Re: [v6 PATCH 07/21] x86/insn-eval: Add utility function to get segment descriptor

[Ricardo Neri]

On Wed, 2017-04-19 at 12:26 +0200, Borislav Petkov wrote:
> On Tue, Mar 07, 2017 at 04:32:40PM -0800, Ricardo Neri wrote:
> > The segment descriptor contains information that is relevant to how linear
> > address need to be computed. It contains the default size of addresses as
> > well as the base address of the segment. Thus, given a segment selector,
> > we ought look at segment descriptor to correctly calculate the linear
> > address.
> > 
> > In protected mode, the segment selector might indicate a segment
> > descriptor from either the global descriptor table or a local descriptor
> > table. Both cases are considered in this function.
> > 
> > This function is the initial implementation for subsequent functions that
> > will obtain the aforementioned attributes of the segment descriptor.
> > 
> > Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> > Cc: Adam Buchbinder <adam.buchbinder@xxxxxxxxx>
> > Cc: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> > Cc: Lorenzo Stoakes <lstoakes@xxxxxxxxx>
> > Cc: Qiaowei Ren <qiaowei.ren@xxxxxxxxx>
> > Cc: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> > Cc: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > Cc: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> > Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> > Cc: Thomas Garnier <thgarnie@xxxxxxxxxx>
> > Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > Cc: Borislav Petkov <bp@xxxxxxx>
> > Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> > Cc: Ravi V. Shankar <ravi.v.shankar@xxxxxxxxx>
> > Cc: x86@xxxxxxxxxx
> > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
> > ---
> >  arch/x86/lib/insn-eval.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 61 insertions(+)
> > 
> > diff --git a/arch/x86/lib/insn-eval.c b/arch/x86/lib/insn-eval.c
> > index 8d45df8..8608adf 100644
> > --- a/arch/x86/lib/insn-eval.c
> > +++ b/arch/x86/lib/insn-eval.c
> > @@ -5,9 +5,13 @@
> >   */
> >  #include <linux/kernel.h>
> >  #include <linux/string.h>
> > +#include <asm/desc_defs.h>
> > +#include <asm/desc.h>
> >  #include <asm/inat.h>
> >  #include <asm/insn.h>
> >  #include <asm/insn-eval.h>
> > +#include <asm/ldt.h>
> > +#include <linux/mmu_context.h>
> >  #include <asm/vm86.h>
> >  
> >  enum reg_type {
> > @@ -294,6 +298,63 @@ static int get_reg_offset(struct insn *insn, struct pt_regs *regs,
> >  }
> >  
> >  /**
> > + * get_desc() - Obtain address of segment descriptor
> > + * @seg:	Segment selector
> 
> Maybe that should be
> 
> @sel
> 
> if it is a sel-ector. :)

It makes sense. I will rename it.
> 
> And using "sel" makes more sense then when you look at:
> 
> 	desc_base = sel & ~(SEGMENT_RPL_MASK | SEGMENT_TI_MASK);
> 
> for example:
> 
> > + * @desc:	Pointer to the selected segment descriptor
> > + *
> > + * Given a segment selector, obtain a memory pointer to the segment
> 
> s/memory //

Will update it.
> 
> > + * descriptor. Both global and local descriptor tables are supported.
> > + * desc will contain the address of the descriptor.
> > + *
> > + * Return: 0 if success, -EINVAL if failure
> 
> Why isn't this function returning the pointer or NULL on error? Maybe
> the later patches have an answer and I'll discover it if I continue
> reviewing :)

After revisiting the code, I don't see why the function cannot return
NULL.
> 
> > + */
> > +static int get_desc(unsigned short seg, struct desc_struct **desc)
> > +{
> > +	struct desc_ptr gdt_desc = {0, 0};
> > +	unsigned long desc_base;
> > +
> > +	if (!desc)
> > +		return -EINVAL;
> > +
> > +	desc_base = seg & ~(SEGMENT_RPL_MASK | SEGMENT_TI_MASK);
> 
> That looks useless as you're doing it below again.

Yes, it is useless. Please see my comment below.
> 
> > +
> > +#ifdef CONFIG_MODIFY_LDT_SYSCALL
> > +	if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
> > +		seg >>= 3;
> > +
> > +		mutex_lock(&current->active_mm->context.lock);
> > +		if (unlikely(!current->active_mm->context.ldt ||
> 
> Is that really a fast path to complicate the if-test with an unlikely()?
> If not, you don't really need it.

I will remove it.
> 
> > +			     seg >= current->active_mm->context.ldt->size)) {
> 
> ldt->size is the size of the descriptor table but you've shifted seg by
> 3. That selector index is shifted by 3 (to the left) to form an offset
> into the descriptor table because the entries there are 8 bytes.

I double-checked the ldt code and it seems to me that size refers to the
number of entries in the table; it is always multiplied by
LDT_ENTRY_SIZE [1], [2]. Am I missing something?

> 
> So I *think* you wanna use the "useless" desc_base above... :)
> 
> > +			*desc = NULL;
> > +			mutex_unlock(&current->active_mm->context.lock);
> > +			return -EINVAL;
> > +		}
> > +
> > +		*desc = &current->active_mm->context.ldt->entries[seg];
> 
> ... and seg here as it is an index into the table.
> 
> > +		mutex_unlock(&current->active_mm->context.lock);
> > +		return 0;
> > +	}
> > +#endif
> > +	native_store_gdt(&gdt_desc);
> > +
> > +	/*
> > +	 * Bits [15:3] of the segment selector contain the index. Such
> > +	 * index needs to be multiplied by 8.
> 
> ... because <insert reason I typed in above>.

I will elaborate on the reason for this.

Thanks and BR,
Ricardo

[1].
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/ldt.c?id=refs/tags/v4.11-rc8#n260
[2].
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/ldt.c?id=refs/tags/v4.11-rc8#n50