proc.c

在x86平台上运行不可信任代码的sandbox。

		if(up->procctl == Proc_exitme && up->closingfgrp)
			forceclosefgrp();
		error(Eintr);
	}

	splx(s);
}

static int
tfn(void *arg)
{
	return up->trend == nil || up->tfn(arg);
}

void
twakeup(Ureg *ureg, Timer *t)
{
	Proc *p;
	Rendez *trend;

	p = t->ta;
	trend = p->trend;
	p->trend = 0;
	if(trend)
		wakeup(trend);
}

void
tsleep(Rendez *r, int (*fn)(void*), void *arg, ulong ms)
{
	if (up->timer.tt){
		print("tsleep: timer active: mode %d, tf 0x%lux\n", up->timer.tmode, up->timer.tf);
		timerdel(&up->timer);
	}
	up->timer.tns = MS2NS(ms);
	up->timer.tf = twakeup;
	up->timer.tmode = Trelative;
	up->timer.ta = up;
	up->trend = r;
	up->tfn = fn;
	timeradd(&up->timer);

	if(waserror()){
		timerdel(&up->timer);
		nexterror();
	}
	sleep(r, tfn, arg);
	if (up->timer.tt)
		timerdel(&up->timer);
	up->timer.twhen = 0;
	poperror();
}

/*
 *  Expects that only one process can call wakeup for any given Rendez.
 *  We hold both locks to ensure that r->p and p->r remain consistent.
 *  Richard Miller has a better solution that doesn't require both to
 *  be held simultaneously, but I'm a paranoid - presotto.
 */
Proc*
wakeup(Rendez *r)
{
	Proc *p;
	int s;

	s = splhi();

	lock(&r->lk);
	p = r->p;

	if(p != nil){
		lock(&p->rlock);
		if(p->state != Wakeme || p->r != r){
			iprint("%p %p %d\n", p->r, r, p->state);
			panic("wakeup: state");
		}
		r->p = nil;
		p->r = nil;
		ready(p);
		unlock(&p->rlock);
	}
	unlock(&r->lk);

	splx(s);

	return p;
}

/*
 *  if waking a sleeping process, this routine must hold both
 *  p->rlock and r->lock.  However, it can't know them in
 *  the same order as wakeup causing a possible lock ordering
 *  deadlock.  We break the deadlock by giving up the p->rlock
 *  lock if we can't get the r->lock and retrying.
 */
int
postnote(Proc *p, int dolock, char *n, int flag)
{
	int s, ret;
	Rendez *r;
	Proc *d, **l;

	if(dolock)
		qlock(&p->debug);

	if(flag != NUser && (p->notify == 0 || p->notified))
		p->nnote = 0;

	ret = 0;
	if(p->nnote < NNOTE) {
		strcpy(p->note[p->nnote].msg, n);
		p->note[p->nnote++].flag = flag;
		ret = 1;
	}
	p->notepending = 1;
	if(dolock)
		qunlock(&p->debug);

	/* this loop is to avoid lock ordering problems. */
	for(;;){
		s = splhi();
		lock(&p->rlock);
		r = p->r;

		/* waiting for a wakeup? */
		if(r == nil)
			break;	/* no */

		/* try for the second lock */
		if(canlock(&r->lk)){
			if(p->state != Wakeme || r->p != p)
				panic("postnote: state %d %d %d", r->p != p, p->r != r, p->state);
			p->r = nil;
			r->p = nil;
			ready(p);
			unlock(&r->lk);
			break;
		}

		/* give other process time to get out of critical section and try again */
		unlock(&p->rlock);
		splx(s);
		sched();
	}
	unlock(&p->rlock);
	splx(s);

	if(p->state != Rendezvous)
		return ret;

	/* Try and pull out of a rendezvous */
	lock(&p->rgrp->ref.lk);
	if(p->state == Rendezvous) {
		p->rendval = ~0;
		l = &REND(p->rgrp, p->rendtag);
		for(d = *l; d; d = d->rendhash) {
			if(d == p) {
				*l = p->rendhash;
				break;
			}
			l = &d->rendhash;
		}
		ready(p);
	}
	unlock(&p->rgrp->ref.lk);
	return ret;
}

