On Fri, May 18, 2018 at 04:27:14PM +0200, Florian Weimer wrote: > This test program: > > #include <errno.h> > #include <stdio.h> > #include <stdlib.h> > #include <unistd.h> > #include <sys/syscall.h> > #include <err.h> > > /* Return the value of the AMR register. */ > static inline unsigned long int > pkey_read (void) > { > unsigned long int result; > __asm__ volatile ("mfspr %0, 13" : "=r" (result)); > return result; > } > > /* Overwrite the AMR register with VALUE. */ > static inline void > pkey_write (unsigned long int value) > { > __asm__ volatile ("mtspr 13, %0" : : "r" (value)); > } > > int > main (int argc, char **argv) > { > printf ("AMR (PID %d): 0x%016lx\n", (int) getpid (), pkey_read()); > if (argc > 1) > { > int key = syscall (__NR_pkey_alloc, 0, 0); > if (key < 0) > err (1, "pkey_alloc"); > printf ("Allocated key (PID %d): %d\n", (int) getpid (), key); > return 0; > } > > pid_t pid = fork (); > if (pid == 0) > { > execl ("/proc/self/exe", argv[0], "subprocess", NULL); > _exit (1); > } > if (pid < 0) > err (1, "fork"); > int status; > if (waitpid (pid, &status, 0) < 0) > err (1, "waitpid"); > > int key = syscall (__NR_pkey_alloc, 0, 0); > if (key < 0) > err (1, "pkey_alloc"); > printf ("Allocated key (PID %d): %d\n", (int) getpid (), key); > > unsigned long int amr = -1; > printf ("Setting AMR: 0x%016lx\n", amr); > pkey_write (amr); > printf ("New AMR value (PID %d, before execl): 0x%016lx\n", > (int) getpid (), pkey_read()); > execl ("/proc/self/exe", argv[0], "subprocess", NULL); > err (1, "exec"); > return 1; > } > > shows that the AMR register value is not reset on execve: > > AMR (PID 112291): 0x0000000000000000 > AMR (PID 112292): 0x0000000000000000 > Allocated key (PID 112292): 2 > Allocated key (PID 112291): 2 > Setting AMR: 0xffffffffffffffff > New AMR value (PID 112291, before execl): 0x0c00000000000000 > AMR (PID 112291): 0x0c00000000000000 > Allocated key (PID 112291): 2 > > I think this is a real bug and needs to be fixed even if the > defaults are kept as-is (see the other thread).
The issue you may be talking about here is that -- "when you set the AMR register to 0xffffffffffffffff, it just sets it to 0x0c00000000000000." To me it looks like, exec/fork are not related to the issue. Or are they also somehow connected to the issue? The reason the AMR register does not get set to 0xffffffffffffffff, is because none of those keys; except key 2, are active. So it ignores all other bits and just sets the bits corresponding to key 2. However the fundamental issue is still the same, as mentioned in the other thread. "Should the permissions on a key be allowed to be changed, if the key is not allocated in the first place?". my answer is NO. Lets debate :) RP