xref: /minix3/minix/servers/vm/main.c (revision 0b98e8aad89f2bd4ba80b523d73cf29e9dd82ce1)
1 
2 #define _SYSTEM		1
3 
4 #include <minix/callnr.h>
5 #include <minix/com.h>
6 #include <minix/config.h>
7 #include <minix/const.h>
8 #include <minix/ds.h>
9 #include <minix/endpoint.h>
10 #include <minix/minlib.h>
11 #include <minix/type.h>
12 #include <minix/ipc.h>
13 #include <minix/sysutil.h>
14 #include <minix/syslib.h>
15 #include <minix/const.h>
16 #include <minix/bitmap.h>
17 #include <minix/rs.h>
18 #include <minix/vfsif.h>
19 
20 #include <sys/exec.h>
21 
22 #include <libexec.h>
23 #include <ctype.h>
24 #include <errno.h>
25 #include <string.h>
26 #include <env.h>
27 #include <stdio.h>
28 #include <assert.h>
29 
30 #define _MAIN 1
31 #include "glo.h"
32 #include "proto.h"
33 #include "util.h"
34 #include "vm.h"
35 #include "sanitycheck.h"
36 
37 extern int missing_spares;
38 
39 #include <machine/archtypes.h>
40 #include <sys/param.h>
41 #include "kernel/const.h"
42 #include "kernel/config.h"
43 #include "kernel/proc.h"
44 
45 #include <signal.h>
46 #include <lib.h>
47 
48 /* Table of calls and a macro to test for being in range. */
49 struct {
50 	int (*vmc_func)(message *);	/* Call handles message. */
51 	const char *vmc_name;			/* Human-readable string. */
52 } vm_calls[NR_VM_CALLS];
53 
54 /* Macro to verify call range and map 'high' range to 'base' range
55  * (starting at 0) in one. Evaluates to zero-based call number if call
56  * number is valid, returns -1 otherwise.
57  */
58 #define CALLNUMBER(c) (((c) >= VM_RQ_BASE && 				\
59 			(c) < VM_RQ_BASE + ELEMENTS(vm_calls)) ?	\
60 			((c) - VM_RQ_BASE) : -1)
61 
62 static int map_service(struct rprocpub *rpub);
63 
64 static struct rprocpub rprocpub[NR_SYS_PROCS];
65 int __vm_init_fresh;
66 
67 /* SEF functions and variables. */
68 static void sef_local_startup(void);
69 static int sef_cb_init_lu_restart(int type, sef_init_info_t *info);
70 static int sef_cb_init_fresh(int type, sef_init_info_t *info);
71 static void sef_cb_signal_handler(int signo);
72 
73 void init_vm(void);
74 
75 int do_sef_init_request(message *);
76 
77 /*===========================================================================*
78  *				is_first_time				     *
79  *===========================================================================*/
80 static int is_first_time(void)
81 {
82 	struct proc rs_proc;
83 	int r;
84 
85 	if ((r = sys_getproc(&rs_proc, RS_PROC_NR)) != OK)
86 		panic("VM: couldn't get RS process data: %d", r);
87 
88 	return RTS_ISSET(&rs_proc, RTS_BOOTINHIBIT);
89 }
90 
91 /*===========================================================================*
92  *				main					     *
93  *===========================================================================*/
94 int main(void)
95 {
96   message msg;
97   int result, who_e, rcv_sts;
98   int caller_slot;
99 
100   /* Initialize system so that all processes are runnable the first time. */
101   if (is_first_time()) {
102 	init_vm();
103 	__vm_init_fresh=1;
104   }
105 
106   /* SEF local startup. */
107   sef_local_startup();
108   __vm_init_fresh=0;
109 
110   SANITYCHECK(SCL_TOP);
111 
112   /* This is VM's main loop. */
113   while (TRUE) {
