Skip to contentSkip to conversation

UBSAN fixes

PATCH In review Oleksii Kurochko 2 participants 13 messages Jun 3 – Jun 25

lore
Share & tools
lore
#1 lore raw

PATCH v3 1/2 libfdt: fix UBSAN null pointer in fdt_property()

In reply to Oleksii Kurochko

fdt_property() unconditionally calls memcpy(ptr, val, len) even when len is zero and val is NULL. This is a legitimate calling convention for adding empty FDT properties such as "interrupt-controller", which carry no payload.

However, compilers that treat memcpy as nonnull on its pointer arguments will fire UBSAN before observing that len is zero.

Guard the memcpy() with a check on len so it is skipped entirely when there is no payload to copy, bringing the code in line with the nonnull contract.

Origin: git://git.kernel.org/pub/scm/utils/dtc/dtc.git f57e7df35df4

Fixes f0ea06558068 ("libfdt: add version 1.3.0")

---
Changes in v3:
- Add Reviewed-by: Michal Orzel <[email protected]>.
- Add Release-Acked-by.
---
Changes in v2:
- Backport patch from https://github.com/dgibson/dtc/commit/f57e7df35df4a301961cbbf9433ba4e85c2ee5ed

---
 xen/common/libfdt/fdt_sw.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
xen/common/libfdt/fdt_sw.c +2 -1
index 4c569ee7eb0d..96d4cf571319 100644
@@ -330,7 +330,8 @@ int fdt_property(void *fdt, const char *name, const void *val, int len)
 	ret = fdt_property_placeholder(fdt, name, len, &ptr);
 	if (ret)
 		return ret;
-	memcpy(ptr, val, len);
+	if (len)
+		memcpy(ptr, val, len);
 	return 0;
 }
 
Signature · 1 line
2.54.0
#2 lore raw

PATCH v3 2/2 xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

In reply to Oleksii Kurochko

On architectures that run guests in dom0less mode without the PV ABI (currently RISC-V), no shared_info page is allocated and d->shared_info remains NULL throughout the domain lifetime. Several places in common code access d->shared_info through the shared_info() macro or directly, causing UBSAN null-pointer errors on such architectures.

Rather than adding runtime NULL guards that are logically unreachable on x86 and Arm (where shared_info is always allocated), introduce a new Kconfig symbol CONFIG_HAS_SHARED_INFO selected by x86 and Arm.

On !HAS_SHARED_INFO the shared_info() macro expands to a dereference of a pointer returned by shared_info_absent(), which is declared but intentionally never defined. Any use of shared_info() that is not dead-code-eliminated will therefore cause a link-time failure, making missed guards impossible to overlook.

The 2L event-channel ops call shared_info() and must not be compiled on architectures without a shared_info page, so event_2l.o is gated on CONFIG_HAS_SHARED_INFO. On such architectures evtchn_init() installs the FIFO ops as a placeholder instead; evtchn_fifo_word_from_port() is guarded against uninitialised d->evtchn_fifo so the FIFO ops are safe before evtchn_fifo_init_control() is called by the guest.

---
With having in Xen:
info->shared_info_frame = INVALID_GFN_RAW;
probably we want to add at least ASSERT() in tools/libs/guest code as xc_map_foreign_range() doesn't check if mfn arguemnt is correct or not.
---
Changes in v3:
- Introduce CONFIG_HAS_SHARED_INFO Kconfig symbol selected by x86 and Arm; RISC-V does not select it.
- Gate shared_info() macro on CONFIG_HAS_SHARED_INFO; on !HAS_SHARED_INFO it calls shared_info_absent() (declared, never defined) so any unguarded use produces a link-time error.
- Replace runtime if (!d->shared_info) guards with IS_ENABLED() at call sites so both branches type-check and dead code is eliminated.
- Guard shared_info_frame assignment in domctl.c.
- Gate event_2l.o on CONFIG_HAS_SHARED_INFO; use FIFO ops as placeholder on !HAS_SHARED_INFO archs instead of dedicated stub ops; guard evtchn_fifo_word_from_port() against uninitialised d->evtchn_fifo.
- Add static inline stubs for evtchn_2l_init() (!HAS_SHARED_INFO) and evtchn_fifo_init_ops() (!EVTCHN_FIFO) so call sites can use IS_ENABLED() without #ifdef.
- Drop inaccurate changelog entry about "only FIFO ABI" migration.
- Update the commit message.
- Drop R-by: Baptiste ... as some extra checks are added.
---
Changes in v2:
- Update commit message + subject.
- Drop Fixes tag.

