kernel/os/main.c

   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 /*
  30  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  31  */
  32
  33 #include <sys/types.h>
  34 #include <sys/param.h>
  35 #include <sys/sysmacros.h>
  36 #include <sys/pcb.h>
  37 #include <sys/systm.h>
  38 #include <sys/signal.h>
  39 #include <sys/cred.h>
  40 #include <sys/user.h>
  41 #include <sys/vfs.h>
  42 #include <sys/vnode.h>
  43 #include <sys/proc.h>
  44 #include <sys/time.h>
  45 #include <sys/file.h>
  46 #include <sys/priocntl.h>
  47 #include <sys/procset.h>
  48 #include <sys/disp.h>
  49 #include <sys/callo.h>
  50 #include <sys/callb.h>
  51 #include <sys/debug.h>
  52 #include <sys/conf.h>
  53 #include <sys/bootconf.h>
  54 #include <sys/utsname.h>
  55 #include <sys/cmn_err.h>
  56 #include <sys/vmparam.h>
  57 #include <sys/modctl.h>
  58 #include <sys/vm.h>
  59 #include <sys/callb.h>
  60 #include <sys/ddi_periodic.h>
  61 #include <sys/sunddi.h>
  62 #include <sys/kmem.h>
  63 #include <sys/vmem.h>
  64 #include <sys/cpuvar.h>
  65 #include <sys/corectl.h>
  66 #include <sys/exec.h>
  67 #include <sys/syscall.h>
  68 #include <sys/reboot.h>
  69 #include <sys/task.h>
  70 #include <sys/exacct.h>
  71 #include <sys/autoconf.h>
  72 #include <sys/errorq.h>
  73 #include <sys/class.h>
  74 #include <sys/stack.h>
  75 #include <sys/brand.h>
  76
  77 #include <vm/as.h>
  78 #include <vm/seg_kmem.h>
  79
  80 #include <c2/audit.h>
  81 #include <sys/bootprops.h>
  82
  83 /* well known processes */
  84 proc_t *proc_sched;             /* memory scheduler */
  85 proc_t *proc_init;              /* init */
  86 proc_t *proc_pageout;           /* pageout daemon */
  87 proc_t *proc_fsflush;           /* fsflush daemon */
  88
  89 pgcnt_t maxmem;         /* Maximum available memory in pages.   */
  90 pgcnt_t freemem;        /* Current available memory in pages.   */
  91 int     interrupts_unleashed;   /* set when we do the first spl0() */
  92
  93 kmem_cache_t *process_cache;    /* kmem cache for proc structures */
  94
  95 /*
  96  * Indicates whether the auditing module (c2audit) is loaded. Possible
  97  * values are:
  98  * 0 - c2audit module is excluded in /etc/system and cannot be loaded
  99  * 1 - c2audit module is not loaded but can be anytime
 100  * 2 - c2audit module is loaded
 101  */
 102 int audit_active = C2AUDIT_DISABLED;
 103
 104 /*
 105  * Process 0's lwp directory and lwpid hash table.
 106  */
 107 lwpdir_t p0_lwpdir[2];
 108 tidhash_t p0_tidhash[2];
 109 lwpent_t p0_lep;
 110
 111 /*
 112  * Machine-independent initialization code
 113  * Called from cold start routine as
 114  * soon as a stack and segmentation
 115  * have been established.
 116  * Functions:
 117  *      clear and free user core
 118  *      turn on clock
 119  *      hand craft 0th process
 120  *      call all initialization routines
 121  *      fork    - process 0 to schedule
 122  *              - process 1 execute bootstrap
 123  *              - process 2 to page out
 124  *      create system threads
 125  */
 126
 127 char initname[INITNAME_SZ] = "/sbin/init";      /* also referenced by zone0 */
 128 char initargs[BOOTARGS_MAX] = "";               /* also referenced by zone0 */
 129
 130 /*
 131  * Construct a stack for init containing the arguments to it, then
 132  * pass control to exec_common.