114 	int r, c;
115 	int type;
116 	int transid = 0;	/* VFS transid if any */
117 
118 	SANITYCHECK(SCL_TOP);
119 	if(missing_spares > 0) {
120 		alloc_cycle();	/* mem alloc code wants to be called */
121 	}
122 
123   	if ((r=sef_receive_status(ANY, &msg, &rcv_sts)) != OK)
124 		panic("sef_receive_status() error: %d", r);
125 
126 	if (is_ipc_notify(rcv_sts)) {
127 		/* Unexpected ipc_notify(). */
128 		printf("VM: ignoring ipc_notify() from %d\n", msg.m_source);
129 		continue;
130 	}
131 	who_e = msg.m_source;
132 	if(vm_isokendpt(who_e, &caller_slot) != OK)
133 		panic("invalid caller %d", who_e);
134 
135 	/* We depend on this being false for the initialized value. */
136 	assert(!IS_VFS_FS_TRANSID(transid));
137 
138 	type = msg.m_type;
139 	c = CALLNUMBER(type);
140 	result = ENOSYS; /* Out of range or restricted calls return this. */
141 
142 	transid = TRNS_GET_ID(msg.m_type);
143 
144 	if((msg.m_source == VFS_PROC_NR) && IS_VFS_FS_TRANSID(transid)) {
145 		/* If it's a request from VFS, it might have a transaction id. */
146 		msg.m_type = TRNS_DEL_ID(msg.m_type);
147 
148 		/* Calls that use the transid */
149 		result = do_procctl(&msg, transid);
150 	} else if(msg.m_type == RS_INIT && msg.m_source == RS_PROC_NR) {
151 		result = do_sef_init_request(&msg);
152 		if(result != OK) panic("do_sef_init_request failed!\n");
153 		result = SUSPEND;	/* do not reply to RS */
154 	} else if (msg.m_type == VM_PAGEFAULT) {
155 		if (!IPC_STATUS_FLAGS_TEST(rcv_sts, IPC_FLG_MSG_FROM_KERNEL)) {
156 			printf("VM: process %d faked VM_PAGEFAULT "
157 					"message!\n", msg.m_source);
158 		}
159 		do_pagefaults(&msg);
160 		/*
161 		 * do not reply to this call, the caller is unblocked by
162 		 * a sys_vmctl() call in do_pagefaults if success. VM panics
163 		 * otherwise
164 		 */
165 		continue;
166 	} else if(c < 0 || !vm_calls[c].vmc_func) {
167 		/* out of range or missing callnr */
168 	} else {
169 		if (acl_check(&vmproc[caller_slot], c) != OK) {
170 			printf("VM: unauthorized %s by %d\n",
171 					vm_calls[c].vmc_name, who_e);
172 		} else {
173 			SANITYCHECK(SCL_FUNCTIONS);
174 			result = vm_calls[c].vmc_func(&msg);
175 			SANITYCHECK(SCL_FUNCTIONS);
176 		}
177 	}
178 
179 	/* Send reply message, unless the return code is SUSPEND,
180 	 * which is a pseudo-result suppressing the reply message.
181 	 */
182 	if(result != SUSPEND) {
183 		msg.m_type = result;
184 
185 		assert(!IS_VFS_FS_TRANSID(transid));
186 
187 		if((r=ipc_send(who_e, &msg)) != OK) {
188 			printf("VM: couldn't send %d to %d (err %d)\n",
189 				msg.m_type, who_e, r);
190 			panic("ipc_send() error");
191 		}
192 	}
193   }
194   return(OK);
195 }
196 
197 static void sef_cb_lu_state_changed(int old_state, int state)
198 {
199 /* Called whenever the live-update state changes. We need to restore certain
200  * state in the old VM instance after a live update has failed, because some
201  * but not all memory is shared between the two VM instances.
202  */
203   struct vmproc *vmp;
204 
205   if (state == SEF_LU_STATE_NULL) {
206 	/* Undo some of the changes that may have been made by the new VM
207 	 * instance.  If the new VM instance is us, nothing happens.
