[CRIU] [PATCH] syscall: add memfd_create() (v3)

[Andrew Vagin]

On Tue, Oct 14, 2014 at 01:11:37PM +0400, Pavel Emelyanov wrote:
> On 10/14/2014 01:05 PM, Andrey Vagin wrote:
> > v2: Follow the kerndat style that "features" are described
> > just by global boolean variables.
> > 
> > v3: close a file descriptor only if memfd_create() was successful
> > Signed-off-by: Andrey Vagin <avagin at openvz.org>
> > ---
> >  arch/arm/syscall.def        |  1 +
> >  arch/x86/syscall-x86-64.def |  1 +
> >  include/kerndat.h           |  2 ++
> >  kerndat.c                   | 22 ++++++++++++++++++++++
> >  4 files changed, 26 insertions(+)
> > 
> > diff --git a/arch/arm/syscall.def b/arch/arm/syscall.def
> > index 02eaa65..3b2d1ed 100644
> > --- a/arch/arm/syscall.def
> > +++ b/arch/arm/syscall.def
> > @@ -96,3 +96,4 @@ kcmp				272	378	(pid_t pid1, pid_t pid2, int type, unsigned long idx1, unsigned
> >  openat				56	322	(int dirfd, const char *pathname, int flags, mode_t mode)
> >  mkdirat				34	323	(int dirfd, const char *pathname, mode_t mode)
> >  unlinkat			35	328	(int dirfd, const char *pathname, int flags)
> > +memfd_create			279	385	(const char *name, unsigned int flags)
> > diff --git a/arch/x86/syscall-x86-64.def b/arch/x86/syscall-x86-64.def
> > index 5d31337..c780c40 100644
> > --- a/arch/x86/syscall-x86-64.def
> > +++ b/arch/x86/syscall-x86-64.def
> > @@ -96,3 +96,4 @@ __NR_prlimit64		302		sys_prlimit64		(pid_t pid, unsigned int resource, const str
> >  __NR_open_by_handle_at	304		sys_open_by_handle_at	(int mountdirfd, struct file_handle *handle, int flags)
> >  __NR_setns		308		sys_setns		(int fd, int nstype)
> >  __NR_kcmp		312		sys_kcmp		(pid_t pid1, pid_t pid2, int type, unsigned long idx1, unsigned long idx2)
> > +__NR_memfd_create	319		sys_memfd_create	(const char *name, unsigned int flags)
> > diff --git a/include/kerndat.h b/include/kerndat.h
> > index 6b9947f..2b0a653 100644
> > --- a/include/kerndat.h
> > +++ b/include/kerndat.h
> > @@ -24,4 +24,6 @@ extern u64 zero_page_pfn;
> >  struct stat;
> >  extern struct stat *kerndat_get_devpts_stat(void);
> >  
> > +extern bool memfd_is_supported;
> > +
> >  #endif /* __CR_KERNDAT_H__ */
> > diff --git a/kerndat.c b/kerndat.c
> > index 3c87f6c..9df04dc 100644
> > --- a/kerndat.c
> > +++ b/kerndat.c
> > @@ -13,6 +13,7 @@
> >  #include "mem.h"
> >  #include "compiler.h"
> >  #include "sysctl.h"
> > +#include "syscall.h"
> >  #include "asm/types.h"
> >  #include "cr_options.h"
> >  #include "util.h"
> > @@ -220,6 +221,25 @@ int get_last_cap(void)
> >  	return sysctl_op(req, CTL_READ);
> >  }
> >  
> > +bool memfd_is_supported;
> > +static bool kerndat_has_memfd_create(void)
> > +{
> > +	int ret;
> > +
> > +	ret = sys_memfd_create("", 0);
> > +
> > +	if (ret == -ENOSYS)
> > +		memfd_is_supported = false;
> 
> Few steps below the memfd_is_supported will be set to true.

Eh. Pls look at the another version of this patch. It's version is v3
too, because it's an alternative to this. It doesn't create a real memfd
descriptor. It gives NULL as name and expect to get EFAUL if
memfd_create() is supported.

> 
> > +	else if (ret < 0) {
> > +		pr_err("Unexpected error %d from memfd_create(NULL, 0)\n", ret);
> > +		return -1;
> > +	} else
> > +		close(ret);
> > +
> > +	memfd_is_supported = true;
> > +	return 0;
> > +}
> > +
> >  int kerndat_init(void)
> >  {
> >  	int ret;
> > @@ -231,6 +251,8 @@ int kerndat_init(void)
> >  		ret = init_zero_page_pfn();
> >  	if (!ret)
> >  		ret = get_last_cap();
> > +	if (!ret)
> > +		ret = kerndat_has_memfd_create();
> >  
> >  	return ret;
> >  }
> > 
>