Skip to contentSkip to conversation

xentrace: fix bug in t_info size

PATCH In review George Dunlap 2 participants 3 messages May 8, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xentrace: fix bug in t_info size

t_info size should be in bytes, not pages. This fixes a bug that crashes the hypervisor if the total number of all pages is more than 1024 but less than 2048.

xen/common/trace.c +1 -1
diff -r caea94988515 -r e633befe28ec xen/common/trace.c
@@ -340,7 +340,7 @@
     case XEN_SYSCTL_TBUFOP_get_info:
         tbc->evt_mask   = tb_event_mask;
         tbc->buffer_mfn = t_info ? virt_to_mfn(t_info) : 0;
-        tbc->size = T_INFO_PAGES;
+        tbc->size = T_INFO_PAGES * PAGE_SIZE;
         break;
     case XEN_SYSCTL_TBUFOP_set_cpu_mask:
         xenctl_cpumap_to_cpumask(&tb_cpu_mask, &tbc->cpu_mask);
#1 lore raw
Quoted 4 lines · George Dunlap

t_info size should be in bytes, not pages. This fixes a bug that crashes the hypervisor if the total number of all pages is more than 1024 but less than 2048.

Could this be causing other memory corruption too?

J

Quoted 21 lines

diff -r caea94988515 -r e633befe28ec xen/common/trace.c

xen/common/trace.c +1 -1
@@ -340,7 +340,7 @@
     case XEN_SYSCTL_TBUFOP_get_info:
         tbc->evt_mask   = tb_event_mask;
         tbc->buffer_mfn = t_info ? virt_to_mfn(t_info) : 0;
-        tbc->size = T_INFO_PAGES;
+        tbc->size = T_INFO_PAGES * PAGE_SIZE;
         break;
     case XEN_SYSCTL_TBUFOP_set_cpu_mask:
         xenctl_cpumap_to_cpumask(&tb_cpu_mask, &tbc->cpu_mask);

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#2 lore raw

I don't think so... The entire xen structure actually is allocated, and the bounds checking makes sure nothing goes off the end of it. It's just that (before this patch) xentrace only maps one of the two pages when it maps t_info. It then happily passes who knows what into xc_map_foreign_range().

Arguably, passing junk into xc_map_foreign_range() shouldn't crash Xen; but that's a slightly different issue.

-George

Quoted 37 lines · Jeremy Fitzhardinge
Quoted 5 lines · George Dunlap

t_info size should be in bytes, not pages. This fixes a bug that crashes the hypervisor if the total number of all pages is more than 1024 but less than 2048.

Could this be causing other memory corruption too?

J

Quoted 22 lines

diff -r caea94988515 -r e633befe28ec xen/common/trace.c

xen/common/trace.c +1 -1
@@ -340,7 +340,7 @@
     case XEN_SYSCTL_TBUFOP_get_info:
         tbc->evt_mask   = tb_event_mask;
         tbc->buffer_mfn = t_info ? virt_to_mfn(t_info) : 0;
-        tbc->size = T_INFO_PAGES;
+        tbc->size = T_INFO_PAGES * PAGE_SIZE;
         break;
     case XEN_SYSCTL_TBUFOP_set_cpu_mask:
         xenctl_cpumap_to_cpumask(&tb_cpu_mask, &tbc->cpu_mask);

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel