Re: mkisofs-1.10 now available.

Eric Youngdale (eric@andante.jic.com)
Sun, 23 Mar 1997 19:10:59 -0500

I finally figured out why people keep getting I/O errors reading
to the end of a cdrom. The page cache causes reads to be generated out from
the end of the file to the next page boundary. A simple patch to the linux
iso9660 filesystem seems to clear up the error.

-Eric

*** fs/isofs/inode.c.~3~ Sat Mar 15 10:05:46 1997
--- fs/isofs/inode.c Sun Mar 23 18:55:25 1997
***************
*** 532,537 ****
--- 532,565 ----
printk("_isofs_bmap: block<0");
return 0;
}
+
+ /*
+ * If we are beyond the end of this file, don't give out any
+ * blocks.
+ */
+ if( (block << ISOFS_BUFFER_BITS(inode)) > inode->i_size )
+ {
+ off_t max_legal_read_offset;
+
+ /*
+ * If we are *way* beyond the end of the file, print a message.
+ * Access beyond the end of the file up to the next page boundary
+ * is normal because of the way the page cache works.
+ * In this case, we just return 0 so that we can properly fill
+ * the page with useless information without generating any
+ * I/O errors.
+ */
+ max_legal_read_offset = (inode->i_size + PAGE_SIZE - 1)
+ & ~(PAGE_SIZE - 1);
+ if( (block << ISOFS_BUFFER_BITS(inode)) >= max_legal_read_offset )
+ {
+
+ printk("_isofs_bmap: block>= EOF(%d, %d)", block,
+ inode->i_size);
+ }
+ return 0;
+ }
+
return (inode->u.isofs_i.i_first_extent >> ISOFS_BUFFER_BITS(inode)) + block;
}