-
Notifications
You must be signed in to change notification settings - Fork 476
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Support a not in-place api's for protect/unprtect #693
Conversation
This PR is set to draft state as it is still a work in progress and can potentially be force updated with breaking changes. |
non_null_buf = tagbuf; | ||
non_null_dst_buf = tagbuf; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I do not know why the original code allowed the source buffer to be NULL
. But now we have a case where the source buffer could be NULL
, but the destination buffer might point to valid memory. It appears this line will cause the output to go into the local tagbuf
, even if a valid dst
was provided. Is that the wanted effect?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, if the src buffer is null, it is an indication that this is being used for authentication only and no data will be encrypted.
The api today does not expect the tag to be written to dst at this point, see #714
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'll admit I'm a little confused, but I guess this is an NSS thing more than a libsrtp thing. If there is no source data, what is being authenticated?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
in this case the entire packet was pass as aad data previously with srtp_aes_gcm_nss_set_aad()
7f7a1a0
to
fa42772
Compare
fa42772
to
8b50c31
Compare
A layer to the srtp driver test is added to be able to run the tests both in-place & not-in-place without duplicating the test. The not-in-place functions are currently a minimal implementation using a memcpy.
The protect/unprotect api's can now operate both in-place or not in-place, depending on the requirements of the caller. The length of the out buffer can now be checked to ensure there is sufficient space. Tests are add to verify validation of the output buffer length.
Also update the minimal code example in README.md to reflect the new api.
With the new api the full length of dst is now passed in, if it is too small it should fail.
non_null_buf = tagbuf; | ||
non_null_dst_buf = tagbuf; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'll admit I'm a little confused, but I guess this is an NSS thing more than a libsrtp thing. If there is no source data, what is being authenticated?
memcpy(c->tag, non_null_buf + (*enc_len - c->tag_size), c->tag_size); | ||
*enc_len -= c->tag_size; | ||
memcpy(c->tag, non_null_dst_buf + (*dst_len - c->tag_size), c->tag_size); | ||
*dst_len -= c->tag_size; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This sets the length in the formal param dst_len
to be whatever it was coming in minus the tag size. What if dst_size
was 0 or 1, for example? Since I cannot see any data coming out when the src
buffer is NULL, is this just indicating how much buffer would remain if a tag had been produced? (I understand a tag is somehow produced for something and stored in c->tag
, but I don't see the relationship with that dst
parameter when the src
is NULL.)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
you are right that this is confusing and if we do #714 "consider merging the gcm encrypt and get tag functions of cipher api" once this is merged it will clean all this up alot.
As it is now though, if the call to srtp_aes_gcm_nss_do_crypto() succeeds then dst_len was large enough for the tag and any encrypted data, it then would have been update to be that size. Since currently the tag is not going to be returned from this function it is truncated by tag size. But I agree that a sanity check against tag size would have be good here.
{ | ||
srtp_err_status_t status = | ||
srtp_aes_gcm_nss_do_crypto(cv, false, buf, enc_len); | ||
uint8_t tagbuf[16]; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
While no less confusing to me than the encrypt()
function, it might be good to replicate much of the comments. That said, it would be good if the comments on encrypt()
would speak about which "NULL buffer" (src or dst). The When we get a non-NULL buffer
doesn't indicate which buffer.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I guess that is an old comment, will update.
This is a safety check to ensure we do not wrap dst len when removing tag size. All of this code would get much cleaner if the tag could just be returned instead of cached. see cisco#714
The protect/unprotect api's can now operate both in-place or not in-place, depending on the requirements of the caller.
The length of the out buffer can now be checked to ensure there is sufficient space.
This addresses #322 "provide an new version of protect function that does not corrupt memory" and #569 "Please add a "non inplace" protect and unprotect API"
srtp_driver has been updated so all existing test can be run either in-place or not in-place.