 133  */
 134 int
 135 exec_init(const char *initpath, const char *args)
 136 {
 137         caddr32_t ucp;
 138         caddr32_t *uap;
 139         caddr32_t *argv;
 140         caddr32_t exec_fnamep;
 141         char *scratchargs;
 142         int i, sarg;
 143         size_t argvlen, alen;
 144         boolean_t in_arg;
 145         int argc = 0;
 146         int error = 0, count = 0;
 147         proc_t *p = ttoproc(curthread);
 148         klwp_t *lwp = ttolwp(curthread);
 149         int brand_action;
 150
 151         if (args == NULL)
 152                 args = "";
 153
 154         alen = strlen(initpath) + 1 + strlen(args) + 1;
 155         scratchargs = kmem_alloc(alen, KM_SLEEP);
 156         (void) snprintf(scratchargs, alen, "%s %s", initpath, args);
 157
 158         /*
 159          * We do a quick two state parse of the string to sort out how big
 160          * argc should be.
 161          */
 162         in_arg = B_FALSE;
 163         for (i = 0; i < strlen(scratchargs); i++) {
 164                 if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
 165                         if (in_arg) {
 166                                 in_arg = B_FALSE;
 167                                 argc++;
 168                         }
 169                 } else {
 170                         in_arg = B_TRUE;
 171                 }
 172         }
 173         argvlen = sizeof (caddr32_t) * (argc + 1);
 174         argv = kmem_zalloc(argvlen, KM_SLEEP);
 175
 176         /*
 177          * We pull off a bit of a hack here.  We work our way through the
 178          * args string, putting nulls at the ends of space delimited tokens
 179          * (boot args don't support quoting at this time).  Then we just
 180          * copy the whole mess to userland in one go.  In other words, we
 181          * transform this: "init -s -r\0" into this on the stack:
 182          *
 183          *      -0x00 \0
 184          *      -0x01 r
 185          *      -0x02 -  <--------.
 186          *      -0x03 \0          |
 187          *      -0x04 s           |
 188          *      -0x05 -  <------. |
 189          *      -0x06 \0        | |
 190          *      -0x07 t         | |
 191          *      -0x08 i         | |
 192          *      -0x09 n         | |
 193          *      -0x0a i  <---.  | |
 194          *      -0x10 NULL   |  | |     (argv[3])
 195          *      -0x14   -----|--|-'     (argv[2])
 196          *      -0x18  ------|--'       (argv[1])
 197          *      -0x1c -------'          (argv[0])
 198          *
 199          * Since we know the value of ucp at the beginning of this process,
 200          * we can trivially compute the argv[] array which we also need to
 201          * place in userland: argv[i] = ucp - sarg(i), where ucp is the
 202          * stack ptr, and sarg is the string index of the start of the
 203          * argument.
 204          */
 205         ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
 206
 207         argc = 0;
 208         in_arg = B_FALSE;
 209         sarg = 0;
 210
 211         for (i = 0; i < alen; i++) {
 212                 if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
 213                         if (in_arg == B_TRUE) {
 214                                 in_arg = B_FALSE;
 215                                 scratchargs[i] = '\0';
 216                                 argv[argc++] = ucp - (alen - sarg);
 217                         }
 218                 } else if (in_arg == B_FALSE) {
 219                         in_arg = B_TRUE;
 220                         sarg = i;
 221                 }
 222         }
 223         ucp -= alen;
 224         error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
 225
 226         uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
 227         uap--;  /* advance to be below the word we're in */
 228         uap -= (argc + 1);      /* advance argc words down, plus one for NULL */
 229         error |= copyout(argv, uap, argvlen);
 230
 231         if (error != 0) {
 232                 zcmn_err(p->p_zone->zone_id, CE_WARN,
 233                     "Could not construct stack for init.\n");
 234                 kmem_free(argv, argvlen);
 235                 kmem_free(scratchargs, alen);
 236                 return (EFAULT);
 237         }
 238
 239         exec_fnamep = argv[0];
 240         kmem_free(argv, argvlen);
 241         kmem_free(scratchargs, alen);
 242
 243         /*
 244          * Point at the arguments.
 245          */
 246         lwp->lwp_ap = lwp->lwp_arg;
 247         lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
 248         lwp->lwp_arg[1] = (uintptr_t)uap;
 249         lwp->lwp_arg[2] = (uintptr_t)NULL;
 250         curthread->t_post_sys = 1;
 251         curthread->t_sysnum = SYS_execve;
 252
 253         /*
 254          * If we are executing init from zsched, we may have inherited its
 255          * parent process's signal mask.  Clear it now so that we behave in
 256          * the same way as when started from the global zone.
