xref: /minix/minix/servers/pm/main.c (revision fb9c64b2)
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/endpoint.h>
17 #include <minix/minlib.h>
18 #include <minix/type.h>
19 #include <minix/vm.h>
20 #include <signal.h>
21 #include <stdlib.h>
22 #include <fcntl.h>
23 #include <sys/resource.h>
24 #include <sys/utsname.h>
25 #include <sys/wait.h>
26 #include <machine/archtypes.h>
27 #include <assert.h>
28 #include "mproc.h"
29 
30 #include "kernel/const.h"
31 #include "kernel/config.h"
32 #include "kernel/proc.h"
33 
34 #if ENABLE_SYSCALL_STATS
35 EXTERN unsigned long calls_stats[NR_PM_CALLS];
36 #endif
37 
38 static int get_nice_value(int queue);
39 static void handle_vfs_reply(void);
40 
41 /* SEF functions and variables. */
42 static void sef_local_startup(void);
43 static int sef_cb_init_fresh(int type, sef_init_info_t *info);
44 
45 /*===========================================================================*
46  *				main					     *
47  *===========================================================================*/
48 int
49 main(void)
50 {
51 /* Main routine of the process manager. */
52   unsigned int call_index;
53   int ipc_status, result;
54 
55   /* SEF local startup. */
56   sef_local_startup();
57 
58   /* This is PM's main loop-  get work and do it, forever and forever. */
59   while (TRUE) {
60 	/* Wait for the next message. */
61 	if (sef_receive_status(ANY, &m_in, &ipc_status) != OK)
62 		panic("PM sef_receive_status error");
63 
64 	/* Check for system notifications first. Special cases. */
65 	if (is_ipc_notify(ipc_status)) {
66 		if (_ENDPOINT_P(m_in.m_source) == CLOCK)
67 			expire_timers(m_in.m_notify.timestamp);
68 
69 		/* done, continue */
70 		continue;
71 	}
72 
73 	/* Extract useful information from the message. */
74 	who_e = m_in.m_source;	/* who sent the message */
75 	if (pm_isokendpt(who_e, &who_p) != OK)
76 		panic("PM got message from invalid endpoint: %d", who_e);
77 	mp = &mproc[who_p];	/* process slot of caller */
78 	call_nr = m_in.m_type;	/* system call number */
79 
80 	/* Drop delayed calls from exiting processes. */
81 	if (mp->mp_flags & EXITING)
82 		continue;
83 
84 	if (IS_VFS_PM_RS(call_nr) && who_e == VFS_PROC_NR) {
85 		handle_vfs_reply();
86 
87 		result = SUSPEND;		/* don't reply */
88 	} else if (call_nr == PROC_EVENT_REPLY) {
89 		result = do_proc_event_reply();
90 	} else if (IS_PM_CALL(call_nr)) {
91 		/* If the system call number is valid, perform the call. */
92 		call_index = (unsigned int) (call_nr - PM_BASE);
93 
94 		if (call_index < NR_PM_CALLS && call_vec[call_index] != NULL) {
95 #if ENABLE_SYSCALL_STATS
96 			calls_stats[call_index]++;
97 #endif
98 
99 			result = (*call_vec[call_index])();
100 		} else
101 			result = ENOSYS;
102 	} else
103 		result = ENOSYS;
104 
105 	/* Send reply. */
106 	if (result != SUSPEND) reply(who_p, result);
107   }
108   return(OK);
109 }
110 
111 /*===========================================================================*
112  *			       sef_local_startup			     *
113  *===========================================================================*/
114 static void
115 sef_local_startup(void)
116 {
117   /* Register init callbacks. */
118   sef_setcb_init_fresh(sef_cb_init_fresh);
119   sef_setcb_init_restart(SEF_CB_INIT_RESTART_STATEFUL);
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 	rmp->mp_sigact = mpsigact[rmp - mproc];
151 	rmp->mp_eventsub = NO_EVENTSUB;
152   }
153 
154   /* Build the set of signals which cause core dumps, and the set of signals
155    * that are by default ignored.
156    */
157   sigemptyset(&core_sset);
158   for (sig_ptr = core_sigs; sig_ptr < core_sigs+sizeof(core_sigs); sig_ptr++)
159 	sigaddset(&core_sset, *sig_ptr);
160   sigemptyset(&ign_sset);
161   for (sig_ptr = ign_sigs; sig_ptr < ign_sigs+sizeof(ign_sigs); sig_ptr++)
162 	sigaddset(&ign_sset, *sig_ptr);
163   sigemptyset(&noign_sset);
164   for (sig_ptr = noign_sigs; sig_ptr < noign_sigs+sizeof(noign_sigs); sig_ptr++)
165 	sigaddset(&noign_sset, *sig_ptr);
166 
167   /* Obtain a copy of the boot monitor parameters.
168    */
169   if ((s=sys_getmonparams(monitor_params, sizeof(monitor_params))) != OK)
170       panic("get monitor params failed: %d", s);
171 
172   /* Initialize PM's process table. Request a copy of the system image table
173    * that is defined at the kernel level to see which slots to fill in.
174    */
175   if (OK != (s=sys_getimage(image)))
176   	panic("couldn't get image table: %d", s);
177   procs_in_use = 0;				/* start populating table */
178   for (ip = &image[0]; ip < &image[NR_BOOT_PROCS]; ip++) {
179   	if (ip->proc_nr >= 0) {			/* task have negative nrs */
180   		procs_in_use += 1;		/* found user process */
181 
182 		/* Set process details found in the image table. */
183 		rmp = &mproc[ip->proc_nr];
184   		strlcpy(rmp->mp_name, ip->proc_name, PROC_NAME_LEN);
185   		(void) sigemptyset(&rmp->mp_ignore);
186   		(void) sigemptyset(&rmp->mp_sigmask);
187   		(void) sigemptyset(&rmp->mp_catch);
188 		if (ip->proc_nr == INIT_PROC_NR) {	/* user process */
189   			/* INIT is root, we make it father of itself. This is
190   			 * not really OK, INIT should have no father, i.e.
