tape.c

早期freebsd实现

			startnewtape(0);
			spcl.c_tapea = savedtapea;
			lastspclrec = savedtapea - 1;
		}
		size = (char *)ntb - (char *)q;
		if (atomic(write, slp->fd, (char *)q, size) != size) {
			perror("  DUMP: error writing command pipe");
			dumpabort(0);
		}
		slp->sent = 1;
		if (++slp >= &slaves[SLAVES])
			slp = &slaves[0];

		q->count = 1;

		if (prev->dblk != 0) {
			/*
			 * If the last one was a disk block, make the 
			 * first of this one be the last bit of that disk 
			 * block...
			 */
			q->dblk = prev->dblk +
				prev->count * (TP_BSIZE / DEV_BSIZE);
			ntb = (union u_spcl *)tslp->tblock;
		} else {
			/*
			 * It wasn't a disk block.  Copy the data to its 
			 * new location in the buffer.
			 */
			q->dblk = 0;
			*((union u_spcl *)tslp->tblock) = *ntb;
			ntb = (union u_spcl *)tslp->tblock[1];
		}
	}
	slp->req[0] = *q;
	nextblock = slp->tblock;
	if (q->dblk == 0)
		nextblock++;
	trecno = 1;

	/*
	 * Clear the first slaves' response.  One hopes that it
	 * worked ok, otherwise the tape is much too short!
	 */
	if (slp->sent) {
		if (atomic(read, slp->fd, (char *)&got, sizeof got)
		    != sizeof got) {
			perror("  DUMP: error reading command pipe in master");
			dumpabort(0);
		}
		slp->sent = 0;

		if (got != writesize) {
			quit("EOT detected at start of the tape!\n");
		}
	}
}

/*
 * We implement taking and restoring checkpoints on the tape level.
 * When each tape is opened, a new process is created by forking; this
 * saves all of the necessary context in the parent.  The child
 * continues the dump; the parent waits around, saving the context.
 * If the child returns X_REWRITE, then it had problems writing that tape;
 * this causes the parent to fork again, duplicating the context, and
 * everything continues as if nothing had happened.
 */
void
startnewtape(top)
	int top;
{
	int	parentpid;
	int	childpid;
	int	status;
	int	waitpid;
	char	*p;
#ifdef sunos
	void	(*interrupt_save)();
#else
	sig_t	interrupt_save;
#endif

	interrupt_save = signal(SIGINT, SIG_IGN);
	parentpid = getpid();

restore_check_point:
	(void)signal(SIGINT, interrupt_save);
	/*
	 *	All signals are inherited...
	 */
	childpid = fork();
	if (childpid < 0) {
		msg("Context save fork fails in parent %d\n", parentpid);
		Exit(X_ABORT);
	}
	if (childpid != 0) {
		/*
		 *	PARENT:
		 *	save the context by waiting
		 *	until the child doing all of the work returns.
		 *	don't catch the interrupt
		 */
		signal(SIGINT, SIG_IGN);
#ifdef TDEBUG
		msg("Tape: %d; parent process: %d child process %d\n",
			tapeno+1, parentpid, childpid);
#endif /* TDEBUG */
		while ((waitpid = wait(&status)) != childpid)
			msg("Parent %d waiting for child %d has another child %d return\n",
				parentpid, childpid, waitpid);
		if (status & 0xFF) {
			msg("Child %d returns LOB status %o\n",
				childpid, status&0xFF);
		}
		status = (status >> 8) & 0xFF;
#ifdef TDEBUG
		switch(status) {
			case X_FINOK:
				msg("Child %d finishes X_FINOK\n", childpid);
				break;
			case X_ABORT:	
				msg("Child %d finishes X_ABORT\n", childpid);
				break;
			case X_REWRITE:
				msg("Child %d finishes X_REWRITE\n", childpid);
				break;
			default:
				msg("Child %d finishes unknown %d\n",
					childpid, status);
				break;
		}
#endif /* TDEBUG */
		switch(status) {
			case X_FINOK:
				Exit(X_FINOK);
			case X_ABORT:
				Exit(X_ABORT);
			case X_REWRITE:
				goto restore_check_point;
			default:
				msg("Bad return code from dump: %d\n", status);
				Exit(X_ABORT);
		}
		/*NOTREACHED*/
	} else {	/* we are the child; just continue */
#ifdef TDEBUG
		sleep(4);	/* allow time for parent's message to get out */
		msg("Child on Tape %d has parent %d, my pid = %d\n",
			tapeno+1, parentpid, getpid());
#endif /* TDEBUG */
		/*
		 * If we have a name like "/dev/rmt0,/dev/rmt1",
		 * use the name before the comma first, and save
		 * the remaining names for subsequent volumes.
		 */
		tapeno++;               /* current tape sequence */
		if (nexttape || index(tape, ',')) {
			if (nexttape && *nexttape)
				tape = nexttape;
			if ((p = index(tape, ',')) != NULL) {
				*p = '\0';
				nexttape = p + 1;
			} else
				nexttape = NULL;
			msg("Dumping volume %d on %s\n", tapeno, tape);
		}
#ifdef RDUMP
		while ((tapefd = (host ? rmtopen(tape, 2) :
			pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
#else
		while ((tapefd = (pipeout ? 1 : 
				  open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
#endif
		    {
			msg("Cannot open output \"%s\".\n", tape);
			if (!query("Do you want to retry the open?"))
				dumpabort(0);
		}

