[pacman-dev] [PATCH] Check for all return values of _alpm_key_in_keychain
This fixes a bug I encountered with a GPG keyring where the key id used to locate a key in the keyring was ambiguous within my keychain. This commit ensures that all valid return values are checked to catch this and related error cases rather than incorrectly taking an error case to mean the key was found, since this is rarely going to be the case. --- lib/libalpm/be_package.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/lib/libalpm/be_package.c b/lib/libalpm/be_package.c index 7e8b7920..1891fa5a 100644 --- a/lib/libalpm/be_package.c +++ b/lib/libalpm/be_package.c @@ -754,10 +754,21 @@ int SYMEXPORT alpm_pkg_load(alpm_handle_t *handle, const char *filename, int ful alpm_list_t *k; for(k = keys; k; k = k->next) { char *key = k->data; - if(_alpm_key_in_keychain(handle, key) == 0) { - if(_alpm_key_import(handle, key) == -1) { + switch(_alpm_key_in_keychain(handle, key)) { + case 1: + /* key is known; proceed */ + break; + case 0: + /* key is unknown; attempt to import */ + if(_alpm_key_import(handle, key) == -1) { + fail = 1; + } + break; + case -1: + /* error finding key in keychain */ + default: fail = 1; - } + break; } } FREELIST(keys); -- 2.12.2
On 21/04/17 14:07, David Phillips wrote:
This fixes a bug I encountered with a GPG keyring where the key id used to locate a key in the keyring was ambiguous within my keychain.
This commit ensures that all valid return values are checked to catch this and related error cases rather than incorrectly taking an error case to mean the key was found, since this is rarely going to be the case.
Can you tell me what happens both before and after this patch with ambiguous key IDs. Also, what happens if you only have one of the keys in the keyring and want to match the other? Perhaps we also need to look at matching the whole key fingerprint... Cheers, Allan
Hi, It now looks like this problem's root cause was a keyring that got corrupted while I was debugging some problems with refreshing pacman's keyring. That is to say, it looks like I do not have a genuine "in the wild" key collision. It looks like my own public key became copied to both me and one of the TUs in the keyring. With that said though, I suppose it's still not impossible for long key IDs to genuinely collide. On Mon, Apr 24, 2017 at 11:06:05AM +1000, Allan McRae wrote:
Can you tell me what happens both before and after this patch with ambiguous key IDs.
Please find attached two log files before and after. As you can see, pacman used to blindly run ahead, trying to check the signature even though it could not get the key from the keychain. A quick look at a backtrace (not attached) shows this to be a segmentation fault after control is passed to libgpgme.
Also, what happens if you only have one of the keys in the keyring and want to match the other?
I cannot test this, since it turns out to be a corrupt keyring where the key in question looks to be attached to two people and is exactly the same.
Perhaps we also need to look at matching the whole key fingerprint...
If the issue with collisions comes up again and doesn't end up being put down to a corrupt keyring, this would be a good idea. For now, it seems like seeing this in the wild too narrow a possibility, but there is still a clear hiccup with tht way pacman behaves before this patch. Thanks David
On 04/21/17 at 04:07pm, David Phillips wrote:
This fixes a bug I encountered with a GPG keyring where the key id used to locate a key in the keyring was ambiguous within my keychain.
This commit ensures that all valid return values are checked to catch this and related error cases rather than incorrectly taking an error case to mean the key was found, since this is rarely going to be the case. --- lib/libalpm/be_package.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-)
diff --git a/lib/libalpm/be_package.c b/lib/libalpm/be_package.c index 7e8b7920..1891fa5a 100644 --- a/lib/libalpm/be_package.c +++ b/lib/libalpm/be_package.c @@ -754,10 +754,21 @@ int SYMEXPORT alpm_pkg_load(alpm_handle_t *handle, const char *filename, int ful alpm_list_t *k; for(k = keys; k; k = k->next) { char *key = k->data; - if(_alpm_key_in_keychain(handle, key) == 0) { - if(_alpm_key_import(handle, key) == -1) { + switch(_alpm_key_in_keychain(handle, key)) { + case 1: + /* key is known; proceed */ + break; + case 0: + /* key is unknown; attempt to import */ + if(_alpm_key_import(handle, key) == -1) { + fail = 1; + } + break; + case -1: + /* error finding key in keychain */ + default: fail = 1;
Just below this, an error message is printed if fail is true, saying that the key is missing. If we got a generic code, that may not be the case, and is the exact opposite of the problem that prompted patch. I'm wondering if this is necessary at all, though. This bit of code is specifically just for importing missing keys. Any other significant errors should be caught during the actual validation attempt, where we already provide the actual gpg error message. If the gpgme segmentation fault is no longer an issue, do we gain anything other than bailing out slightly sooner by adding the extra check and complexity for printing an appropriate error message here?
- } + break; } } FREELIST(keys); -- 2.12.2
participants (3)
-
Allan McRae
-
Andrew Gregory
-
David Phillips