mbox series

[v5,0/3] hugetlbfs: close race between MADV_DONTNEED and page fault

Message ID 20231001005659.2185316-1-riel@surriel.com
Headers
Series hugetlbfs: close race between MADV_DONTNEED and page fault |

Message

Rik van Riel Oct. 1, 2023, 12:55 a.m. UTC 
  v5: somehow a __vma_private_lock(vma) test failed to make it from my tree into the v4 series, fix that
v4: fix unmap_vmas locking issue pointed out by Mike Kravetz, and resulting lockdep fallout
v3: fix compile error w/ lockdep and test case errors with patch 3
v2: fix the locking bug found with the libhugetlbfs tests.

Malloc libraries, like jemalloc and tcalloc, take decisions on when
to call madvise independently from the code in the main application.

This sometimes results in the application page faulting on an address,
right after the malloc library has shot down the backing memory with
MADV_DONTNEED.

Usually this is harmless, because we always have some 4kB pages
sitting around to satisfy a page fault. However, with hugetlbfs
systems often allocate only the exact number of huge pages that
the application wants.

Due to TLB batching, hugetlbfs MADV_DONTNEED will free pages outside of
any lock taken on the page fault path, which can open up the following
race condition:

       CPU 1                            CPU 2

       MADV_DONTNEED
       unmap page
       shoot down TLB entry
                                       page fault
                                       fail to allocate a huge page
                                       killed with SIGBUS
       free page

Fix that race by extending the hugetlb_vma_lock locking scheme to also
cover private hugetlb mappings (with resv_map), and pulling the locking 
from __unmap_hugepage_final_range into helper functions called from
zap_page_range_single. This ensures page faults stay locked out of
the MADV_DONTNEED VMA until the huge pages have actually been freed.

The third patch in the series is more of an RFC. Using the
invalidate_lock instead of the hugetlb_vma_lock greatly simplifies
the code, but at the cost of turning a per-VMA lock into a lock
per backing hugetlbfs file, which could slow things down when
multiple processes are mapping the same hugetlbfs file.

Comments

Andrew Morton Oct. 1, 2023, 2:54 a.m. UTC | #1 
On Sat, 30 Sep 2023 20:55:47 -0400 riel@surriel.com wrote:

> v5: somehow a __vma_private_lock(vma) test failed to make it from my tree into the v4 series, fix that
> v4: fix unmap_vmas locking issue pointed out by Mike Kravetz, and resulting lockdep fallout
> v3: fix compile error w/ lockdep and test case errors with patch 3
> v2: fix the locking bug found with the libhugetlbfs tests.
> 
> Malloc libraries, like jemalloc and tcalloc, take decisions on when
> to call madvise independently from the code in the main application.
> 
> This sometimes results in the application page faulting on an address,
> right after the malloc library has shot down the backing memory with
> MADV_DONTNEED.
> 
> Usually this is harmless, because we always have some 4kB pages
> sitting around to satisfy a page fault. However, with hugetlbfs
> systems often allocate only the exact number of huge pages that
> the application wants.
> 
> Due to TLB batching, hugetlbfs MADV_DONTNEED will free pages outside of
> any lock taken on the page fault path, which can open up the following
> race condition:
> 
>        CPU 1                            CPU 2
> 
>        MADV_DONTNEED
>        unmap page
>        shoot down TLB entry
>                                        page fault
>                                        fail to allocate a huge page
>                                        killed with SIGBUS
>        free page
> 
> Fix that race by extending the hugetlb_vma_lock locking scheme to also
> cover private hugetlb mappings (with resv_map), and pulling the locking 
> from __unmap_hugepage_final_range into helper functions called from
> zap_page_range_single. This ensures page faults stay locked out of
> the MADV_DONTNEED VMA until the huge pages have actually been freed.

Didn't we decide that [1/3] and [2/3] should be cc:stable?

> The third patch in the series is more of an RFC. Using the
> invalidate_lock instead of the hugetlb_vma_lock greatly simplifies
> the code, but at the cost of turning a per-VMA lock into a lock
> per backing hugetlbfs file, which could slow things down when
> multiple processes are mapping the same hugetlbfs file.

"could slow things down" is testable-for?

This third one I'd queue up for testing for a 6.7-rc1 merge, so I'll split
the series apart.  Not a problem, but it would be a little better if
things were originally packaged that way.