Re: [lvm-devel] Re: lvm 0.9.1-beta1 still segfaults vgexport

[Andrea Arcangeli]

On Sun, Jan 14, 2001 at 05:32:34PM +0100, Andrea Arcangeli wrote:
> BTW, I can easily reproduce. I was near to go into it yesterday but got
> interrupted by other issues (like the merging of the 0.9.1-beta1 kernel driver
> and extraction of the strictly necessary fixes from the 0.9.1-beta1 userspace
> against 0.9).

This looks the right fix for the vgexport segfault trivially reproducible
on 0.9 and 0.9.1_beta1 lvmtools. Now that I see the details of the bug
it was possible to reproduce it also with `vgdisplay -D xxxxxxx' where
xxxxxxx is just a random name of a not existent VG.

--- ./tools/lib/pv_read_all_pv_of_vg.c.~1~	Mon Jan 15 03:35:51 2001
+++ ./tools/lib/pv_read_all_pv_of_vg.c	Mon Jan 15 04:57:00 2001
@@ -137,6 +137,11 @@
          while ( pv_this[np] != NULL) np++;
       }
 
+      if ( np == 0) {
+         ret = -LVM_EPV_READ_ALL_PV_OF_VG_NP;
+         goto pv_read_all_pv_of_vg_end;
+      }
+
       /* avoid multiple access pathes */
       for ( p = 0; pv_this[p] != NULL; p++) {
             /* avoid multiple access pathes for now (2.4.0-test8)

I also got a reminder from Marco d'Itri to integrate this hack for
some more non-x86 platform:

--- ./tools/lib/pv_get_size.c.~1~	Mon Jan 15 03:35:51 2001
+++ ./tools/lib/pv_get_size.c	Mon Jan 15 04:04:03 2001
@@ -58,7 +58,7 @@
 #define read_le(x) (x)
 #endif
 
-#if !defined(__alpha__) && !defined(__s390__)
+#ifdef __i386__
 int pv_get_size ( char *dev_name, struct partition *part_ptr) {
    int i = 0;
    int dir_cache_count = 0;

Andrea