Re: [RFC PATCH 15/19] tools/vhost-user-rpmb: implement VIRTIO_RPMB_REQ_DATA_WRITE

[Joakim Bech]

On Mon, Sep 28, 2020 at 03:56:19PM +0100, Alex Bennée wrote:
> 
> Joakim Bech <joakim.bech@linaro.org> writes:
> 
> > On Fri, Sep 25, 2020 at 01:51:43PM +0100, Alex Bennée wrote:
> >> With this command we are finally updating data to the backing store
> >> and cycling the write_count and each successful write. We also include
> >> the write count in all response frames as the spec is a little unclear
> >> but the example test code expected it.
> >> 
> >> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
> >> ---
> >>  tools/vhost-user-rpmb/main.c | 111 +++++++++++++++++++++++++++++++++--
> >>  1 file changed, 105 insertions(+), 6 deletions(-)
> >> 
> >> diff --git a/tools/vhost-user-rpmb/main.c b/tools/vhost-user-rpmb/main.c
> >> index 88747c50fa44..a17c3b4bcc4e 100644
> >> --- a/tools/vhost-user-rpmb/main.c
> >> +++ b/tools/vhost-user-rpmb/main.c
> >> @@ -62,6 +62,7 @@ enum {
> >>  #define KiB     (1UL << 10)
> >>  #define MAX_RPMB_SIZE (KiB * 128 * 256)
> >>  #define RPMB_KEY_MAC_SIZE 32
> >> +#define RPMB_BLOCK_SIZE 256
> >>  
> >>  /* RPMB Request Types */
> >>  #define VIRTIO_RPMB_REQ_PROGRAM_KEY        0x0001
> >> @@ -100,7 +101,7 @@ struct virtio_rpmb_config {
> >>  struct virtio_rpmb_frame {
> >>      uint8_t stuff[196];
> >>      uint8_t key_mac[RPMB_KEY_MAC_SIZE];
> >> -    uint8_t data[256];
> >> +    uint8_t data[RPMB_BLOCK_SIZE];
> >>      uint8_t nonce[16];
> >>      /* remaining fields are big-endian */
> >>      uint32_t write_counter;
> >> @@ -124,6 +125,7 @@ typedef struct VuRpmb {
> >>      uint8_t  last_nonce[16];
> >>      uint16_t last_result;
> >>      uint16_t last_reqresp;
> >> +    uint16_t last_address;
> >>      uint32_t write_count;
> >>  } VuRpmb;
> >>  
> >> @@ -239,17 +241,30 @@ vrpmb_set_config(VuDev *dev, const uint8_t *data,
> >>   * which itself uses a 3 clause BSD chunk of code.
> >>   */
> >>  
> >> +static const int rpmb_frame_dlen = (sizeof(struct virtio_rpmb_frame) -
> >> +                                    offsetof(struct virtio_rpmb_frame, 
> >> data));
> >> +
> >>  static void vrpmb_update_mac_in_frame(VuRpmb *r, struct virtio_rpmb_frame 
> >> *frm)
> >>  {
> >>      hmac_sha256_ctx ctx;
> >> -    static const int dlen = (sizeof(struct virtio_rpmb_frame) -
> >> -                             offsetof(struct virtio_rpmb_frame, data));
> >>  
> >>      hmac_sha256_init(&ctx, r->key, RPMB_KEY_MAC_SIZE);
> >> -    hmac_sha256_update(&ctx, frm->data, dlen);
> >> +    hmac_sha256_update(&ctx, frm->data, rpmb_frame_dlen);
> >>      hmac_sha256_final(&ctx, &frm->key_mac[0], 32);
> >>  }
> >>  
> >> +static bool vrpmb_verify_mac_in_frame(VuRpmb *r, struct virtio_rpmb_frame 
> >> *frm)
> >> +{
> >> +    hmac_sha256_ctx ctx;
> >> +    uint8_t calculated_mac[RPMB_KEY_MAC_SIZE];
> >> +
> >> +    hmac_sha256_init(&ctx, r->key, RPMB_KEY_MAC_SIZE);
> >> +    hmac_sha256_update(&ctx, frm->data, rpmb_frame_dlen);
> >> +    hmac_sha256_final(&ctx, calculated_mac, RPMB_KEY_MAC_SIZE);
> >> +
> >> +    return memcmp(calculated_mac, frm->key_mac, RPMB_KEY_MAC_SIZE) == 0;
> >>
> > I'd prefer using a constant time comparison function for this one
> > instead of memcmp (regardless if it's used for simulation or not) to
> > prevent eventual timing attacks.
> 
> Could you recommend such a function for this?
> 
Here is such an implementation:
https://github.com/OP-TEE/optee_os/blob/master/lib/libutils/ext/consttime_memcmp.c

