xref: /titanic_44/usr/src/uts/common/os/main.c (revision 3081df9ddbd41d6eca52e0d1aee5d61ea71cdcf5)
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/kmem.h>
62 #include <sys/vmem.h>
63 #include <sys/cpuvar.h>
64 #include <sys/cladm.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 #include <sys/dc_ki.h>
81 
82 #include <c2/audit.h>
83 #include <sys/bootprops.h>
84 
85 /* well known processes */
86 proc_t *proc_sched;		/* memory scheduler */
87 proc_t *proc_init;		/* init */
88 proc_t *proc_pageout;		/* pageout daemon */
89 proc_t *proc_fsflush;		/* fsflush daemon */
90 
91 pgcnt_t	maxmem;		/* Maximum available memory in pages.	*/
92 pgcnt_t	freemem;	/* Current available memory in pages.	*/
93 int	interrupts_unleashed;	/* set when we do the first spl0() */
94 
95 kmem_cache_t *process_cache;	/* kmem cache for proc structures */
96 
97 /*
98  * Indicates whether the auditing module (c2audit) is loaded. Possible
99  * values are:
100  * 0 - c2audit module is excluded in /etc/system and cannot be loaded
101  * 1 - c2audit module is not loaded but can be anytime
102  * 2 - c2audit module is loaded
103  */
104 int audit_active = C2AUDIT_DISABLED;
105 
106 /*
107  * Process 0's lwp directory and lwpid hash table.
108  */
109 lwpdir_t p0_lwpdir[2];
110 tidhash_t p0_tidhash[2];
111 lwpent_t p0_lep;
112 
113 /*
114  * Machine-independent initialization code
115  * Called from cold start routine as
116  * soon as a stack and segmentation
117  * have been established.
118  * Functions:
119  *	clear and free user core
120  *	turn on clock
121  *	hand craft 0th process
122  *	call all initialization routines
123  *	fork	- process 0 to schedule
124  *		- process 1 execute bootstrap
125  *		- process 2 to page out
126  *	create system threads
127  */
128 
129 int cluster_bootflags = 0;
130 
131 void
cluster_wrapper(void)132 cluster_wrapper(void)
133 {
134 	cluster();
135 	panic("cluster()  returned");
136 }
137 
138 char initname[INITNAME_SZ] = "/sbin/init";	/* also referenced by zone0 */
139 char initargs[BOOTARGS_MAX] = "";		/* also referenced by zone0 */
140 
141 /*
142  * Construct a stack for init containing the arguments to it, then
143  * pass control to exec_common.
144  */
145 int
exec_init(const char * initpath,const char * args)146 exec_init(const char *initpath, const char *args)
147 {
148 	caddr32_t ucp;
149 	caddr32_t *uap;
150 	caddr32_t *argv;
151 	caddr32_t exec_fnamep;
152 	char *scratchargs;
153 	int i, sarg;
154 	size_t argvlen, alen;
155 	boolean_t in_arg;
156 	int argc = 0;
157 	int error = 0, count = 0;
158 	proc_t *p = ttoproc(curthread);
159 	klwp_t *lwp = ttolwp(curthread);
160 	int brand_action;
161 
162 	if (args == NULL)
163 		args = "";
164 
165 	alen = strlen(initpath) + 1 + strlen(args) + 1;
166 	scratchargs = kmem_alloc(alen, KM_SLEEP);
167 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
168 
169 	/*
170 	 * We do a quick two state parse of the string to sort out how big
171 	 * argc should be.
172 	 */
173 	in_arg = B_FALSE;
174 	for (i = 0; i < strlen(scratchargs); i++) {
175 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
176 			if (in_arg) {
177 				in_arg = B_FALSE;
178 				argc++;
179 			}
180 		} else {
181 			in_arg = B_TRUE;
182 		}
183 	}
184 	argvlen = sizeof (caddr32_t) * (argc + 1);
185 	argv = kmem_zalloc(argvlen, KM_SLEEP);
186 
187 	/*
188 	 * We pull off a bit of a hack here.  We work our way through the
189 	 * args string, putting nulls at the ends of space delimited tokens
190 	 * (boot args don't support quoting at this time).  Then we just
191 	 * copy the whole mess to userland in one go.  In other words, we
192 	 * transform this: "init -s -r\0" into this on the stack:
193 	 *
194 	 *	-0x00 \0
195 	 *	-0x01 r
196 	 *	-0x02 -  <--------.
