1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25 
  26 /*      Copyright (c) 1988 AT&T     */
  27 /*        All Rights Reserved   */
  28 
  29 #include <sys/types.h>
  30 #include <sys/param.h>
  31 #include <sys/sysmacros.h>
  32 #include <sys/pcb.h>
  33 #include <sys/systm.h>
  34 #include <sys/signal.h>
  35 #include <sys/cred.h>
  36 #include <sys/user.h>
  37 #include <sys/vfs.h>
  38 #include <sys/vnode.h>
  39 #include <sys/proc.h>
  40 #include <sys/time.h>
  41 #include <sys/file.h>
  42 #include <sys/priocntl.h>
  43 #include <sys/procset.h>
  44 #include <sys/disp.h>
  45 #include <sys/callo.h>
  46 #include <sys/callb.h>
  47 #include <sys/debug.h>
  48 #include <sys/conf.h>
  49 #include <sys/bootconf.h>
  50 #include <sys/utsname.h>
  51 #include <sys/cmn_err.h>
  52 #include <sys/vmparam.h>
  53 #include <sys/modctl.h>
  54 #include <sys/vm.h>
  55 #include <sys/callb.h>
  56 #include <sys/ddi_timer.h>
  57 #include <sys/kmem.h>
  58 #include <sys/vmem.h>
  59 #include <sys/cpuvar.h>
  60 #include <sys/cladm.h>
  61 #include <sys/corectl.h>
  62 #include <sys/exec.h>
  63 #include <sys/syscall.h>
  64 #include <sys/reboot.h>
  65 #include <sys/task.h>
  66 #include <sys/exacct.h>
  67 #include <sys/autoconf.h>
  68 #include <sys/errorq.h>
  69 #include <sys/class.h>
  70 #include <sys/stack.h>
  71 #include <sys/brand.h>
  72 #include <sys/mmapobj.h>
  73 
  74 #include <vm/as.h>
  75 #include <vm/seg_kmem.h>
  76 #include <sys/dc_ki.h>
  77 
  78 #include <c2/audit.h>
  79 #include <sys/bootprops.h>
  80 
  81 /* well known processes */
  82 proc_t *proc_sched;             /* memory scheduler */
  83 proc_t *proc_init;              /* init */
  84 proc_t *proc_pageout;           /* pageout daemon */
  85 proc_t *proc_fsflush;           /* fsflush daemon */
  86 
  87 pgcnt_t maxmem;         /* Maximum available memory in pages.   */
  88 pgcnt_t freemem;        /* Current available memory in pages.   */
  89 int     interrupts_unleashed;   /* set when we do the first spl0() */
  90 
  91 kmem_cache_t *process_cache;    /* kmem cache for proc structures */
  92 
  93 /*
  94  * Indicates whether the auditing module (c2audit) is loaded. Possible
  95  * values are:
  96  * 0 - c2audit module is excluded in /etc/system and cannot be loaded
  97  * 1 - c2audit module is not loaded but can be anytime
  98  * 2 - c2audit module is loaded
  99  */
 100 int audit_active = C2AUDIT_DISABLED;
 101 
 102 /*
 103  * Process 0's lwp directory and lwpid hash table.
 104  */
 105 lwpdir_t p0_lwpdir[2];
 106 tidhash_t p0_tidhash[2];
 107 lwpent_t p0_lep;
 108 
 109 /*
 110  * Machine-independent initialization code
 111  * Called from cold start routine as
 112  * soon as a stack and segmentation
 113  * have been established.
 114  * Functions:
 115  *      clear and free user core
 116  *      turn on clock
 117  *      hand craft 0th process
 118  *      call all initialization routines
 119  *      fork    - process 0 to schedule
 120  *              - process 1 execute bootstrap
 121  *              - process 2 to page out
 122  *      create system threads
 123  */
 124 
 125 int cluster_bootflags = 0;
 126 
 127 void
 128 cluster_wrapper(void)
 129 {
 130         cluster();
 131         panic("cluster()  returned");
 132 }
 133 
 134 char initname[INITNAME_SZ] = "/sbin/init";      /* also referenced by zone0 */
 135 char initargs[BOOTARGS_MAX] = "";               /* also referenced by zone0 */
 136 
 137 /*
 138  * Construct a stack for init containing the arguments to it, then
 139  * pass control to exec_common.
