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 #if DEVELOPMENT || DEBUG
  88 extern boolean_t kdp_has_polled_corefile(void);
  89 #endif /* DEVELOPMENT || DEBUG */
  90
  91 struct sd_filterargs{
  92         int delayterm;
  93         int shutdownstate;
  94 };
  95
  96
  97 struct sd_iterargs {
  98         int signo;              /* the signal to be posted */
  99         int setsdstate;         /* shutdown state to be set */
 100         int countproc;          /* count processes on action */
 101         int activecount;        /* number of processes on which action was done */
 102 };
 103
 104 static vnode_t sd_logvp = NULLVP;
 105 static off_t sd_log_offset = 0;
 106
 107
 108 static int sd_filt1(proc_t, void *);
 109 static int sd_filt2(proc_t, void *);
 110 static int sd_callback1(proc_t p, void * arg);
 111 static int sd_callback2(proc_t p, void * arg);
 112 static int sd_callback3(proc_t p, void * arg);
 113
 114 extern boolean_t panic_include_zprint;
 115 extern vm_offset_t panic_kext_memory_info;
 116 extern vm_size_t panic_kext_memory_size;
 117
 118 static void
 119 kernel_hwm_panic_info(void)
 120 {
 121         mach_memory_info_t      *memory_info;
 122         unsigned int            num_sites;
 123         kern_return_t           kr;
 124
 125         panic_include_zprint = TRUE;
 126         panic_kext_memory_info = 0;
 127         panic_kext_memory_size = 0;
 128
 129         num_sites = VM_KERN_MEMORY_COUNT + VM_KERN_COUNTER_COUNT;
 130         panic_kext_memory_size = round_page(num_sites * sizeof(mach_zone_info_t));
 131
 132         kr = kmem_alloc(kernel_map, (vm_offset_t *) &panic_kext_memory_info, panic_kext_memory_size, VM_KERN_MEMORY_OSFMK);
 133         if (kr != KERN_SUCCESS) {
 134                 panic_kext_memory_info = 0;
 135                 return;
 136         }
 137         memory_info = (mach_memory_info_t *)panic_kext_memory_info;
 138         vm_page_diagnose(memory_info, num_sites, 0);
 139         return;
 140 }
 141
 142 int
 143 reboot_kernel(int howto, char *message)
 144 {
 145         int hostboot_option=0;
 146
 147         if (!OSCompareAndSwap(0, 1, &system_inshutdown)) {
 148                 if ( (howto&RB_QUICK) == RB_QUICK)
 149                         goto force_reboot;
 150                 return (EBUSY);
 151         }
 152         /*
 153          * Temporary hack to notify the power management root domain
 154          * that the system will shut down.
 155          */
 156         IOSystemShutdownNotification();
 157
 158         if ((howto&RB_QUICK)==RB_QUICK) {
 159                 printf("Quick reboot...\n");
 160                 if ((howto&RB_NOSYNC)==0) {
 161                         sync((proc_t)NULL, (void *)NULL, (int *)NULL);
 162                 }
 163         }
 164         else if ((howto&RB_NOSYNC)==0) {
 165                 int iter, nbusy;
 166
 167                 printf("syncing disks... ");
 168
 169                 /*
 170                  * Release vnodes held by texts before sync.
 171                  */
 172
 173                 /* handle live procs (deallocate their root and current directories), suspend initproc */
 174                 proc_shutdown();
 175
 176 #if CONFIG_AUDIT
 177                 audit_shutdown();
 178 #endif
 179
 180                 if (unmountroot_pre_hook != NULL)
 181                         unmountroot_pre_hook();
 182
 183                 sync((proc_t)NULL, (void *)NULL, (int *)NULL);
 184
 185                 if (kdebug_enable)
 186                         kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace");
 187
 188                 /*
 189                  * Unmount filesystems
 190                  */
 191
 192 #if DEVELOPMENT || DEBUG
 193                 if (!(howto & RB_PANIC) || !kdp_has_polled_corefile())
 194 #endif /* DEVELOPMENT || DEBUG */
 195                 {
 196                         vfs_unmountall();
 197                 }
 198
 199                 /* Wait for the buffer cache to clean remaining dirty buffers */
 200                 for (iter = 0; iter < 100; iter++) {
 201                         nbusy = count_busy_buffers();
 202                         if (nbusy == 0)
 203                                 break;
 204                         printf("%d ", nbusy);
 205                         delay_for_interval( 1 * nbusy, 1000 * 1000);
 206                 }
 207                 if (nbusy)
 208                         printf("giving up\n");
 209                 else
 210                         printf("done\n");
 211         }
 212 #if NETWORKING
 213         /*
 214          * Can't just use an splnet() here to disable the network
 215          * because that will lock out softints which the disk
 216          * drivers depend on to finish DMAs.