208 	 */
209 	vmp = &vmproc[VM_PROC_NR];
210 
211 	/* Rebind page tables. */
212 	pt_bind(&vmp->vm_pt, vmp);
213 	pt_clearmapcache();
214 
215 	/* Readjust process references. */
216 	adjust_proc_refs();
217   }
218 }
219 
220 static void sef_local_startup(void)
221 {
222 	/* Register init callbacks. */
223 	sef_setcb_init_fresh(sef_cb_init_fresh);
224 	sef_setcb_init_lu(sef_cb_init_lu_restart);
225 	sef_setcb_init_restart(sef_cb_init_lu_restart);
226 	/* In order to avoid a deadlock at boot time, send the first RS_INIT
227 	 * reply to RS asynchronously. After that, use sendrec as usual.
228 	 */
229 	if (__vm_init_fresh)
230 		sef_setcb_init_response(sef_cb_init_response_rs_asyn_once);
231 
232 	/* Register live update callbacks. */
233 	sef_setcb_lu_state_changed(sef_cb_lu_state_changed);
234 
235 	/* Register signal callbacks. */
236 	sef_setcb_signal_handler(sef_cb_signal_handler);
237 
238 	/* Let SEF perform startup. */
239 	sef_startup();
240 }
241 
242 static int sef_cb_init_fresh(int type, sef_init_info_t *info)
243 {
244 	int s, i;
245 
246 	/* Map all the services in the boot image. */
247 	if((s = sys_safecopyfrom(RS_PROC_NR, info->rproctab_gid, 0,
248 		(vir_bytes) rprocpub, sizeof(rprocpub))) != OK) {
249 		panic("vm: sys_safecopyfrom (rs) failed: %d", s);
250 	}
251 
252 	for(i=0;i < NR_BOOT_PROCS;i++) {
253 		if(rprocpub[i].in_use) {
254 			if((s = map_service(&rprocpub[i])) != OK) {
255 				panic("unable to map service: %d", s);
256 			}
257 		}
258 	}
259 
260 	return(OK);
261 }
262 
263 static struct vmproc *init_proc(endpoint_t ep_nr)
264 {
265 	struct boot_image *ip;
266 
267 	for (ip = &kernel_boot_info.boot_procs[0];
268 		ip < &kernel_boot_info.boot_procs[NR_BOOT_PROCS]; ip++) {
269 		struct vmproc *vmp;
270 
271 		if(ip->proc_nr != ep_nr) continue;
272 
273 		if(ip->proc_nr >= _NR_PROCS || ip->proc_nr < 0)
274 			panic("proc: %d", ip->proc_nr);
275 
276 		vmp = &vmproc[ip->proc_nr];
277 		assert(!(vmp->vm_flags & VMF_INUSE));	/* no double procs */
278 		clear_proc(vmp);
279 		vmp->vm_flags = VMF_INUSE;
280 		vmp->vm_endpoint = ip->endpoint;
281 		vmp->vm_boot = ip;
282 
283 		return vmp;
284 	}
285 
286 	panic("no init_proc");
287 }
288 
289 struct vm_exec_info {
290 	struct exec_info execi;
291 	struct boot_image *ip;
292 	struct vmproc *vmp;
293 };
294 
295 static int libexec_copy_physcopy(struct exec_info *execi,
296 	off_t off, vir_bytes vaddr, size_t len)
297 {
298 	vir_bytes end;
299 	struct vm_exec_info *ei = execi->opaque;
300 	end = ei->ip->start_addr + ei->ip->len;
301 	assert(ei->ip->start_addr + off + len <= end);
302 	return sys_physcopy(NONE, ei->ip->start_addr + off,
303 		execi->proc_e, vaddr, len, 0);
304 }
305 
306 static void boot_alloc(struct exec_info *execi, off_t vaddr,
307 	size_t len, int flags)
308 {
309 	struct vmproc *vmp = ((struct vm_exec_info *) execi->opaque)->vmp;
310 
311 	if(!(map_page_region(vmp, vaddr, 0, len,
312 		VR_ANON | VR_WRITABLE | VR_UNINITIALIZED, flags,
313 		&mem_type_anon))) {
314 		panic("VM: exec: map_page_region for boot process failed");
315 	}
316 }
317 
318 static int libexec_alloc_vm_prealloc(struct exec_info *execi,