 140  */
 141 int
 142 exec_init(const char *initpath, const char *args)
 143 {
 144         caddr32_t ucp;
 145         caddr32_t *uap;
 146         caddr32_t *argv;
 147         caddr32_t exec_fnamep;
 148         char *scratchargs;
 149         int i, sarg;
 150         size_t argvlen, alen;
 151         boolean_t in_arg;
 152         int argc = 0;
 153         int error = 0, count = 0;
 154         proc_t *p = ttoproc(curthread);
 155         klwp_t *lwp = ttolwp(curthread);
 156         int brand_action;
 157 
 158         if (args == NULL)
 159                 args = "";
 160 
 161         alen = strlen(initpath) + 1 + strlen(args) + 1;
 162         scratchargs = kmem_alloc(alen, KM_SLEEP);
 163         (void) snprintf(scratchargs, alen, "%s %s", initpath, args);
 164 
 165         /*
 166          * We do a quick two state parse of the string to sort out how big
 167          * argc should be.
 168          */
 169         in_arg = B_FALSE;
 170         for (i = 0; i < strlen(scratchargs); i++) {
 171                 if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
 172                         if (in_arg) {
 173                                 in_arg = B_FALSE;
 174                                 argc++;
 175                         }
 176                 } else {
 177                         in_arg = B_TRUE;
 178                 }
 179         }
 180         argvlen = sizeof (caddr32_t) * (argc + 1);
 181         argv = kmem_zalloc(argvlen, KM_SLEEP);
 182 
 183         /*
 184          * We pull off a bit of a hack here.  We work our way through the
 185          * args string, putting nulls at the ends of space delimited tokens
 186          * (boot args don't support quoting at this time).  Then we just
 187          * copy the whole mess to userland in one go.  In other words, we
 188          * transform this: "init -s -r\0" into this on the stack:
 189          *
 190          *      -0x00 \0
 191          *      -0x01 r
 192          *      -0x02 -  <--------.
 193          *      -0x03 \0          |
 194          *      -0x04 s           |
 195          *      -0x05 -  <------. |
 196          *      -0x06 \0        | |
 197          *      -0x07 t         | |
 198          *      -0x08 i         | |
 199          *      -0x09 n         | |
 200          *      -0x0a i  <---.  | |
 201          *      -0x10 NULL   |  | |     (argv[3])
 202          *      -0x14   -----|--|-'     (argv[2])
 203          *      -0x18  ------|--'       (argv[1])
 204          *      -0x1c -------'          (argv[0])
 205          *
 206          * Since we know the value of ucp at the beginning of this process,
 207          * we can trivially compute the argv[] array which we also need to
 208          * place in userland: argv[i] = ucp - sarg(i), where ucp is the
 209          * stack ptr, and sarg is the string index of the start of the
 210          * argument.
 211          */
 212         ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
 213 
 214         argc = 0;
 215         in_arg = B_FALSE;
 216         sarg = 0;
 217 
 218         for (i = 0; i < alen; i++) {
 219                 if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
 220                         if (in_arg == B_TRUE) {
 221                                 in_arg = B_FALSE;
 222                                 scratchargs[i] = '\0';
 223                                 argv[argc++] = ucp - (alen - sarg);
 224                         }
 225                 } else if (in_arg == B_FALSE) {
 226                         in_arg = B_TRUE;
 227                         sarg = i;
 228                 }
 229         }
 230         ucp -= alen;
 231         error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
 232 
 233         uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
 234         uap--;  /* advance to be below the word we're in */
 235         uap -= (argc + 1);      /* advance argc words down, plus one for NULL */
 236         error |= copyout(argv, uap, argvlen);
 237 
 238         if (error != 0) {
 239                 zcmn_err(p->p_zone->zone_id, CE_WARN,
 240                     "Could not construct stack for init.\n");
 241                 kmem_free(argv, argvlen);
 242                 kmem_free(scratchargs, alen);
 243                 return (EFAULT);
 244         }
 245 
 246         exec_fnamep = argv[0];
 247         kmem_free(argv, argvlen);
 248         kmem_free(scratchargs, alen);
 249 
 250         /*
 251          * Point at the arguments.
 252          */
 253         lwp->lwp_ap = lwp->lwp_arg;
 254         lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
 255         lwp->lwp_arg[1] = (uintptr_t)uap;
 256         lwp->lwp_arg[2] = NULL;
 257         curthread->t_post_sys = 1;
 258         curthread->t_sysnum = SYS_execve;
 259 
 260         /*
 261          * If we are executing init from zsched, we may have inherited its
 262          * parent process's signal mask.  Clear it now so that we behave in
 263          * the same way as when started from the global zone.