 217          */
 218         if_down_all();
 219 #endif /* NETWORKING */
 220
 221 force_reboot:
 222
 223         if (howto & RB_PANIC) {
 224                 if (strncmp(message, "Kernel memory has exceeded limits", 33) == 0) {
 225                         kernel_hwm_panic_info();
 226                 }
 227                 panic ("userspace panic: %s", message);
 228         }
 229
 230         if (howto & RB_POWERDOWN)
 231                 hostboot_option = HOST_REBOOT_HALT;
 232         if (howto & RB_HALT)
 233                 hostboot_option = HOST_REBOOT_HALT;
 234
 235         if (howto & RB_UPSDELAY) {
 236                 hostboot_option = HOST_REBOOT_UPSDELAY;
 237         }
 238
 239         host_reboot(host_priv_self(), hostboot_option);
 240         /*
 241          * should not be reached
 242          */
 243         return (0);
 244 }
 245
 246 static int
 247 sd_openlog(vfs_context_t ctx)
 248 {
 249         int error = 0;
 250         struct timeval tv;
 251
 252         /* Open shutdown log */
 253         if ((error = vnode_open(PROC_SHUTDOWN_LOG, (O_CREAT | FWRITE | O_NOFOLLOW), 0644, 0, &sd_logvp, ctx))) {
 254                 printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG, error);
 255                 sd_logvp = NULLVP;
 256                 return error;
 257         }
 258
 259         vnode_setsize(sd_logvp, (off_t)0, 0, ctx);
 260
 261         /* Write a little header */
 262         microtime(&tv);
 263         sd_log(ctx, "Process shutdown log.  Current time is %lu (in seconds).\n\n", tv.tv_sec);
 264
 265         return 0;
 266 }
 267
 268 static int
 269 sd_closelog(vfs_context_t ctx)
 270 {
 271         int error = 0;
 272         if (sd_logvp != NULLVP) {
 273                 VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx);
 274                 error = vnode_close(sd_logvp, FWRITE, ctx);
 275         }
 276
 277         return error;
 278 }
 279
 280 static void
 281 sd_log(vfs_context_t ctx, const char *fmt, ...)
 282 {
 283         int resid, log_error, len;
 284         char logbuf[100];
 285         va_list arglist;
 286
 287         /* If the log isn't open yet, open it */
 288         if (sd_logvp == NULLVP) {
 289                 if (sd_openlog(ctx) != 0) {
 290                         /* Couldn't open, we fail out */
 291                         return;
 292                 }
 293         }
 294
 295         va_start(arglist, fmt);
 296         len = vsnprintf(logbuf, sizeof(logbuf), fmt, arglist);
 297         log_error = vn_rdwr(UIO_WRITE, sd_logvp, (caddr_t)logbuf, len, sd_log_offset,
 298                         UIO_SYSSPACE, IO_UNIT | IO_NOAUTH, vfs_context_ucred(ctx), &resid, vfs_context_proc(ctx));
 299         if (log_error == EIO || log_error == 0) {
 300                 sd_log_offset += (len - resid);
 301         }
 302
 303         va_end(arglist);
 304
 305 }
 306
 307 static int
 308 sd_filt1(proc_t p, void * args)
 309 {
 310         proc_t self = current_proc();
 311         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 312         int delayterm = sf-> delayterm;
 313         int shutdownstate = sf->shutdownstate;
 314
 315         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 316                 ||(p == self) || (p->p_stat == SZOMB)
 317                 || (p->p_shutdownstate != shutdownstate)
 318                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))
 319                 || ((p->p_sigcatch & sigmask(SIGTERM))== 0)) {
 320                         return(0);
 321                 }
 322         else
 323                 return(1);
 324 }
 325
 326
 327 static int
 328 sd_callback1(proc_t p, void * args)
 329 {
 330         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 331         int signo = sd->signo;
 332         int setsdstate = sd->setsdstate;
 333         int countproc = sd->countproc;
 334
 335         proc_lock(p);
 336         p->p_shutdownstate = setsdstate;
 337         if (p->p_stat != SZOMB) {
 338                 proc_unlock(p);
 339                 if (countproc != 0) {
 340                         proc_list_lock();
 341                         p->p_listflag |= P_LIST_EXITCOUNT;
 342                         proc_shutdown_exitcount++;
 343                         proc_list_unlock();
 344                 }
 345
