Today I happened to use mkfs.minix from util-linux-ng-2.16.1 and notice that it is broken. Funny enough the symptoms are identical to those reported in http://lists.busybox.net/pipermail/busybox/2000-September/034736.html when mkfs.minix was broken in busybox. The problem is that the code for the bitops is no good. The code from the old days had #include "bitops.h" where the bit(), setbit(), clrbit() functions return a truth value (0 or 1). static inline int bit(char * addr,unsigned int nr) { return (addr[nr >> 3] & (1<<(nr & 7))) != 0; } and it is still like this in util-linux-ng-2.14.1. The mkfs.minix and fsck.minix code actually uses the integer value of this boolean. For example, there are tests like if (zone_in_use(i) == zone_count[i]) with zone_in_use(i) defined as #define zone_in_use(x) (bit(zone_map,(x)-FIRSTZONE+1)) and zone_count[i] an integer. Nowadays, I see in 2.16.1: #define zone_in_use(x) (isset(zone_map,(x)-FIRSTZONE+1)) which uses the undocumented isset() function - bad. On this machine these macros are taken from <sys/param.h> and say #define isset(a,i) ((a)[(i)/NBBY] & (1<<((i)%NBBY))) in other words are not 0-1. The easy solution is to revert to earlier code in mkfs.minix and fsck.minix. It is always a bad idea to edit the code of little-used utilities. It may take a long time before someone notices that they are broken now. Andries It is this commit: commit 95356e8b744439336925eeb36f01399f1ee8a5e9 Author: Karel Zak <kzak@xxxxxxxxxx> Date: Thu Feb 5 15:44:17 2009 +0000 mkfs.minix: remove local implementation of {set,clr}bit We needn't to duplicate stuff from <sys/param.h>. Signed-off-by: Karel Zak <kzak@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
