harvey/sys/src/cmd/disk/kfs/main.c

/*
 * This file is part of the UCB release of Plan 9. It is subject to the license
 * terms in the LICENSE file found in the top-level directory of this
 * distribution and at http://akaros.cs.berkeley.edu/files/Plan9License. No
 * part of the UCB release of Plan 9, including this file, may be copied,
 * modified, propagated, or distributed except according to the terms contained
 * in the LICENSE file.
 */

#include	"all.h"

int	sfd;
int	cmdmode = 0660;
int	rfd;
int	chat;
extern	char *wrenfile;
extern	int nwren;
char	*myname;
int	cmdfd;
int	writeallow;	/* never on; for compatibility with fs */
int	wstatallow;
int	writegroup;
int	allownone;
int	noatime;
int	srvfd(char*, int, int);
void	usage(void);
void	confinit(void);
Chan	*chaninit(char*);
void	consinit(void);
void	forkserve(void);

void
main(int argc, char *argv[])
{
	Filsys *fs;
	int ream, fsok;
	int newbufsize, nocheck;
	char buf[NAMELEN];
	int pid, ctl;

	progname = "kfs";
	procname = "init";

	/*
	 * insulate from invoker's environment and keep it from swapping
	 */
	rfork(RFNAMEG|RFNOTEG|RFREND);

	confinit();
	sfd = -1;
	ream = 0;
	newbufsize = 0;
	nocheck = 0;
	wrenfile = "/dev/sdC0/fs";

	pid = getpid();
	snprint(buf, sizeof buf, "/proc/%d/ctl", pid);
	ctl = open(buf, OWRITE);
	fprint(ctl, "noswap\n");
	close(ctl);

	buf[0] = '\0';

	ARGBEGIN{
	case 'b':
		newbufsize = atol(ARGF());
		break;
	case 'c':
		nocheck = 1;
		break;
	case 'f':
		wrenfile = ARGF();
		break;
	case 'm':
		nwren = atol(ARGF());
		break;
	case 'n':
		strncpy(buf, ARGF(), NAMELEN-1);
		buf[NAMELEN-1] = '\0';
		break;
	case 'p':
		cmdmode = atol(ARGF());
		break;
	case 'r':
		ream = 1;
		break;
	case 's':
		sfd = 0;
		rfd = dup(1, -1);
		close(1);
		if(open("/dev/cons", OWRITE) < 0)
			open("#c/cons", OWRITE);
		break;
	case 'B':
		conf.niobuf = strtoul(ARGF(), 0, 0);
		break;
	case 'C':
		chat = 1;
		break;
	default:
		usage();
	}ARGEND

	if(argc != 0)
		usage();

	cmdfd = 2;

	if (access(wrenfile, AREAD|AWRITE) == -1)
		sysfatal("%s cannot access device", wrenfile);

	formatinit();
	sublockinit();

	if(buf[0])
		sprint(service, "kfs.%s", buf);
	else
		strcpy(service, "kfs");
	chan = chaninit(service);
	consinit();
	tlocks = ialloc(NTLOCK * sizeof *tlocks);
	uid = ialloc(conf.nuid * sizeof(*uid));
	uidspace = ialloc(conf.uidspace * sizeof(*uidspace));
	gidspace = ialloc(conf.gidspace * sizeof(*gidspace));

	/*
	 * init global locks
	 */
	wlock(&mainlock); wunlock(&mainlock);

	/*
	 * init the file system, ream it if needed, and get the block sizes
	 */
	ream = fsinit(ream, newbufsize);
	iobufinit();
	for(fs=filesys; fs->name; fs++)
		if(fs->flags & FREAM){		/* set by fsinit if reamed */
			ream++;
			rootream(fs->dev, getraddr(fs->dev));
			superream(fs->dev, superaddr(fs->dev));
		}

	boottime = time(nil);

	consserve();
	fsok = superok(filesys[0].dev, superaddr(filesys[0].dev), 0);
	if(!nocheck && !ream && !fsok)
		cmd_exec("check fq");

	startproc(forkserve, "srv");
	startproc(syncproc, "sync");

	exits(0);
}

void
forkserve(void)
{
	serve(chan);
}

static
struct
{
	int	nfilter;
	Filter*	filters[100];
}f;

int alarmed;

void
catchalarm(void *regs, char *msg)
{
	USED(regs);
	USED(msg);
	if(strcmp(msg, "alarm") == 0){
		alarmed = 1;
		noted(NCONT);
	} else
		noted(NDFLT);
}

/*
 * process to synch blocks
 * it puts out a block/line every second
 * it waits 10 seconds if catches up.
 * in both cases, it takes about 10 seconds
 * to get up-to-date.
 *
 * it also updates the filter stats
 * and executes commands
 */
void
syncproc(void)
{
	char buf[4*1024];
	Filter *ft;
	uint32_t c0, c1;
	int32_t t, n, d;
	int i, p[2];

