xref: /minix3/minix/servers/pm/main.c (revision 433d6423c39e34ec4b79c950597bb2d236f886be)
1 /* This file contains the main program of the process manager and some related
2  * procedures.  When MINIX starts up, the kernel runs for a little while,
3  * initializing itself and its tasks, and then it runs PM and VFS.  Both PM
4  * and VFS initialize themselves as far as they can. PM asks the kernel for
5  * all free memory and starts serving requests.
6  *
7  * The entry points into this file are:
8  *   main:	starts PM running
9  *   reply:	send a reply to a process making a PM system call
10  */
11 
12 #include "pm.h"
13 #include <minix/callnr.h>
14 #include <minix/com.h>
15 #include <minix/ds.h>
16 #include <minix/type.h>
17 #include <minix/endpoint.h>
18 #include <minix/minlib.h>
19 #include <minix/type.h>
20 #include <minix/vm.h>
21 #include <signal.h>
22 #include <stdlib.h>
23 #include <fcntl.h>
24 #include <sys/resource.h>
25 #include <sys/utsname.h>
26 #include <sys/wait.h>
27 #include <machine/archtypes.h>
28 #include <env.h>
29 #include <assert.h>
30 #include "mproc.h"
31 
32 #include "kernel/const.h"
33 #include "kernel/config.h"
34 #include "kernel/proc.h"
35 
36 #if ENABLE_SYSCALL_STATS
37 EXTERN unsigned long calls_stats[NR_PM_CALLS];
38 #endif
39 
40 static int get_nice_value(int queue);
41 static void handle_vfs_reply(void);
42 
43 /* SEF functions and variables. */
44 static void sef_local_startup(void);
45 static int sef_cb_init_fresh(int type, sef_init_info_t *info);
46 
47 /*===========================================================================*
48  *				main					     *
49  *===========================================================================*/
50 int main()
51 {
52 /* Main routine of the process manager. */
53   unsigned int call_index;
54   int ipc_status, result;
55 
56   /* SEF local startup. */
57   sef_local_startup();
58 
59   /* This is PM's main loop-  get work and do it, forever and forever. */
60   while (TRUE) {
61 	/* Wait for the next message. */
62 	if (sef_receive_status(ANY, &m_in, &ipc_status) != OK)
63 		panic("PM sef_receive_status error");
64 
65 	/* Check for system notifications first. Special cases. */
66 	if (is_ipc_notify(ipc_status)) {
67 		if (_ENDPOINT_P(m_in.m_source) == CLOCK)
68 			expire_timers(m_in.m_notify.timestamp);
69 
70 		/* done, continue */
71 		continue;
72 	}
73 
74 	/* Extract useful information from the message. */
75 	who_e = m_in.m_source;	/* who sent the message */
76 	if (pm_isokendpt(who_e, &who_p) != OK)
77 		panic("PM got message from invalid endpoint: %d", who_e);
78 	mp = &mproc[who_p];	/* process slot of caller */
79 	call_nr = m_in.m_type;	/* system call number */
80 
81 	/* Drop delayed calls from exiting processes. */
82 	if (mp->mp_flags & EXITING)
83 		continue;
84 
85 	if (IS_VFS_PM_RS(call_nr) && who_e == VFS_PROC_NR) {
86 		handle_vfs_reply();
87 
88 		result = SUSPEND;		/* don't reply */
89 	} else if (IS_PM_CALL(call_nr)) {
90 		/* If the system call number is valid, perform the call. */
91 		call_index = (unsigned int) (call_nr - PM_BASE);
92 
93 		if (call_index < NR_PM_CALLS && call_vec[call_index] != NULL) {
94 #if ENABLE_SYSCALL_STATS
95 			calls_stats[call_index]++;
96 #endif
97 
98 			result = (*call_vec[call_index])();
99 		} else
100 			result = ENOSYS;
101 	} else
102 		result = ENOSYS;
103 
104 	/* Send reply. */
105 	if (result != SUSPEND) reply(who_p, result);
106   }
107   return(OK);
108 }
109 
