On 5/5/26 4:18 PM, Zhuoying Cai wrote:
[...]

+}
+
+static int get_key_type(const S390IPLCertificate *cert)
+{
+    int rc;
+    Error *err = NULL;
+
+    rc = qcrypto_x509_check_ecc_curve_p521(cert->raw, cert->size, &err);
+    if (rc == -1) {
+        error_report_err(err);
+        return -1;
+    }
+
+    return (rc == 1) ? DIAG_320_VCE_KEYTYPE_ECDSA_P521 :
+                       DIAG_320_VCE_KEYTYPE_SELF_DESCRIBING;
+}
+
+static int build_vce_header(VCEntry *vce, const S390IPLCertificate *cert, int 
idx)
+{
+    int key_type;
+
+    vce->vce_hdr.len = cpu_to_be32(sizeof(VCEntryHeader));
+    vce->vce_hdr.cert_idx = cpu_to_be16(idx + 1);
+    memcpy(vce->vce_hdr.name, cert->name, CERT_NAME_MAX_LEN);
+
+    key_type = get_key_type(cert);
+    if (key_type == -1) {
+        return -1;
+    }
+    vce->vce_hdr.key_type = key_type;
+
+    return 0;
+}
+
+static int build_vce_data(VCEntry *vce, const S390IPLCertificate *cert)
+{
+    uint16_t keyid_field_len;
+    uint16_t hash_field_len;
+    uint32_t cert_field_len;
+    uint32_t vce_len;
+    int rc;
+
+    rc = handle_key_id(vce, cert);
+    if (rc) {
+        return -1;
+    }
+    keyid_field_len = ROUND_UP(be16_to_cpu(vce->vce_hdr.keyid_len), 4);
Maybe return the converted length from the handle_* functions, with 0 being the
error condition?

e.g. (untested)
    keyid_len = handle_key_id(vce, cert);
    if (!keyid_len) {
        return -1;
    }

Just a nit, but I think it would improve readability.

+
+    rc = handle_hash(vce, cert, keyid_field_len);
+    if (rc) {
+        return -1;
+    }
+    hash_field_len = ROUND_UP(be16_to_cpu(vce->vce_hdr.hash_len), 4);
+
+    rc = handle_cert(vce, cert, hash_field_len);
+    if (rc || !is_cert_valid(cert)) {
+        return -1;
+    }
+    cert_field_len = ROUND_UP(be32_to_cpu(vce->vce_hdr.cert_len), 4);
+
+    vce_len = sizeof(VCEntryHeader) + keyid_field_len + hash_field_len + 
cert_field_len;
+    if (vce_len > be32_to_cpu(vce->vce_hdr.len)) {
+        return -1;
+    }
We already used the maximum size when we allocated space for the VCE, so this
check is redundant I think?

+
+    vce->vce_hdr.flags |= DIAG_320_VCE_FLAGS_VALID;
+
+    /* Update vce length to reflect the actual size used by vce */
+    vce->vce_hdr.len = cpu_to_be32(vce_len);
+
+    return 0;
+}
+
+static VCEntry *diag_320_build_vce(const S390IPLCertificate *cert, int idx)
+{
+    g_autofree VCEntry *vce = NULL;
+    uint32_t vce_max_size;
+    int rc;
+
+    /*
+     * Each field of the VCE is word-aligned.
+     * Allocate enough space for the largest possible size for this VCE.
+     * As the certificate fields (key-id, hash, data) are parsed, the
+     * VCE's length field will be updated accordingly.
+     */
+    vce_max_size = sizeof(VCEntryHeader) +
+                   ROUND_UP(CERT_KEY_ID_LEN, 4) +
+                   ROUND_UP(CERT_HASH_LEN, 4) +
+                   ROUND_UP(cert->der_size, 4);
In Patch 10, we define MAX_ENTRY_SIZE.  Is that something different than this?
Let's use the constant here unless there is a specific reason not to.