	/*
	 * make a pipe for commands
	 */
	if(pipe(p) < 0)
		panic("command pipe");
	sprint(buf, "#s/%s.cmd", service);
	srvfd(buf, cmdmode, p[0]);
	close(p[0]);
	cmdfd = p[1];
	notify(catchalarm);

	t = time(nil);
	for(;;){
		i = syncblock();
		alarmed = 0;
		alarm(i ? 1000: 10000);
		n = read(cmdfd, buf, sizeof buf - 1);
		if(n <= 0 && !alarmed)
			sleep(i ? 1000: 10000);
		alarm(0);
		if(n > 0){
			buf[n] = '\0';
			if(cmd_exec(buf))
				fprint(cmdfd, "done");
			else
				fprint(cmdfd, "unknown command");
		}
		n = time(nil);
		d = n - t;
		if(d < 0 || d > 5*60)
			d = 0;
		while(d >= 1) {
			d -= 1;
			for(i=0; i<f.nfilter; i++) {
				ft = f.filters[i];
				c0 = ft->count;
				c1 = c0 - ft->oldcount;
				ft->oldcount = c0;
				ft->filter[0] = famd(ft->filter[0], c1, 59, 60);
				ft->filter[1] = famd(ft->filter[1], c1, 599, 600);
				ft->filter[2] = famd(ft->filter[2], c1, 5999, 6000);
			}
		}
		t = n;
	}
}

void
dofilter(Filter *ft)
{
	int i;

	i = f.nfilter;
	if(i >= sizeof f.filters / sizeof f.filters[0]) {
		print("dofilter: too many filters\n");
		return;
	}
	f.filters[i] = ft;
	f.nfilter = i+1;
}

void
startproc(void (*f)(void), char *name)
{
	switch(rfork(RFMEM|RFFDG|RFPROC)){
	case -1:
		panic("can't fork");
	case 0:
		break;
	default:
		return;
	}
	procname = name;
	f();
	_exits(nil);
}

void
confinit(void)
{
	conf.niobuf = 0;
	conf.nuid = 600;
	conf.nserve = 2;
	conf.uidspace = conf.nuid*6;
	conf.gidspace = conf.nuid*3;
	cons.flags = 0;
}

static void
dochaninit(Chan *cp, int fd)
{
	cp->chan = fd;
	fileinit(cp);
	wlock(&cp->reflock);
	wunlock(&cp->reflock);
	lock(&cp->flock);
	unlock(&cp->flock);
}

Chan*
chaninit(char *server)
{
	Chan *cp;
	char buf[3*NAMELEN];
	int p[2];

	sprint(buf, "#s/%s", server);
	if(sfd < 0){
		if(pipe(p) < 0)
			panic("can't make a pipe");
		sfd = p[0];
		rfd = p[1];
	}
	srvfd(buf, 0666, sfd);
	close(sfd);
	cp = ialloc(sizeof *cp);
	cons.srvchan = cp;
	dochaninit(cp, rfd);
	return cp;
}

int
netserve(char *netaddr)
{
	int afd, lfd, fd;
	char adir[2*NAMELEN], ldir[2*NAMELEN];
	Chan *netchan;

	if(access("/net/tcp/clone", 0) < 0)
		bind("#I", "/net", MAFTER);
	if(access("/net.alt/tcp/clone", 0) < 0)
		bind("#I1", "/net.alt", MAFTER);

	afd = announce(netaddr, adir);
	if (afd < 0)
		return -1;
	switch (rfork(RFMEM|RFFDG|RFPROC)) {
	case -1:
		return -1;
	case 0:
		break;
	default:
		return 0;
	}
	for (;;) {
		lfd = listen(adir, ldir);
		if (lfd < 0)
			continue;
		fd = accept(lfd, ldir);
		if (fd < 0) {
			close(lfd);
			continue;
		}
		netchan = mallocz(sizeof(Chan), 1);
		if(netchan == nil)
			panic("out of memory");
		dochaninit(netchan, fd);
		switch (rfork(RFMEM|RFFDG|RFPROC)) {
		case -1:
			panic("can't fork");
		case 0:
			close(afd);
			close(lfd);
			serve(netchan);
			free(netchan);
			exits(0);
		default:
			close(fd);
			close(lfd);
			continue;
		}
	}
}

int
srvfd(char *s, int mode, int sfd)
{
	int fd;
	char buf[32];

	fd = create(s, ORCLOSE|OWRITE, mode);
	if(fd < 0){
		remove(s);
		fd = create(s, ORCLOSE|OWRITE, mode);
		if(fd < 0)
			panic(s);
	}
	sprint(buf, "%d", sfd);
	if(write(fd, buf, strlen(buf)) != strlen(buf))
		panic("srv write");
	return sfd;
}

void
consinit(void)
{
	int i;

	cons.chan = ialloc(sizeof(Chan));
	wlock(&cons.chan->reflock);
	wunlock(&cons.chan->reflock);
	lock(&cons.chan->flock);
	unlock(&cons.chan->flock);
	dofilter(&cons.work);
	dofilter(&cons.rate);
	dofilter(&cons.bhit);
	dofilter(&cons.bread);
	dofilter(&cons.binit);
	for(i = 0; i < MAXTAG; i++)
		dofilter(&cons.tags[i]);
}