191   			 * a father with pid NO_PID. But PM currently assumes
192   			 * that mp_parent always points to a valid slot number.
193   			 */
194   			rmp->mp_parent = INIT_PROC_NR;
195   			rmp->mp_procgrp = rmp->mp_pid = INIT_PID;
196 			rmp->mp_flags |= IN_USE;
197 
198 			/* Set scheduling info */
199 			rmp->mp_scheduler = KERNEL;
200 			rmp->mp_nice = get_nice_value(USR_Q);
201 		}
202 		else {					/* system process */
203   			if(ip->proc_nr == RS_PROC_NR) {
204   				rmp->mp_parent = INIT_PROC_NR;
205   			}
206   			else {
207   				rmp->mp_parent = RS_PROC_NR;
208   			}
209   			rmp->mp_pid = get_free_pid();
210 			rmp->mp_flags |= IN_USE | PRIV_PROC;
211 
212 			/* RS schedules this process */
213 			rmp->mp_scheduler = NONE;
214 			rmp->mp_nice = get_nice_value(SRV_Q);
215 		}
216 
217 		/* Get kernel endpoint identifier. */
218 		rmp->mp_endpoint = ip->endpoint;
219 
220 		/* Tell VFS about this system process. */
221 		memset(&mess, 0, sizeof(mess));
222 		mess.m_type = VFS_PM_INIT;
223 		mess.VFS_PM_SLOT = ip->proc_nr;
224 		mess.VFS_PM_PID = rmp->mp_pid;
225 		mess.VFS_PM_ENDPT = rmp->mp_endpoint;
226   		if (OK != (s=ipc_send(VFS_PROC_NR, &mess)))
227 			panic("can't sync up with VFS: %d", s);
228   	}
229   }
230 
231   /* Tell VFS that no more system processes follow and synchronize. */
232   memset(&mess, 0, sizeof(mess));
233   mess.m_type = VFS_PM_INIT;
234   mess.VFS_PM_ENDPT = NONE;
235   if (ipc_sendrec(VFS_PROC_NR, &mess) != OK || mess.m_type != OK)
236 	panic("can't sync up with VFS");
237 
238  system_hz = sys_hz();
239 
240   /* Initialize user-space scheduling. */
241   sched_init();
242 
243   return(OK);
244 }
245 
246 /*===========================================================================*
247  *				reply					     *
248  *===========================================================================*/
249 void
250 reply(
251 	int proc_nr,			/* process to reply to */
252 	int result			/* result of call (usually OK or error #) */
253 )
254 {
255 /* Send a reply to a user process.  System calls may occasionally fill in other
256  * fields, this is only for the main return value and for sending the reply.
257  */
258   struct mproc *rmp;
259   int r;
260 
261   if(proc_nr < 0 || proc_nr >= NR_PROCS)
262       panic("reply arg out of range: %d", proc_nr);
263 
264   rmp = &mproc[proc_nr];
265   rmp->mp_reply.m_type = result;
266 
267   if ((r = ipc_sendnb(rmp->mp_endpoint, &rmp->mp_reply)) != OK)
268 	printf("PM can't reply to %d (%s): %d\n", rmp->mp_endpoint,
269 		rmp->mp_name, r);
270 }
271 
272 /*===========================================================================*
273  *				get_nice_value				     *
274  *===========================================================================*/
275 static int
276 get_nice_value(
277 	int queue				/* store mem chunks here */
278 )
279 {
280 /* Processes in the boot image have a priority assigned. The PM doesn't know
281  * about priorities, but uses 'nice' values instead. The priority is between
282  * MIN_USER_Q and MAX_USER_Q. We have to scale between PRIO_MIN and PRIO_MAX.
283  */
284   int nice_val = (queue - USER_Q) * (PRIO_MAX-PRIO_MIN+1) /
285       (MIN_USER_Q-MAX_USER_Q+1);
286   if (nice_val > PRIO_MAX) nice_val = PRIO_MAX;	/* shouldn't happen */
287   if (nice_val < PRIO_MIN) nice_val = PRIO_MIN;	/* shouldn't happen */
288   return nice_val;
289 }
290 
291 /*===========================================================================*
292  *				handle_vfs_reply       			     *
293  *===========================================================================*/
294 static void
295 handle_vfs_reply(void)
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_CORE_REPLY:
357 	if (m_in.VFS_PM_STATUS == OK)
358 		rmp->mp_sigstatus |= WCOREFLAG;
359 
360 	/* FALLTHROUGH */
361   case VFS_PM_EXIT_REPLY:
362 	assert(rmp->mp_flags & EXITING);
363 
364 	/* Publish the exit event. Continue exiting the process after that. */
365 	publish_event(rmp);
366 
367 	return; /* do not take the default action */
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 	/* Publish the signal event. Continue with signals only after that. */
413 	publish_event(rmp);
414 
415 	return; /* do not take the default action */
416 
417   default:
418 	panic("handle_vfs_reply: unknown reply code: %d", call_nr);
419   }
420 
421   /* Now that the process is idle again, look at pending signals */
422   if ((rmp->mp_flags & (IN_USE | EXITING)) == IN_USE)
423 	  restart_sigs(rmp);
424 }
425