---
 xen/arch/arm/Kconfig       |  1 +
 xen/arch/x86/Kconfig       |  1 +
 xen/common/Kconfig         |  3 +++
 xen/common/Makefile        |  2 +-
 xen/common/domain.c        |  6 +++---
 xen/common/domctl.c        | 11 ++++++++---
 xen/common/event_channel.c | 14 +++++++++++---
 xen/common/event_channel.h |  6 ++++++
 xen/common/event_fifo.c    | 11 ++++++++++-
 xen/common/time.c          |  2 ++
 xen/include/xen/shared.h   |  7 +++++++
 xen/include/xen/time.h     |  4 ++++
 12 files changed, 57 insertions(+), 11 deletions(-)
xen/arch/arm/Kconfig +1 -0
index 79622b46a10d..23ff49842c90 100644
@@ -20,6 +20,7 @@ config ARM
 	select HAS_DEVICE_TREE_DISCOVERY
 	select HAS_DOM0LESS
 	select HAS_GRANT_CACHE_FLUSH if GRANT_TABLE
+	select HAS_SHARED_INFO
 	select HAS_STACK_PROTECTOR
 	select HAS_UBSAN
 
xen/arch/x86/Kconfig +1 -0
index 2ce4747f6ea7..49697b795259 100644
@@ -29,6 +29,7 @@ config X86
 	select HAS_PCI_MSI
 	select HAS_PIRQ
 	select HAS_SCHED_GRANULARITY
+	select HAS_SHARED_INFO
 	imply HAS_SOFT_RESET
 	select HAS_UBSAN
 	select HAS_VMAP
xen/common/Kconfig +3 -0
index 5ff71480eebe..8b48d84c79e8 100644
@@ -158,6 +158,9 @@ config HAS_PMAP
 config HAS_SCHED_GRANULARITY
 	bool
 
+config HAS_SHARED_INFO
+	bool
+
 config HAS_SOFT_RESET
 	bool
 
xen/common/Makefile +1 -1
index 6018e256147f..f69d47d18934 100644
@@ -12,7 +12,7 @@ obj-$(CONFIG_DEVICE_TREE_PARSE) += device-tree/
 obj-$(CONFIG_IOREQ_SERVER) += dm.o
 obj-y += domain.o
 obj-y += domid.o
-obj-y += event_2l.o
+obj-$(CONFIG_HAS_SHARED_INFO) += event_2l.o
 obj-y += event_channel.o
 obj-$(CONFIG_EVTCHN_FIFO) += event_fifo.o
 obj-$(CONFIG_GRANT_TABLE) += grant_table.o
xen/common/domain.c +3 -3
index bb9e210c2895..caf3b1e1ba0b 100644
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
     struct domain *d = v->domain;
 
     v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
 }
 
 static struct domain *alloc_domain_struct(void)
xen/common/domctl.c +8 -3
index b969f5ada6cc..edeb8ec32b3b 100644
@@ -104,9 +104,14 @@ void getdomaininfo(struct domain *d, struct xen_domctl_getdomaininfo *info)
 #ifdef CONFIG_MEM_PAGING
     info->paged_pages       = atomic_read(&d->paged_pages);
 #endif
-    info->shared_info_frame =
-        gfn_x(mfn_to_gfn(d, _mfn(virt_to_mfn(d->shared_info))));
-    BUG_ON(SHARED_M2P(info->shared_info_frame));
+    if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+    {
+        info->shared_info_frame =
+            gfn_x(mfn_to_gfn(d, _mfn(virt_to_mfn(d->shared_info))));
+        BUG_ON(SHARED_M2P(info->shared_info_frame));
+    }
+    else
+        info->shared_info_frame = INVALID_GFN_RAW;
 
     info->cpupool = cpupool_get_id(d);
 
xen/common/event_channel.c +11 -3
index a3d18bc464e8..d8ff0122e86f 100644
@@ -1323,9 +1323,13 @@ int evtchn_reset(struct domain *d, bool resuming)
         rc = -EAGAIN;
     else if ( d->evtchn_fifo )
     {
-        /* Switching back to 2-level ABI. */
         evtchn_fifo_destroy(d);
-        evtchn_2l_init(d);
+
+        if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+            /* Switching back to 2-level ABI. */
+            evtchn_2l_init(d);
+        else
+            evtchn_fifo_init_ops(d);
     }
 
     write_unlock(&d->event_lock);
@@ -1624,7 +1628,11 @@ void evtchn_check_pollers(struct domain *d, unsigned int port)
 
 int evtchn_init(struct domain *d, unsigned int max_port)
 {
-    evtchn_2l_init(d);
+    if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+        evtchn_2l_init(d);
+    else
+        evtchn_fifo_init_ops(d);
+
     d->max_evtchn_port = min_t(unsigned int, max_port, INT_MAX);
 
     d->evtchn = alloc_evtchn_bucket(d, 0);
xen/common/event_channel.h +6 -0
index dc94a43cc2dd..91f6d7b3d4bf 100644
@@ -44,7 +44,11 @@ static inline void evtchn_port_print_state(struct domain *d,
 
