fix cd permissions when user belongs to folder group (#9531)

# Description

Fixes: #9498

Actually we don't need this pr if the upstream pr is merged:
https://github.com/ogham/rust-users/pull/45

But it doesn't have any commit since 2021, and the author seems not
active on github now, I think we have to copy the function into nushell
to get relative issue fixed...

crates/nu-command/src/filesystem/cd.rs

@@ -272,27 +272,74 @@ fn have_permission(dir: impl AsRef<Path>) -> PermissionResult<'static> {
     }
 }
 
+// NOTE: it's a re-export of https://github.com/ogham/rust-users crate
+// we need to use it because the upstream pr isn't merged: https://github.com/ogham/rust-users/pull/45
+// once it's merged, we can remove this module
+#[cfg(unix)]
+mod nu_users {
+    use libc::c_int;
+    use std::ffi::{CString, OsStr};
+    use std::os::unix::ffi::OsStrExt;
+    use users::{gid_t, Group};
+    /// Returns groups for a provided user name and primary group id.
+    ///
+    /// # libc functions used
+    ///
+    /// - [`getgrouplist`](https://docs.rs/libc/*/libc/fn.getgrouplist.html)
+    ///
+    /// # Examples
+    ///
+    /// ```no_run
+    /// use users::get_user_groups;
+    ///
+    /// for group in get_user_groups("stevedore", 1001).expect("Error looking up groups") {
+    ///     println!("User is a member of group #{} ({:?})", group.gid(), group.name());
+    /// }
+    /// ```
+    pub fn get_user_groups<S: AsRef<OsStr> + ?Sized>(
+        username: &S,
+        gid: gid_t,
+    ) -> Option<Vec<Group>> {
+        // MacOS uses i32 instead of gid_t in getgrouplist for unknown reasons
+        #[cfg(all(unix, target_os = "macos"))]
+        let mut buff: Vec<i32> = vec![0; 1024];
+        #[cfg(all(unix, not(target_os = "macos")))]
+        let mut buff: Vec<gid_t> = vec![0; 1024];
+        let name = CString::new(username.as_ref().as_bytes()).expect("OsStr is guaranteed to be zero-free, which is the condition for CString::new to succeed");
+        let mut count = buff.len() as c_int;
+        // MacOS uses i32 instead of gid_t in getgrouplist for unknown reasons
+        // SAFETY:
+        // int getgrouplist(const char *user, gid_t group, gid_t *groups, int *ngroups);
+        //
+        // `name` is valid CStr to be `const char*` for `user`
+        // every valid value will be accepted for `group`
+        // The capacity for `*groups` is passed in as `*ngroups` which is the buffer max length/capacity (as we initialize with 0)
+        // Following reads from `*groups`/`buff` will only happen after `buff.truncate(*ngroups)`
+        #[cfg(all(unix, target_os = "macos"))]
+        let res =
+            unsafe { libc::getgrouplist(name.as_ptr(), gid as i32, buff.as_mut_ptr(), &mut count) };
+        #[cfg(all(unix, not(target_os = "macos")))]
+        let res = unsafe { libc::getgrouplist(name.as_ptr(), gid, buff.as_mut_ptr(), &mut count) };
+        if res < 0 {
+            None
+        } else {
+            buff.truncate(count as usize);
+            buff.sort_unstable();
+            buff.dedup();
+            // allow trivial cast: on macos i is i32, on linux it's already gid_t
+            #[allow(trivial_numeric_casts)]
+            buff.into_iter()
+                .filter_map(|i| users::get_group_by_gid(i as gid_t))
+                .collect::<Vec<_>>()
+                .into()
+        }
+    }
+}
+
 #[cfg(unix)]
 fn any_group(current_user_gid: gid_t, owner_group: u32) -> bool {
     users::get_current_username()
-        .map(|name| {
+        .map(|name| nu_users::get_user_groups(&name, current_user_gid).unwrap_or_default())
-            users::get_user_groups(&name, current_user_gid)
-                .map(|mut groups| {
-                    // Fixes https://github.com/ogham/rust-users/issues/44
-                    // If a user isn't in more than one group then this fix won't work,
-                    // However its common for a user to be in more than one group, so this should work for most.
-                    if groups.len() == 2 && groups[1].gid() == 0 {
-                        // We have no way of knowing if this is due to the issue or the user is actually in the root group
-                        // So we will assume they are in the root group and leave it.
-                        // It's not the end of the world if we are wrong, they will just get a permission denied error once inside.
-                    } else {
-                        groups.pop();
-                    }
-
-                    groups
-                })
-                .unwrap_or_default()
-        })
         .unwrap_or_default()
         .into_iter()
         .any(|group| group.gid() == owner_group)