110 /*===========================================================================*
111  *			       sef_local_startup			     *
112  *===========================================================================*/
113 static void sef_local_startup()
114 {
115   /* Register init callbacks. */
116   sef_setcb_init_fresh(sef_cb_init_fresh);
117   sef_setcb_init_restart(sef_cb_init_fail);
118 
119   /* No live update support for now. */
120 
121   /* Register signal callbacks. */
122   sef_setcb_signal_manager(process_ksig);
123 
124   /* Let SEF perform startup. */
125   sef_startup();
126 }
127 
128 /*===========================================================================*
129  *		            sef_cb_init_fresh                                *
130  *===========================================================================*/
131 static int sef_cb_init_fresh(int UNUSED(type), sef_init_info_t *UNUSED(info))
132 {
133 /* Initialize the process manager. */
134   int s;
135   static struct boot_image image[NR_BOOT_PROCS];
136   register struct boot_image *ip;
137   static char core_sigs[] = { SIGQUIT, SIGILL, SIGTRAP, SIGABRT,
138 				SIGEMT, SIGFPE, SIGBUS, SIGSEGV };
139   static char ign_sigs[] = { SIGCHLD, SIGWINCH, SIGCONT, SIGINFO };
140   static char noign_sigs[] = { SIGILL, SIGTRAP, SIGEMT, SIGFPE,
141 				SIGBUS, SIGSEGV };
142   register struct mproc *rmp;
143   register char *sig_ptr;
144   message mess;
145 
146   /* Initialize process table, including timers. */
147   for (rmp=&mproc[0]; rmp<&mproc[NR_PROCS]; rmp++) {
148 	init_timer(&rmp->mp_timer);
149 	rmp->mp_magic = MP_MAGIC;
150   }
151 
152   /* Build the set of signals which cause core dumps, and the set of signals
153    * that are by default ignored.
154    */
155   sigemptyset(&core_sset);
156   for (sig_ptr = core_sigs; sig_ptr < core_sigs+sizeof(core_sigs); sig_ptr++)
157 	sigaddset(&core_sset, *sig_ptr);
158   sigemptyset(&ign_sset);
159   for (sig_ptr = ign_sigs; sig_ptr < ign_sigs+sizeof(ign_sigs); sig_ptr++)
160 	sigaddset(&ign_sset, *sig_ptr);
161   sigemptyset(&noign_sset);
162   for (sig_ptr = noign_sigs; sig_ptr < noign_sigs+sizeof(noign_sigs); sig_ptr++)
163 	sigaddset(&noign_sset, *sig_ptr);
164 
165   /* Obtain a copy of the boot monitor parameters.
166    */
167   if ((s=sys_getmonparams(monitor_params, sizeof(monitor_params))) != OK)
168       panic("get monitor params failed: %d", s);
169 
170   /* Initialize PM's process table. Request a copy of the system image table
171    * that is defined at the kernel level to see which slots to fill in.
172    */
173   if (OK != (s=sys_getimage(image)))
174   	panic("couldn't get image table: %d", s);
175   procs_in_use = 0;				/* start populating table */
176   for (ip = &image[0]; ip < &image[NR_BOOT_PROCS]; ip++) {
177   	if (ip->proc_nr >= 0) {			/* task have negative nrs */
178   		procs_in_use += 1;		/* found user process */
179 
180 		/* Set process details found in the image table. */
181 		rmp = &mproc[ip->proc_nr];
182   		strlcpy(rmp->mp_name, ip->proc_name, PROC_NAME_LEN);
183   		(void) sigemptyset(&rmp->mp_ignore);
184   		(void) sigemptyset(&rmp->mp_sigmask);
185   		(void) sigemptyset(&rmp->mp_catch);
186 		if (ip->proc_nr == INIT_PROC_NR) {	/* user process */
187   			/* INIT is root, we make it father of itself. This is
188   			 * not really OK, INIT should have no father, i.e.
189   			 * a father with pid NO_PID. But PM currently assumes
190   			 * that mp_parent always points to a valid slot number.