 /* 2-level */
 
+#ifdef CONFIG_HAS_SHARED_INFO
 void evtchn_2l_init(struct domain *d);
+#else
+static inline void evtchn_2l_init(struct domain *d) {}
+#endif
 
 /* FIFO */
 
@@ -55,6 +59,7 @@ struct evtchn_expand_array;
 int evtchn_fifo_init_control(struct evtchn_init_control *init_control);
 int evtchn_fifo_expand_array(const struct evtchn_expand_array *expand_array);
 void evtchn_fifo_destroy(struct domain *d);
+void evtchn_fifo_init_ops(struct domain *d);
 #else
 static inline int evtchn_fifo_init_control(struct evtchn_init_control *init_control)
 {
@@ -68,6 +73,7 @@ static inline void evtchn_fifo_destroy(struct domain *d)
 {
     return;
 }
+static inline void evtchn_fifo_init_ops(struct domain *d) {}
 #endif /* CONFIG_EVTCHN_FIFO */
 
 #endif /* EVENT_CHANNEL_H */
xen/common/event_fifo.c +10 -1
index 37cba9bc4564..4aa34ca13a56 100644
@@ -62,6 +62,9 @@ static inline event_word_t *evtchn_fifo_word_from_port(const struct domain *d,
      */
     smp_rmb();
 
+    if ( unlikely(!d->evtchn_fifo) )
+        return NULL;
+
     if ( unlikely(port >= d->evtchn_fifo->num_evtchns) )
         return NULL;
 
@@ -420,6 +423,11 @@ static const struct evtchn_port_ops evtchn_port_ops_fifo =
     .print_state   = evtchn_fifo_print_state,
 };
 
+void evtchn_fifo_init_ops(struct domain *d)
+{
+    d->evtchn_port_ops = &evtchn_port_ops_fifo;
+}
+
 static int map_guest_page(struct domain *d, uint64_t gfn, void **virt)
 {
     struct page_info *p;
@@ -562,7 +570,8 @@ static void setup_ports(struct domain *d, unsigned int prev_evtchns)
 
         evtchn = evtchn_from_port(d, port);
 
-        if ( guest_test_bit(d, port, &shared_info(d, evtchn_pending)) )
+        if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) &&
+             guest_test_bit(d, port, &shared_info(d, evtchn_pending)) )
             evtchn->pending = true;
 
         evtchn_fifo_set_priority(d, evtchn, EVTCHN_FIFO_PRIORITY_DEFAULT);
xen/common/time.c +2 -0
index 04a65f00b35c..cdfdc53b6a17 100644
@@ -89,6 +89,7 @@ struct tm gmtime(unsigned long t)
     return tbuf;
 }
 
+#ifdef CONFIG_HAS_SHARED_INFO
 void update_domain_wallclock_time(struct domain *d)
 {
     uint32_t *wc_version;
@@ -117,6 +118,7 @@ void update_domain_wallclock_time(struct domain *d)
 
     spin_unlock(&wc_lock);
 }
+#endif /* CONFIG_HAS_SHARED_INFO */
 
 /* Set clock to <secs,usecs> after 00:00:00 UTC, 1 January, 1970. */
 void do_settime(u64 secs, unsigned int nsecs, u64 system_time_base)
xen/include/xen/shared.h +7 -0
index 5b71342cab32..a938312170e5 100644
@@ -43,7 +43,14 @@ typedef struct vcpu_info vcpu_info_t;
 
 extern vcpu_info_t dummy_vcpu_info;
 
+#ifdef CONFIG_HAS_SHARED_INFO
 #define shared_info(d, field)      __shared_info(d, (d)->shared_info, field)
+#else
+void * shared_info_absent(void);
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())
+#endif /* CONFIG_HAS_SHARED_INFO */
+
 #define vcpu_info(v, field)        \
         __vcpu_info(v, (vcpu_info_t *)(v)->vcpu_info_area.map, field)
 
xen/include/xen/time.h +4 -0
index e9c0822e6f31..2f872f580ffc 100644
@@ -66,7 +66,11 @@ struct tm wallclock_time(uint64_t *ns);
 #define version_update_begin(v) (((v) + 1) | 1)
 #define version_update_end(v)   ((v) + 1)
 extern void update_vcpu_system_time(struct vcpu *v);
+#ifdef CONFIG_HAS_SHARED_INFO
 extern void update_domain_wallclock_time(struct domain *d);
+#else
+static inline void update_domain_wallclock_time(struct domain *d) {}
+#endif
 
 extern void do_settime(
     u64 secs, unsigned int nsecs, u64 system_time_base);
Signature · 1 line
2.54.0
#0 lore raw

v3 cover UBSAN fixes

During Baptiste Le Duc's work on adding CI stuff for RISC-V several UBSAN issues were found.