319 	vir_bytes vaddr, size_t len)
320 {
321 	boot_alloc(execi, vaddr, len, MF_PREALLOC);
322 	return OK;
323 }
324 
325 static int libexec_alloc_vm_ondemand(struct exec_info *execi,
326 	vir_bytes vaddr, size_t len)
327 {
328 	boot_alloc(execi, vaddr, len, 0);
329 	return OK;
330 }
331 
332 static void exec_bootproc(struct vmproc *vmp, struct boot_image *ip)
333 {
334 	struct vm_exec_info vmexeci;
335 	struct exec_info *execi = &vmexeci.execi;
336 	char hdr[VM_PAGE_SIZE];
337 
338 	size_t frame_size = 0;	/* Size of the new initial stack. */
339 	int argc = 0;		/* Argument count. */
340 	int envc = 0;		/* Environment count */
341 	char overflow = 0;	/* No overflow yet. */
342 	struct ps_strings *psp;
343 
344 	int vsp = 0;	/* (virtual) Stack pointer in new address space. */
345 	char *argv[] = { ip->proc_name, NULL };
346 	char *envp[] = { NULL };
347 	char *path = ip->proc_name;
348 	char frame[VM_PAGE_SIZE];
349 
350 	memset(&vmexeci, 0, sizeof(vmexeci));
351 
352 	if(pt_new(&vmp->vm_pt) != OK)
353 		panic("VM: no new pagetable");
354 
355 	if(pt_bind(&vmp->vm_pt, vmp) != OK)
356 		panic("VM: pt_bind failed");
357 
358 	if(sys_physcopy(NONE, ip->start_addr, SELF,
359 		(vir_bytes) hdr, sizeof(hdr), 0) != OK)
360 		panic("can't look at boot proc header");
361 
362 	execi->stack_high = kernel_boot_info.user_sp;
363 	execi->stack_size = DEFAULT_STACK_LIMIT;
364 	execi->proc_e = vmp->vm_endpoint;
365 	execi->hdr = hdr;
366 	execi->hdr_len = sizeof(hdr);
367 	strlcpy(execi->progname, ip->proc_name, sizeof(execi->progname));
368 	execi->frame_len = 0;
369 	execi->opaque = &vmexeci;
370 	execi->filesize = ip->len;
371 
372 	vmexeci.ip = ip;
373 	vmexeci.vmp = vmp;
374 
375 	/* callback functions and data */
376 	execi->copymem = libexec_copy_physcopy;
377 	execi->clearproc = NULL;
378 	execi->clearmem = libexec_clear_sys_memset;
379 	execi->allocmem_prealloc_junk = libexec_alloc_vm_prealloc;
380 	execi->allocmem_prealloc_cleared = libexec_alloc_vm_prealloc;
381 	execi->allocmem_ondemand = libexec_alloc_vm_ondemand;
382 
383 	if (libexec_load_elf(execi) != OK)
384 		panic("vm: boot process load of process %s (ep=%d) failed\n",
385 			execi->progname, vmp->vm_endpoint);
386 
387 	/* Setup a minimal stack. */
388 	minix_stack_params(path, argv, envp, &frame_size, &overflow, &argc,
389 		&envc);
390 
391 	/* The party is off if there is an overflow, or it is too big for our
392 	 * pre-allocated space. */
393 	if(overflow || frame_size > sizeof(frame))
394 		panic("vm: could not alloc stack for boot process %s (ep=%d)\n",
395 			execi->progname, vmp->vm_endpoint);
396 
397 	minix_stack_fill(path, argc, argv, envc, envp, frame_size, frame, &vsp,
398 		&psp);
399 
400 	if(handle_memory_once(vmp, vsp, frame_size, 1) != OK)
401 		panic("vm: could not map stack for boot process %s (ep=%d)\n",
402 			execi->progname, vmp->vm_endpoint);
403 
404 	if(sys_datacopy(SELF, (vir_bytes)frame, vmp->vm_endpoint, vsp, frame_size) != OK)
405 		panic("vm: could not copy stack for boot process %s (ep=%d)\n",
406 			execi->progname, vmp->vm_endpoint);
407 
408 	if(sys_exec(vmp->vm_endpoint, (vir_bytes)vsp,
