rpctest: filter tar members on extraction; document two API contracts

include/dogecoin/bip39.h

@@ -173,6 +173,11 @@ int verify_mnemonic_sentence(const char* mnemonic, const char* wordlist[], const
 /* size of the generated mnemonic in bytes (output) */
 /* generated mnemonic (output) */
 /* returns 0 (success), -1 (fail) */
+/* entropy_out receives the entropy as a hex string and must have room for
+   MAX_ENTROPY_STRING_SIZE bytes -- the type is HEX_ENTROPY. The signature says
+   char*, so the requirement is not visible at the call site; entropy_size is
+   validated to 128..256 bits, which bounds the write to 64 hex characters plus
+   a terminator. */
 LIBDOGECOIN_API int dogecoin_generate_mnemonic (const ENTROPY_SIZE entropy_size, const char* language, const char* space, const char* entropy, const char* filename, char* entropy_out, size_t* size, char* words);
 
 /* Verifies the mnemonic phrase */

include/dogecoin/utils.h

@@ -73,6 +73,10 @@ LIBDOGECOIN_API char* to_string(uint8_t* x);
 LIBDOGECOIN_API char* hash_to_string(uint8_t* x);
 LIBDOGECOIN_API uint8_t* hash_to_bytes(uint8_t* x);
 LIBDOGECOIN_API void* safe_malloc(size_t size);
+/* Fast, NON-CRYPTOGRAPHIC randomness, seeded from the wall clock. For values
+   that only need to be unlikely to repeat -- P2P nonces are the in-tree use.
+   Never for keys, seeds, entropy or anything an attacker benefits from
+   predicting: use dogecoin_random_bytes(), which fails closed. */
 LIBDOGECOIN_API void dogecoin_cheap_random_bytes(uint8_t* buf, size_t len);
 LIBDOGECOIN_API void dogecoin_get_default_datadir(cstring* path_out);
 LIBDOGECOIN_API void dogecoin_file_commit(FILE* file);

rpctest/fetch.py

@@ -91,7 +91,33 @@
     zipfile.extractall(os.getcwd())
 else:
     with tarfile.open(fileobj=BytesIO(req.content), mode='r:gz') as tar:
-        tar.extractall(os.getcwd())
+        # Filter members rather than trusting the archive. Without this a
+        # crafted tarball can write outside the destination via '../' members,
+        # absolute paths, or symlinks pointing out of the tree. The checksum
+        # check above narrows who can supply such an archive, but it does not
+        # make extraction safe on its own -- and zipfile.extractall (the branch
+        # above) already sanitises member paths, so only tar was exposed.
+        if hasattr(tarfile, 'data_filter'):
+            tar.extractall(os.getcwd(), filter='data')
+        else:
+            # PEP 706 filters land in 3.12 and were backported to security
+            # releases of 3.8+. Where they are unavailable, check by hand
+            # rather than silently extracting unfiltered.
+            dest = os.path.abspath(os.getcwd())
+            for member in tar.getmembers():
+                target = os.path.abspath(os.path.join(dest, member.name))
+                if not (target == dest or target.startswith(dest + os.sep)):
+                    print("\033[31m> refusing tar member outside destination: "
+                          + member.name + "\033[0m")
+                    exit(1)
+                if member.issym() or member.islnk():
+                    link = os.path.abspath(os.path.join(
+                        os.path.dirname(target), member.linkname))
+                    if not (link == dest or link.startswith(dest + os.sep)):
+                        print("\033[31m> refusing tar link outside destination: "
+                              + member.name + "\033[0m")
+                        exit(1)
+            tar.extractall(dest)
         tar.close()
 
 deps_path = ["dogecoind"]