This patch series resolves the found issues.

CI tests: https://gitlab.com/xen-project/people/olkur/xen/-/pipelines/2573666715

---
Changes in v3:
- Reorder patches in the way that ready to be merged go first.
- Merged to upstream/staging:
xen/riscv: fix switch_stack_and_jump() for range beyond 1M
- Add R-by for patch "libfdt: fix UBSAN null pointer in fdt_property()".
- Address the comments from ML.
---
Changes in v2:
- Address comments from ML.
---

Oleksii Kurochko (2):
libfdt: fix UBSAN null pointer in fdt_property()
xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

 xen/arch/arm/Kconfig       |  1 +
 xen/arch/x86/Kconfig       |  1 +
 xen/common/Kconfig         |  3 +++
 xen/common/Makefile        |  2 +-
 xen/common/domain.c        |  6 +++---
 xen/common/domctl.c        | 11 ++++++++---
 xen/common/event_channel.c | 14 +++++++++++---
 xen/common/event_channel.h |  6 ++++++
 xen/common/event_fifo.c    | 11 ++++++++++-
 xen/common/libfdt/fdt_sw.c |  3 ++-
 xen/common/time.c          |  2 ++
 xen/include/xen/shared.h   |  7 +++++++
 xen/include/xen/time.h     |  4 ++++
 13 files changed, 59 insertions(+), 12 deletions(-)
Signature · 1 line
2.54.0
#3 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

In reply to Oleksii Kurochko

xen/common/domain.c void vcpu_info_reset(struct vcpu *v) · line 326

Quoted 13 lines · Oleksii Kurochko
xen/common/domain.c +3 -3
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
     struct domain *d = v->domain;
 
     v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
 }

While the change here is likely okay, it points at possible further omissions. You've dealt with all uses of shared_info(), but you've left alone all uses of vcpu_info() (and __vcpu_info()). Reads are presumably okay, but writes to dummy_vcpu_info open a side channel for possible info leaks. Looking over uses in common code, no code changes may be needed; extending the description may be all that's wanted here.

Quoted 6 lines
xen/common/event_channel.c +0 -0
@@ -1323,9 +1323,13 @@ int evtchn_reset(struct domain *d, bool resuming)
         rc = -EAGAIN;
     else if ( d->evtchn_fifo )
     {

xen/common/event_channel.c int evtchn_reset(struct domain *d, bool resuming) · line 1333

Quoted 10 lines
xen/common/event_channel.c +6 -2
-        /* Switching back to 2-level ABI. */
         evtchn_fifo_destroy(d);
-        evtchn_2l_init(d);
+
+        if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+            /* Switching back to 2-level ABI. */
+            evtchn_2l_init(d);
+        else
+            evtchn_fifo_init_ops(d);
     }

The "else" part isn't needed, is it? evtchn_fifo_destroy() doesn't undo what ...

xen/common/event_channel.c void evtchn_check_pollers(struct domain *d, unsigned int port) · line 1634

Quoted 9 lines
xen/common/event_channel.c +4 -1
@@ -1624,7 +1628,11 @@ void evtchn_check_pollers(struct domain *d, unsigned int port)
 
 int evtchn_init(struct domain *d, unsigned int max_port)
 {
-    evtchn_2l_init(d);
+    if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+        evtchn_2l_init(d);
+    else
+        evtchn_fifo_init_ops(d);

... was done here.

xen/include/xen/shared.h typedef struct vcpu_info vcpu_info_t; · line 49

Quoted 10 lines
xen/include/xen/shared.h +3 -0
@@ -43,7 +43,14 @@ typedef struct vcpu_info vcpu_info_t;
 
 extern vcpu_info_t dummy_vcpu_info;
 
+#ifdef CONFIG_HAS_SHARED_INFO
 #define shared_info(d, field)      __shared_info(d, (d)->shared_info, field)
+#else
+void * shared_info_absent(void);

Nit: Stray blank after '*'.

xen/include/xen/shared.h typedef struct vcpu_info vcpu_info_t; · line 51

Quoted 2 lines
xen/include/xen/shared.h +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