409 		   (vir_bytes)execi->progname, execi->pc,
410 		   vsp + ((int)psp - (int)frame)) != OK)
411 		panic("vm: boot process exec of process %s (ep=%d) failed\n",
412 			execi->progname,vmp->vm_endpoint);
413 
414 	/* make it runnable */
415 	if(sys_vmctl(vmp->vm_endpoint, VMCTL_BOOTINHIBIT_CLEAR, 0) != OK)
416 		panic("VMCTL_BOOTINHIBIT_CLEAR failed");
417 }
418 
419 static int do_procctl_notrans(message *msg)
420 {
421 	int transid = 0;
422 
423 	assert(!IS_VFS_FS_TRANSID(transid));
424 
425 	return do_procctl(msg, transid);
426 }
427 
428 void init_vm(void)
429 {
430 	int s, i;
431 	static struct memory mem_chunks[NR_MEMS];
432 	struct boot_image *ip;
433 	extern void __minix_init(void);
434 	multiboot_module_t *mod;
435 	vir_bytes kern_dyn, kern_static;
436 
437 #if SANITYCHECKS
438 	incheck = nocheck = 0;
439 #endif
440 
441 	/* Retrieve various crucial boot parameters */
442 	if(OK != (s=sys_getkinfo(&kernel_boot_info))) {
443 		panic("couldn't get bootinfo: %d", s);
444 	}
445 
446 	/* Turn file mmap on? */
447 	enable_filemap=1;	/* yes by default */
448 	env_parse("filemap", "d", 0, &enable_filemap, 0, 1);
449 
450 	/* Sanity check */
451 	assert(kernel_boot_info.mmap_size > 0);
452 	assert(kernel_boot_info.mods_with_kernel > 0);
453 
454 	/* Get chunks of available memory. */
455 	get_mem_chunks(mem_chunks);
456 
457 	/* Set table to 0. This invalidates all slots (clear VMF_INUSE). */
458 	memset(vmproc, 0, sizeof(vmproc));
459 
460 	for(i = 0; i < ELEMENTS(vmproc); i++) {
461 		vmproc[i].vm_slot = i;
462 	}
463 
464 	/* Initialize ACL data structures. */
465 	acl_init();
466 
467 	/* region management initialization. */
468 	map_region_init();
469 
470 	/* Initialize tables to all physical memory. */
471 	mem_init(mem_chunks);
472 
473 	/* Architecture-dependent initialization. */
474 	init_proc(VM_PROC_NR);
475 	pt_init();
476 
477 	/* Acquire kernel ipc vectors that weren't available
478 	 * before VM had determined kernel mappings
479 	 */
480 	__minix_init();
481 
482 	/* The kernel's freelist does not include boot-time modules; let
483 	 * the allocator know that the total memory is bigger.
484 	 */
485 	for (mod = &kernel_boot_info.module_list[0];
486 		mod < &kernel_boot_info.module_list[kernel_boot_info.mods_with_kernel-1]; mod++) {
487 		phys_bytes len = mod->mod_end-mod->mod_start+1;
488 		len = roundup(len, VM_PAGE_SIZE);
489 		mem_add_total_pages(len/VM_PAGE_SIZE);
490 	}
491 
492 	kern_dyn = kernel_boot_info.kernel_allocated_bytes_dynamic;
493 	kern_static = kernel_boot_info.kernel_allocated_bytes;
494 	kern_static = roundup(kern_static, VM_PAGE_SIZE);
495 	mem_add_total_pages((kern_dyn + kern_static)/VM_PAGE_SIZE);
496 
497 	/* Give these processes their own page table. */
498 	for (ip = &kernel_boot_info.boot_procs[0];
499 		ip < &kernel_boot_info.boot_procs[NR_BOOT_PROCS]; ip++) {
500 		struct vmproc *vmp;
501 
502 		if(ip->proc_nr < 0) continue;
503 
504 		assert(ip->start_addr);
505 
506 		/* VM has already been set up by the kernel and pt_init().
507 		 * Any other boot process is already in memory and is set up
508 		 * here.
