2017-08-07 13:12:40

by Mike Rapoport

Subject: [PATCH] userfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage

When the process exit races with outstanding mcopy_atomic, it would be
better to return ESRCH error. When such race occurs the process and it's mm
are going away and returning "no such process" to the uffd monitor seems
better fit than ENOSPC.

Suggested-by: Michal Hocko <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: "Dr. David Alan Gilbert" <[email protected]>
Cc: Pavel Emelyanov <[email protected]>
Cc: Mike Kravetz <[email protected]>
Signed-off-by: Mike Rapoport <[email protected]>
---
The man-pages update is ready and I'll send it out once the patch is
merged.

fs/userfaultfd.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 06ea26b8c996..b0d5897bc4e6 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -1600,7 +1600,7 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
uffdio_copy.len);
mmput(ctx->mm);
} else {
- return -ENOSPC;
+ return -ESRCH;
}
if (unlikely(put_user(ret, &user_uffdio_copy->copy)))
return -EFAULT;
@@ -1647,7 +1647,7 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
uffdio_zeropage.range.len);
mmput(ctx->mm);
} else {
- return -ENOSPC;
+ return -ESRCH;
}
if (unlikely(put_user(ret, &user_uffdio_zeropage->zeropage)))
return -EFAULT;
--
2.7.4

2017-08-08 06:08:28

by Mike Rapoport

Subject: Re: [PATCH] userfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage

(adding Michal)

On Mon, Aug 07, 2017 at 04:12:25PM +0300, Mike Rapoport wrote:
> When the process exit races with outstanding mcopy_atomic, it would be
> better to return ESRCH error. When such race occurs the process and it's mm
> are going away and returning "no such process" to the uffd monitor seems
> better fit than ENOSPC.
>
> Suggested-by: Michal Hocko <[email protected]>
> Cc: Andrea Arcangeli <[email protected]>
> Cc: "Dr. David Alan Gilbert" <[email protected]>
> Cc: Pavel Emelyanov <[email protected]>
> Cc: Mike Kravetz <[email protected]>
> Signed-off-by: Mike Rapoport <[email protected]>
> ---
> The man-pages update is ready and I'll send it out once the patch is
> merged.
>
> fs/userfaultfd.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> index 06ea26b8c996..b0d5897bc4e6 100644
> --- a/fs/userfaultfd.c
> +++ b/fs/userfaultfd.c
> @@ -1600,7 +1600,7 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
> uffdio_copy.len);
> mmput(ctx->mm);
> } else {
> - return -ENOSPC;
> + return -ESRCH;
> }
> if (unlikely(put_user(ret, &user_uffdio_copy->copy)))
> return -EFAULT;
> @@ -1647,7 +1647,7 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
> uffdio_zeropage.range.len);
> mmput(ctx->mm);
> } else {
> - return -ENOSPC;
> + return -ESRCH;
> }
> if (unlikely(put_user(ret, &user_uffdio_zeropage->zeropage)))
> return -EFAULT;
> --
> 2.7.4
>

--
Sincerely yours,
Mike.

2017-08-10 09:33:53

by Michal Hocko

Subject: Re: [PATCH] userfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage

On Tue 08-08-17 09:08:17, Mike Rapoport wrote:
> (adding Michal)

Thanks

> On Mon, Aug 07, 2017 at 04:12:25PM +0300, Mike Rapoport wrote:
> > When the process exit races with outstanding mcopy_atomic, it would be
> > better to return ESRCH error. When such race occurs the process and it's mm
> > are going away and returning "no such process" to the uffd monitor seems
> > better fit than ENOSPC.

Not only the error message would be less confusing I also think that
error handling should be more straightforward. Although I cannot find
any guidelines for ENOSPC handling I've considered this errno as
potentially temporary and retry might be feasible while ESRCH is a
terminal error. I do not expect any userfaultfd users would retry on
error but who knows how the interface will be used in future so better
be prepared.

> > Suggested-by: Michal Hocko <[email protected]>
> > Cc: Andrea Arcangeli <[email protected]>
> > Cc: "Dr. David Alan Gilbert" <[email protected]>
> > Cc: Pavel Emelyanov <[email protected]>
> > Cc: Mike Kravetz <[email protected]>
> > Signed-off-by: Mike Rapoport <[email protected]>

Acked-by: Michal Hocko <[email protected]>

> > ---
> > The man-pages update is ready and I'll send it out once the patch is
> > merged.
> >
> > fs/userfaultfd.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> > index 06ea26b8c996..b0d5897bc4e6 100644
> > --- a/fs/userfaultfd.c
> > +++ b/fs/userfaultfd.c
> > @@ -1600,7 +1600,7 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
> > uffdio_copy.len);
> > mmput(ctx->mm);
> > } else {
> > - return -ENOSPC;
> > + return -ESRCH;
> > }
> > if (unlikely(put_user(ret, &user_uffdio_copy->copy)))
> > return -EFAULT;
> > @@ -1647,7 +1647,7 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
> > uffdio_zeropage.range.len);
> > mmput(ctx->mm);
> > } else {
> > - return -ENOSPC;
> > + return -ESRCH;
> > }
> > if (unlikely(put_user(ret, &user_uffdio_zeropage->zeropage)))
> > return -EFAULT;
> > --
> > 2.7.4
> >
>
> --
> Sincerely yours,
> Mike.
>

--
Michal Hocko
SUSE Labs