Jan

#4 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 6 lines · Jan Beulich
Quoted 5 lines · Oleksii Kurochko
xen/common/domain.c +0 -0
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
      struct domain *d = v->domain;
 
Quoted 15 lines

xen/common/domain.c void vcpu_info_reset(struct vcpu *v) · line 326

Quoted 8 lines
xen/common/domain.c +3 -3
      v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
  }

While the change here is likely okay, it points at possible further omissions. You've dealt with all uses of shared_info(), but you've left alone all uses of vcpu_info() (and __vcpu_info()). Reads are presumably okay, but writes to dummy_vcpu_info open a side channel for possible info leaks. Looking over uses in common code, no code changes may be needed; extending the description may be all that's wanted here.

Isn't there already a side channel that could allow leaks, even without this change? The change here just made it worsen because now info leak will happen for all vCPUs when CONFIG_HAS_SHARED_INFO=n.

I will add to the description the following:

```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs.  Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious will happen if it will be leaked; the event_2l.c paths are unreachable because the preceding shared_info() call would trap first; the write in vcpu_info_populate() targets the new mapping buffer, not dummy_vcpu_info; all remaining writes are x86 PV-specific for which CONFIG_HAS_SHARED_INFO=y.  No code changes are needed.
```
Quoted 12 lines
Quoted 12 lines
xen/common/event_channel.c +2 -2
@@ -1323,9 +1323,13 @@ int evtchn_reset(struct domain *d, bool resuming)
          rc = -EAGAIN;
      else if ( d->evtchn_fifo )
      {
-        /* Switching back to 2-level ABI. */
          evtchn_fifo_destroy(d);
-        evtchn_2l_init(d);
+
+        if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
Quoted 8 lines

xen/common/event_channel.c int evtchn_reset(struct domain *d, bool resuming) · line 1333

Quoted 5 lines
xen/common/event_channel.c +4 -0
+            /* Switching back to 2-level ABI. */
+            evtchn_2l_init(d);
+        else
+            evtchn_fifo_init_ops(d);
      }

The "else" part isn't needed, is it? evtchn_fifo_destroy() doesn't undo what ...

Agree, it isn't really needed. I will drop "else" part.

Quoted 29 lines

xen/common/event_channel.c void evtchn_check_pollers(struct domain *d, unsigned int port) · line 1634

Quoted 10 lines
xen/common/event_channel.c +4 -1
@@ -1624,7 +1628,11 @@ void evtchn_check_pollers(struct domain *d, unsigned int port)
 
  int evtchn_init(struct domain *d, unsigned int max_port)
  {
-    evtchn_2l_init(d);
+    if ( IS_ENABLED(CONFIG_HAS_SHARED_INFO) )
+        evtchn_2l_init(d);
+    else
+        evtchn_fifo_init_ops(d);

... was done here.

xen/include/xen/shared.h typedef struct vcpu_info vcpu_info_t; · line 49

Quoted 10 lines
xen/include/xen/shared.h +3 -0
@@ -43,7 +43,14 @@ typedef struct vcpu_info vcpu_info_t;
 
  extern vcpu_info_t dummy_vcpu_info;
 
+#ifdef CONFIG_HAS_SHARED_INFO
  #define shared_info(d, field)      __shared_info(d, (d)->shared_info, field)
+#else
+void * shared_info_absent(void);

Nit: Stray blank after '*'.

xen/include/xen/shared.h typedef struct vcpu_info vcpu_info_t; · line 51

Quoted 2 lines
xen/include/xen/shared.h +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())
Quoted 6 lines

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

It would be better. I will apply that.

Thanks.

~ Oleksii

#5 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 23 lines · Oleksii Kurochko
Quoted 6 lines · Jan Beulich
Quoted 5 lines · Oleksii Kurochko
xen/common/domain.c +0 -0
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
      struct domain *d = v->domain;
 
Quoted 15 lines

xen/common/domain.c void vcpu_info_reset(struct vcpu *v) · line 326

Quoted 8 lines
xen/common/domain.c +3 -3
      v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
  }

While the change here is likely okay, it points at possible further omissions. You've dealt with all uses of shared_info(), but you've left alone all uses of vcpu_info() (and __vcpu_info()). Reads are presumably okay, but writes to dummy_vcpu_info open a side channel for possible info leaks. Looking over uses in common code, no code changes may be needed; extending the description may be all that's wanted here.

Quoted 2 lines

Isn't there already a side channel that could allow leaks, even without this change?

