zfs_dbgmsg doesn't print valid pointers

[shartse]

System information

Type	Version/Name
Distribution Name	Ubuntu
Distribution Version	18.04
Linux Kernel
Architecture	x86_64
ZFS Version	0.8.0-rc3
SPL Version

Problem

There are several places where we use zfs_dbgmsg and %p to print pointers (see metaslab.c:2371:metaslab_condense() or zio.c:1887:zio_deadman_impl()). In the Linux kernel, these values "are hashed to give a unique identifier without leaking kernel addresses to user
space. On 64 bit machines the first 32 bits are zeroed." (https://www.kernel.org/doc/Documentation/printk-formats.txt)

This means the pointers aren't very useful for debugging crash dumps since you can't find the location of the struct they refer to.

It is possible to print the actual pointers using %pK ("For printing kernel pointers which should be hidden from unprivileged users") or %px ("For printing pointers when you really want to print the address").

Note that on Linux, /proc/spl/kstat/zfs/dbgmsg is world-readable (on Illumos it's root only).

Would restricting the permissions for /proc/spl/kstat/zfs/dbgmsg and then printing the actual pointers be an appealing option for people?

And if we don't want to print the actual pointers is there any benefit to continue printing the scrambled version of them?

Describe how to reproduce the problem

You can see this in the metaslab.c:2371:metaslab_condense() entries in zfsdbg_msg.

[ahrens]

@behlendorf would be good to get your opinion on this.

[behlendorf]

Would restricting the permissions for /proc/spl/kstat/zfs/dbgmsg and then printing the actual pointers be an appealing option for people?

Yes that makes sense, I definitely think we should start by updating /proc/spl/kstat/zfs/dbgmsg to only be accessible by root. At which point I think it would be reasonable to start printing the real pointers.

[shartse]

Sounds good. I'll keep investigating this.