 346                 psignal(p, signo);
 347                 if (countproc !=  0)
 348                         sd->activecount++;
 349         } else {
 350                 proc_unlock(p);
 351         }
 352
 353         return PROC_RETURNED;
 354 }
 355
 356 static int
 357 sd_filt2(proc_t p, void * args)
 358 {
 359         proc_t self = current_proc();
 360         struct sd_filterargs * sf = (struct sd_filterargs *)args;
 361         int delayterm = sf-> delayterm;
 362         int shutdownstate = sf->shutdownstate;
 363
 364         if (((p->p_flag&P_SYSTEM) != 0) || (p->p_ppid == 0)
 365                 ||(p == self) || (p->p_stat == SZOMB)
 366                 || (p->p_shutdownstate == shutdownstate)
 367                 ||((delayterm == 0) && ((p->p_lflag& P_LDELAYTERM) == P_LDELAYTERM))) {
 368                         return(0);
 369                 }
 370         else
 371                 return(1);
 372 }
 373
 374 static int
 375 sd_callback2(proc_t p, void * args)
 376 {
 377         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 378         int signo = sd->signo;
 379         int setsdstate = sd->setsdstate;
 380         int countproc = sd->countproc;
 381
 382         proc_lock(p);
 383         p->p_shutdownstate = setsdstate;
 384         if (p->p_stat != SZOMB) {
 385                 proc_unlock(p);
 386                 if (countproc !=  0) {
 387                         proc_list_lock();
 388                         p->p_listflag |= P_LIST_EXITCOUNT;
 389                         proc_shutdown_exitcount++;
 390                         proc_list_unlock();
 391                 }
 392                 psignal(p, signo);
 393                 if (countproc !=  0)
 394                         sd->activecount++;
 395         } else {
 396                 proc_unlock(p);
 397         }
 398
 399         return PROC_RETURNED;
 400 }
 401
 402 static int
 403 sd_callback3(proc_t p, void * args)
 404 {
 405         struct sd_iterargs * sd = (struct sd_iterargs *)args;
 406         vfs_context_t ctx = vfs_context_current();
 407
 408         int setsdstate = sd->setsdstate;
 409
 410         proc_lock(p);
 411         p->p_shutdownstate = setsdstate;
 412         if (p->p_stat != SZOMB) {
 413                /*
 414                 * NOTE: following code ignores sig_lock and plays
 415                 * with exit_thread correctly.  This is OK unless we
 416                 * are a multiprocessor, in which case I do not
 417                 * understand the sig_lock.  This needs to be fixed.
 418                 * XXX
 419                 */
 420                 if (p->exit_thread) {   /* someone already doing it */
 421                         proc_unlock(p);
 422                         /* give him a chance */
 423                         thread_block(THREAD_CONTINUE_NULL);
 424                 } else {
 425                         p->exit_thread = current_thread();
 426                         printf(".");
 427
 428                         sd_log(ctx, "%s[%d] had to be forced closed with exit1().\n", p->p_comm, p->p_pid);
 429
 430                         proc_unlock(p);
 431                         KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_FRCEXIT) | DBG_FUNC_NONE,
 432                                               p->p_pid, 0, 1, 0, 0);
 433                         sd->activecount++;
 434                         exit1(p, 1, (int *)NULL);
 435                 }
 436         } else {
 437                 proc_unlock(p);
 438         }
 439
 440         return PROC_RETURNED;
 441 }
 442
 443
 444 /*
 445  * proc_shutdown()
 446  *
 447  *      Shutdown down proc system (release references to current and root
 448  *      dirs for each process).
 449  *
 450  * POSIX modifications:
 451  *
 452  *      For POSIX fcntl() file locking call vno_lockrelease() on
 453  *      the file to release all of its record locks, if any.
 454  */
 455
 456 static void
 457 proc_shutdown(void)
 458 {
 459         vfs_context_t ctx = vfs_context_current();
 460         struct proc *p, *self;
 461         int delayterm = 0;
 462         struct sd_filterargs sfargs;
 463         struct sd_iterargs sdargs;
 464         int error = 0;
 465         struct timespec ts;
 466
 467         /*
 468          *      Kill as many procs as we can.  (Except ourself...)