191   			 */
192   			rmp->mp_parent = INIT_PROC_NR;
193   			rmp->mp_procgrp = rmp->mp_pid = INIT_PID;
194 			rmp->mp_flags |= IN_USE;
195 
196 			/* Set scheduling info */
197 			rmp->mp_scheduler = KERNEL;
198 			rmp->mp_nice = get_nice_value(USR_Q);
199 		}
200 		else {					/* system process */
201   			if(ip->proc_nr == RS_PROC_NR) {
202   				rmp->mp_parent = INIT_PROC_NR;
203   			}
204   			else {
205   				rmp->mp_parent = RS_PROC_NR;
206   			}
207   			rmp->mp_pid = get_free_pid();
208 			rmp->mp_flags |= IN_USE | PRIV_PROC;
209 
210 			/* RS schedules this process */
211 			rmp->mp_scheduler = NONE;
212 			rmp->mp_nice = get_nice_value(SRV_Q);
213 		}
214 
215 		/* Get kernel endpoint identifier. */
216 		rmp->mp_endpoint = ip->endpoint;
217 
218 		/* Tell VFS about this system process. */
219 		memset(&mess, 0, sizeof(mess));
220 		mess.m_type = VFS_PM_INIT;
221 		mess.VFS_PM_SLOT = ip->proc_nr;
222 		mess.VFS_PM_PID = rmp->mp_pid;
223 		mess.VFS_PM_ENDPT = rmp->mp_endpoint;
224   		if (OK != (s=ipc_send(VFS_PROC_NR, &mess)))
225 			panic("can't sync up with VFS: %d", s);
226   	}
227   }
228 
229   /* Tell VFS that no more system processes follow and synchronize. */
230   memset(&mess, 0, sizeof(mess));
231   mess.m_type = VFS_PM_INIT;
232   mess.VFS_PM_ENDPT = NONE;
233   if (ipc_sendrec(VFS_PROC_NR, &mess) != OK || mess.m_type != OK)
234 	panic("can't sync up with VFS");
235 
236 #if defined(__i386__)
237         uts_val.machine[0] = 'i';
238         strcpy(uts_val.machine + 1, itoa(getprocessor()));
239 #elif defined(__arm__)
240         strcpy(uts_val.machine, "arm");
241 #endif
242 
243  system_hz = sys_hz();
244 
245   /* Initialize user-space scheduling. */
246   sched_init();
247 
248   return(OK);
249 }
250 
251 /*===========================================================================*
252  *				reply					     *
253  *===========================================================================*/
254 void reply(proc_nr, result)
255 int proc_nr;			/* process to reply to */
256 int result;			/* result of call (usually OK or error #) */
257 {
258 /* Send a reply to a user process.  System calls may occasionally fill in other
259  * fields, this is only for the main return value and for sending the reply.
260  */
261   struct mproc *rmp;
262   int r;
263 
264   if(proc_nr < 0 || proc_nr >= NR_PROCS)
265       panic("reply arg out of range: %d", proc_nr);
266 
267   rmp = &mproc[proc_nr];
268   rmp->mp_reply.m_type = result;
269 
270   if ((r = ipc_sendnb(rmp->mp_endpoint, &rmp->mp_reply)) != OK)
271 	printf("PM can't reply to %d (%s): %d\n", rmp->mp_endpoint,
272 		rmp->mp_name, r);
273 }
274 
275 /*===========================================================================*
276  *				get_nice_value				     *
277  *===========================================================================*/
278 static int get_nice_value(queue)
279 int queue;				/* store mem chunks here */
280 {
281 /* Processes in the boot image have a priority assigned. The PM doesn't know
282  * about priorities, but uses 'nice' values instead. The priority is between
283  * MIN_USER_Q and MAX_USER_Q. We have to scale between PRIO_MIN and PRIO_MAX.
284  */
285   int nice_val = (queue - USER_Q) * (PRIO_MAX-PRIO_MIN+1) /
286       (MIN_USER_Q-MAX_USER_Q+1);
287   if (nice_val > PRIO_MAX) nice_val = PRIO_MAX;	/* shouldn't happen */
288   if (nice_val < PRIO_MIN) nice_val = PRIO_MIN;	/* shouldn't happen */
289   return nice_val;
290 }
291 
292 /*===========================================================================*
293  *				handle_vfs_reply       			     *
294  *===========================================================================*/
295 static void handle_vfs_reply()
296 {
297   struct mproc *rmp;
298   endpoint_t proc_e;
299   int r, proc_n, new_parent;
300 
301   /* VFS_PM_REBOOT is the only request not associated with a process.
302    * Handle its reply first.
303    */
304   if (call_nr == VFS_PM_REBOOT_REPLY) {
305 	/* Ask the kernel to abort. All system services, including
306 	 * the PM, will get a HARD_STOP notification. Await the
307 	 * notification in the main loop.