+
+    vce = g_malloc0(vce_max_size);
+    rc = build_vce_header(vce, cert, idx);
+    if (rc) {
+        /*
+         * Error occurs - VCE does not contain a valid certificate.
+         * Bit 0 of the VCE flags is 0 and the VCE length is set.
+         */
+        vce->vce_hdr.len = cpu_to_be32(VCE_INVALID_LEN);
+        goto out;
+    }
+
+    vce->vce_hdr.len = cpu_to_be32(vce_max_size);
It is not obvious later in the code if the VCE length is a true value or a
place holder for the max size, so MAX_ENTRY_SIZE would help clarify also.

+    rc = build_vce_data(vce, cert);
+    if (rc) {
+        vce->vce_hdr.len = cpu_to_be32(VCE_INVALID_LEN);
+    }
+
+out:
+    return g_steal_pointer(&vce);
+}
Actually, I wonder if we might be able to remove the diag_320_build_vce()
function since it is basically just a wrapper for building the header and data.

If we introduce the MAX_ENTRY_SIZE constant and allocate memory before calling
the build functions, I think we can simplify the whole thing.

See notes below in the loop to build VCEs...

+
+static int handle_diag320_store_vc(S390CPU *cpu, uint64_t addr, uint64_t r1, 
uintptr_t ra,
+                                   S390IPLCertificateStore *cs)
+{
+    g_autofree VCBlock *vcb = NULL;
+    size_t remaining_space;
+    uint16_t first_vc_index;
+    uint16_t last_vc_index;
+    int cs_start_index;
+    int cs_end_index;
+    uint32_t in_len;
+
+    vcb = g_new0(VCBlock, 1);
+    if (s390_cpu_virt_mem_read(cpu, addr, r1, vcb, sizeof(*vcb))) {
+        s390_cpu_virt_mem_handle_exc(cpu, ra);
+        return -1;
+    }
This is only the VCB header, right?  That's all the space is allocated for
anyway.  I find it confusing, here is my understanding, please correct if wrong:

We are allocating a VCB and updating the header values, but we never populate
the VCB's buffer area because each VCE being built has it's own malloc call?

Meaning when we read/write the VCB it is actually just the header, and the VCEs
are written independently?

If that is correct, then we never really build a full VCB right? Could we just
use the VCBHeader struct here instead to make that clear?

+
+    in_len = be32_to_cpu(vcb->vcb_hdr.in_len);
+    first_vc_index = be16_to_cpu(vcb->vcb_hdr.first_vc_index);
+    last_vc_index = be16_to_cpu(vcb->vcb_hdr.last_vc_index);
+
+    if (in_len % TARGET_PAGE_SIZE != 0) {
+        return DIAG_320_RC_INVAL_VCB_LEN;
+    }
+
+    if (first_vc_index > last_vc_index) {
+        return DIAG_320_RC_BAD_RANGE;
+    }
+
+    vcb->vcb_hdr.out_len = sizeof(VCBlockHeader);
+
+    /*
+     * DIAG 320 subcode 2 expects to query a certificate store that
+     * maintains an index origin of 1. However, the S390IPLCertificateStore
+     * maintains an index origin of 0. Thus, the indices must be adjusted
+     * for correct access into the cert store. A couple of special cases
+     * must also be accounted for.
+     */
+
+    /* Both indices are 0; return header with no certs */
+    if (first_vc_index == 0 && last_vc_index == 0) {
+        goto out;
+    }
+
+    /* Normalize indices */
+    cs_start_index = (first_vc_index == 0) ? 0 : first_vc_index - 1;
+    cs_end_index = last_vc_index - 1;
+
+    /* Requested range is outside the cert store; return header with no certs 
*/
+    if (cs_start_index >= cs->count || cs_end_index >= cs->count) {
+        goto out;
+    }
+
+    remaining_space = in_len - sizeof(VCBlockHeader);
+
+    for (int i = cs_start_index; i <= cs_end_index; i++) {
+        VCEntry *vce;
What about g_autofree VCEntry *vce = g_malloc0(MAX_ENTRY_SIZE);

Collin already mentioned using g_autofree, but in addition to that if we can
allocate the memory up front that would allow us to call the build
functions directly from here e.g. (untested)

    if (build_vce_header(vce, cert, i) || build_vce_data(vce, cert) {
        vce->vce_hdr.len = cpu_to_be32(VCE_INVALID_LEN);
    }

That would avoid the frees and steal pointer too.

