Update Information from Qualys

[exim-website.git] / templates / static / doc / security / CVE-2020-qualys / patches3.txt

diff --git a/templates/static/doc/security/CVE-2020-qualys/patches3.txt b/templates/static/doc/security/CVE-2020-qualys/patches3.txt
new file mode 100644 (file)
index 0000000..dab93f7
--- /dev/null
+++ b/templates/static/doc/security/CVE-2020-qualys/patches3.txt
@@ -0,0 +1,69 @@
+Hi all,
+
+We noticed the branch "hs/qualys-2020" and reviewed the patches. We just
+have three comments and one question:
+
+========================================================================
+commit 4f0ac4ad70d38a13fb3f248c3ae2b66b0e1fe7d3
+
+    Safeguard against relative names for msglog files.
+
+Is there a reason for the extra "&& (p == filename || *(p-1) == '/')"?
+
+Because of this extra condition, our exploit against CVE-2019-15846 for
+example would work despite this patch: we overwrite the message id (the
+argument for deliver_message()) with "/../../../../../../../etc/passwd"
+but depending on the alignment of our memory corruption, the beginning
+of id can be "./../../" (for example) and:
+
+- (p = Ustrstr(filename, US"/../")) is true;
+
+- (p == filename) is false, because filename always starts with
+  spool_directory, an absolute path;
+
+- (*(p-1) == '/') is false, because the character before the first
+  "/../" is '.';
+
+as a result, log_write(LOG_PANIC_DIE) is not called and we can modify
+/etc/passwd.
+
+========================================================================
+commit e5cb5e615a63a4c97d3e2e88903eaaadfb254bcb
+
+    Check overrun rcpt_count integer
+
+      if (rcpt_count+1 < 0
+
+Actually, such a signed integer overflow is undefined behavior in C and
+an optimizing compiler may therefore remove this check completely (some
+do). The check should be made against INT_MAX (before the int overflow)
+and should be made earlier, when rcpt_count is incremented (at the very
+beginning of the case RCPT_CMD): otherwise, rcpt_count can be increased
+without going through the recipients_max check (e.g., via the breaks in
+"if (sender_address == NULL)" and "if (!recipient_domain)").
+
+========================================================================
+commit 54895bc3ffdf5ecebcbafb2e6041fa52d6f5e5fb
+
+    smtp_out: Leave a clean input buffer, even in case of read error
+
++      inblock->ptr = ptr;
+       return -1;
+
+There is another return -1 in this function, should it also get fixed?
+Or maybe a common error codepath should be added instead?
+
+========================================================================
+
+We saw 20-patches2.txt.gpg in cve-2020-qualys/ and just in case: did you
+also receive/read patches1.txt?
+
+========================================================================
+
+Thank you very much for all your work on Exim! We are at your disposal
+for questions, comments, and further discussions.
+
+With best regards,
+
+-- 
+the Qualys Security Advisory team