 600         printf("continuing\n");
 601 }
 602