All of lore.kernel.org
 help / color / mirror / Atom feed
From: Stefano Garzarella <sgarzare@redhat.com>
To: Luigi Leonardi <leonardi@redhat.com>
Cc: qemu-devel@nongnu.org, Gerd Hoffmann <kraxel@redhat.com>,
	 Ani Sinha <anisinha@redhat.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	 Zhao Liu <zhao1.liu@intel.com>,
	qemu-stable@nongnu.org
Subject: Re: [PATCH v3 1/2] igvm: validate byte_offset before using it in parameter directives
Date: Thu, 10 Sep 2026 11:35:30 +0200	[thread overview]
Message-ID: <aqJ3PFF1a8yexOAz@sgarzare-redhat> (raw)
In-Reply-To: <20260907-fix_offset-v3-1-bcffcefc0985@redhat.com>

On Mon, Sep 07, 2026 at 04:58:55PM +0200, Luigi Leonardi wrote:
>None of the directive handlers that place data at a byte_offset within
>a parameter area validated that byte_offset actually falls within the
>parameter area's size. A malformed IGVM file with byte_offset > size would
>underflow the "size - byte_offset" computation used to determine
>remaining space, wrapping to a huge value and defeating the size
>check, then write out of bounds through param_entry->data +
>byte_offset.
>
>Add qigvm_get_param_data(), which looks up the parameter area,
>checks byte_offset against its size, and returns the offset-adjusted
>data pointer directly (NULL on failure), with the remaining space
>returned via an output parameter, instead of the raw
>QIgvmParameterData entry. This keeps the byte_offset arithmetic in
>one place instead of repeating param_entry->data + byte_offset and
>param_entry->size - byte_offset at every call site.
>
>Use it in the vp-count and environment-info handlers, adding an
>explicit check that the fixed-size write fits in the remaining space.
>

Should we add a Fixes tag?

>Signed-off-by: Luigi Leonardi <leonardi@redhat.com>
>---
> backends/igvm.c                | 70 +++++++++++++++++++++++++++++++++++-------
> include/system/igvm-internal.h |  6 ++++
> 2 files changed, 65 insertions(+), 11 deletions(-)
>
>diff --git a/backends/igvm.c b/backends/igvm.c
>index 7b7bdc72b7..a8340d7d55 100644
>--- a/backends/igvm.c
>+++ b/backends/igvm.c
>@@ -101,6 +101,39 @@ qigvm_find_param_entry(QIgvm *igvm, uint32_t parameter_area_index,
>     return NULL;
> }
>
>+/*
>+ * Look up a parameter area and check that param->byte_offset falls within
>+ * it, in one step. On success, returns the offset-adjusted write location
>+ * within the parameter area and sets *param_size to the remaining space
>+ * there, so callers never need to touch param->byte_offset themselves.
>+ * Returns NULL on failure.
>+ */

How about something a little more concise and straight to the point?

/*
  * Get parameter area data at byte_offset with bounds validation.
  * On success, returns offset-adjusted data pointer and sets param_size
  * to remaining space.
  * Returns NULL on failure.
  */

>+uint8_t *
>+qigvm_get_param_data(QIgvm *igvm, uint32_t parameter_area_index,

IIUC all callers are just passing param->parameter_area_index, since we 
have `param`, can we avoid passsing `parameter_area_index` as parameter?

>+                     const IGVM_VHS_PARAMETER *param,
>+                     uint32_t *param_size,
>+                     Error **errp)
>+{
>+    QIgvmParameterData *param_entry;
>+
>+    assert(param_size);
>+
>+    param_entry = qigvm_find_param_entry(igvm, parameter_area_index, errp);
>+    if (!param_entry) {
>+        return NULL;
>+    }
>+
>+    if (param->byte_offset > param_entry->size) {
>+        error_setg(errp,
>+                   "IGVM: byte_offset 0x%x exceeds parameter area size 0x%x",
>+                   param->byte_offset, param_entry->size);
>+        return NULL;
>+    }
>+
>+    *param_size = param_entry->size - param->byte_offset;
>+    return param_entry->data + param->byte_offset;
>+}
>+
> static int qigvm_directive_page_data(QIgvm *ctx, const uint8_t *header_data,
>                                      Error **errp);
> static int qigvm_directive_vp_context(QIgvm *ctx, const uint8_t *header_data,
>@@ -682,17 +715,25 @@ static int qigvm_directive_vp_count(QIgvm *ctx, const uint8_t *header_data,
>                                     Error **errp)
> {
>     const IGVM_VHS_PARAMETER *param = (const IGVM_VHS_PARAMETER *)header_data;
>-    QIgvmParameterData *param_entry;
>+    uint8_t *param_data;
>+    uint32_t param_size;
>     uint32_t *vp_count;
>     CPUState *cpu;
>
>-    param_entry = qigvm_find_param_entry(ctx,
>-                                         param->parameter_area_index, errp);
>-    if (param_entry == NULL) {
>+    param_data = qigvm_get_param_data(ctx, param->parameter_area_index,