197 	 *	-0x03 \0	  |
198 	 *	-0x04 s		  |
199 	 *	-0x05 -  <------. |
200 	 *	-0x06 \0	| |
201 	 *	-0x07 t		| |
202 	 *	-0x08 i 	| |
203 	 *	-0x09 n		| |
204 	 *	-0x0a i  <---.  | |
205 	 *	-0x10 NULL   |  | |	(argv[3])
206 	 *	-0x14   -----|--|-'	(argv[2])
207 	 *	-0x18  ------|--'	(argv[1])
208 	 *	-0x1c -------'		(argv[0])
209 	 *
210 	 * Since we know the value of ucp at the beginning of this process,
211 	 * we can trivially compute the argv[] array which we also need to
212 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
213 	 * stack ptr, and sarg is the string index of the start of the
214 	 * argument.
215 	 */
216 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
217 
218 	argc = 0;
219 	in_arg = B_FALSE;
220 	sarg = 0;
221 
222 	for (i = 0; i < alen; i++) {
223 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
224 			if (in_arg == B_TRUE) {
225 				in_arg = B_FALSE;
226 				scratchargs[i] = '\0';
227 				argv[argc++] = ucp - (alen - sarg);
228 			}
229 		} else if (in_arg == B_FALSE) {
230 			in_arg = B_TRUE;
231 			sarg = i;
232 		}
233 	}
234 	ucp -= alen;
235 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
236 
237 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
238 	uap--;	/* advance to be below the word we're in */
239 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
240 	error |= copyout(argv, uap, argvlen);
241 
242 	if (error != 0) {
243 		zcmn_err(p->p_zone->zone_id, CE_WARN,
244 		    "Could not construct stack for init.\n");
245 		kmem_free(argv, argvlen);
246 		kmem_free(scratchargs, alen);
247 		return (EFAULT);
248 	}
249 
250 	exec_fnamep = argv[0];
251 	kmem_free(argv, argvlen);
252 	kmem_free(scratchargs, alen);
253 
254 	/*
255 	 * Point at the arguments.
256 	 */
257 	lwp->lwp_ap = lwp->lwp_arg;
258 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
259 	lwp->lwp_arg[1] = (uintptr_t)uap;
260 	lwp->lwp_arg[2] = NULL;
261 	curthread->t_post_sys = 1;
262 	curthread->t_sysnum = SYS_execve;
263 
264 	/*
265 	 * If we are executing init from zsched, we may have inherited its
266 	 * parent process's signal mask.  Clear it now so that we behave in
267 	 * the same way as when started from the global zone.
268 	 */
269 	sigemptyset(&curthread->t_hold);
270 
271 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
272 again:
273 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
274 	    (const char **)(uintptr_t)uap, NULL, brand_action);
275 
276 	/*
277 	 * Normally we would just set lwp_argsaved and t_post_sys and
278 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
279 	 * exec_init isn't always called from a system call.  Instead
280 	 * of making a mess of trap_cleanup, we just reset the args
281 	 * pointer here.
282 	 */
283 	reset_syscall_args();
284 
285 	switch (error) {
286 	case 0:
287 		return (0);
288 
289 	case ENOENT:
290 		zcmn_err(p->p_zone->zone_id, CE_WARN,
291 		    "exec(%s) failed (file not found).\n", initpath);
292 		return (ENOENT);
293 
294 	case EAGAIN:
295 	case EINTR:
296 		++count;
297 		if (count < 5) {
298 			zcmn_err(p->p_zone->zone_id, CE_WARN,
299 			    "exec(%s) failed with errno %d.  Retrying...\n",
300 			    initpath, error);
301 			goto again;
302 		}
303 	}
304 
305 	zcmn_err(p->p_zone->zone_id, CE_WARN,
306 	    "exec(%s) failed with errno %d.", initpath, error);
307 	return (error);
308 }
309 
310 /*
311  * This routine does all of the common setup for invoking init; global
312  * and non-global zones employ this routine for the functionality which is
313  * in common.
