Is there any safe strcmp?

In some cases std::strncmp can solve your problem:

int strncmp ( const char * str1, const char * str2, size_t num ); 

It compares up to num characters of the C string str1 to those of the C string str2.

---

Also, take a look, what the US DHS National Cyber Security Division recommends on this matter:

Ensure that strings are null terminated before passing into strcmp. This can be enforced by always placing a \0 in the last allocated byte of the buffer.

char str1[] ="something";
char str2[] = "another thing";
/* In this case we know strings are null terminated. Pretend we don't. */
str1[sizeof(str1)-1] = '\0';
str2[sizeof(str2)-1] = '\0';
/* Now the following is safe. */
if (strcmp(str1, str2)) { /* do something */ } else { /* do something else */ }

If you are passing strings to strcmp() that are not null terminated you have already lost. The fact that you have a string that is not null terminated (but should be) indicates that you have deeper issues in your code. You cannot change strcmp() to safely deal with this problem.

You should be writing your code so that can never happen. Start by using the string class. At the boundaries where you take data into your code you need to make sure you deal with the exceptional cases; if you get too much data you need to Do The Right Thing. That does not involve running off the end of your buffer. If you must perform I/O into a C style buffer, use functions where you specify the length of the buffer and detect and deal with cases where the buffer is not large enough at that point.

No, there's no (standard) way to tell whether a char * actually points to valid memory.

In your situation, it is better to use std::string rather than char *s for all your strings, along with the overloaded == operator. If you do this, the compiler would enforce type safety.

EDIT: As per the comments below if you find yourself in a situation where you sometimes pass char *s that may or may not be valid strings to functions that expect null-terminated strings then something is fundamentally wrong with your approach, so basically @janm's answer below.