 257          */
 258         sigemptyset(&curthread->t_hold);
 259
 260         brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
 261 again:
 262         error = exec_common((const char *)(uintptr_t)exec_fnamep,
 263             (const char **)(uintptr_t)uap, NULL, brand_action);
 264
 265         /*
 266          * Normally we would just set lwp_argsaved and t_post_sys and
 267          * let post_syscall reset lwp_ap for us.  Unfortunately,
 268          * exec_init isn't always called from a system call.  Instead
 269          * of making a mess of trap_cleanup, we just reset the args
 270          * pointer here.
 271          */
 272         reset_syscall_args();
 273
 274         switch (error) {
 275         case 0:
 276                 return (0);
 277
 278         case ENOENT:
 279                 zcmn_err(p->p_zone->zone_id, CE_WARN,
 280                     "exec(%s) failed (file not found).\n", initpath);
 281                 return (ENOENT);
 282
 283         case EAGAIN:
 284         case EINTR:
 285                 ++count;
 286                 if (count < 5) {
 287                         zcmn_err(p->p_zone->zone_id, CE_WARN,
 288                             "exec(%s) failed with errno %d.  Retrying...\n",
 289                             initpath, error);
 290                         goto again;
 291                 }
 292         }
 293
 294         zcmn_err(p->p_zone->zone_id, CE_WARN,
 295             "exec(%s) failed with errno %d.", initpath, error);
 296         return (error);
 297 }
 298
 299 /*
 300  * This routine does all of the common setup for invoking init; global
 301  * and non-global zones employ this routine for the functionality which is
 302  * in common.
 303  *
 304  * This program (init, presumably) must be a 32-bit process.
 305  */
 306 int
 307 start_init_common()
 308 {
 309         proc_t *p = curproc;
 310         ASSERT_STACK_ALIGNED();
 311         p->p_zone->zone_proc_initpid = p->p_pid;
 312
 313         p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
 314         p->p_usrstack = (caddr_t)USRSTACK32;
 315         p->p_model = DATAMODEL_ILP32;
 316         p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
 317         p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
 318         p->p_stk_ctl = INT32_MAX;
 319
 320         p->p_as = as_alloc();
 321         p->p_as->a_proc = p;
 322         p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
 323         (void) hat_setup(p->p_as->a_hat, HAT_INIT);
 324
 325         init_core();
 326
 327         init_mstate(curthread, LMS_SYSTEM);
 328         return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
 329 }
 330
 331 /*
 332  * Start the initial user process for the global zone; once running, if
 333  * init should subsequently fail, it will be automatically be caught in the
 334  * exit(2) path, and restarted by restart_init().
 335  */
 336 static void
 337 start_init(void)
 338 {
 339         proc_init = curproc;
 340
 341         ASSERT(curproc->p_zone->zone_initname != NULL);
 342
 343         if (start_init_common() != 0)
 344                 halt("unix: Could not start init");
 345         lwp_rtt();
 346 }
 347
 348 void
 349 main(void)
 350 {
 351         proc_t          *p = ttoproc(curthread);        /* &p0 */
 352         int             (**initptr)();
 353         extern void     sched();
 354         extern void     fsflush();
 355         extern int      (*init_tbl[])();
 356         extern int      (*mp_init_tbl[])();
 357         extern id_t     syscid, defaultcid;
 358         extern int      swaploaded;
 359         extern int      netboot;
 360         extern ib_boot_prop_t *iscsiboot_prop;
 361         extern void     vm_init(void);
 362         extern void     cbe_init_pre(void);
 363         extern void     cbe_init(void);
 364         extern void     clock_tick_init_pre(void);
 365         extern void     clock_tick_init_post(void);
 366         extern void     clock_init(void);
 367         extern void     physio_bufs_init(void);
 368         extern void     pm_cfb_setup_intr(void);
 369         extern int      pm_adjust_timestamps(dev_info_t *, void *);
 370         extern void     start_other_cpus(int);
 371         extern void     sysevent_evc_thrinit();
 372         extern kmutex_t ualock;
 373 #if defined(__x86)
 374         extern void     fastboot_post_startup(void);
 375         extern void     progressbar_start(void);
 376 #endif
 377         /*
 378          * In the horrible world of x86 in-lines, you can't get symbolic
 379          * structure offsets a la genassym.  This assertion is here so
 380          * that the next poor slob who innocently changes the offset of
 381          * cpu_thread doesn't waste as much time as I just did finding
 382          * out that it's hard-coded in i86/ml/i86.il.  Similarly for
 383          * curcpup.  You're welcome.