/*
 * always called with mainlock locked
 */
void
syncall(void)
{
	for(;;)
		if(!syncblock())
			return;
}

int
askream(Filsys *fs)
{
	char c;

	print("File system %s inconsistent\n", fs->name);
	print("Would you like to ream it (y/n)? ");
	read(0, &c, 1);
	return c == 'y';
}

uint32_t
memsize(void)
{
	char *p, buf[128];
	int fd, n, by2pg, secs;

	by2pg = 4*1024;
	p = getenv("cputype");
	if(p && strcmp(p, "68020") == 0)
		by2pg = 8*1024;

	secs = 4*1024*1024;
	
	fd = open("/dev/swap", OREAD);
	if(fd < 0)
		return secs;
	n = read(fd, buf, sizeof(buf)-1);
	close(fd);
	if(n <= 0)
		return secs;
	buf[n] = 0;
	p = strchr(buf, '/');
	if(p)
		secs = strtoul(p+1, 0, 0)*by2pg;
	return secs;
}

/*
 * init the devices
 * wipe some of the file systems, or all if ream is set
 * this code really assumes that only one file system exists
 */
int
fsinit(int ream, int newbufsize)
{
	Filsys *fs;

	RBUFSIZE = 4 * 1024;
	for(fs=filesys; fs->name; fs++)
		(*devcall[fs->dev.type].init)(fs->dev);
	if(newbufsize == 0)
		newbufsize = RBUFSIZE;

	if(conf.niobuf == 0) {
		conf.niobuf = memsize()/10;
		if(conf.niobuf > 2*1024*1024)
			conf.niobuf = 2*1024*1024;
		conf.niobuf /= newbufsize;
		if(conf.niobuf < 30)
			conf.niobuf = 30;
	}

	BUFSIZE = RBUFSIZE - sizeof(Tag);

	for(fs=filesys; fs->name; fs++)
		if(ream || (*devcall[fs->dev.type].check)(fs->dev) && askream(fs)){
			RBUFSIZE = newbufsize;
			BUFSIZE = RBUFSIZE - sizeof(Tag);
			(*devcall[fs->dev.type].ream)(fs->dev);
			fs->flags |= FREAM;
			ream = 1;
		}

	/*
	 * set up the block size dependant variables
	 */
	BUFSIZE = RBUFSIZE - sizeof(Tag);
	DIRPERBUF = BUFSIZE / sizeof(Dentry);
	INDPERBUF = BUFSIZE / sizeof(int32_t);
	INDPERBUF2 = INDPERBUF * INDPERBUF;
	FEPERBUF = (BUFSIZE - sizeof(Super1) - sizeof(int32_t)) / sizeof(int32_t);
	return ream;
}

/*
 * allocate rest of mem
 * for io buffers.
 */
#define	HWIDTH	5	/* buffers per hash */
void
iobufinit(void)
{
	int32_t i;
	Iobuf *p, *q;
	Hiob *hp;

	i = conf.niobuf*RBUFSIZE;
	niob = i / (sizeof(Iobuf) + RBUFSIZE + sizeof(Hiob)/HWIDTH);
	nhiob = niob / HWIDTH;
	while(!prime(nhiob))
		nhiob++;
	if(chat)
		print("	%ld buffers; %ld hashes\n", niob, nhiob);
	hiob = ialloc(nhiob * sizeof(Hiob));
	hp = hiob;
	for(i=0; i<nhiob; i++) {
		lock(hp);
		unlock(hp);
		hp++;
	}
	p = ialloc(niob * sizeof(Iobuf));
	hp = hiob;
	for(i=0; i<niob; i++) {
		qlock(p);
		qunlock(p);
		if(hp == hiob)
			hp = hiob + nhiob;
		hp--;
		q = hp->link;
		if(q) {
			p->fore = q;
			p->back = q->back;
			q->back = p;
			p->back->fore = p;
		} else {
			hp->link = p;
			p->fore = p;
			p->back = p;
		}
		p->dev = devnone;
		p->addr = -1;
		p->xiobuf = ialloc(RBUFSIZE);
		p->iobuf = (char*)-1;
		p++;
	}
}

void
usage(void)
{
	fprint(2, "usage: kfs [-cCr] [-b bufsize] [-s infd outfd] [-f fsfile]\n");
	exits(0);
}