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