From: James Bottomley Date: Fri, 15 May 2020 23:23:09 +0000 Subject: Re: [PATCH v9 0/8] TPM 2.0 trusted keys with attached policy Message-Id: <1589584989.30847.20.camel@HansenPartnership.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit List-Id: References: <20200507231147.27025-1-James.Bottomley@HansenPartnership.com> <23639de13874c00e6bb2b816b4db0b586c9a074c.camel@linux.intel.com> <483c4f1af7be41c8d091b11d4484b606ebd319b7.camel@linux.intel.com> <1589514263.5759.25.camel@HansenPartnership.com> <20200515084702.GA3404@linux.intel.com> <20200515191758.ieojyk5xhsx2hzzd@cantor> <1589571278.3653.22.camel@HansenPartnership.com> <1589573417.3653.28.camel@HansenPartnership.com> <56688CD4-A4A5-4D98-8724-6CBA10C7E1CE@unh.edu> <1589581169.30847.5.camel@HansenPartnership.com> In-Reply-To: <1589581169.30847.5.camel@HansenPartnership.com> To: "Kayaalp, Mehmet" Cc: Jerry Snitselaar , Jarkko Sakkinen , "linux-integrity@vger.kernel.org" , Mimi Zohar , David Woodhouse , "keyrings@vger.kernel.org" , David Howells On Fri, 2020-05-15 at 15:19 -0700, James Bottomley wrote: > On Fri, 2020-05-15 at 21:03 +0000, Kayaalp, Mehmet wrote: > > > On May 15, 2020, at 4:10 PM, James Bottomley > > se > > > npartnership.com> wrote: > > > > > > I think that means the solution is not to run the smoke test > > > under sudo but to do sudo -s and then run it. > > > > > > James > > > > How about "sudo -i": > > > > https://askubuntu.com/questions/376199/sudo-su-vs-sudo-i-vs-sudo- > > bin-bash-when-does-it-matter-which-is-used > > Actually, no that doesn't work either: > > jejb@testdeb> sudo -i keyctl list @s > 1 key in keyring: > 1041514063: ---lswrv 1000 65534 keyring: _uid.1000 > > I suspect this might be a very subtle bug to do with when you get a > new session keyring. It turns out to be incredibly subtle, but I'm not sure if it's a bug or not. the util-linux sudo like commands have special pam profiles /etc/pam.d/su-l /etc/pam.d/runuser-l These special profiles call pam_keyinit.so with flags to install a new session keyring. sudo doesn't have this, so it never, on its own, even when called with -i or -s, installs a new session keyring. This does strike me as a bizarre oversight which leads directly to this weird keyctl pipe behaviour. I'm also not sure the keyctl pipe behaviour is correct: why should keyctl pipe deny access to root to read a key just because it's in a different session keyring? It defintely seems intentional, because if I create a key as a non root user and try to print it by id as root I get EPERM. The weirdest behaviour, though seems to be keyctl: keyctl add ... @u will add to the session keyrings of the actual uid regardless of what session keyring the creator actually has, which is why keyctl add ... @u works under sudo but you get EPERM back trying to pipe it by id. James