 264          */
 265         sigemptyset(&curthread->t_hold);
 266 
 267         brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
 268 again:
 269         error = exec_common((const char *)(uintptr_t)exec_fnamep,
 270             (const char **)(uintptr_t)uap, NULL, brand_action);
 271 
 272         /*
 273          * Normally we would just set lwp_argsaved and t_post_sys and
 274          * let post_syscall reset lwp_ap for us.  Unfortunately,
 275          * exec_init isn't always called from a system call.  Instead
 276          * of making a mess of trap_cleanup, we just reset the args
 277          * pointer here.
 278          */
 279         reset_syscall_args();
 280 
 281         switch (error) {
 282         case 0:
 283                 return (0);
 284 
 285         case ENOENT:
 286                 zcmn_err(p->p_zone->zone_id, CE_WARN,
 287                     "exec(%s) failed (file not found).\n", initpath);
 288                 return (ENOENT);
 289 
 290         case EAGAIN:
 291         case EINTR:
 292                 ++count;
 293                 if (count < 5) {
 294                         zcmn_err(p->p_zone->zone_id, CE_WARN,
 295                             "exec(%s) failed with errno %d.  Retrying...\n",
 296                             initpath, error);
 297                         goto again;
 298                 }
 299         }
 300 
 301         zcmn_err(p->p_zone->zone_id, CE_WARN,
 302             "exec(%s) failed with errno %d.", initpath, error);
 303         return (error);
 304 }
 305 
 306 /*
 307  * This routine does all of the common setup for invoking init; global
 308  * and non-global zones employ this routine for the functionality which is
 309  * in common.
 310  *
 311  * This program (init, presumably) must be a 32-bit process.
 312  */
 313 int
 314 start_init_common()
 315 {
 316         proc_t *p = curproc;
 317         ASSERT_STACK_ALIGNED();
 318         p->p_zone->zone_proc_initpid = p->p_pid;
 319 
 320         p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
 321         p->p_usrstack = (caddr_t)USRSTACK32;
 322         p->p_model = DATAMODEL_ILP32;
 323         p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
 324         p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
 325         p->p_stk_ctl = INT32_MAX;
 326 
 327         p->p_as = as_alloc();
 328         p->p_as->a_proc = p;
 329         p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
 330         (void) hat_setup(p->p_as->a_hat, HAT_INIT);
 331 
 332         init_core();
 333 
 334         init_mstate(curthread, LMS_SYSTEM);
 335         return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
 336 }
 337 
 338 /*
 339  * Start the initial user process for the global zone; once running, if
 340  * init should subsequently fail, it will be automatically be caught in the
 341  * exit(2) path, and restarted by restart_init().
 342  */
 343 static void
 344 start_init(void)
 345 {
 346         proc_init = curproc;
 347 
 348         ASSERT(curproc->p_zone->zone_initname != NULL);
 349 
 350         if (start_init_common() != 0)
 351                 halt("unix: Could not start init");
 352         lwp_rtt();
 353 }
 354 
 355 void
 356 main(void)
 357 {
 358         proc_t          *p = ttoproc(curthread);        /* &p0 */
 359         int             (**initptr)();
 360         extern void     sched();
 361         extern void     fsflush();
 362         extern int      (*init_tbl[])();
 363         extern int      (*mp_init_tbl[])();
 364         extern id_t     syscid, defaultcid;
 365         extern int      swaploaded;
 366         extern int      netboot;
 367         extern ib_boot_prop_t *iscsiboot_prop;
 368         extern void     vm_init(void);
 369         extern void     cbe_init_pre(void);
 370         extern void     cbe_init(void);
 371         extern void     clock_tick_init_pre(void);
 372         extern void     clock_tick_init_post(void);
 373         extern void     clock_init(void);
 374         extern void     physio_bufs_init(void);
 375         extern void     pm_cfb_setup_intr(void);
 376         extern int      pm_adjust_timestamps(dev_info_t *, void *);
 377         extern void     start_other_cpus(int);
 378         extern void     sysevent_evc_thrinit();
 379         extern kmutex_t ualock;
 380 #if defined(__x86)
 381         extern void     fastboot_post_startup(void);
 382         extern void     progressbar_start(void);
 383 #endif
 384         /*
 385          * In the horrible world of x86 in-lines, you can't get symbolic
 386          * structure offsets a la genassym.  This assertion is here so
 387          * that the next poor slob who innocently changes the offset of
 388          * cpu_thread doesn't waste as much time as I just did finding
 389          * out that it's hard-coded in i86/ml/i86.il.  Similarly for
 390          * curcpup.  You're welcome.