Can we avoid `param_data` at all and assing it directly to `vp_count` ?

>+                                       param, &param_size, errp);
>+    if (!param_data) {
>         return -1;
>     }
>
>-    vp_count = (uint32_t *)(param_entry->data + param->byte_offset);
>+    if (sizeof(*vp_count) > param_size) {
>+        error_setg(errp,
>+                   "IGVM: vp-count parameter exceeds parameter area "
>+                   "defined in IGVM file");
>+        return -1;
>+    }
>+
>+    vp_count = (uint32_t *)param_data;
>     *vp_count = 0;
>     CPU_FOREACH(cpu)
>     {
>@@ -707,17 +748,24 @@ static int qigvm_directive_environment_info(QIgvm *ctx,
>                                             Error **errp)
> {
>     const IGVM_VHS_PARAMETER *param = (const IGVM_VHS_PARAMETER *)header_data;
>-    QIgvmParameterData *param_entry;
>+    uint8_t *param_data;
>+    uint32_t param_size;
>     IgvmEnvironmentInfo *environmental_state;
>
>-    param_entry = qigvm_find_param_entry(ctx,
>-                                         param->parameter_area_index, errp);
>-    if (param_entry == NULL) {
>+    param_data = qigvm_get_param_data(ctx, param->parameter_area_index,
>+                                       param, &param_size, errp);

Ditto.

>+    if (!param_data) {
>+        return -1;
>+    }
>+
>+    if (sizeof(*environmental_state) > param_size) {
>+        error_setg(errp,
>+                   "IGVM: environment-info parameter exceeds parameter area "
>+                   "defined in IGVM file");
>         return -1;
>     }
>
>-    environmental_state =
>-        (IgvmEnvironmentInfo *)(param_entry->data + param->byte_offset);
>+    environmental_state = (IgvmEnvironmentInfo *)param_data;
>     environmental_state->memory_is_shared = 1;
>
>     return 0;
>diff --git a/include/system/igvm-internal.h b/include/system/igvm-internal.h
>index 9e9fa1d9af..f5ee2d5b1b 100644
>--- a/include/system/igvm-internal.h
>+++ b/include/system/igvm-internal.h
>@@ -81,4 +81,10 @@ QIgvmParameterData*
> qigvm_find_param_entry(QIgvm *igvm, uint32_t parameter_area_index,
>                        Error **errp);
>
>+uint8_t *
>+qigvm_get_param_data(QIgvm *igvm, uint32_t parameter_area_index,
>+                     const IGVM_VHS_PARAMETER *param,
>+                     uint32_t *param_size,
>+                     Error **errp);
>+
> #endif
>
>-- 
>2.55.0
>



  reply	other threads:[~2026-09-10  9:36 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-07 14:58 [PATCH v3 0/2] igvm: fix byte_offset handling in parameter directives Luigi Leonardi
2026-09-07 14:58 ` [PATCH v3 1/2] igvm: validate byte_offset before using it " Luigi Leonardi
2026-09-10  9:35   ` Stefano Garzarella [this message]
2026-09-07 14:58 ` [PATCH v3 2/2] igvm: honor byte_offset when writing memory map, MADT and device tree Luigi Leonardi
2026-09-10  9:47   ` Stefano Garzarella

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=aqJ3PFF1a8yexOAz@sgarzare-redhat \
    --to=sgarzare@redhat.com \
    --cc=anisinha@redhat.com \
    --cc=kraxel@redhat.com \
    --cc=leonardi@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=qemu-stable@nongnu.org \
    --cc=zhao1.liu@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.