There are multiple aspects here. First, for PV secondary vCPU-s cannot be launched when their vcpu-info still points at dummy_vcpu_info. HVM vCPU-s make very limited use of vcpu-info fields. Writes look to be limited to the evtchn_upcall_{mask,pending} fields, which isn't really an info leak.

My main point here is: None of this goes without making clear that the necessary auditing was properly done.

Quoted 4 lines

The change here just made it worsen because now info leak will happen for all vCPUs when CONFIG_HAS_SHARED_INFO=n.

I will add to the description the following:

Quoted 11 lines
```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs.  Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious will happen 
if it will be leaked; the event_2l.c paths are unreachable because the 
preceding shared_info() call would trap first; the write in 
vcpu_info_populate() targets the new mapping buffer, not 
dummy_vcpu_info; all remaining writes are x86 PV-specific for which 
CONFIG_HAS_SHARED_INFO=y.  No code changes are needed.
```

As you start with "common code", how come the "x86 PV-specific" part is still there (i.e. relevant)? Isn't all PV stuff in x86-specific code?

Jan

#6 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 51 lines · Jan Beulich
Quoted 23 lines · Oleksii Kurochko
Quoted 6 lines · Jan Beulich
Quoted 5 lines · Oleksii Kurochko
xen/common/domain.c +0 -0
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
       struct domain *d = v->domain;
 
Quoted 15 lines

xen/common/domain.c void vcpu_info_reset(struct vcpu *v) · line 326

Quoted 8 lines
xen/common/domain.c +3 -3
       v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
   }

While the change here is likely okay, it points at possible further omissions. You've dealt with all uses of shared_info(), but you've left alone all uses of vcpu_info() (and __vcpu_info()). Reads are presumably okay, but writes to dummy_vcpu_info open a side channel for possible info leaks. Looking over uses in common code, no code changes may be needed; extending the description may be all that's wanted here.

Quoted 2 lines

Isn't there already a side channel that could allow leaks, even without this change?

There are multiple aspects here. First, for PV secondary vCPU-s cannot be launched when their vcpu-info still points at dummy_vcpu_info. HVM vCPU-s make very limited use of vcpu-info fields. Writes look to be limited to the evtchn_upcall_{mask,pending} fields, which isn't really an info leak.

My main point here is: None of this goes without making clear that the necessary auditing was properly done.

Quoted 4 lines

The change here just made it worsen because now info leak will happen for all vCPUs when CONFIG_HAS_SHARED_INFO=n.

I will add to the description the following:

Quoted 11 lines
```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs.  Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious will happen
if it will be leaked; the event_2l.c paths are unreachable because the
preceding shared_info() call would trap first; the write in
vcpu_info_populate() targets the new mapping buffer, not
dummy_vcpu_info; all remaining writes are x86 PV-specific for which
CONFIG_HAS_SHARED_INFO=y.  No code changes are needed.
```
Quoted 2 lines

As you start with "common code", how come the "x86 PV-specific" part is still there (i.e. relevant)? Isn't all PV stuff in x86-specific code?

Good point. The "x86 PV-specific" part is not part of the review of common code. I mentioned it separately to complete the audit of all write paths reachable through vcpu_info(). The intent was:

* the writes discussed before the semicolon are the common-code paths;
* the writes after the semicolon are outside common code and are x86 PV-specific, where CONFIG_HAS_SHARED_INFO=y anyway.

To avoid the ambiguity, I can reword the sentence to make that separation explicit:

```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs. Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious would happen
if it were leaked; the event_2l.c paths are unreachable because the
preceding shared_info() call would trap first; the write in
vcpu_info_populate() targets the new mapping buffer, not
dummy_vcpu_info.

Outside common code, the remaining writes are x86 PV-specific, for which
CONFIG_HAS_SHARED_INFO=y. No code changes are needed.
```

Would that wording work for you?

Thanks.

~ Oleksii

#7 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 81 lines · Oleksii Kurochko
Quoted 51 lines · Jan Beulich
Quoted 23 lines · Oleksii Kurochko
Quoted 6 lines · Jan Beulich
Quoted 5 lines · Oleksii Kurochko
xen/common/domain.c +0 -0
@@ -320,9 +320,9 @@ void vcpu_info_reset(struct vcpu *v)
       struct domain *d = v->domain;
 
Quoted 15 lines

xen/common/domain.c void vcpu_info_reset(struct vcpu *v) · line 326

Quoted 8 lines
xen/common/domain.c +3 -3
       v->vcpu_info_area.map =
-        ((v->vcpu_id < XEN_LEGACY_MAX_VCPUS)
-         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
-         : &dummy_vcpu_info);
+        IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < XEN_LEGACY_MAX_VCPUS
+        ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+        : &dummy_vcpu_info;
   }

While the change here is likely okay, it points at possible further omissions. You've dealt with all uses of shared_info(), but you've left alone all uses of vcpu_info() (and __vcpu_info()). Reads are presumably okay, but writes to dummy_vcpu_info open a side channel for possible info leaks. Looking over uses in common code, no code changes may be needed; extending the description may be all that's wanted here.