 384          */
 385         ASSERT(CPU == CPU->cpu_self);
 386         ASSERT(curthread == CPU->cpu_thread);
 387         ASSERT_STACK_ALIGNED();
 388
 389         /*
 390          * We take the ualock until we have completed the startup
 391          * to prevent kadmin() from disrupting this work. In particular,
 392          * we don't want kadmin() to bring the system down while we are
 393          * trying to start it up.
 394          */
 395         mutex_enter(&ualock);
 396
 397         /*
 398          * Setup root lgroup and leaf lgroup for CPU 0
 399          */
 400         lgrp_init(LGRP_INIT_STAGE2);
 401
 402         vmobject_init(&kvps[KV_KVP].v_object, &kvps[KV_KVP]);
 403         vmobject_init(&kvps[KV_ZVP].v_object, &kvps[KV_ZVP]);
 404
 405         /*
 406          * Once 'startup()' completes, the thread_reaper() daemon would be
 407          * created(in thread_init()). After that, it is safe to create threads
 408          * that could exit. These exited threads will get reaped.
 409          */
 410         startup();
 411         segkmem_gc();
 412         callb_init();
 413         cbe_init_pre(); /* x86 must initialize gethrtimef before timer_init */
 414         ddi_periodic_init();
 415         cbe_init();
 416         callout_init(); /* callout table MUST be init'd after cyclics */
 417         clock_tick_init_pre();
 418         clock_init();
 419
 420 #if defined(__x86)
 421         /*
 422          * The progressbar thread uses cv_reltimedwait() and hence needs to be
 423          * started after the callout mechanism has been initialized.
 424          */
 425         progressbar_start();
 426 #endif
 427         /*
 428          * On some platforms, clkinitf() changes the timing source that
 429          * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
 430          * clkinitf(), so re-initialize the microstate counters after the
 431          * timesource has been chosen.
 432          */
 433         init_mstate(&t0, LMS_SYSTEM);
 434         init_cpu_mstate(CPU, CMS_SYSTEM);
 435
 436         /*
 437          * May need to probe to determine latencies from CPU 0 after
 438          * gethrtime() comes alive in cbe_init() and before enabling interrupts
 439          * and copy and release any temporary memory allocated with BOP_ALLOC()
 440          * before release_bootstrap() frees boot memory
 441          */
 442         lgrp_init(LGRP_INIT_STAGE3);
 443
 444         /*
 445          * Call all system initialization functions.
 446          */
 447         for (initptr = &init_tbl[0]; *initptr; initptr++)
 448                 (**initptr)();
 449         /*
 450          * Load iSCSI boot properties
 451          */
 452         ld_ib_prop();
 453         /*
 454          * initialize vm related stuff.
 455          */
 456         vm_init();
 457
 458         /*
 459          * initialize buffer pool for raw I/O requests
 460          */
 461         physio_bufs_init();
 462
 463         ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
 464
 465         /*
 466          * Drop the interrupt level and allow interrupts.  At this point
 467          * the DDI guarantees that interrupts are enabled.
 468          */
 469         (void) spl0();
 470         interrupts_unleashed = 1;
 471
 472         /*
 473          * Create kmem cache for proc structures
 474          */
 475         process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
 476             0, NULL, NULL, NULL, NULL, NULL, 0);
 477
 478         vfs_mountroot();        /* Mount the root file system */
 479         errorq_init();          /* after vfs_mountroot() so DDI root is ready */
 480         cpu_kstat_init(CPU);    /* after vfs_mountroot() so TOD is valid */
 481         ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
 482                                 /* after vfs_mountroot() so hrestime is valid */
 483
 484         post_startup();
 485         swaploaded = 1;
 486
 487         /*
 488          * Initialize Solaris Audit Subsystem
 489          */
 490         audit_init();
 491
 492         /*
 493          * Plumb the protocol modules and drivers only if we are not
 494          * networked booted, in this case we already did it in rootconf().
