Skip to contentSkip to conversation

remap_file_pages doesn't like MAP_ANONYMOUS (but used to?)

DISCUSSION Kenny Simpson 2 participants 4 messages Mar 17, 2011 – Mar 21, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hello,
  In older kernels (2.6.15-2.6.18?) I was able to make an anonymous mapping and remap part of it back over itself to create a circular buffer:
  p = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0)
  remap_file_pages(p + sz/2, sz/2, 0/*prot*/, 0/*offset*/, 0/*flags*/)

I find the same method fails with EINVAL on the remap_file_pages call in 2.6.35/38. Making a dummy file or shared memory object works, but seems a bit more messy.

Why drop support for ANONYMOUS? Is there another way to get the same effect?

thanks,
-Kenny

maintainer
#1 lore raw
Quoted 8 lines · Kenny Simpson

Hello,
  In older kernels (2.6.15-2.6.18?) I was able to make an anonymous mapping and remap part of it back over itself to create a circular buffer:
  p = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0)
  remap_file_pages(p + sz/2, sz/2, 0/*prot*/, 0/*offset*/, 0/*flags*/)

I find the same method fails with EINVAL on the remap_file_pages call in 2.6.35/38. Making a dummy file or shared memory object works, but seems a bit more messy.

Why drop support for ANONYMOUS? Is there another way to get the same effect?

Thanks for reporting. It was a simple oversight: and until you came along, nobody noticed. Here's the patch, I'll pass it upstream shortly.

[PATCH] shmem: let shared anonymous be nonlinear again

Up to 2.6.22, you could use remap_file_pages(2) on a tmpfs file or a shared mapping of /dev/zero or a shared anonymous mapping. In 2.6.23 we disabled it by default, but set VM_CAN_NONLINEAR to enable it on safe mappings. We made sure to set it in shmem_mmap() for tmpfs files, but missed it in shmem_zero_setup() for the others. Fix that at last.

---

 mm/shmem.c |    1 +
 1 file changed, 1 insertion(+)
mm/shmem.c +1 -0
@@ -2791,5 +2791,6 @@ int shmem_zero_setup(struct vm_area_stru
 		fput(vma->vm_file);
 	vma->vm_file = file;
 	vma->vm_ops = &shmem_vm_ops;
+	vma->vm_flags |= VM_CAN_NONLINEAR;
 	return 0;
 }
#2 lore raw
Quoted 4 lines · Hugh Dickins

Thanks for reporting.  It was a simple oversight: and until you came
along, nobody noticed.  Here's the patch, I'll pass it upstream shortly.

Since this bug has existed for a while, could it be pushed back to the various stable/longterm branches as well? (as they all are affected)

Many thanks on the quick response! Glad it was a simple oversight with an easy and obvious fix.

thanks again!
-Kenny

maintainer
#3 lore raw
Quoted 6 lines · Kenny Simpson
Quoted 4 lines · Hugh Dickins

Thanks for reporting.  It was a simple oversight: and until you came
along, nobody noticed.  Here's the patch, I'll pass it upstream shortly.

Quoted 6 lines

Since this bug has existed for a while, could it be pushed back to the various stable/longterm branches as well? (as they all are affected)

Many thanks on the quick response! Glad it was a simple oversight with an easy and obvious fix.

thanks again!
-Kenny

It's a rare pleasure to be presented with such a straightforward bug!

stable/longterm: well, I don't mind asking akpm to add a Cc: stable, but don't be surprised if he prefers not: although it was a regression back in 2.6.23, it's not a security issue, and the lack of interest between then and now doesn't make a strong argument for it.

Andrew, if you're so inclined, please add a Cc: [email protected] to [PATCH] shmem: let shared anonymous be nonlinear again

Thanks,
Hugh