Quoted 2 lines

Isn't there already a side channel that could allow leaks, even without this change?

There are multiple aspects here. First, for PV secondary vCPU-s cannot be launched when their vcpu-info still points at dummy_vcpu_info. HVM vCPU-s make very limited use of vcpu-info fields. Writes look to be limited to the evtchn_upcall_{mask,pending} fields, which isn't really an info leak.

My main point here is: None of this goes without making clear that the necessary auditing was properly done.

Quoted 4 lines

The change here just made it worsen because now info leak will happen for all vCPUs when CONFIG_HAS_SHARED_INFO=n.

I will add to the description the following:

Quoted 11 lines
```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs.  Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious will happen
if it will be leaked; the event_2l.c paths are unreachable because the
preceding shared_info() call would trap first; the write in
vcpu_info_populate() targets the new mapping buffer, not
dummy_vcpu_info; all remaining writes are x86 PV-specific for which
CONFIG_HAS_SHARED_INFO=y.  No code changes are needed.
```
Quoted 2 lines

As you start with "common code", how come the "x86 PV-specific" part is still there (i.e. relevant)? Isn't all PV stuff in x86-specific code?

Good point. The "x86 PV-specific" part is not part of the review of common code. I mentioned it separately to complete the audit of all write paths reachable through vcpu_info(). The intent was:

* the writes discussed before the semicolon are the common-code paths;
* the writes after the semicolon are outside common code and are x86 PV-specific, where CONFIG_HAS_SHARED_INFO=y anyway.

To avoid the ambiguity, I can reword the sentence to make that separation explicit:

```
With CONFIG_HAS_SHARED_INFO=n all vCPUs fall back to the global
dummy_vcpu_info, so writes through vcpu_info() could leak data between
vCPUs. Reviewing the write paths in common code: the write in
map_guest_area() stores the constant ~0 so nothing serious would happen
if it were leaked; the event_2l.c paths are unreachable because the
preceding shared_info() call would trap first; the write in
vcpu_info_populate() targets the new mapping buffer, not
dummy_vcpu_info.

Outside common code, the remaining writes are x86 PV-specific, for which
CONFIG_HAS_SHARED_INFO=y. No code changes are needed.
```
Quoted 1 line

Would that wording work for you?

I think so, yes.

Jan

#8 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

In reply to Jan Beulich
Quoted 8 lines · Jan Beulich

Review

Quoted 2 lines
(quoted diff) +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

This could lead to compilation error:

common/domain.c: In function 'vcpu_info_reset':
common/domain.c:316:20: error: unused variable 'd' [-Werror=unused-variable]
316 | struct domain *d = v->domain;
| ^
cc1: all warnings being treated as errors

One of fixes could be just drop usage of local variable d in vcpu_info_reset():

xen/common/domain.c +0 -2
index fba8e9161937..d3b0bd571609 100644
@@ -313,11 +313,9 @@ static void vcpu_check_shutdown(struct vcpu *v)
 
 void vcpu_info_reset(struct vcpu *v)
 {
-    struct domain *d = v->domain;
-
     v->vcpu_info_area.map =
         IS_ENABLED(CONFIG_HAS_SHARED_INFO) && v->vcpu_id < 

XEN_LEGACY_MAX_VCPUS
- ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])
+ ? (vcpu_info_t *)&shared_info(v->domain, vcpu_info[v->vcpu_id])

         : &dummy_vcpu_info;

}

OR shared_info() defintion should be updated to:

#define shared_info(d, field) (*((void)(d), &shared_info_absent->field))

IMO, an update of macros defintion looks a little bit more common.

Which one do you prefer? Any better suggestion?

Thanks.

~ oleksii

#9 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 10 lines · Oleksii Kurochko
Quoted 8 lines · Jan Beulich

Review

Quoted 2 lines
(quoted diff) +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

Quoted 7 lines

This could lead to compilation error:

common/domain.c: In function 'vcpu_info_reset':
common/domain.c:316:20: error: unused variable 'd' [-Werror=unused-variable]
316 | struct domain *d = v->domain;
| ^
cc1: all warnings being treated as errors

#define shared_info(d, field) ((void)(d), shared_info_absent->field)

Which it should have been anyway, to make sure the first argument is evaluated.