 495          */
 496         if (netboot == 0 && iscsiboot_prop == NULL)
 497                 (void) strplumb();
 498
 499         gethrestime(&PTOU(curproc)->u_start);
 500         curthread->t_start = PTOU(curproc)->u_start.tv_sec;
 501         p->p_mstart = gethrtime();
 502
 503         /*
 504          * Perform setup functions that can only be done after root
 505          * and swap have been set up.
 506          */
 507         consconfig();
 508         release_bootstrap();
 509
 510         /*
 511          * attach drivers with ddi-forceattach prop
 512          * It must be done early enough to load hotplug drivers (e.g.
 513          * pcmcia nexus) so that devices enumerated via hotplug is
 514          * available before I/O subsystem is fully initialized.
 515          */
 516         i_ddi_forceattach_drivers();
 517
 518         /*
 519          * Set the scan rate and other parameters of the paging subsystem.
 520          */
 521         setupclock(0);
 522
 523         /*
 524          * Initialize process 0's lwp directory and lwpid hash table.
 525          */
 526         p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
 527         p->p_lwpdir->ld_next = p->p_lwpdir + 1;
 528         p->p_lwpdir_sz = 2;
 529         p->p_tidhash = p0_tidhash;
 530         p->p_tidhash_sz = 2;
 531         p0_lep.le_thread = curthread;
 532         p0_lep.le_lwpid = curthread->t_tid;
 533         p0_lep.le_start = curthread->t_start;
 534         lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
 535
 536         /*
 537          * Initialize extended accounting.
 538          */
 539         exacct_init();
 540
 541         /*
 542          * Initialize threads of sysevent event channels
 543          */
 544         sysevent_evc_thrinit();
 545
 546         /*
 547          * This must be done after post_startup() but before
 548          * start_other_cpus()
 549          */
 550         lgrp_init(LGRP_INIT_STAGE4);
 551
 552         /*
 553          * Perform MP initialization, if any.
 554          */
 555         start_other_cpus(0);
 556
 557
 558         /*
 559          * Finish lgrp initialization after all CPUS are brought online.
 560          */
 561         lgrp_init(LGRP_INIT_STAGE5);
 562
 563         /*
 564          * After mp_init(), number of cpus are known (this is
 565          * true for the time being, when there are actually
 566          * hot pluggable cpus then this scheme  would not do).
 567          * Any per cpu initialization is done here.
 568          */
 569         kmem_mp_init();
 570         vmem_update(NULL);
 571
 572         clock_tick_init_post();
 573
 574         for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
 575                 (**initptr)();
 576
 577         /*
 578          * These must be called after start_other_cpus
 579          */
 580         pm_cfb_setup_intr();
 581 #if defined(__x86)
 582         fastboot_post_startup();
 583 #endif
 584
 585         /*
 586          * Make init process; enter scheduling loop with system process.
 587          *
 588          * Note that we manually assign the pids for these processes, for
 589          * historical reasons.  If more pre-assigned pids are needed,
 590          * FAMOUS_PIDS will have to be updated.
 591          */
 592
 593         /* create init process */
 594         if (newproc(start_init, NULL, defaultcid, 59, NULL,
 595             FAMOUS_PID_INIT))
 596                 panic("main: unable to fork init.");
 597
 598         /* create pageout daemon */
 599         if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
 600             FAMOUS_PID_PAGEOUT))
 601                 panic("main: unable to fork pageout()");
 602
 603         /* create fsflush daemon */
 604         if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
 605             FAMOUS_PID_FSFLUSH))
 606                 panic("main: unable to fork fsflush()");
 607
 608         /*
 609          * Create system threads (threads are associated with p0)
 610          */
 611
 612         /* create module uninstall daemon */
 613         /* BugID 1132273. If swapping over NFS need a bigger stack */
 614         (void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
 615             NULL, 0, &p0, TS_RUN, minclsyspri);
 616
 617         (void) thread_create(NULL, 0, seg_pasync_thread,
 618             NULL, 0, &p0, TS_RUN, minclsyspri);
 619
 620         pid_setmin();
 621
 622         /* system is now ready */
 623         mutex_exit(&ualock);
 624
 625         bcopy("sched", PTOU(curproc)->u_psargs, 6);
 626         bcopy("sched", PTOU(curproc)->u_comm, 5);
 627         sched();
 628         /* NOTREACHED */
 629 }