314  *
315  * This program (init, presumably) must be a 32-bit process.
316  */
317 int
start_init_common()318 start_init_common()
319 {
320 	proc_t *p = curproc;
321 	ASSERT_STACK_ALIGNED();
322 	p->p_zone->zone_proc_initpid = p->p_pid;
323 
324 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
325 	p->p_usrstack = (caddr_t)USRSTACK32;
326 	p->p_model = DATAMODEL_ILP32;
327 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
328 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
329 	p->p_stk_ctl = INT32_MAX;
330 
331 	p->p_as = as_alloc();
332 	p->p_as->a_proc = p;
333 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
334 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
335 
336 	init_core();
337 
338 	init_mstate(curthread, LMS_SYSTEM);
339 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
340 }
341 
342 /*
343  * Start the initial user process for the global zone; once running, if
344  * init should subsequently fail, it will be automatically be caught in the
345  * exit(2) path, and restarted by restart_init().
346  */
347 static void
start_init(void)348 start_init(void)
349 {
350 	proc_init = curproc;
351 
352 	ASSERT(curproc->p_zone->zone_initname != NULL);
353 
354 	if (start_init_common() != 0)
355 		halt("unix: Could not start init");
356 	lwp_rtt();
357 }
358 
359 void
main(void)360 main(void)
361 {
362 	proc_t		*p = ttoproc(curthread);	/* &p0 */
363 	int		(**initptr)();
364 	extern void	sched();
365 	extern void	fsflush();
366 	extern int	(*init_tbl[])();
367 	extern int	(*mp_init_tbl[])();
368 	extern id_t	syscid, defaultcid;
369 	extern int	swaploaded;
370 	extern int	netboot;
371 	extern ib_boot_prop_t *iscsiboot_prop;
372 	extern void	vm_init(void);
373 	extern void	cbe_init_pre(void);
374 	extern void	cbe_init(void);
375 	extern void	clock_tick_init_pre(void);
376 	extern void	clock_tick_init_post(void);
377 	extern void	clock_init(void);
378 	extern void	physio_bufs_init(void);
379 	extern void	pm_cfb_setup_intr(void);
380 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
381 	extern void	start_other_cpus(int);
382 	extern void	sysevent_evc_thrinit();
383 	extern kmutex_t	ualock;
384 #if defined(__x86)
385 	extern void	fastboot_post_startup(void);
386 	extern void	progressbar_start(void);
387 #endif
388 	/*
389 	 * In the horrible world of x86 in-lines, you can't get symbolic
390 	 * structure offsets a la genassym.  This assertion is here so
391 	 * that the next poor slob who innocently changes the offset of
392 	 * cpu_thread doesn't waste as much time as I just did finding
393 	 * out that it's hard-coded in i86/ml/i86.il.  Similarly for
394 	 * curcpup.  You're welcome.
395 	 */
396 	ASSERT(CPU == CPU->cpu_self);
397 	ASSERT(curthread == CPU->cpu_thread);
398 	ASSERT_STACK_ALIGNED();
399 
400 	/*
401 	 * We take the ualock until we have completed the startup
402 	 * to prevent kadmin() from disrupting this work. In particular,
403 	 * we don't want kadmin() to bring the system down while we are
404 	 * trying to start it up.
405 	 */
406 	mutex_enter(&ualock);
407 
408 	/*
409 	 * Setup root lgroup and leaf lgroup for CPU 0
410 	 */
411 	lgrp_init(LGRP_INIT_STAGE2);
412 
413 	/*
414 	 * Once 'startup()' completes, the thread_reaper() daemon would be
415 	 * created(in thread_init()). After that, it is safe to create threads
416 	 * that could exit. These exited threads will get reaped.