509 		 */
510 		if(ip->proc_nr == VM_PROC_NR) continue;
511 
512 		vmp = init_proc(ip->proc_nr);
513 
514 		exec_bootproc(vmp, ip);
515 
516 		/* Free the file blob */
517 		assert(!(ip->start_addr % VM_PAGE_SIZE));
518 		ip->len = roundup(ip->len, VM_PAGE_SIZE);
519 		free_mem(ABS2CLICK(ip->start_addr), ABS2CLICK(ip->len));
520 	}
521 
522 	/* Set up table of calls. */
523 #define CALLMAP(code, func) { int _cmi;		      \
524 	_cmi=CALLNUMBER(code);				\
525 	assert(_cmi >= 0);					\
526 	assert(_cmi < NR_VM_CALLS);		\
527 	vm_calls[_cmi].vmc_func = (func); 	      \
528 	vm_calls[_cmi].vmc_name = #code;	      \
529 }
530 
531 	/* Set call table to 0. This invalidates all calls (clear
532 	 * vmc_func).
533 	 */
534 	memset(vm_calls, 0, sizeof(vm_calls));
535 
536 	/* Basic VM calls. */
537 	CALLMAP(VM_MMAP, do_mmap);
538 	CALLMAP(VM_MUNMAP, do_munmap);
539 	CALLMAP(VM_MAP_PHYS, do_map_phys);
540 	CALLMAP(VM_UNMAP_PHYS, do_munmap);
541 
542 	/* Calls from PM. */
543 	CALLMAP(VM_EXIT, do_exit);
544 	CALLMAP(VM_FORK, do_fork);
545 	CALLMAP(VM_BRK, do_brk);
546 	CALLMAP(VM_WILLEXIT, do_willexit);
547 	CALLMAP(VM_NOTIFY_SIG, do_notify_sig);
548 
549 	CALLMAP(VM_PROCCTL, do_procctl_notrans);
550 
551 	/* Calls from VFS. */
552 	CALLMAP(VM_VFS_REPLY, do_vfs_reply);
553 	CALLMAP(VM_VFS_MMAP, do_vfs_mmap);
554 
555 	/* Calls from RS */
556 	CALLMAP(VM_RS_SET_PRIV, do_rs_set_priv);
557 	CALLMAP(VM_RS_PREPARE, do_rs_prepare);
558 	CALLMAP(VM_RS_UPDATE, do_rs_update);
559 	CALLMAP(VM_RS_MEMCTL, do_rs_memctl);
560 
561 	/* Generic calls. */
562 	CALLMAP(VM_REMAP, do_remap);
563 	CALLMAP(VM_REMAP_RO, do_remap);
564 	CALLMAP(VM_GETPHYS, do_get_phys);
565 	CALLMAP(VM_SHM_UNMAP, do_munmap);
566 	CALLMAP(VM_GETREF, do_get_refcount);
567 	CALLMAP(VM_INFO, do_info);
568 	CALLMAP(VM_QUERY_EXIT, do_query_exit);
569 	CALLMAP(VM_WATCH_EXIT, do_watch_exit);
570 
571 	/* Cache blocks. */
572 	CALLMAP(VM_MAPCACHEPAGE, do_mapcache);
573 	CALLMAP(VM_SETCACHEPAGE, do_setcache);
574 	CALLMAP(VM_FORGETCACHEPAGE, do_forgetcache);
575 	CALLMAP(VM_CLEARCACHE, do_clearcache);
576 
577 	/* getrusage */
578 	CALLMAP(VM_GETRUSAGE, do_getrusage);
579 
580 	/* Initialize the structures for queryexit */
581 	init_query_exit();
582 
583 	/* Mark VM instances. */
584 	num_vm_instances = 1;
585 	vmproc[VM_PROC_NR].vm_flags |= VMF_VM_INSTANCE;
586 
587 	/* Let SEF know about VM mmapped regions. */
588 	s = sef_llvm_add_special_mem_region((void*)VM_OWN_HEAPBASE,
589 	    VM_OWN_MMAPTOP-VM_OWN_HEAPBASE, "%MMAP_ALL");
590 	if(s < 0) {
591 	    printf("VM: st_add_special_mmapped_region failed %d\n", s);
592 	}
593 }
594 
595 /*===========================================================================*
596  *			      sef_cb_init_vm_multi_lu			     *
597  *===========================================================================*/
598 static int sef_cb_init_vm_multi_lu(int type, sef_init_info_t *info)
599 {
600 	message m;
601 	int i, r;
602 	ipc_filter_el_t ipc_filter[IPCF_MAX_ELEMENTS];
603 	int num_elements;
604 
605 	if(type != SEF_INIT_LU || !(info->flags & SEF_LU_MULTI)) {
606 	    return OK;
607 	}
608 
609 	/* If this is a multi-component update, we need to perform the update
610 	 * for services that need to be updated. In addition, make sure VM
611 	 * can only receive messages from RS, tasks, and other services being
612 	 * updated until RS specifically sends a special update cancel message.