		enslave();  /* Share open tape file descriptor with slaves */

		asize = 0;
		blocksthisvol = 0;
		if (top)
			newtape++;		/* new tape signal */
		spcl.c_count = slp->count; 
		/*
		 * measure firstrec in TP_BSIZE units since restore doesn't
		 * know the correct ntrec value...
		 */
		spcl.c_firstrec = slp->firstrec;
		spcl.c_volume++;
		spcl.c_type = TS_TAPE;
		spcl.c_flags |= DR_NEWHEADER;
		writeheader((ino_t)slp->inode);
		spcl.c_flags &=~ DR_NEWHEADER;
		if (tapeno > 1)
			msg("Volume %d begins with blocks from inode %d\n",
				tapeno, slp->inode);
	}
}

void
dumpabort(signo)
	int signo;
{

	if (master != 0 && master != getpid())
		/* Signals master to call dumpabort */
		(void) kill(master, SIGTERM);
	else {
		killall();
		msg("The ENTIRE dump is aborted.\n");
	}
#ifdef RDUMP
	rmtclose();
#endif
	Exit(X_ABORT);
}

__dead void
Exit(status)
	int status;
{

#ifdef TDEBUG
	msg("pid = %d exits with status %d\n", getpid(), status);
#endif /* TDEBUG */
	exit(status);
}

/*
 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
 */
void
proceed(signo)
	int signo;
{

	if (ready)
		longjmp(jmpbuf, 1);
	caught++;
}

void
enslave()
{
	int cmd[2];
	register int i, j;

	master = getpid();

	signal(SIGTERM, dumpabort);  /* Slave sends SIGTERM on dumpabort() */
	signal(SIGPIPE, sigpipe);
	signal(SIGUSR1, tperror);    /* Slave sends SIGUSR1 on tape errors */
	signal(SIGUSR2, proceed);    /* Slave sends SIGUSR2 to next slave */

	for (i = 0; i < SLAVES; i++) {
		if (i == slp - &slaves[0]) {
			caught = 1;
		} else {
			caught = 0;
		}

		if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
		    (slaves[i].pid = fork()) < 0)
			quit("too many slaves, %d (recompile smaller): %s\n",
			    i, strerror(errno));

		slaves[i].fd = cmd[1];
		slaves[i].sent = 0;
		if (slaves[i].pid == 0) { 	    /* Slave starts up here */
			for (j = 0; j <= i; j++)
			        (void) close(slaves[j].fd);
			signal(SIGINT, SIG_IGN);    /* Master handles this */
			doslave(cmd[0], i);
			Exit(X_FINOK);
		}
	}
	
	for (i = 0; i < SLAVES; i++)
		(void) atomic(write, slaves[i].fd, 
			      (char *) &slaves[(i + 1) % SLAVES].pid, 
		              sizeof slaves[0].pid);
		
	master = 0; 
}

void
killall()
{
	register int i;

	for (i = 0; i < SLAVES; i++)
		if (slaves[i].pid > 0)
			(void) kill(slaves[i].pid, SIGKILL);
}