However, cross check the license so it's OK to use. If not there are
similar implementations out there under other licenses.

> >
> >> +}
> >> +
> >>  /*
> >>   * Handlers for individual control messages
> >>   */
> >> @@ -324,6 +339,82 @@ vrpmb_handle_get_write_counter(VuDev *dev, struct 
> >> virtio_rpmb_frame *frame)
> >>      return resp;
> >>  }
> >>  
> >> +/*
> >> + * vrpmb_handle_write:
> >> + *
> >> + * We will report the success/fail on receipt of
> >> + * VIRTIO_RPMB_REQ_RESULT_READ. Returns the number of extra frames
> >> + * processed in the request.
> >> + */
> >> +static int vrpmb_handle_write(VuDev *dev, struct virtio_rpmb_frame *frame)
> >> +{
> >> +    VuRpmb *r = container_of(dev, VuRpmb, dev.parent);
> >> +    int extra_frames = 0;
> >> +    uint16_t block_count = be16toh(frame->block_count);
> >> +    uint32_t write_counter = be32toh(frame->write_counter);
> >> +    size_t offset;
> >> +
> >> +    r->last_reqresp = VIRTIO_RPMB_RESP_DATA_WRITE;
> >> +    r->last_address = be16toh(frame->address);
> >> +    offset =  r->last_address * RPMB_BLOCK_SIZE;
> >> +
> >> +    /*
> >> +     * Run the checks from:
> >> +     * 5.12.6.1.3 Device Requirements: Device Operation: Data Write
> >> +     */
> >> +    if (!r->key) {
> >> +        g_warning("no key programmed");
> >> +        r->last_result = VIRTIO_RPMB_RES_NO_AUTH_KEY;
> >> +    } else if (block_count == 0 ||
> >> +               block_count > r->virtio_config.max_wr_cnt) {
> >> +        r->last_result = VIRTIO_RPMB_RES_GENERAL_FAILURE;
> >> +    } else if (false /* what does an expired write counter mean? */) {
> >>
> > IIRC, the counter has room for a 32-bit value and the counter will never
> > wrap around. So once the counter have reached max for uint32_t, then
> > there is an additional bit set (permanently) in the operation result.
> > I.e., writes are no longer possible once that has happened. That is
> > probably what you're referring to here I suppose.
> 
> That would make sense. I'll see if I can make the spec language a bit
> clearer as well.
> 
> >
> >> +        r->last_result = VIRTIO_RPMB_RES_WRITE_COUNTER_EXPIRED;
> >> +    } else if (offset > (r->virtio_config.capacity * (128 * KiB))) {
> >> +        r->last_result = VIRTIO_RPMB_RES_ADDR_FAILURE;
> >> +    } else if (!vrpmb_verify_mac_in_frame(r, frame)) {
> >> +        r->last_result = VIRTIO_RPMB_RES_AUTH_FAILURE;
> >> +    } else if (write_counter != r->write_count) {
> >> +        r->last_result = VIRTIO_RPMB_RES_COUNT_FAILURE;
> >> +    } else {
> >> +        int i;
> >> +        /* At this point we have a valid authenticated write request
> >> +         * so the counter can incremented and we can attempt to
> >> +         * update the backing device.
> >> +         */
> >> +        r->write_count++;
> >> +        for (i = 0; i < block_count; i++) {
> >> +            void *blk = r->flash_map + offset;
> >> +            g_debug("%s: writing block %d", __func__, i);