308 	 */
309 	sys_abort(abort_flag);
310 
311 	return;
312   }
313 
314   /* Get the process associated with this call */
315   proc_e = m_in.VFS_PM_ENDPT;
316 
317   if (pm_isokendpt(proc_e, &proc_n) != OK) {
318 	panic("handle_vfs_reply: got bad endpoint from VFS: %d", proc_e);
319   }
320 
321   rmp = &mproc[proc_n];
322 
323   /* Now that VFS replied, mark the process as VFS-idle again */
324   if (!(rmp->mp_flags & VFS_CALL))
325 	panic("handle_vfs_reply: reply without request: %d", call_nr);
326 
327   new_parent = rmp->mp_flags & NEW_PARENT;
328   rmp->mp_flags &= ~(VFS_CALL | NEW_PARENT);
329 
330   if (rmp->mp_flags & UNPAUSED)
331   	panic("handle_vfs_reply: UNPAUSED set on entry: %d", call_nr);
332 
333   /* Call-specific handler code */
334   switch (call_nr) {
335   case VFS_PM_SETUID_REPLY:
336   case VFS_PM_SETGID_REPLY:
337   case VFS_PM_SETGROUPS_REPLY:
338 	/* Wake up the original caller */
339 	reply(rmp-mproc, OK);
340 
341 	break;
342 
343   case VFS_PM_SETSID_REPLY:
344 	/* Wake up the original caller */
345 	reply(rmp-mproc, rmp->mp_procgrp);
346 
347 	break;
348 
349   case VFS_PM_EXEC_REPLY:
350 	exec_restart(rmp, m_in.VFS_PM_STATUS, (vir_bytes)m_in.VFS_PM_PC,
351 		(vir_bytes)m_in.VFS_PM_NEWSP,
352 		(vir_bytes)m_in.VFS_PM_NEWPS_STR);
353 
354 	break;
355 
356   case VFS_PM_EXIT_REPLY:
357 	exit_restart(rmp, FALSE /*dump_core*/);
358 
359 	break;
360 
361   case VFS_PM_CORE_REPLY:
362 	if (m_in.VFS_PM_STATUS == OK)
363 		rmp->mp_sigstatus |= WCOREFLAG;
364 
365 	exit_restart(rmp, TRUE /*dump_core*/);
366 
367 	break;
368 
369   case VFS_PM_FORK_REPLY:
370 	/* Schedule the newly created process ... */
371 	r = OK;
372 	if (rmp->mp_scheduler != KERNEL && rmp->mp_scheduler != NONE) {
373 		r = sched_start_user(rmp->mp_scheduler, rmp);
374 	}
375 
376 	/* If scheduling the process failed, we want to tear down the process
377 	 * and fail the fork */
378 	if (r != OK) {
379 		/* Tear down the newly created process */
380 		rmp->mp_scheduler = NONE; /* don't try to stop scheduling */
381 		exit_proc(rmp, -1, FALSE /*dump_core*/);
382 
383 		/* Wake up the parent with a failed fork (unless dead) */
384 		if (!new_parent)
385 			reply(rmp->mp_parent, -1);
386 	}
387 	else {
388 		/* Wake up the child */
389 		reply(proc_n, OK);
390 
391 		/* Wake up the parent, unless the parent is already dead */
392 		if (!new_parent)
393 			reply(rmp->mp_parent, rmp->mp_pid);
394 	}
395 
396 	break;
397 
398   case VFS_PM_SRV_FORK_REPLY:
399 	/* Nothing to do */
400 
401 	break;
402 
403   case VFS_PM_UNPAUSE_REPLY:
404 	/* The target process must always be stopped while unpausing; otherwise
405 	 * it could just end up pausing itself on a new call afterwards.
406 	 */
407 	assert(rmp->mp_flags & PROC_STOPPED);
408 
409 	/* Process is now unpaused */
410 	rmp->mp_flags |= UNPAUSED;
411 
412 	break;
413 
414   default:
415 	panic("handle_vfs_reply: unknown reply code: %d", call_nr);
416   }
417 
418   /* Now that the process is idle again, look at pending signals */
419   if ((rmp->mp_flags & (IN_USE | EXITING)) == IN_USE)
420 	  restart_sigs(rmp);
421 }
422