minor bug in dump, can cause system to hang.

[Dave Truesdell]

Index:  dump/dumptraverse.c--etc

Description:
	Dump would hang the system while doing incremental dumps (levels > 0),
	during pass II.

	When doing dumps from a raw device, dump does not force all reads
	to be done in multiples of DEV_BSIZE byte chunks.  In most cases
	drivers seem to handle this correctly, but one obscure case has
	caused one of our systems (a VAX 11/785 running 4.2BSD) to hang.


Repeat-By:
	Arrange for a raw read (size != n*512) of a directory to fail.

	In our case, a empty directory (containing ".", and "..") occupied
	a block which was forwarded by the HP driver.   When dump attempted to
	read the directory entry ( 24 bytes long ), the system hung.

Fix:

	Force bread to do all reads in multiples of DEV_BSIZE byte blocks.
	However, for efficiency, I have added a seperate version of bread
	called raw_bread, that is used in pass II.

RCS file: RCS/dumptraverse.c,v
retrieving revision 1.3
retrieving revision 1.4
diff -c3 -r1.3 -r1.4
*** /tmp/,RCSt1018693	Tue Sep 24 12:40:03 1985
--- /tmp/,RCSt2018693	Tue Sep 24 12:40:05 1985
***************
*** 265,271
  		return;
  	if (filesize > size)
  		filesize = size;
! 	bread(fsbtodb(sblock, d), dblk, filesize);
  	for (loc = 0; loc < filesize; ) {
  		dp = (struct direct *)(dblk + loc);
  		if (dp->d_reclen == 0) {

--- 269,275 -----
  		return;
  	if (filesize > size)
  		filesize = size;
! 	raw_bread(fsbtodb(sblock, d), dblk, filesize);
  	for (loc = 0; loc < filesize; ) {
  		dp = (struct direct *)(dblk + loc);
  		if (dp->d_reclen == 0) {
***************
*** 338,343
  		goto loop;
  	}
  	msg("(This should not happen)bread from %s [block %d]: count=%d, got=%d\n",
  		disk, da, cnt, n);
  	if (++breaderrors > BREADEMAX){
  		msg("More than %d block read errors from %d\n",

--- 342,400 -----
  		goto loop;
  	}
  	msg("(This should not happen)bread from %s [block %d]: count=%d, got=%d\n",
+ 		disk, da, cnt, n);
+ 	if (++breaderrors > BREADEMAX){
+ 		msg("More than %d block read errors from %d\n",
+ 			BREADEMAX, disk);
+ 		broadcast("DUMP IS AILING!\n");
+ 		msg("This is an unrecoverable error.\n");
+ 		if (!query("Do you want to attempt to continue?")){
+ 			dumpabort();
+ 			/*NOTREACHED*/
+ 		} else
+ 			breaderrors = 0;
+ 	}
+ }
+ 
+ /*
+  *     This version of bread forces reads to be done in multiples of DEV_BSIZE
+  * (normally 512) bytes.  This is required for proper operation on raw I/O
+  * devices.  The lack of enforcement, has caused one of our systems to hang
+  * during incremental dumps.
+  */
+ raw_bread(da, ba, cnt)
+ 	daddr_t da;
+ 	char *ba;
+ 	int	cnt;	
+ {
+ 	int n;
+ 	int size;
+ 
+ 	size = cnt;
+ 	/*
+ 	 * Note:  This implicitly assumes that DEV_BSIZE is a power of 2.
+ 	 *	If, for whatever reason, this is not true, an alternate
+ 	 *	function is incuded, but ifdef'd out.
+ 	 */
+ 	if ((size & (DEV_BSIZE - 1)) != 0)
+ 		size = (size + (DEV_BSIZE - 1)) &~ (DEV_BSIZE - 1);
+ #ifdef	not_def
+ 	if ((size % DEV_BSIZE) != 0)
+ 		size += DEV_BSIZE - (size % DEV_BSIZE);
+ #endif
+ loop:
+ 	if (lseek(fi, (long)(da * DEV_BSIZE), 0) < 0){
+ 		msg("raw_bread: lseek fails\n");
+ 	}
+ 	n = read(fi, ba, size);
+ 	if (n == size || n >= cnt)
+ 		return;
+ 	if (da + (size / DEV_BSIZE) > fsbtodb(sblock, sblock->fs_size)) {
+ 		size -= DEV_BSIZE;
+ 		cnt  -= DEV_BSIZE;
+ 		goto loop;
+ 	}
+ 	msg("(This should not happen)raw_bread from %s [block %d]: count=%d, got=%d\n",
  		disk, da, cnt, n);
  	if (++breaderrors > BREADEMAX){
  		msg("More than %d block read errors from %d\n",
--
    David A. Truesdell
    Programmer/Analyst

    The RAND Corporation
    1700 Main Street
    P.O. Box 2138
    CSD/1
    Santa Monica, CA 90406-2138

Phone:
    (213) 393-0411 ext. 7173
ARPAnet:
    davet at rand-unix
UUCP/usenet:
    cbosg ----
    decvax ---|
    fortune --|
    harpo ----|- randvax!davet
    trw-unix -|
    ucla-cs --|
    vortex ---