Thread (1 message) 1 message, 1 author, 2019-10-30

Re: mbind() breaks its API definition since v5.2 by commit d883544515aa (mm: mempolicy: make the behavior consistent when MPOL_MF_MOVE* and MPOL_MF_STRICT were specified)

From: Yang Shi <hidden>
Date: 2019-10-30 02:50:24
Also in: linux-mm, lkml
Subsystem: memory management, memory management - memory policy and migration, the rest · Maintainers: Andrew Morton, David Hildenbrand, Linus Torvalds

Possibly related (same subject, not in this thread)


On 10/29/19 7:27 PM, Li Xinhai wrote:
quoted hunk
One change in do_mbind() of this commit has suspicious usage of return value of
queue_pages_range(), excerpt as below:

---
@@ -1243,10 +1265,15 @@ static long do_mbind(unsigned long start, unsigned long len,
  	if (err)
  		goto mpol_out;
  
-	err = queue_pages_range(mm, start, end, nmask,
+	ret = queue_pages_range(mm, start, end, nmask,
  			  flags | MPOL_MF_INVERT, &pagelist);
-	if (!err)
-		err = mbind_range(mm, start, end, new);
+
+	if (ret < 0) {      /////// convert to all possible 'ret' to '-EIO' <<<<
+		err = -EIO;
+		goto up_out;
+	}
+
+	err = mbind_range(mm, start, end, new);
  
  	if (!err) {
  		int nr_failed = 0;
---
Note that inside queue_pages_range(), the call to walk_page_range() may return
errors from 'test_walk' of 'struct mm_walk_ops', e.g. -EFAULT. Now, those error
codes are no longer reported to user space application.

 From user space, the mbind() call need to reported error, with EFAULT, as example:
EFAULT
Part or all of the memory range specified by nodemask and maxnode points
outside your accessible address space. Or, there was an unmapped hole in the
specified memory range specified by addr and len.
Thanks for catching this. That commit was aimed to correct the return 
values for some corner cases in mbind(), but it should not alter the 
errno for other failure cases, i.e. -EFAULT.

Could you please try the below patch (build test only)?
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 4ae967b..99df43a 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1286,7 +1286,7 @@ static long do_mbind(unsigned long start, unsigned 
long len,
                           flags | MPOL_MF_INVERT, &pagelist);

         if (ret < 0) {
-               err = -EIO;
+               err = ret;
                 goto up_out;
         }

Please correct me if this is the intended change(and will have updated API
definition), or something was misunderstood.

-Xinhai
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help