613 	 * This is necessary to limit the number of VM state changes to support
614 	 * rollback. Allow only safe message types for safe updates.
615 	 */
616 	memset(ipc_filter, 0, sizeof(ipc_filter));
617 	num_elements = 0;
618 	ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE;
619 	ipc_filter[num_elements++].m_source = RS_PROC_NR;
620 	if((r = sys_safecopyfrom(RS_PROC_NR, info->rproctab_gid, 0,
621 	    (vir_bytes) rprocpub, NR_SYS_PROCS*sizeof(struct rprocpub))) != OK) {
622 	    panic("sys_safecopyfrom failed: %d", r);
623 	}
624 	m.m_source = VM_PROC_NR;
625 	for(i=0;i < NR_SYS_PROCS;i++) {
626 	    if(rprocpub[i].in_use && rprocpub[i].old_endpoint != NONE) {
627 	        if(num_elements <= IPCF_MAX_ELEMENTS-5) {
628                     /* VM_BRK is needed for normal operation during the live
629                      * update.  VM_INFO is needed for state transfer in the
630                      * light of holes.  Pagefaults and handle-memory requests
631                      * are blocked intentionally, as handling these would
632                      * prevent VM from being able to roll back.
633                      */
634 	            ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE | IPCF_MATCH_M_TYPE;
635 	            ipc_filter[num_elements].m_source = rprocpub[i].old_endpoint;
636 	            ipc_filter[num_elements++].m_type = VM_BRK;
637 	            ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE | IPCF_MATCH_M_TYPE;
638 	            ipc_filter[num_elements].m_source = rprocpub[i].new_endpoint;
639 	            ipc_filter[num_elements++].m_type = VM_BRK;
640 	            ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE | IPCF_MATCH_M_TYPE;
641 	            ipc_filter[num_elements].m_source = rprocpub[i].old_endpoint;
642 	            ipc_filter[num_elements++].m_type = VM_INFO;
643 	            ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE | IPCF_MATCH_M_TYPE;
644 	            ipc_filter[num_elements].m_source = rprocpub[i].new_endpoint;
645 	            ipc_filter[num_elements++].m_type = VM_INFO;
646 	            /* Make sure we can talk to any RS instance. */
647 	            if(rprocpub[i].old_endpoint == RS_PROC_NR) {
648 	                ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE;
649 	                ipc_filter[num_elements++].m_source = rprocpub[i].new_endpoint;
650 	            }
651 	            else if(rprocpub[i].new_endpoint == RS_PROC_NR) {
652 	                ipc_filter[num_elements].flags = IPCF_MATCH_M_SOURCE;
653 	                ipc_filter[num_elements++].m_source = rprocpub[i].old_endpoint;
654 	            }
655 	        }
656 	        else {
657 	            printf("sef_cb_init_vm_multi_lu: skipping ipc filter elements for %d and %d\n",
658 	                rprocpub[i].old_endpoint, rprocpub[i].new_endpoint);
659 	        }
660 	        if(rprocpub[i].sys_flags & SF_VM_UPDATE) {
661 	            m.m_lsys_vm_update.src = rprocpub[i].new_endpoint;
662 	            m.m_lsys_vm_update.dst = rprocpub[i].old_endpoint;
663 	            m.m_lsys_vm_update.flags = rprocpub[i].sys_flags;
664 	            r = do_rs_update(&m);
665 	            if(r != OK && r != SUSPEND) {
666 	                printf("sef_cb_init_vm_multi_lu: do_rs_update failed: %d", r);
667 	            }
668 	        }
669 	    }
670 	}
671 