/*
 * weird thing: keep at most NBROKEN around
 */
#define	NBROKEN 4
struct
{
	QLock lk;
	int	n;
	Proc	*p[NBROKEN];
}broken;

void
addbroken(Proc *p)
{
	qlock(&broken.lk);
	if(broken.n == NBROKEN) {
		ready(broken.p[0]);
		memmove(&broken.p[0], &broken.p[1], sizeof(Proc*)*(NBROKEN-1));
		--broken.n;
	}
	broken.p[broken.n++] = p;
	qunlock(&broken.lk);

	p->state = Broken;
	p->psstate = 0;
	sched();
}

void
unbreak(Proc *p)
{
	int b;

	qlock(&broken.lk);
	for(b=0; b < broken.n; b++)
		if(broken.p[b] == p) {
			broken.n--;
			memmove(&broken.p[b], &broken.p[b+1],
					sizeof(Proc*)*(NBROKEN-(b+1)));
			ready(p);
			break;
		}
	qunlock(&broken.lk);
}

int
freebroken(void)
{
	int i, n;

	qlock(&broken.lk);
	n = broken.n;
	for(i=0; i<n; i++) {
		ready(broken.p[i]);
		broken.p[i] = 0;
	}
	broken.n = 0;
	qunlock(&broken.lk);
	return n;
}

void
pexit(char *exitstr, int freemem)
{
	Proc *p;
	Segment **s, **es;
	long utime, stime;
	Waitq *wq, *f, *next;
	Fgrp *fgrp;
	Egrp *egrp;
	Rgrp *rgrp;
	Pgrp *pgrp;
	Chan *dot;
	void (*pt)(Proc*, int, vlong);

	up->alarm = 0;
	if (up->timer.tt)
		timerdel(&up->timer);
	pt = proctrace;
	if(pt)
		pt(up, SDead, 0);

	/* nil out all the resources under lock (free later) */
	qlock(&up->debug);
	fgrp = up->fgrp;
	up->fgrp = nil;
	egrp = up->egrp;
	up->egrp = nil;
	rgrp = up->rgrp;
	up->rgrp = nil;
	pgrp = up->pgrp;
	up->pgrp = nil;
	dot = up->dot;
	up->dot = nil;
	qunlock(&up->debug);

	if(fgrp)
		closefgrp(fgrp);
	if(egrp)
		closeegrp(egrp);
	if(rgrp)
		closergrp(rgrp);
	if(dot)
		cclose(dot);
	if(pgrp)
		closepgrp(pgrp);

	/*
	 * if not a kernel process and have a parent,
	 * do some housekeeping.
	 */
	if(up->kp == 0) {
		p = up->parent;
		if(p == 0) {
			if(exitstr == 0)
				exitstr = "unknown";
			panic("boot process died: %s", exitstr);
		}

		while(waserror())
			;

		wq = smalloc(sizeof(Waitq));
		poperror();

		wq->w.pid = up->pid;
		utime = up->time[TUser] + up->time[TCUser];
		stime = up->time[TSys] + up->time[TCSys];
		wq->w.time[TUser] = tk2ms(utime);
		wq->w.time[TSys] = tk2ms(stime);
		wq->w.time[TReal] = tk2ms(msec() - up->time[TReal]);
		if(exitstr && exitstr[0])
			snprint(wq->w.msg, sizeof(wq->w.msg), "%s %lud: %s", up->text, up->pid, exitstr);
		else
			wq->w.msg[0] = '\0';