417 	 */
418 	startup();
419 	segkmem_gc();
420 	callb_init();
421 	cbe_init_pre();	/* x86 must initialize gethrtimef before timer_init */
422 	ddi_periodic_init();
423 	cbe_init();
424 	callout_init();	/* callout table MUST be init'd after cyclics */
425 	clock_tick_init_pre();
426 	clock_init();
427 
428 #if defined(__x86)
429 	/*
430 	 * The progressbar thread uses cv_reltimedwait() and hence needs to be
431 	 * started after the callout mechanism has been initialized.
432 	 */
433 	progressbar_start();
434 #endif
435 	/*
436 	 * On some platforms, clkinitf() changes the timing source that
437 	 * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
438 	 * clkinitf(), so re-initialize the microstate counters after the
439 	 * timesource has been chosen.
440 	 */
441 	init_mstate(&t0, LMS_SYSTEM);
442 	init_cpu_mstate(CPU, CMS_SYSTEM);
443 
444 	/*
445 	 * May need to probe to determine latencies from CPU 0 after
446 	 * gethrtime() comes alive in cbe_init() and before enabling interrupts
447 	 * and copy and release any temporary memory allocated with BOP_ALLOC()
448 	 * before release_bootstrap() frees boot memory
449 	 */
450 	lgrp_init(LGRP_INIT_STAGE3);
451 
452 	/*
453 	 * Call all system initialization functions.
454 	 */
455 	for (initptr = &init_tbl[0]; *initptr; initptr++)
456 		(**initptr)();
457 	/*
458 	 * Load iSCSI boot properties
459 	 */
460 	ld_ib_prop();
461 	/*
462 	 * initialize vm related stuff.
463 	 */
464 	vm_init();
465 
466 	/*
467 	 * initialize buffer pool for raw I/O requests
468 	 */
469 	physio_bufs_init();
470 
471 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
472 
473 	/*
474 	 * Drop the interrupt level and allow interrupts.  At this point
475 	 * the DDI guarantees that interrupts are enabled.
476 	 */
477 	(void) spl0();
478 	interrupts_unleashed = 1;
479 
480 	/*
481 	 * Create kmem cache for proc structures
482 	 */
483 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
484 	    0, NULL, NULL, NULL, NULL, NULL, 0);
485 
486 	vfs_mountroot();	/* Mount the root file system */
487 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
488 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
489 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
490 				/* after vfs_mountroot() so hrestime is valid */
491 
492 	post_startup();
493 	swaploaded = 1;
494 
495 	/*
496 	 * Initialize Solaris Audit Subsystem
497 	 */
498 	audit_init();
499 
500 	/*
501 	 * Start the periodic hash rescale for all vmem arenas before we load
502 	 * protocol modules and drivers via strplumb() below.  Some drivers
503 	 * might rely on heavy vmem operations that could hurt performance
504 	 * without the rescale.
505 	 */
506 	vmem_update(NULL);
507 
508 	/*
509 	 * Plumb the protocol modules and drivers only if we are not
510 	 * networked booted, in this case we already did it in rootconf().
511 	 */
512 	if (netboot == 0 && iscsiboot_prop == NULL)
513 		(void) strplumb();
514 
515 	gethrestime(&PTOU(curproc)->u_start);
516 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
517 	p->p_mstart = gethrtime();
518 
519 	/*
520 	 * Perform setup functions that can only be done after root
521 	 * and swap have been set up.
522 	 */
523 	consconfig();
524 #ifndef	__sparc
525 	release_bootstrap();
526 #endif
527 
528 	/*
529 	 * attach drivers with ddi-forceattach prop
530 	 * It must be done early enough to load hotplug drivers (e.g.
531 	 * pcmcia nexus) so that devices enumerated via hotplug is
532 	 * available before I/O subsystem is fully initialized.
