> From: Al Viro [mailto:viro@xxxxxxxxxxxxxxxx] On Behalf Of Al Viro > Sent: Monday, November 16, 2020 7:09 PM > On Mon, Nov 16, 2020 at 09:37:32AM -0800, Linus Torvalds wrote: > > On Mon, Nov 16, 2020 at 8:47 AM Mimi Zohar <zohar@xxxxxxxxxxxxx> > wrote: > > > > > > This discussion seems to be going down the path of requiring an IMA > > > filesystem hook for reading the file, again. That solution was > > > rejected, not by me. What is new this time? > > > > You can't read a non-read-opened file. Not even IMA can. > > > > So don't do that then. > > > > IMA is doing something wrong. Why would you ever read a file that can't > be read? > > > > Fix whatever "open" function instead of trying to work around the fact > > that you opened it wrong. > > IMA pulls that crap on _every_ open(2), including O_WRONLY. As far as I'm > concerned, the only sane answer is not enabling that thing on your builds; > they are deeply special and I hadn't been able to reason with them no > matter how much I tried ;-/ A file-based protection mechanism against offline attacks would require to verify the current HMAC also before writing and to update the HMAC after the write. One of the reasons why dentry_open() cannot be used and IMA switches to the old method of changing the mode of the current file descriptor is that the current process does not have enough privileges to do the operation. If we find a way to read the file that always works, without reducing the security, the old method can be removed. Roberto HUAWEI TECHNOLOGIES Duesseldorf GmbH, HRB 56063 Managing Director: Li Peng, Li Jian, Shi Yanli