		lock(&p->exl);
		/*
		 * Check that parent is still alive.
		 */
		if(p->pid == up->parentpid && p->state != Broken) {
			p->nchild--;
			p->time[TCUser] += utime;
			p->time[TCSys] += stime;
			/*
			 * If there would be more than 128 wait records
			 * processes for my parent, then don't leave a wait
			 * record behind.  This helps prevent badly written
			 * daemon processes from accumulating lots of wait
			 * records.
		 	 */
			if(p->nwait < 128) {
				wq->next = p->waitq;
				p->waitq = wq;
				p->nwait++;
				wq = nil;
				wakeup(&p->waitr);
			}
		}
		unlock(&p->exl);
		if(wq)
			free(wq);
	}

	if(!freemem)
		addbroken(up);

	qlock(&up->seglock);
	es = &up->seg[NSEG];
	for(s = up->seg; s < es; s++) {
		if(*s) {
			putseg(*s);
			*s = 0;
		}
	}
	qunlock(&up->seglock);

	lock(&up->exl);		/* Prevent my children from leaving waits */
	pidunhash(up);
	up->pid = 0;
	wakeup(&up->waitr);
	unlock(&up->exl);

	for(f = up->waitq; f; f = next) {
		next = f->next;
		free(f);
	}

	/* release debuggers */
	qlock(&up->debug);
	if(up->pdbg) {
		wakeup(&up->pdbg->sleep);
		up->pdbg = 0;
	}
	qunlock(&up->debug);

	/* Sched must not loop for these locks */
	lock(&procalloc.lk);
	lock(&palloc.lk);

	up->state = Moribund;
	sched();
	panic("pexit");
}

int
haswaitq(void *x)
{
	Proc *p;

	p = (Proc *)x;
	return p->waitq != 0;
}

ulong
pwait(Waitmsg *w)
{
	ulong cpid;
	Waitq *wq;

	if(!canqlock(&up->qwaitr))
		error(Einuse);

	if(waserror()) {
		qunlock(&up->qwaitr);
		nexterror();
	}

	lock(&up->exl);
	if(up->nchild == 0 && up->waitq == 0) {
		unlock(&up->exl);
		error(Enochild);
	}
	unlock(&up->exl);

	sleep(&up->waitr, haswaitq, up);

	lock(&up->exl);
	wq = up->waitq;
	up->waitq = wq->next;
	up->nwait--;
	unlock(&up->exl);

	qunlock(&up->qwaitr);
	poperror();

	if(w)
		memmove(w, &wq->w, sizeof(Waitmsg));
	cpid = wq->w.pid;
	free(wq);
	return cpid;
}

Proc*
proctab(int i)
{
	return &procalloc.arena[i];
}

void
dumpaproc(Proc *p)
{
	ulong bss;
	char *s;

	if(p == 0)
		return;

	bss = 0;
	if(p->seg[BSEG])
		bss = p->seg[BSEG]->top;

	s = p->psstate;
	if(s == 0)
		s = statename[p->state];
	print("%3lud:%10s pc %8lux dbgpc %8lux  %8s (%s) ut %ld st %ld bss %lux qpc %lux nl %lud nd %lud lpc %lux pri %lud\n",
		p->pid, p->text, p->pc, dbgpc(p),  s, statename[p->state],
		p->time[0], p->time[1], bss, p->qpc, p->nlocks.ref, p->delaysched, p->lastlock ? p->lastlock->pc : 0, p->priority);
}

void
procdump(void)
{
	int i;
	Proc *p;

	if(up)
		print("up %lud\n", up->pid);
	else
		print("no current process\n");
	for(i=0; i<conf.nproc; i++) {
		p = &procalloc.arena[i];
		if(p->state == Dead)
			continue;

		dumpaproc(p);
	}
}

/*
 *  wait till all processes have flushed their mmu
 *  state about segement s
 */
void
procflushseg(Segment *s)
{
	int i, ns, nm, nwait;
	Proc *p;