/*
 * Synchronization - each process has a lockfile, and shares file
 * descriptors to the following process's lockfile.  When our write
 * completes, we release our lock on the following process's lock-
 * file, allowing the following process to lock it and proceed. We
 * get the lock back for the next cycle by swapping descriptors.
 */
static void
doslave(cmd, slave_number)
	register int cmd;
        int slave_number;
{
	register int nread;
	int nextslave, size, wrote, eot_count;

	/*
	 * Need our own seek pointer.
	 */
	(void) close(diskfd);
	if ((diskfd = open(disk, O_RDONLY)) < 0)
		quit("slave couldn't reopen disk: %s\n", strerror(errno));

	/*
	 * Need the pid of the next slave in the loop...
	 */
	if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave))
	    != sizeof nextslave) {
		quit("master/slave protocol botched - didn't get pid of next slave.\n");
	}

	/*
	 * Get list of blocks to dump, read the blocks into tape buffer
	 */
	while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) {
		register struct req *p = slp->req;

		for (trecno = 0; trecno < ntrec;
		     trecno += p->count, p += p->count) {
			if (p->dblk) {
				bread(p->dblk, slp->tblock[trecno],
					p->count * TP_BSIZE);
			} else {
				if (p->count != 1 || atomic(read, cmd,
				    (char *)slp->tblock[trecno], 
				    TP_BSIZE) != TP_BSIZE)
				       quit("master/slave protocol botched.\n");
			}
		}
		if (setjmp(jmpbuf) == 0) {
			ready = 1;
			if (!caught)
				(void) pause();
		}
		ready = 0;
		caught = 0;

		/* Try to write the data... */
		eot_count = 0;
		size = 0;

		while (eot_count < 10 && size < writesize) {
#ifdef RDUMP
			if (host)
				wrote = rmtwrite(slp->tblock[0]+size,
				    writesize-size);
			else
#endif
				wrote = write(tapefd, slp->tblock[0]+size,
				    writesize-size);
#ifdef WRITEDEBUG
			printf("slave %d wrote %d\n", slave_number, wrote);
#endif
			if (wrote < 0) 
				break;
			if (wrote == 0)
				eot_count++;
			size += wrote;
		}

#ifdef WRITEDEBUG
		if (size != writesize) 
		 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
		     slave_number, size, writesize);
#endif

		if (eot_count > 0)
			size = 0;

		/*
		 * fixme: Pyramids running OSx return ENOSPC
		 * at EOT on 1/2 inch drives.
		 */
		if (size < 0) {
			(void) kill(master, SIGUSR1);
			for (;;)
				(void) sigpause(0);
		} else {
			/*
			 * pass size of write back to master
			 * (for EOT handling)
			 */
			(void) atomic(write, cmd, (char *)&size, sizeof size);
		} 

		/*
		 * If partial write, don't want next slave to go.
		 * Also jolts him awake.
		 */
		(void) kill(nextslave, SIGUSR2);
	}
	if (nread != 0)
		quit("error reading command pipe: %s\n", strerror(errno));
}

/*
 * Since a read from a pipe may not return all we asked for,
 * or a write may not write all we ask if we get a signal,
 * loop until the count is satisfied (or error).
 */
static int
atomic(func, fd, buf, count)
	int (*func)(), fd;
	char *buf;
	int count;
{
	int got, need = count;

	while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0)
		buf += got;
	return (got < 0 ? got : count - need);
}