Fix region lost in /proc/self/smaps

Commit Message

Xiao Guangrong Sept. 9, 2016, 8:19 a.m. UTC

On 09/08/2016 10:05 PM, Dave Hansen wrote:
> On 09/07/2016 08:36 PM, Xiao Guangrong wrote:>> The user will see two
> VMAs in their output:
>>>
>>>     A: 0x1000->0x2000
>>>     C: 0x1000->0x3000
>>>
>>> Will it confuse them to see the same virtual address range twice?  Or is
>>> there something preventing that happening that I'm missing?
>>>
>>
>> You are right. Nothing can prevent it.
>>
>> However, it is not easy to handle the case that the new VMA overlays
>> with the old VMA
>> already got by userspace. I think we have some choices:
>> 1: One way is completely skipping the new VMA region as current kernel
>> code does but i
>>    do not think this is good as the later VMAs will be dropped.
>>
>> 2: show the un-overlayed portion of new VMA. In your case, we just show
>> the region
>>    (0x2000 -> 0x3000), however, it can not work well if the VMA is a new
>> created
>>    region with different attributions.
>>
>> 3: completely show the new VMA as this patch does.
>>
>> Which one do you prefer?
>
> I'd be willing to bet that #3 will break *somebody's* tooling.
> Addresses going backwards is certainly screwy.  Imagine somebody using
> smaps to search for address holes and doing hole_size=0x1000-0x2000.
>
> #1 can lies about there being no mapping in place where there there may
> have _always_ been a mapping and is very similar to the bug you were
> originally fixing.  I think that throws it out.
>
> #2 is our best bet, I think.  It's unfortunately also the most code.
> It's also a bit of a fib because it'll show a mapping that never
> actually existed, but I think this is OK.  I'm not sure what the
> downside is that you're referring to, though.  Can you explain?

Yes. I was talking the case as follows:
    1: read() #1: prints vma-A(0x1000 -> 0x2000)
    2: unmap vma-A(0x1000 -> 0x2000)
    3: create vma-B(0x80 -> 0x3000) on other file with different permission
       (w, r, x)
    4: read #2: prints vma-B(0x2000 -> 0x3000)

Then userspace will get just a portion of vma-B. well, maybe it is not too bad. :)

How about this changes:

--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Dave Hansen Sept. 9, 2016, 4:47 p.m. UTC | #1

On 09/09/2016 01:19 AM, Xiao Guangrong wrote:
> 
> Yes. I was talking the case as follows:
>    1: read() #1: prints vma-A(0x1000 -> 0x2000)
>    2: unmap vma-A(0x1000 -> 0x2000)
>    3: create vma-B(0x80 -> 0x3000) on other file with different permission
>       (w, r, x)
>    4: read #2: prints vma-B(0x2000 -> 0x3000)
> 
> Then userspace will get just a portion of vma-B. well, maybe it is not
> too bad. :)

Yeah, I think this is the way to go.  Feel free to add my ack.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 187d84e..10ca648 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -147,7 +147,7 @@  m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
  static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
  {
         if (m->count < m->size) /* vma is copied successfully */
-               m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
+               m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
  }

  static void *m_start(struct seq_file *m, loff_t *ppos)
@@ -176,14 +176,14 @@  static void *m_start(struct seq_file *m, loff_t *ppos)

         if (last_addr) {
                 vma = find_vma(mm, last_addr);
-               if (vma && (vma = m_next_vma(priv, vma)))
+               if (vma)
                         return vma;
         }

         m->version = 0;
         if (pos < mm->map_count) {
                 for (vma = mm->mmap; pos; pos--) {
-                       m->version = vma->vm_start;
+                       m->version = vma->vm_end;
                         vma = vma->vm_next;
                 }
                 return vma;
@@ -293,7 +293,7 @@  show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
         vm_flags_t flags = vma->vm_flags;
         unsigned long ino = 0;
         unsigned long long pgoff = 0;
-       unsigned long start, end;
+       unsigned long end, start = m->version;
         dev_t dev = 0;
         const char *name = NULL;

@@ -304,8 +304,13 @@  show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
         }

+       /*
+        * the region [0, m->version) has already been handled, do not
+        * handle it doubly.
+        */
+       start = max(vma->vm_start, start);
+
         /* We don't show the stack guard page in /proc/maps */
-       start = vma->vm_start;
         if (stack_guard_page_start(vma, start))
                 start += PAGE_SIZE;
         end = vma->vm_end;