 469          */
 470         self = (struct proc *)current_proc();
 471
 472         /*
 473          * Signal the init with SIGTERM so that he does not launch
 474          * new processes
 475          */
 476         p = proc_find(1);
 477         if (p && p != self) {
 478                 psignal(p, SIGTERM);
 479         }
 480         proc_rele(p);
 481
 482         printf("Killing all processes ");
 483
 484 sigterm_loop:
 485         /*
 486          * send SIGTERM to those procs interested in catching one
 487          */
 488         sfargs.delayterm = delayterm;
 489         sfargs.shutdownstate = 0;
 490         sdargs.signo = SIGTERM;
 491         sdargs.setsdstate = 1;
 492         sdargs.countproc = 1;
 493         sdargs.activecount = 0;
 494
 495         error = 0;
 496         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 497         proc_rebootscan(sd_callback1, (void *)&sdargs, sd_filt1, (void *)&sfargs);
 498
 499         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 500                 proc_list_lock();
 501                 if (proc_shutdown_exitcount != 0) {
 502                         /*
 503                         * now wait for up to 30 seconds to allow those procs catching SIGTERM
 504                         * to digest it
 505                         * as soon as these procs have exited, we'll continue on to the next step
 506                         */
 507                         ts.tv_sec = 30;
 508                         ts.tv_nsec = 0;
 509                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 510                         if (error != 0) {
 511                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 512                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 513                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 514                                 }
 515                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 516                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 517                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 518                                 }
 519                         }
 520
 521                 }
 522                 proc_list_unlock();
 523         }
 524         if (error == ETIMEDOUT) {
 525                 /*
 526                  * log the names of the unresponsive tasks
 527                  */
 528
 529
 530                 proc_list_lock();
 531
 532                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 533                         if (p->p_shutdownstate == 1) {
 534                                 printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 535                                 sd_log(ctx, "%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
 536                         }
 537                 }
 538
 539                 proc_list_unlock();
 540
 541                 delay_for_interval(1000 * 5, 1000 * 1000);
 542         }
 543
 544         /*
 545          * send a SIGKILL to all the procs still hanging around
 546          */
 547         sfargs.delayterm = delayterm;
 548         sfargs.shutdownstate = 2;
 549         sdargs.signo = SIGKILL;
 550         sdargs.setsdstate = 2;
 551         sdargs.countproc = 1;
 552         sdargs.activecount = 0;
 553
 554         /* post a SIGKILL to all that catch SIGTERM and not marked for delay */
 555         proc_rebootscan(sd_callback2, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 556
 557         if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
 558                 proc_list_lock();
 559                 if (proc_shutdown_exitcount != 0) {
 560                         /*
 561                         * wait for up to 60 seconds to allow these procs to exit normally
 562                         *
 563                         * History:      The delay interval was changed from 100 to 200
 564                         *               for NFS requests in particular.
 565                         */
 566                         ts.tv_sec = 60;
 567                         ts.tv_nsec = 0;
 568                         error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
 569                         if (error != 0) {
 570                                 for (p = allproc.lh_first; p; p = p->p_list.le_next) {
 571                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 572                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 573                                 }
 574                                 for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
 575                                         if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
 576                                                 p->p_listflag &= ~P_LIST_EXITCOUNT;
 577                                 }
 578                         }
 579                 }
 580                 proc_list_unlock();
 581         }
 582
 583         /*
 584          * if we still have procs that haven't exited, then brute force 'em
 585          */
 586         sfargs.delayterm = delayterm;
 587         sfargs.shutdownstate = 3;
 588         sdargs.signo = 0;
 589         sdargs.setsdstate = 3;
 590         sdargs.countproc = 0;
 591         sdargs.activecount = 0;
 592
 593         /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
 594         proc_rebootscan(sd_callback3, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 595         printf("\n");
 596
 597         /* Now start the termination of processes that are marked for delayed termn */
 598         if (delayterm == 0) {
 599                 delayterm = 1;
 600                 goto  sigterm_loop;
 601         }
 602
 603         sd_closelog(ctx);
 604
 605         /*
 606          * Now that all other processes have been terminated, suspend init
 607          */
 608         task_suspend_internal(initproc->task);
 609
 610         /* drop the ref on initproc */
 611         proc_rele(initproc);
 612         printf("continuing\n");
 613 }
 614