tools/misc/Makefile | 2 +- tools/misc/xen-hvmcrash.c | 90 ++++++++------------------------------- 2 files changed, 19 insertions(+), 73 deletions(-)
xen-hvmcrash would previously save records, overwrite the instruction
pointer with a bogus value, and then restore them to crash a domain
just enough to cause the guest OS to memdump.
This approach is found to be unreliable when tested on a guest running
Windows 10 x64, with some executions doing nothing at all.
Another approach would be to trigger NMIs. This approach is found to be
unreliable when tested on Linux (Ubuntu 22.04), as Linux will ignore
NMIs if it is not configured to handle such.
Injecting a double fault abort to all vCPUs is found to be more
reliable at crashing and invoking memdumps from Windows and Linux
domains.
This patch modifies the xen-hvmcrash tool to inject #DF to all vCPUs
belonging to the specified domain, instead of overwriting RIP.
Signed-off-by: Matthew Barnes <matthew.barnes@cloud.com>
---
Changes in v2:
- Use xendevicemodel API instead of legacy devicemodel API
- Return error status code if no vcpus could be injected
- Use existing exception vector macro defined in `x86-defns.h`
- Clean up unnecessary headers
- Fix NULL build errors
---
tools/misc/Makefile | 2 +-
tools/misc/xen-hvmcrash.c | 90 ++++++++-------------------------------
2 files changed, 19 insertions(+), 73 deletions(-)
diff --git a/tools/misc/Makefile b/tools/misc/Makefile
index 66d0d6b09029..c26e544e8393 100644
--- a/tools/misc/Makefile
+++ b/tools/misc/Makefile
@@ -81,7 +81,7 @@ xen-hvmctx: xen-hvmctx.o
$(CC) $(LDFLAGS) -o $@ $< $(LDLIBS_libxenctrl) $(APPEND_LDFLAGS)
xen-hvmcrash: xen-hvmcrash.o
- $(CC) $(LDFLAGS) -o $@ $< $(LDLIBS_libxenctrl) $(APPEND_LDFLAGS)
+ $(CC) $(LDFLAGS) -o $@ $< $(LDLIBS_libxenctrl) $(LDLIBS_libxendevicemodel) $(APPEND_LDFLAGS)
xen-memshare: xen-memshare.o
$(CC) $(LDFLAGS) -o $@ $< $(LDLIBS_libxenctrl) $(APPEND_LDFLAGS)
diff --git a/tools/misc/xen-hvmcrash.c b/tools/misc/xen-hvmcrash.c
index 1d058fa40a47..efa47c9dfec8 100644
--- a/tools/misc/xen-hvmcrash.c
+++ b/tools/misc/xen-hvmcrash.c
@@ -24,36 +24,25 @@
* DEALINGS IN THE SOFTWARE.
*/
-#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
-#include <stddef.h>
-#include <stdint.h>
-#include <unistd.h>
#include <string.h>
#include <errno.h>
-#include <limits.h>
-
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <arpa/inet.h>
#include <xenctrl.h>
-#include <xen/xen.h>
-#include <xen/domctl.h>
-#include <xen/hvm/save.h>
+#include <xendevicemodel.h>
+
+#include <xen/asm/x86-defns.h>
int
main(int argc, char **argv)
{
int domid;
xc_interface *xch;
+ xendevicemodel_handle *dmod;
xc_domaininfo_t dominfo;
- int ret;
- uint32_t len;
- uint8_t *buf;
- uint32_t off;
- struct hvm_save_descriptor *descriptor;
+ int vcpu_id, ret;
+ bool injected = false;
if (argc != 2 || !argv[1] || (domid = atoi(argv[1])) < 0) {
fprintf(stderr, "usage: %s <domid>\n", argv[0]);
@@ -77,65 +66,22 @@ main(int argc, char **argv)
exit(1);
}
- ret = xc_domain_pause(xch, domid);
- if (ret < 0) {
- perror("xc_domain_pause");
- exit(-1);
- }
-
- /*
- * Calling with zero buffer length should return the buffer length
- * required.
- */
- ret = xc_domain_hvm_getcontext(xch, domid, 0, 0);
- if (ret < 0) {
- perror("xc_domain_hvm_getcontext");
- exit(1);
- }
-
- len = ret;
- buf = malloc(len);
- if (buf == NULL) {
- perror("malloc");
- exit(1);
- }
-
- ret = xc_domain_hvm_getcontext(xch, domid, buf, len);
- if (ret < 0) {
- perror("xc_domain_hvm_getcontext");
- exit(1);
- }
-
- off = 0;
-
- while (off < len) {
- descriptor = (struct hvm_save_descriptor *)(buf + off);
-
- off += sizeof (struct hvm_save_descriptor);
-
- if (descriptor->typecode == HVM_SAVE_CODE(CPU)) {
- HVM_SAVE_TYPE(CPU) *cpu;
+ dmod = xc_interface_dmod_handle(xch);
- /* Overwrite EIP/RIP with some recognisable but bogus value */
- cpu = (HVM_SAVE_TYPE(CPU) *)(buf + off);
- printf("CPU[%d]: RIP = %" PRIx64 "\n", descriptor->instance, cpu->rip);
- cpu->rip = 0xf001;
- } else if (descriptor->typecode == HVM_SAVE_CODE(END)) {
- break;
+ for (vcpu_id = 0; vcpu_id <= dominfo.max_vcpu_id; vcpu_id++) {
+ printf("Injecting #DF to vcpu ID #%d...\n", vcpu_id);
+ ret = xendevicemodel_inject_event(dmod, domid, vcpu_id,
+ X86_EXC_DF,
+ XEN_DMOP_EVENT_hw_exc, 0, 0, 0);
+ if (ret < 0) {
+ fprintf(stderr, "Could not inject #DF to vcpu ID #%d: %s\n", vcpu_id, strerror(errno));
+ continue;
}
-
- off += descriptor->length;
- }
-
- ret = xc_domain_hvm_setcontext(xch, domid, buf, len);
- if (ret < 0) {
- perror("xc_domain_hvm_setcontext");
- exit(1);
+ injected = true;
}
- ret = xc_domain_unpause(xch, domid);
- if (ret < 0) {
- perror("xc_domain_unpause");
+ if (!injected) {
+ fprintf(stderr, "Could not inject #DF to any vcpu!\n");
exit(1);
}
--
2.34.1
On 01/07/2024 3:47 pm, Matthew Barnes wrote: > diff --git a/tools/misc/xen-hvmcrash.c b/tools/misc/xen-hvmcrash.c > index 1d058fa40a47..efa47c9dfec8 100644 > --- a/tools/misc/xen-hvmcrash.c > +++ b/tools/misc/xen-hvmcrash.c > @@ -77,65 +66,22 @@ main(int argc, char **argv) > exit(1); > } > > - ret = xc_domain_pause(xch, domid); > - if (ret < 0) { > - perror("xc_domain_pause"); > - exit(-1); > - } You can't really drop the xc_domain_{un,}pause() calls. Previously, from the point of view of the guest, we atomically corrupted all vCPUs at once. Without the pause, we're corrupting each vCPU in turn, with the guest able to run in the meantime. ~Andrew
© 2016 - 2024 Red Hat, Inc.