On Thu, Oct 10, 2019 at 10:12:51AM +0200, Peter Zijlstra wrote:
> On Wed, Oct 09, 2019 at 04:36:57PM -0400, Joel Fernandes (Google) wrote:
> > In currentl mainline, the degree of access to perf_event_open(2) system
> > call depends on the perf_event_paranoid sysctl. This has a number of
> > limitations:
> >
> > 1. The sysctl is only a single value. Many types of accesses are controlled
> > based on the single value thus making the control very limited and
> > coarse grained.
> > 2. The sysctl is global, so if the sysctl is changed, then that means
> > all processes get access to perf_event_open(2) opening the door to
> > security issues.
> >
> > This patch adds LSM and SELinux access checking which will be used in
> > Android to access perf_event_open(2) for the purposes of attaching BPF
> > programs to tracepoints, perf profiling and other operations from
> > userspace. These operations are intended for production systems.
> >
> > 5 new LSM hooks are added:
> > 1. perf_event_open: This controls access during the perf_event_open(2)
> > syscall itself. The hook is called from all the places that the
> > perf_event_paranoid sysctl is checked to keep it consistent with the
> > systctl. The hook gets passed a 'type' argument which controls CPU,
> > kernel and tracepoint accesses (in this context, CPU, kernel and
> > tracepoint have the same semantics as the perf_event_paranoid sysctl).
> > Additionally, I added an 'open' type which is similar to
> > perf_event_paranoid sysctl == 3 patch carried in Android and several other
> > distros but was rejected in mainline [1] in 2016.
> >
> > 2. perf_event_alloc: This allocates a new security object for the event
> > which stores the current SID within the event. It will be useful when
> > the perf event's FD is passed through IPC to another process which may
> > try to read the FD. Appropriate security checks will limit access.
> >
> > 3. perf_event_free: Called when the event is closed.
> >
> > 4. perf_event_read: Called from the read(2) system call path for the event.
>
> + mmap()
> >
> > 5. perf_event_write: Called from the read(2) system call path for the event.
>
> - read() + ioctl()
Fixed.
>
> fresh from the keyboard.. but maybe consoldate things a little.
Looks great to me, I folded it into the patch. Thanks Peter! Just one comment
on change in existing logic of the code, below:
[snip]
> --- a/arch/x86/events/intel/p4.c
> +++ b/arch/x86/events/intel/p4.c
> @@ -8,7 +8,6 @@
> */
>
> #include <linux/perf_event.h>
> -#include <linux/security.h>
>
> #include <asm/perf_event_p4.h>
> #include <asm/hardirq.h>
> @@ -777,10 +776,7 @@ static int p4_validate_raw_event(struct
> * the user needs special permissions to be able to use it
> */
> if (p4_ht_active() && p4_event_bind_map[v].shared) {
> - if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
> - return -EACCES;
> -
> - v = security_perf_event_open(&event->attr, PERF_SECURITY_CPU);
> + v = perf_allow_cpu(&event->attr);
> if (v)
> return v;
> }
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -56,6 +56,7 @@ struct perf_guest_info_callbacks {
> #include <linux/perf_regs.h>
> #include <linux/cgroup.h>
> #include <linux/refcount.h>
> +#include <linux/security.h>
> #include <asm/local.h>
>
> struct perf_callchain_entry {
> @@ -1244,19 +1245,28 @@ extern int perf_cpu_time_max_percent_han
> int perf_event_max_stack_handler(struct ctl_table *table, int write,
> void __user *buffer, size_t *lenp, loff_t *ppos);
>
> -static inline bool perf_paranoid_tracepoint_raw(void)
> +static inline int perf_allow_kernel(struct perf_event_attr *attr)
> {
> - return sysctl_perf_event_paranoid > -1;
> + if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
> + return -EACCES;
> +
> + return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
> }
>
> -static inline bool perf_paranoid_cpu(void)
> +static inline int perf_allow_cpu(struct perf_event_attr *attr)
> {
> - return sysctl_perf_event_paranoid > 0;
> + if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
> + return -EACCES;
> +
> + return security_perf_event_open(attr, PERF_SECURITY_CPU);
> }
>
> -static inline bool perf_paranoid_kernel(void)
> +static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
> {
> - return sysctl_perf_event_paranoid > 1;
> + if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
> + return -EPERM;
> +
Here the sysctl check of > -1 also is now coupled with a CAP_SYS_ADMIN check.
However..
> + return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);
> }
>
> extern void perf_event_init(void);
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4229,10 +4229,7 @@ find_get_context(struct pmu *pmu, struct
>
> if (!task) {
> /* Must be root to operate on a CPU event: */
> - if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
> - return ERR_PTR(-EACCES);
> -
> - err = security_perf_event_open(&event->attr, PERF_SECURITY_CPU);
> + err = perf_allow_cpu(&event->attr);
> if (err)
> return ERR_PTR(err);
>
> @@ -5862,14 +5859,8 @@ static int perf_mmap(struct file *file,
> lock_limit >>= PAGE_SHIFT;
> locked = atomic64_read(&vma->vm_mm->pinned_vm) + extra;
>
> - if (locked > lock_limit) {
> - if (perf_paranoid_tracepoint_raw() && !capable(CAP_IPC_LOCK)) {
> - ret = -EPERM;
> - goto unlock;
> - }
> -
> - ret = security_perf_event_open(&event->attr,
> - PERF_SECURITY_TRACEPOINT);
> + if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> + ret = perf_allow_tracepoint(&event->attr);
In previous code, this check did not involve a check for CAP_SYS_ADMIN.
I am Ok with adding the CAP_SYS_ADMIN check as well which does make sense to
me for tracepoint access. But it is still a change in the logic so I wanted
to bring it up.
Let me know any other thoughts and then I'll post a new patch.
thanks,
- Joel
[snip]