	/*
	 *  tell all processes with this
	 *  segment to flush their mmu's
	 */
	nwait = 0;
	for(i=0; i<conf.nproc; i++) {
		p = &procalloc.arena[i];
		if(p->state == Dead)
			continue;
		for(ns = 0; ns < NSEG; ns++)
			if(p->seg[ns] == s){
				p->newtlb = 1;
				for(nm = 0; nm < conf.nmach; nm++){
					if(MACHP(nm)->proc == p){
						MACHP(nm)->flushmmu = 1;
						nwait++;
					}
				}
				break;
			}
	}

	if(nwait == 0)
		return;

	/*
	 *  wait for all processors to take a clock interrupt
	 *  and flush their mmu's
	 */
	for(nm = 0; nm < conf.nmach; nm++)
		if(MACHP(nm) != m)
			while(MACHP(nm)->flushmmu)
				sched();
}

void
scheddump(void)
{
	Proc *p;
	Schedq *rq;

	for(rq = &runq[Nrq-1]; rq >= runq; rq--){
		if(rq->head == 0)
			continue;
		print("rq%ld:", rq-runq);
		for(p = rq->head; p; p = p->rnext)
			print(" %lud(%lud)", p->pid, msec() - p->readytime);
		print("\n");
		delay(150);
	}
	print("nrdy %d\n", nrdy);
}

void
kproc(char *name, void (*func)(void *), void *arg)
{
	Proc *p;
	static Pgrp *kpgrp;

	p = newproc();
	p->psstate = 0;
	p->procmode = 0640;
	p->kp = 1;
	p->noswap = 1;

	if(up){
		p->fpsave = up->fpsave;
		p->scallnr = up->scallnr;
		p->s = up->s;
		p->slash = up->slash;
		p->dot = up->dot;
		if(p->dot)
			incref(&p->dot->ref);

		memmove(p->note, up->note, sizeof(p->note));
		p->nnote = up->nnote;
		p->lastnote = up->lastnote;
		p->notify = up->notify;
	}

 	p->notified = 0;
 	p->ureg = 0;
 	p->dbgreg = 0;
	p->nerrlab = 0;

	procpriority(p, PriKproc, 0);

	kprocchild(p, func, arg);

	kstrdup(&p->user, eve);
	kstrdup(&p->text, name);
	if(kpgrp == 0)
		kpgrp = newpgrp();
	p->pgrp = kpgrp;
	incref(&kpgrp->ref);

	memset(p->time, 0, sizeof(p->time));
	p->time[TReal] = msec();
	ready(p);
}

/*
 *  called splhi() by notify().  See comment in notify for the
 *  reasoning.
 */
void
procctl(Proc *p)
{
	char *state;
	ulong s;

	switch(p->procctl) {
	case Proc_exitbig:
		spllo();
		pexit("Killed: Insufficient physical memory", 1);

	case Proc_exitme:
		spllo();		/* pexit has locks in it */
		pexit("Killed", 1);

	case Proc_traceme:
		if(p->nnote == 0)
			return;
		/* No break */

	case Proc_stopme:
		p->procctl = 0;
		state = p->psstate;
		p->psstate = "Stopped";
		/* free a waiting debugger */
		s = spllo();
		qlock(&p->debug);
		if(p->pdbg) {
			wakeup(&p->pdbg->sleep);
			p->pdbg = 0;
		}
		qunlock(&p->debug);
		splhi();
		p->state = Stopped;
		sched();
		p->psstate = state;
		splx(s);
		return;
	}
}

#include "errstr.h"

void
error(char *err)
{
	spllo();

	assert(up->nerrlab < NERR);
	kstrcpy(up->errstr, err, ERRMAX);
	setlabel(&up->errlab[NERR-1]);
	nexterror();
}

void
nexterror(void)
{
	gotolabel(&up->errlab[--up->nerrlab]);
}

void
exhausted(char *resource)
{
	char buf[ERRMAX];

	sprint(buf, "no free %s", resource);
	iprint("%s\n", buf);
	error(buf);
}

void
killbig(char *why)
{
	int i;
	Segment *s;
	ulong l, max;
	Proc *p, *ep, *kp;

