On 1/30/20 8:13 AM, Christian Brauner wrote: > On Thu, Jan 30, 2020 at 07:11:08AM -0700, Jens Axboe wrote: >> On 1/30/20 3:26 AM, Christian Brauner wrote: >>> On Thu, Jan 30, 2020 at 11:11:58AM +0100, Jann Horn wrote: >>>> On Thu, Jan 30, 2020 at 2:08 AM Jens Axboe <axboe@xxxxxxxxx> wrote: >>>>> On 1/29/20 10:34 AM, Jens Axboe wrote: >>>>>> On 1/29/20 7:59 AM, Jann Horn wrote: >>>>>>> On Tue, Jan 28, 2020 at 8:42 PM Jens Axboe <axboe@xxxxxxxxx> wrote: >>>>>>>> On 1/28/20 11:04 AM, Jens Axboe wrote: >>>>>>>>> On 1/28/20 10:19 AM, Jens Axboe wrote: >>>>>>> [...] >>>>>>>>>> #1 adds support for registering the personality of the invoking task, >>>>>>>>>> and #2 adds support for IORING_OP_USE_CREDS. Right now it's limited to >>>>>>>>>> just having one link, it doesn't support a chain of them. >>>>>>> [...] >>>>>>>> I didn't like it becoming a bit too complicated, both in terms of >>>>>>>> implementation and use. And the fact that we'd have to jump through >>>>>>>> hoops to make this work for a full chain. >>>>>>>> >>>>>>>> So I punted and just added sqe->personality and IOSQE_PERSONALITY. >>>>>>>> This makes it way easier to use. Same branch: >>>>>>>> >>>>>>>> https://git.kernel.dk/cgit/linux-block/log/?h=for-5.6/io_uring-vfs-creds >>>>>>>> >>>>>>>> I'd feel much better with this variant for 5.6. >>>>>>> >>>>>>> Some general feedback from an inspectability/debuggability perspective: >>>>>>> >>>>>>> At some point, it might be nice if you could add a .show_fdinfo >>>>>>> handler to the io_uring_fops that makes it possible to get a rough >>>>>>> overview over the state of the uring by reading /proc/$pid/fdinfo/$fd, >>>>>>> just like e.g. eventfd (see eventfd_show_fdinfo()). It might be >>>>>>> helpful for debugging to be able to see information about the fixed >>>>>>> files and buffers that have been registered. Same for the >>>>>>> personalities; that information might also be useful when someone is >>>>>>> trying to figure out what privileges a running process actually has. >>>>>> >>>>>> Agree, that would be a very useful addition. I'll take a look at it. >>>>> >>>>> Jann, how much info are you looking for? Here's a rough start, just >>>>> shows the number of registered files and buffers, and lists the >>>>> personalities registered. We could also dump the buffer info for >>>>> each of them, and ditto for the files. Not sure how much verbosity >>>>> is acceptable in fdinfo? >>>> >>>> At the moment, I personally am just interested in this from the >>>> perspective of being able to audit the state of personalities, to make >>>> important information about the security state of processes visible. >>>> >>>> Good point about verbosity in fdinfo - I'm not sure about that myself either. > > Afaik, there's no rule here. I would expect that it shouldn't exceed > 4096kb just because that is the limit that seems to be enforced for > writes to proc files atm; other than that it should be the wild west. > The fdinfo files are mostly interesting for anon_inode fds imho and the > ones that come to mind right now simply don't have a lot of information > to provide: > > eventfd > timerfd > seccomp_notify_fd > > Potentially, the mount fds from David could be extended in the future. 4MB is huge, I'd not get anywhere near that. So I'd say the current format is probably fine. I honed it a little bit to be prettier, looks good to me. I'll send it out for review. > (Side note: One thing that comes to mind is that we should probably > enforce^Wdocument that all fdinfo files use CamelCase?) Fine with me, seems to already be the norm, would be nice to have it documented. -- Jens Axboe