672 	r = sys_statectl(SYS_STATE_ADD_IPC_WL_FILTER, ipc_filter, num_elements*sizeof(ipc_filter_el_t));
673 	if(r != OK) {
674 	    printf("sef_cb_init_vm_multi_lu: sys_statectl failed: %d", r);
675 	}
676 
677 	return OK;
678 }
679 
680 /*===========================================================================*
681  *			     sef_cb_init_lu_restart			     *
682  *===========================================================================*/
683 static int sef_cb_init_lu_restart(int type, sef_init_info_t *info)
684 {
685 /* Restart the vm server. */
686         int r;
687         endpoint_t old_e;
688         int old_p;
689         struct vmproc *old_vmp, *new_vmp;
690 
691         /* Perform default state transfer first. */
692         if(type == SEF_INIT_LU) {
693 		sef_setcb_init_restart(SEF_CB_INIT_RESTART_STATEFUL);
694 		r = SEF_CB_INIT_LU_DEFAULT(type, info);
695         }
696         else {
697 		r = SEF_CB_INIT_RESTART_STATEFUL(type, info);
698         }
699         if(r != OK) {
700 		return r;
701         }
702 
703 	/* Lookup slots for old process. */
704 	old_e = info->old_endpoint;
705 	if(vm_isokendpt(old_e, &old_p) != OK) {
706 		printf("sef_cb_init_lu_restart: bad old endpoint %d\n", old_e);
707 		return EINVAL;
708 	}
709 	old_vmp = &vmproc[old_p];
710 	new_vmp = &vmproc[VM_PROC_NR];
711 
712 	/* Swap proc slots and dynamic data. */
713 	if((r = swap_proc_slot(old_vmp, new_vmp)) != OK) {
714 		printf("sef_cb_init_lu_restart: swap_proc_slot failed\n");
715 		return r;
716 	}
717         if((r = swap_proc_dyn_data(old_vmp, new_vmp, 0)) != OK) {
718 		printf("sef_cb_init_lu_restart: swap_proc_dyn_data failed\n");
719 		return r;
720 	}
721 
722 	/* Rebind page tables. */
723 	pt_bind(&new_vmp->vm_pt, new_vmp);
724 	pt_bind(&old_vmp->vm_pt, old_vmp);
725 	pt_clearmapcache();
726 
727 	/* Adjust process references. */
728 	adjust_proc_refs();
729 
730 	/* Handle multi-component live update when necessary. */
731 	return sef_cb_init_vm_multi_lu(type, info);
732 }
733 
734 /*===========================================================================*
735  *                         sef_cb_signal_handler                             *
736  *===========================================================================*/
737 static void sef_cb_signal_handler(int signo)
738 {
739 	/* Check for known kernel signals, ignore anything else. */
740 	switch(signo) {
741 		/* There is a pending memory request from the kernel. */
742 		case SIGKMEM:
743 			do_memory();
744 		break;
745 	}
746 
747 	/* It can happen that we get stuck receiving signals
748 	 * without sef_receive() returning. We could need more memory
749 	 * though.
750 	 */
751 	if(missing_spares > 0) {
752 		alloc_cycle();	/* pagetable code wants to be called */
753 	}
754 
755 	pt_clearmapcache();
756 }
757 
758 /*===========================================================================*
759  *                             map_service                                   *
760  *===========================================================================*/
761 static int map_service(struct rprocpub *rpub)
762 {
763 /* Map a new service by initializing its call mask. */
764 	int r, proc_nr;
765 
766 	if ((r = vm_isokendpt(rpub->endpoint, &proc_nr)) != OK) {
767 		return r;
768 	}
769 
770 	/* Copy the call mask. */
771 	acl_set(&vmproc[proc_nr], rpub->vm_call_mask, !IS_RPUB_BOOT_USR(rpub));
772 
773 	return(OK);
774 }
775