On Jul 19, 2022, at 3:45 PM, Axel Rasmussen axelrasmussen@google.com wrote:
On Tue, Jul 19, 2022 at 3:32 PM Nadav Amit namit@vmware.com wrote:
On Jul 19, 2022, at 12:56 PM, Axel Rasmussen axelrasmussen@google.com wrote:
Historically, it has been shown that intercepting kernel faults with userfaultfd (thereby forcing the kernel to wait for an arbitrary amount of time) can be exploited, or at least can make some kinds of exploits easier. So, in 37cd0575b8 "userfaultfd: add UFFD_USER_MODE_ONLY" we changed things so, in order for kernel faults to be handled by userfaultfd, either the process needs CAP_SYS_PTRACE, or this sysctl must be configured so that any unprivileged user can do it.
In a typical implementation of a hypervisor with live migration (take QEMU/KVM as one such example), we do indeed need to be able to handle kernel faults. But, both options above are less than ideal:
- Toggling the sysctl increases attack surface by allowing any
unprivileged user to do it.
- Granting the live migration process CAP_SYS_PTRACE gives it this
ability, but *also* the ability to "observe and control the execution of another process [...], and examine and change [its] memory and registers" (from ptrace(2)). This isn't something we need or want to be able to do, so granting this permission violates the "principle of least privilege".
This is all a long winded way to say: we want a more fine-grained way to grant access to userfaultfd, without granting other additional permissions at the same time.
To achieve this, add a /dev/userfaultfd misc device. This device provides an alternative to the userfaultfd(2) syscall for the creation of new userfaultfds. The idea is, any userfaultfds created this way will be able to handle kernel faults, without the caller having any special capabilities. Access to this mechanism is instead restricted using e.g. standard filesystem permissions.
Are there any other “devices" that when opened by different processes provide such isolated interfaces in each process? I.e., devices that if you read from them in different processes you get completely unrelated data? (putting aside namespaces).
It all sounds so wrong to me, that I am going to try again to pushback (sorry).
No need to be sorry. :)
From a semantic point of view - userfaultfd is process specific. It is therefore similar to /proc/[pid]/mem (or /proc/[pid]/pagemap and so on).
So why can’t we put it there? I saw that you argued against it in your cover-letter, and I think that your argument is you would need CAP_SYS_PTRACE if you want to access userfaultfd of other processes. But this is EXACTLY the way opening /proc/[pid]/mem is performed - see proc_mem_open().
So instead of having some strange device that behaves differently in the context of each process, you can just have /proc/[pid]/userfaultfd and then use mm_access() to check if you have permissions to access userfaultfd (just like proc_mem_open() does). This would be more intuitive for users as it is similar to other /proc/[pid]/X, and would cover both local and remote use-cases.
Ah, so actually I find this argument much more compelling.
I don't find it persuasive that we should put it in /proc for the purpose of supporting cross-process memory manipulation, because I think the syscall works better for that, and in that case we don't mind depending on CAP_SYS_PTRACE.
But, what you've argued here I do find persuasive. :) You are right, I can't think of any other example of a device node in /dev that works like this, where it is completely independent on a per-process basis. The closest I could come up with was /dev/zero or /dev/null or similar. You won't affect any other process by touching these, but I don't think these are good examples.
I'll send a v5 which does this. I do worry that cross-process support is probably complex to get right, so I might leave that out and only allow a process to open its own device for now.
So I didn’t want to get into it, and I am fine that you leave it out, since such an interface would still enable to support it later.
Anyhow, I do want to clarify a bit about the “cross-process support” userfaultfd situation. Basically, you can already get cross-process support today, by using calling userfaultfd() on the controlled process and calling pidfd_open() from another process. It does work and I do not remember any issues that it introduced (in contrast, for instance, to io-uring, that would break if you use userfaultfd+iouring+fork today).
Thanks for your reconsideration.
Regards, Nadav