> >> +            if (mprotect(blk, RPMB_BLOCK_SIZE, PROT_WRITE) != 0) {
> >> +                r->last_result =  VIRTIO_RPMB_RES_WRITE_FAILURE;
> >> +                break;
> >> +            }
> >> +            memcpy(blk, frame[i].data, RPMB_BLOCK_SIZE);
> >> +            if (msync(blk, RPMB_BLOCK_SIZE, MS_SYNC) != 0) {
> >> +                g_warning("%s: failed to sync update", __func__);
> >> +                r->last_result = VIRTIO_RPMB_RES_WRITE_FAILURE;
> >> +                break;
> >> +            }
> >> +            if (mprotect(blk, RPMB_BLOCK_SIZE, PROT_READ) != 0) {
> >> +                g_warning("%s: failed to re-apply read protection", 
> >> __func__);
> >> +                r->last_result = VIRTIO_RPMB_RES_GENERAL_FAILURE;
> >> +                break;
> >> +            }
> >> +            offset += RPMB_BLOCK_SIZE;
> >> +        }
> >> +        r->last_result = VIRTIO_RPMB_RES_OK;
> >> +        extra_frames = i - 1;
> >> +    }
> >> +
> >> +    g_info("%s: %s (%x, %d extra frames processed), write_count=%d", 
> >> __func__,
> >> +           r->last_result == VIRTIO_RPMB_RES_OK ? "successful":"failed",
> >> +           r->last_result, extra_frames, r->write_count);
> >> +
> >> +    return extra_frames;
> >> +}
> >> +
> >> +
> >>  /*
> >>   * Return the result of the last message. This is only valid if the
> >>   * previous message was VIRTIO_RPMB_REQ_PROGRAM_KEY or
> >> @@ -339,10 +430,14 @@ static struct virtio_rpmb_frame * 
> >> vrpmb_handle_result_read(VuDev *dev)
> >>      g_info("%s: for request:%x result:%x", __func__,
> >>             r->last_reqresp, r->last_result);
> >>  
> >> -    if (r->last_reqresp == VIRTIO_RPMB_RESP_PROGRAM_KEY ||
> >> -        r->last_reqresp == VIRTIO_RPMB_REQ_DATA_WRITE) {
> >> +    if (r->last_reqresp == VIRTIO_RPMB_RESP_PROGRAM_KEY) {
> >>          resp->result = htobe16(r->last_result);
> >>          resp->req_resp = htobe16(r->last_reqresp);
> >> +    } else if (r->last_reqresp == VIRTIO_RPMB_RESP_DATA_WRITE) {
> >> +        resp->result = htobe16(r->last_result);
> >> +        resp->req_resp = htobe16(r->last_reqresp);
> >> +        resp->write_counter = htobe32(r->write_count);
> >> +        resp->address = htobe16(r->last_address);
> >>      } else {
> >>          resp->result = htobe16(VIRTIO_RPMB_RES_GENERAL_FAILURE);
> >>      }
> >> @@ -445,6 +540,10 @@ vrpmb_handle_ctrl(VuDev *dev, int qidx)
> >>                                __func__);
> >>                  }
> >>                  break;
> >> +            case VIRTIO_RPMB_REQ_DATA_WRITE:
> >> +                /* we can have multiple blocks handled */
> >> +                n += vrpmb_handle_write(dev, f);
> >> +                break;
> >>              default:
> >>                  g_debug("un-handled request: %x", f->req_resp);
> >>                  break;
> >> -- 
> >> 2.20.1
> >> 
> 
> Thanks!
> 
> -- 
> Alex Bennée

-- 
Regards,
Joakim