Re: [PATCH 3/4] KASLR: Handle memory limit specified by memmap and mem option

[Baoquan He]

On 04/18/17 at 01:36pm, Kees Cook wrote:
> On Mon, Apr 17, 2017 at 6:34 AM, Baoquan He <bhe@xxxxxxxxxx> wrote:
> > Option mem= will limit the max address system can use. Any memory
> > region above the limit will be removed. And memmap=nn[KMG] which
> > has no offset specified has the same behaviour as mem=. KASLR need
> > consider this when choose the random position for decompressing
> > kernel. Do it in this patch.
> >
> > Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
> > Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> > Cc: x86@xxxxxxxxxx
> > Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> > Cc: Yinghai Lu <yinghai@xxxxxxxxxx>
> > Cc: Borislav Petkov <bp@xxxxxxx>
> > ---
> >  arch/x86/boot/compressed/kaslr.c | 53 +++++++++++++++++++++++++++++++---------
> >  1 file changed, 42 insertions(+), 11 deletions(-)
> >
> > diff --git a/arch/x86/boot/compressed/kaslr.c b/arch/x86/boot/compressed/kaslr.c
> > index 36ab429..5361abd 100644
> > --- a/arch/x86/boot/compressed/kaslr.c
> > +++ b/arch/x86/boot/compressed/kaslr.c
> > @@ -67,6 +67,10 @@ int mem_avoid_memmap_index;
> >  extern unsigned long get_cmd_line_ptr(void);
> >
> >
> > +/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
> > +unsigned long long mem_limit = ULLONG_MAX;
> 
> I would either make this 0 or ULLONG_MAX - 1 (see below).

I prefer "ULLONG_MAX-1". Please see reason below

> 
> > +
> > +
> >  enum mem_avoid_index {
> >         MEM_AVOID_ZO_RANGE = 0,
> >         MEM_AVOID_INITRD,
> > @@ -117,15 +121,18 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
> >                 return -EINVAL;
> >
> >         switch (*p) {
> > -       case '@':
> > -               /* Skip this region, usable */
> > -               *size = 0;
> > -               *start = 0;
> >         case '#':
> >         case '$':
> >         case '!':
> >                 *start = memparse(p + 1, &p);
> >                 return 0;
> > +       case '@':
> > +               /* Skip this region, usable */
> > +               *size = 0;
> 
> Now it looks like we're intentionally falling through. A comment
> should be included to indicate it.

Yes, as I replied in patch 2/4, it's intended. Will add comment to
indicate.

> 
> > +       default:
> > +               /* Avoid the region which is above the amount limit */
> > +               *start = 0;
> > +               return 0;
> >         }
> >
> >         return -EINVAL;
> > @@ -151,9 +158,14 @@ static void mem_avoid_memmap(char *str)
> >                 if (rc < 0)
> >                         break;
> >                 str = k;
> > -               /* A usable region that should not be skipped */
> > -               if (size == 0)
> > +
> > +               if (start == 0) {
> > +                       /* Store the specified memory limit if size > 0 */
> > +                       if (size > 0)
> > +                               mem_limit = size;
> > +
> >                         continue;
> > +               }
> >
> >                 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
> >                 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
> > @@ -173,6 +185,7 @@ static int handle_mem_memmap(void)
> >         char tmp_cmdline[COMMAND_LINE_SIZE];
> >         size_t len = strlen((char *)args);
> >         char *param, *val;
> > +       u64 mem_size;
> >
> >         len = (len >= COMMAND_LINE_SIZE) ? COMMAND_LINE_SIZE - 1 : len;
> >         memcpy(tmp_cmdline, args, len);
> > @@ -195,8 +208,18 @@ static int handle_mem_memmap(void)
> >                         return -1;
> >                 }
> >
> > -               if (!strcmp(param, "memmap"))
> > +               if (!strcmp(param, "memmap")) {
> >                         mem_avoid_memmap(val);
> > +               } else if (!strcmp(param, "mem")) {
> > +                       char *p = val;
> > +
> > +                       if (!strcmp(p, "nopentium"))
> > +                               continue;
> > +                       mem_size = memparse(p, &p);
> > +                       if (mem_size == 0)
> > +                               return -EINVAL;
> > +                       mem_limit = mem_size;
> > +               }
> >         }
> >
> >         return 0;
> > @@ -432,7 +455,8 @@ static void process_e820_entry(struct e820entry *entry,
> >  {
> >         struct mem_vector region, overlap;
> >         struct slot_area slot_area;
> > -       unsigned long start_orig;
> > +       unsigned long start_orig, end;
> > +       struct e820entry cur_entry;
> >
> >         /* Skip non-RAM entries. */
> >         if (entry->type != E820_RAM)
> > @@ -446,8 +470,15 @@ static void process_e820_entry(struct e820entry *entry,
> >         if (entry->addr + entry->size < minimum)
> >                 return;
> >
> > -       region.start = entry->addr;
> > -       region.size = entry->size;
> > +       /* Ignore entries above memory limit */
> > +       end = min(entry->size + entry->addr - 1, mem_limit);
> > +       if (entry->addr >= end)
> > +               return;
> > +       cur_entry.addr = entry->addr;
> > +       cur_entry.size = end - entry->addr + 1;
> > +
> > +       region.start = cur_entry.addr;
> > +       region.size = cur_entry.size;
> 
> I find the manipulation of entry->addr +/- 1 confusing; it should just
> be mem_limit that is adjusted:
> 
>     end = min(entry->size + entry->addr, mem_limit + 1);

The first one is preferred. Since if no memory limit specifed, we still have
a limit, 64T-1 or UULONG_MAX -1, in logic more understandable. And in
code lines it's less.

Will change with your suggestion.

Thanks a log for your comments!

> 
> And maybe to avoid mem_limit being giant by default, maybe have "0" be special?
> 
>     cur_entry.addr = entry->addr;
>     if (mem_limit) {
>          unsigned long end = min(entry->size + entry->addr, mem_limit + 1);
>          if (entry->addr > end)
>              return;
>          cur_entry.size = end - entry->addr;
>     } else {
>           cur_entry.size = entry->size;
>     }
> 
> or something... and maybe move the whole thing earlier so other tests
> that examine entry->size are checked with the new adjusted value.
> 
> -Kees
> 
> >
> >         /* Give up if slot area array is full. */
> >         while (slot_area_index < MAX_SLOT_AREA) {
> > @@ -461,7 +492,7 @@ static void process_e820_entry(struct e820entry *entry,
> >                 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
> >
> >                 /* Did we raise the address above this e820 region? */
> > -               if (region.start > entry->addr + entry->size)
> > +               if (region.start > cur_entry.addr + cur_entry.size)
> >                         return;
> >
> >                 /* Reduce size by any delta from the original address. */
> > --
> > 2.5.5
> >
> 
> 
> 
> -- 
> Kees Cook
> Pixel Security