+        const S390IPLCertificate *cert = &cs->certs[i];
+
+        /*
+         * Bit 0 of the VCE flags indicates whether the certificate is valid.
+         * The caller of DIAG320 subcode 2 is responsible for verifying that
+         * the VCE contains a valid certificate.
+         */
+        vce = diag_320_build_vce(cert, i);
+
+        /*
+         * If there is no more space to store the cert,
+         * set the remaining verification cert count and
+         * break early.
+         */
+        if (remaining_space < vce->vce_hdr.len) {
+            vcb->vcb_hdr.remain_ct = cpu_to_be16(last_vc_index - i);
+            g_free(vce);
+            break;
+        }
+
+        /* Write VCE */
+        if (s390_cpu_virt_mem_write(cpu, addr + vcb->vcb_hdr.out_len, r1,
+                                    vce, vce->vce_hdr.len)) {
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
+            g_free(vce);
+            return -1;
+        }
+
+        vcb->vcb_hdr.out_len += vce->vce_hdr.len;
+        remaining_space -= vce->vce_hdr.len;
+        vcb->vcb_hdr.stored_ct++;
+
+        g_free(vce);
+    }
+    vcb->vcb_hdr.stored_ct = cpu_to_be16(vcb->vcb_hdr.stored_ct);
+
+out:
+    vcb->vcb_hdr.out_len = cpu_to_be32(vcb->vcb_hdr.out_len);
+
+    if (s390_cpu_virt_mem_write(cpu, addr, r1, vcb, sizeof(VCBlockHeader))) {
+        s390_cpu_virt_mem_handle_exc(cpu, ra);
+        return -1;
+    }
+
+    return DIAG_320_RC_OK;
+}
+
  QEMU_BUILD_BUG_MSG(sizeof(VCStorageSizeBlock) != VCSSB_MIN_LEN,
                     "size of VCStorageSizeBlock is wrong");
+QEMU_BUILD_BUG_MSG(sizeof(VCBlock) != 64, "size of VCBlock is wrong");
+QEMU_BUILD_BUG_MSG(sizeof(VCEntry) != 128, "size of VCEntry is wrong");
void handle_diag_320(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
  {
@@ -270,7 +594,8 @@ void handle_diag_320(CPUS390XState *env, uint64_t r1, 
uint64_t r3, uintptr_t ra)
           * for now.
           */
          uint32_t ism_word0 = cpu_to_be32(DIAG_320_ISM_QUERY_SUBCODES |
-                                         DIAG_320_ISM_QUERY_VCSI);
+                                         DIAG_320_ISM_QUERY_VCSI |
+                                         DIAG_320_ISM_STORE_VC);
if (s390_cpu_virt_mem_write(cpu, addr, r1, &ism_word0, sizeof(ism_word0))) {
              s390_cpu_virt_mem_handle_exc(cpu, ra);
@@ -296,6 +621,18 @@ void handle_diag_320(CPUS390XState *env, uint64_t r1, 
uint64_t r3, uintptr_t ra)
          }
          env->regs[r1 + 1] = rc;
          break;
+    case DIAG_320_SUBC_STORE_VC:
+        if (addr & ~TARGET_PAGE_MASK) {
+            s390_program_interrupt(env, PGM_SPECIFICATION, ra);
+            return;
+        }
+
+        rc = handle_diag320_store_vc(cpu, addr, r1, ra, cs);
+        if (rc == -1) {
+            return;
+        }
+        env->regs[r1 + 1] = rc;
+        break;
      default:
          env->regs[r1 + 1] = DIAG_320_RC_NOT_SUPPORTED;
          break;
Regards,
Jared Rossi

Reply via email to