Le 23/02/2018 à 05:57, Shea Levy a écrit :
> Signed-off-by: Shea Levy <s...@shealevy.com>
> ---
> configure | 21 +++++++++++++++++++
> linux-user/syscall.c | 3 +++
> linux-user/syscall_defs.h | 53
> +++++++++++++++++++++++++++++++++++++++++++++++
> 3 files changed, 77 insertions(+)
>
> diff --git a/configure b/configure
> index 913e14839d..91082aa1dc 100755
> --- a/configure
> +++ b/configure
> @@ -5303,6 +5303,23 @@ if compile_prog "" "" ; then
> have_utmpx=yes
> fi
>
> +##########################################
> +# Check for newer fields of struct statfs on Linux
> +
> +if test "$linux_user" = "yes"; then
> + cat > $TMPC <<EOF
> +#include <sys/vfs.h>
> +
> +int main(void) {
> + struct statfs fs;
> + fs.f_flags = 0;
> + return fs.f_flags;
> +}
> +EOF
> + if compile_object ; then
> + have_statfs_flags=yes
> + fi
> +fi
I don't think you need this.
statfs is provided by glibc and glibc as a flag to tell code we have
this member:
_STATFS_F_FLAGS
(see glibc commit:
3cdaa6adb1 f_flags in Linux statfs implementation.
(Wed Aug 11 14:07:28 2010 -0700)
since glibc 2.13)
> ##########################################
> # checks for sanitizers
>
> @@ -6518,6 +6535,10 @@ if test "$have_utmpx" = "yes" ; then
> echo "HAVE_UTMPX=y" >> $config_host_mak
> fi
>
> +if test "$have_statfs_flags" = "yes" ; then
> + echo "HAVE_STATFS_FLAGS=y" >> $config_host_mak
> +fi
> +
> if test "$ivshmem" = "yes" ; then
> echo "CONFIG_IVSHMEM=y" >> $config_host_mak
> fi
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 82b35a6bdf..77481eca2c 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -9534,6 +9534,9 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
> arg1,
> __put_user(stfs.f_fsid.__val[1], &target_stfs->f_fsid.val[1]);
> __put_user(stfs.f_namelen, &target_stfs->f_namelen);
> __put_user(stfs.f_frsize, &target_stfs->f_frsize);
> +#ifdef HAVE_STATFS_FLAGS
#ifdef _STATFS_F_FLAGS
> + __put_user(stfs.f_flags, &target_stfs->f_flags);
I think you should define f_flags in target_statfs unconditionally (see
below), so you should clear target_stfs->f_flags if host kernel doesn't
support it:
glibc will check for "ST_VALID" flag to see if this field is implemented
by the kernel, and we must be sure in this case it is not set.
> +#endif
> memset(target_stfs->f_spare, 0, sizeof(target_stfs->f_spare));
> unlock_user_struct(target_stfs, arg2, 1);
> }
> diff --git a/linux-user/syscall_defs.h b/linux-user/syscall_defs.h
> index a35c52a60a..64aa49d3c5 100644
> --- a/linux-user/syscall_defs.h
> +++ b/linux-user/syscall_defs.h
> @@ -362,7 +362,14 @@ struct kernel_statfs {
> int f_ffree;
> kernel_fsid_t f_fsid;
> int f_namelen;
> +#ifdef HAVE_STATFS_FLAGS
> + int f_frsize;
> + int f_flags;
> + int f_spare[4];
> +#else
> int f_spare[6];
> +#endif
> +
> };
I think the kernel_statfs structure can be removed totally, as it is not
used anymore since:
56c8f68f1d statfs fix (Mon Nov 28 22:28:41 2005)
>
> struct target_dirent {
> @@ -2223,7 +2230,12 @@ struct target_statfs {
I think you can update unconditionally all target_statfs to be in sync
with current kernel structures, that have been updated by:
commit 365b18189789bfa1acd9939e6312b8a4b4577b28
Author: Christoph Hellwig <h...@lst.de>
Date: Wed Jul 7 18:53:25 2010 +0200
add f_flags to struct statfs(64)
Add a flags field to help glibc implementing statvfs(3) efficiently.
We copy the flag values from glibc, and add a new ST_VALID flag to
denote that f_flags is implemented.
Thanks,
Laurent