 391          */
 392         ASSERT(CPU == CPU->cpu_self);
 393         ASSERT(curthread == CPU->cpu_thread);
 394         ASSERT_STACK_ALIGNED();
 395 
 396         /*
 397          * We take the ualock until we have completed the startup
 398          * to prevent kadmin() from disrupting this work. In particular,
 399          * we don't want kadmin() to bring the system down while we are
 400          * trying to start it up.
 401          */
 402         mutex_enter(&ualock);
 403 
 404         /*
 405          * Setup root lgroup and leaf lgroup for CPU 0
 406          */
 407         lgrp_init(LGRP_INIT_STAGE2);
 408 
 409         /*
 410          * Once 'startup()' completes, the thread_reaper() daemon would be
 411          * created(in thread_init()). After that, it is safe to create threads
 412          * that could exit. These exited threads will get reaped.
 413          */
 414         startup();
 415         segkmem_gc();
 416         callb_init();
 417         cbe_init_pre(); /* x86 must initialize gethrtimef before timer_init */
 418         timer_init();   /* timer must be initialized before cyclic starts */
 419         cbe_init();
 420         callout_init(); /* callout table MUST be init'd after cyclics */
 421         clock_tick_init_pre();
 422         clock_init();
 423 
 424 #if defined(__x86)
 425         /*
 426          * The progressbar thread uses cv_reltimedwait() and hence needs to be
 427          * started after the callout mechanism has been initialized.
 428          */
 429         progressbar_start();
 430 #endif
 431         /*
 432          * On some platforms, clkinitf() changes the timing source that
 433          * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
 434          * clkinitf(), so re-initialize the microstate counters after the
 435          * timesource has been chosen.
 436          */
 437         init_mstate(&t0, LMS_SYSTEM);
 438         init_cpu_mstate(CPU, CMS_SYSTEM);
 439 
 440         /*
 441          * May need to probe to determine latencies from CPU 0 after
 442          * gethrtime() comes alive in cbe_init() and before enabling interrupts
 443          * and copy and release any temporary memory allocated with BOP_ALLOC()
 444          * before release_bootstrap() frees boot memory
 445          */
 446         lgrp_init(LGRP_INIT_STAGE3);
 447 
 448         /*
 449          * Call all system initialization functions.
 450          */
 451         for (initptr = &init_tbl[0]; *initptr; initptr++)
 452                 (**initptr)();
 453         /*
 454          * Load iSCSI boot properties
 455          */
 456         ld_ib_prop();
 457         /*
 458          * initialize vm related stuff.
 459          */
 460         vm_init();
 461 
 462         /*
 463          * initialize buffer pool for raw I/O requests
 464          */
 465         physio_bufs_init();
 466 
 467         ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
 468 
 469         /*
 470          * Drop the interrupt level and allow interrupts.  At this point
 471          * the DDI guarantees that interrupts are enabled.
 472          */
 473         (void) spl0();
 474         interrupts_unleashed = 1;
 475 
 476         /*
 477          * Create kmem cache for proc structures
 478          */
 479         process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
 480             0, NULL, NULL, NULL, NULL, NULL, 0);
 481 
 482         vfs_mountroot();        /* Mount the root file system */
 483         errorq_init();          /* after vfs_mountroot() so DDI root is ready */
 484         cpu_kstat_init(CPU);    /* after vfs_mountroot() so TOD is valid */
 485         ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
 486                                 /* after vfs_mountroot() so hrestime is valid */
 487 
 488         post_startup();
 489         swaploaded = 1;
 490 
 491         /*
 492          * Initialize Solaris Audit Subsystem
 493          */
 494         audit_init();
 495 
 496         /*
 497          * Plumb the protocol modules and drivers only if we are not
 498          * networked booted, in this case we already did it in rootconf().
 499          */
 500         if (netboot == 0 && iscsiboot_prop == NULL)
 501                 (void) strplumb();
 502 
 503         gethrestime(&PTOU(curproc)->u_start);
 504         curthread->t_start = PTOU(curproc)->u_start.tv_sec;
 505         p->p_mstart = gethrtime();
 506 
 507         /*
 508          * Perform setup functions that can only be done after root
 509          * and swap have been set up.