533 	 */
534 	i_ddi_forceattach_drivers();
535 
536 	/*
537 	 * Set the scan rate and other parameters of the paging subsystem.
538 	 */
539 	setupclock(0);
540 
541 	/*
542 	 * Initialize process 0's lwp directory and lwpid hash table.
543 	 */
544 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
545 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
546 	p->p_lwpdir_sz = 2;
547 	p->p_tidhash = p0_tidhash;
548 	p->p_tidhash_sz = 2;
549 	p0_lep.le_thread = curthread;
550 	p0_lep.le_lwpid = curthread->t_tid;
551 	p0_lep.le_start = curthread->t_start;
552 	lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
553 
554 	/*
555 	 * Initialize extended accounting.
556 	 */
557 	exacct_init();
558 
559 	/*
560 	 * Initialize threads of sysevent event channels
561 	 */
562 	sysevent_evc_thrinit();
563 
564 	/*
565 	 * This must be done after post_startup() but before
566 	 * start_other_cpus()
567 	 */
568 	lgrp_init(LGRP_INIT_STAGE4);
569 
570 	/*
571 	 * Perform MP initialization, if any.
572 	 */
573 	start_other_cpus(0);
574 
575 #ifdef	__sparc
576 	/*
577 	 * Release bootstrap here since PROM interfaces are
578 	 * used to start other CPUs above.
579 	 */
580 	release_bootstrap();
581 #endif
582 
583 	/*
584 	 * Finish lgrp initialization after all CPUS are brought online.
585 	 */
586 	lgrp_init(LGRP_INIT_STAGE5);
587 
588 	/*
589 	 * After mp_init(), number of cpus are known (this is
590 	 * true for the time being, when there are actually
591 	 * hot pluggable cpus then this scheme  would not do).
592 	 * Any per cpu initialization is done here.
593 	 */
594 	kmem_mp_init();
595 
596 	clock_tick_init_post();
597 
598 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
599 		(**initptr)();
600 
601 	/*
602 	 * These must be called after start_other_cpus
603 	 */
604 	pm_cfb_setup_intr();
605 #if defined(__x86)
606 	fastboot_post_startup();
607 #endif
608 
609 	/*
610 	 * Make init process; enter scheduling loop with system process.
611 	 *
612 	 * Note that we manually assign the pids for these processes, for
613 	 * historical reasons.  If more pre-assigned pids are needed,
614 	 * FAMOUS_PIDS will have to be updated.
615 	 */
616 
617 	/* create init process */
618 	if (newproc(start_init, NULL, defaultcid, 59, NULL,
619 	    FAMOUS_PID_INIT))
620 		panic("main: unable to fork init.");
621 
622 	/* create pageout daemon */
623 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
624 	    FAMOUS_PID_PAGEOUT))
625 		panic("main: unable to fork pageout()");
626 
627 	/* create fsflush daemon */
628 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
629 	    FAMOUS_PID_FSFLUSH))
630 		panic("main: unable to fork fsflush()");
631 
632 	/* create cluster process if we're a member of one */
633 	if (cluster_bootflags & CLUSTER_BOOTED) {
634 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri,
635 		    NULL, 0)) {
636 			panic("main: unable to fork cluster()");
637 		}
638 	}
639 
640 	/*
641 	 * Create system threads (threads are associated with p0)
642 	 */
643 
644 	/* create module uninstall daemon */
645 	/* BugID 1132273. If swapping over NFS need a bigger stack */
646 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
647 	    NULL, 0, &p0, TS_RUN, minclsyspri);
648 
649 	(void) thread_create(NULL, 0, seg_pasync_thread,
650 	    NULL, 0, &p0, TS_RUN, minclsyspri);
651 
652 	pid_setmin();
653 
654 	/* system is now ready */
655 	mutex_exit(&ualock);
656 
657 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
658 	bcopy("sched", PTOU(curproc)->u_comm, 5);
659 	sched();
660 	/* NOTREACHED */
661 }
662