Jan

#10 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 19 lines · Jan Beulich
Quoted 10 lines · Oleksii Kurochko
Quoted 8 lines · Jan Beulich

Review

Quoted 2 lines
(quoted diff) +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

Quoted 7 lines

This could lead to compilation error:

common/domain.c: In function 'vcpu_info_reset':
common/domain.c:316:20: error: unused variable 'd' [-Werror=unused-variable]

    316 |     struct domain *d = v->domain;
        |                    ^

cc1: all warnings being treated as errors

Quoted 3 lines

#define shared_info(d, field) ((void)(d), shared_info_absent->field)

Which it should have been anyway, to make sure the first argument is evaluated.

I expected that as shared_info() is used as an lvalue whose address is taken (&shared_info(d, vcpu_info[...])). To "consume" d we can't just prepend a comma expression as &(comma_expr) is illegal since a comma result isn't an lvalue.

Trying what you suggested leads to:
common/domain.c: In function 'vcpu_info_reset':
common/domain.c:320:26: error: lvalue required as unary '&' operand
320 | ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])

Thereby this option:
#define shared_info(d, field) (*((void)(d), &shared_info_absent->field))
will be better.

~ Oleksii

#11 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 31 lines · Oleksii Kurochko
Quoted 19 lines · Jan Beulich
Quoted 10 lines · Oleksii Kurochko
Quoted 8 lines · Jan Beulich

Review

Quoted 2 lines
(quoted diff) +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

Quoted 7 lines

This could lead to compilation error:

common/domain.c: In function 'vcpu_info_reset':
common/domain.c:316:20: error: unused variable 'd' [-Werror=unused-variable]

    316 |     struct domain *d = v->domain;
        |                    ^

cc1: all warnings being treated as errors

Quoted 3 lines

#define shared_info(d, field) ((void)(d), shared_info_absent->field)

Which it should have been anyway, to make sure the first argument is evaluated.

I expected that as shared_info() is used as an lvalue whose address is taken (&shared_info(d, vcpu_info[...])). To "consume" d we can't just prepend a comma expression as &(comma_expr) is illegal since a comma result isn't an lvalue.

Quoted 4 lines

Trying what you suggested leads to:
common/domain.c: In function 'vcpu_info_reset':
common/domain.c:320:26: error: lvalue required as unary '&' operand
320 | ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])

Ah yes.

Quoted 3 lines

Thereby this option:

   #define shared_info(d, field) (*((void)(d), &shared_info_absent->field))

will be better.

Or maybe

#define shared_info(d, field) (((void)(d), shared_info_absent)->field)

?

Jan

#12 lore raw

Re: [PATCH v3 2/2] xen: introduce CONFIG_HAS_SHARED_INFO for archs without a shared page

Quoted 43 lines · Jan Beulich
Quoted 31 lines · Oleksii Kurochko
Quoted 19 lines · Jan Beulich
Quoted 10 lines · Oleksii Kurochko
Quoted 8 lines · Jan Beulich

Review

Quoted 2 lines
(quoted diff) +2 -0
+#define shared_info(d, field) \
+    (*(typeof(__shared_info(d, (d)->shared_info, field)) *)shared_info_absent())

How about the simpler

extern struct shared_info *shared_info_absent; #define shared_info(d, field) (shared_info_absent->field)

?

Quoted 7 lines

This could lead to compilation error:

common/domain.c: In function 'vcpu_info_reset':
common/domain.c:316:20: error: unused variable 'd' [-Werror=unused-variable]

     316 |     struct domain *d = v->domain;
         |                    ^

cc1: all warnings being treated as errors

Quoted 3 lines

#define shared_info(d, field) ((void)(d), shared_info_absent->field)

Which it should have been anyway, to make sure the first argument is evaluated.

I expected that as shared_info() is used as an lvalue whose address is taken (&shared_info(d, vcpu_info[...])). To "consume" d we can't just prepend a comma expression as &(comma_expr) is illegal since a comma result isn't an lvalue.

Quoted 4 lines

Trying what you suggested leads to:
common/domain.c: In function 'vcpu_info_reset':
common/domain.c:320:26: error: lvalue required as unary '&' operand

    320 |         ? (vcpu_info_t *)&shared_info(d, vcpu_info[v->vcpu_id])

Ah yes.

Quoted 3 lines

Thereby this option:

    #define shared_info(d, field) (*((void)(d), &shared_info_absent->field))

will be better.

Quoted 5 lines

Or maybe

#define shared_info(d, field) (((void)(d), shared_info_absent)->field)

?

This option looks better. Ill apply it.

Thanks.

~ Oleksii