Stack buffer overflow in tcmu_cdb_print_info

[chrisccoulson]

tcmu_cdb_print_info has a fixed size 193 bytes scratch buffer on the stack, defined here:

#define CDB_TO_BUF_SIZE(bytes) ((bytes) * 3 + 1)
#define CDB_FIX_BYTES 64 /* 64 bytes for default */
#define CDB_FIX_SIZE CDB_TO_BUF_SIZE(CDB_FIX_BYTES)
void tcmu_cdb_print_info(struct tcmu_device *dev,
                         const struct tcmulib_cmd *cmd,
                         const char *info)
{
        int i, n, bytes;
        char fix[CDB_FIX_SIZE], *buf;

        buf = fix;

Further down, it uses sprintf to write some bytes in 2-digit hex format to buf, with each byte occupying 3 bytes in buf:

        for (i = 0, n = 0; i < bytes; i++) {
                n += sprintf(buf + n, "%x ", cmd->cdb[i]);
        }

To avoid overflowing the stack buffer, there is a check earlier in the function which performs a heap allocation if bytes is too large:

        if (bytes > CDB_FIX_SIZE) {
                buf = malloc(CDB_TO_BUF_SIZE(bytes));
                if (!buf) {
                        tcmu_dev_err(dev, "out of memory\n");
                        return;
                }
        }

But this check looks wrong. If bytes > 64, then the previously mentioned for loop will overflow the 193 byte stack buffer. I think it should be:

        if (bytes > CDB_FIX_BYTES) {

Correcting this isn't sufficient on its own though, as this check doesn't take in to account that an additional string (info) is written to buf and can also result in the stack buffer overflowing:

        if (info)
                n += sprintf(buf + n, "%s", info);

I would normally report this kind of issue privately, but it looks like triggering this code relies on changing the log level and AFAICT there isn't really a mechanism for an attacker to exploit this anyway.

[mikechristie]

Thanks! Agree on both cases. I made a PR that should hopefully address all your issues here:

#615