	max = 0;
	kp = 0;
	ep = procalloc.arena+conf.nproc;
	for(p = procalloc.arena; p < ep; p++) {
		if(p->state == Dead || p->kp)
			continue;
		l = 0;
		for(i=1; i<NSEG; i++) {
			s = p->seg[i];
			if(s != 0)
				l += s->top - s->base;
		}
		if(l > max && ((p->procmode&0222) || strcmp(eve, p->user)!=0)) {
			kp = p;
			max = l;
		}
	}

	print("%lud: %s killed: %s\n", kp->pid, kp->text, why);
	for(p = procalloc.arena; p < ep; p++) {
		if(p->state == Dead || p->kp)
			continue;
		if(p != kp && p->seg[BSEG] && p->seg[BSEG] == kp->seg[BSEG])
			p->procctl = Proc_exitbig;
	}
	kp->procctl = Proc_exitbig;
	for(i = 0; i < NSEG; i++) {
		s = kp->seg[i];
		if(s != 0 && canqlock(&s->lk)) {
			mfreeseg(s, s->base, (s->top - s->base)/BY2PG);
			qunlock(&s->lk);
		}
	}
}

/*
 *  change ownership to 'new' of all processes owned by 'old'.  Used when
 *  eve changes.
 */
void
renameuser(char *old, char *new)
{
	Proc *p, *ep;

	ep = procalloc.arena+conf.nproc;
	for(p = procalloc.arena; p < ep; p++)
		if(p->user!=nil && strcmp(old, p->user)==0)
			kstrdup(&p->user, new);
}

/*
 *  time accounting called by clock() splhi'd
 */
void
accounttime(void)
{
	Proc *p;
	ulong n, per;
	static ulong nrun;

	p = m->proc;
	if(p) {
		nrun++;
		p->time[p->insyscall]++;
	}

	/* calculate decaying duty cycles */
	n = perfticks();
	per = n - m->perf.last;
	m->perf.last = n;
	per = (m->perf.period*(HZ-1) + per)/HZ;
	if(per != 0)
		m->perf.period = per;

	m->perf.avg_inidle = (m->perf.avg_inidle*(HZ-1)+m->perf.inidle)/HZ;
	m->perf.inidle = 0;

	m->perf.avg_inintr = (m->perf.avg_inintr*(HZ-1)+m->perf.inintr)/HZ;
	m->perf.inintr = 0;

	/* only one processor gets to compute system load averages */
	if(m->machno != 0)
		return;

	/*
	 * calculate decaying load average.
	 * if we decay by (n-1)/n then it takes
	 * n clock ticks to go from load L to .36 L once
	 * things quiet down.  it takes about 5 n clock
	 * ticks to go to zero.  so using HZ means this is
	 * approximately the load over the last second,
	 * with a tail lasting about 5 seconds.
	 */
	n = nrun;
	nrun = 0;
	n = (nrdy+n)*1000;
	m->load = (m->load*(HZ-1)+n)/HZ;
}

static void
pidhash(Proc *p)
{
	int h;

	h = p->pid % nelem(procalloc.ht);
	lock(&procalloc.lk);
	p->pidhash = procalloc.ht[h];
	procalloc.ht[h] = p;
	unlock(&procalloc.lk);
}

static void
pidunhash(Proc *p)
{
	int h;
	Proc **l;

	h = p->pid % nelem(procalloc.ht);
	lock(&procalloc.lk);
	for(l = &procalloc.ht[h]; *l != nil; l = &(*l)->pidhash)
		if(*l == p){
			*l = p->pidhash;
			break;
		}
	unlock(&procalloc.lk);
}

int
procindex(ulong pid)
{
	Proc *p;
	int h;
	int s;

	s = -1;
	h = pid % nelem(procalloc.ht);
	lock(&procalloc.lk);
	for(p = procalloc.ht[h]; p != nil; p = p->pidhash)
		if(p->pid == pid){
			s = p - procalloc.arena;
			break;
		}
	unlock(&procalloc.lk);
	return s;
}