 510          */
 511         consconfig();
 512 #ifndef __sparc
 513         release_bootstrap();
 514 #endif
 515 
 516         /*
 517          * attach drivers with ddi-forceattach prop
 518          * It must be done early enough to load hotplug drivers (e.g.
 519          * pcmcia nexus) so that devices enumerated via hotplug is
 520          * available before I/O subsystem is fully initialized.
 521          */
 522         i_ddi_forceattach_drivers();
 523 
 524         /*
 525          * Set the scan rate and other parameters of the paging subsystem.
 526          */
 527         setupclock(0);
 528 
 529         /*
 530          * Initialize process 0's lwp directory and lwpid hash table.
 531          */
 532         p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
 533         p->p_lwpdir->ld_next = p->p_lwpdir + 1;
 534         p->p_lwpdir_sz = 2;
 535         p->p_tidhash = p0_tidhash;
 536         p->p_tidhash_sz = 2;
 537         p0_lep.le_thread = curthread;
 538         p0_lep.le_lwpid = curthread->t_tid;
 539         p0_lep.le_start = curthread->t_start;
 540         lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
 541 
 542         /*
 543          * Initialize extended accounting.
 544          */
 545         exacct_init();
 546 
 547         /*
 548          * Initialize threads of sysevent event channels
 549          */
 550         sysevent_evc_thrinit();
 551 
 552         /*
 553          * This must be done after post_startup() but before
 554          * start_other_cpus()
 555          */
 556         lgrp_init(LGRP_INIT_STAGE4);
 557 
 558         /*
 559          * Perform MP initialization, if any.
 560          */
 561         start_other_cpus(0);
 562 
 563 #ifdef  __sparc
 564         /*
 565          * Release bootstrap here since PROM interfaces are
 566          * used to start other CPUs above.
 567          */
 568         release_bootstrap();
 569 #endif
 570 
 571         /*
 572          * Finish lgrp initialization after all CPUS are brought online.
 573          */
 574         lgrp_init(LGRP_INIT_STAGE5);
 575 
 576         /*
 577          * After mp_init(), number of cpus are known (this is
 578          * true for the time being, when there are actually
 579          * hot pluggable cpus then this scheme  would not do).
 580          * Any per cpu initialization is done here.
 581          */
 582         kmem_mp_init();
 583         vmem_update(NULL);
 584 
 585         clock_tick_init_post();
 586 
 587         for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
 588                 (**initptr)();
 589 
 590         /*
 591          * These must be called after start_other_cpus
 592          */
 593         pm_cfb_setup_intr();
 594 #if defined(__x86)
 595         fastboot_post_startup();
 596 #endif
 597 
 598         /*
 599          * Make init process; enter scheduling loop with system process.
 600          *
 601          * Note that we manually assign the pids for these processes, for
 602          * historical reasons.  If more pre-assigned pids are needed,
 603          * FAMOUS_PIDS will have to be updated.
 604          */
 605 
 606         /* create init process */
 607         if (newproc(start_init, NULL, defaultcid, 59, NULL,
 608             FAMOUS_PID_INIT))
 609                 panic("main: unable to fork init.");
 610 
 611         /* create pageout daemon */
 612         if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
 613             FAMOUS_PID_PAGEOUT))
 614                 panic("main: unable to fork pageout()");
 615 
 616         /* create fsflush daemon */
 617         if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
 618             FAMOUS_PID_FSFLUSH))
 619                 panic("main: unable to fork fsflush()");
 620 
 621         /* create cluster process if we're a member of one */
 622         if (cluster_bootflags & CLUSTER_BOOTED) {
 623                 if (newproc(cluster_wrapper, NULL, syscid, minclsyspri,
 624                     NULL, 0)) {
 625                         panic("main: unable to fork cluster()");
 626                 }
 627         }
 628 
 629         /*
 630          * Create system threads (threads are associated with p0)
 631          */
 632 
 633         /* create module uninstall daemon */
 634         /* BugID 1132273. If swapping over NFS need a bigger stack */
 635         (void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
 636             NULL, 0, &p0, TS_RUN, minclsyspri);
 637 
 638         (void) thread_create(NULL, 0, seg_pasync_thread,
 639             NULL, 0, &p0, TS_RUN, minclsyspri);
 640 
 641         pid_setmin();
 642 
 643         /* system is now ready */
 644         mutex_exit(&ualock);
 645 
 646         bcopy("sched", PTOU(curproc)->u_psargs, 6);
 647         bcopy("sched", PTOU(curproc)->u_comm, 5);
 648         sched();
 649         /* NOTREACHED */
 650 }