Move SSH key generation script from pam.d to /etc/profile

When FSx Lustre is configured with the new root_squash feature,
and ParallelCluster is configured with Active Directory with
home folders within the FSx mount, pam_exec.so is unable to
properly run the SSH key generation script. This is because
pam_exec.so runs the script as root, but root does not have
access to any home folders to manipulate the files due to the
fact that root is regarded as nobody/nogroup within the
root_squash'd FSx mount point.

Using su in the generation script to impersonate the user does
not work around the problem, as su itself would trigger
pam_exec.so, and trigger a loop, which doesn't look trivial to
avoid to me.

Instead, I suggest moving the key generation to /etc/profile,
which is executed by default for every interactive shells, by
the connecting user, and serves the purpose.

CHANGELOG.md

@@ -30,6 +30,7 @@ This file is used to list changes made in each version of the AWS ParallelCluste
 - Change the logic to number the routing tables when an instance have multiple NICs.
 - Upgrade Python from 3.7.13 to 3.9.13.
 - Upgrade Slurm to version 22.05.3.
+- Move SSH key generation from pam.d to /etc/profile to enable root_squash'd FSx
 
 3.2.0
 ------

cookbooks/aws-parallelcluster-config/recipes/directory_service.rb

@@ -155,38 +155,45 @@
     sensitive true
   end
 
-  pam_services = %w(sudo su sshd)
-  pam_config_dir = "/etc/pam.d"
   generate_ssh_key_path = "#{node['cluster']['scripts_dir']}/generate_ssh_key.sh"
-  ssh_key_generator_pam_config_line = "session    optional     pam_exec.so log=/var/log/parallelcluster/pam_ssh_key_generator.log #{generate_ssh_key_path}"
-  if node['cluster']["directory_service"]["generate_ssh_keys_for_users"] == 'true'
+  ssh_key_generator_profile_config_line = "bash #{generate_ssh_key_path} >> /var/log/parallelcluster/pam_ssh_key_generator.log 2>&1"
+  if node['cluster']["directory_service"]["generate_ssh_keys_for_users"] == 'true' 
     template generate_ssh_key_path do
       source 'directory_service/generate_ssh_key.sh.erb'
       owner 'root'
       group 'root'
       mode '0755'
     end
-    pam_services.each do |pam_service|
-      pam_config_file = "#{pam_config_dir}/#{pam_service}"
-      append_if_no_line "Ensure PAM service #{pam_service} is configured to call SSH key generation script" do
-        path pam_config_file
-        line ssh_key_generator_pam_config_line
-      end
+    file "/var/log/parallelcluster/pam_ssh_key_generator.log" do
+      action :touch
+      mode '0777'
+    end
+    append_if_no_line "Ensure /etc/profile is configured to call SSH key generation script" do
+      path "/etc/profile"
+      line ssh_key_generator_profile_config_line
     end
   else
-    # Remove script used to generate key if it exists and ensure PAM is not configured to try to call it
+    # Remove script used to generate key if it exists and ensure /etc/profile is not configured to try to call it
     file generate_ssh_key_path do
       action :delete
       only_if { ::File.exist? generate_ssh_key_path }
     end
-
-    pam_services.each do |pam_service|
-      pam_config_file = "#{pam_config_dir}/#{pam_service}"
-      delete_lines "Ensure PAM service #{pam_service} is not configured to call SSH key generation script" do
-        path pam_config_file
-        pattern %r{session\s+optional\s+pam_exec\.so\s+log=/var/log/parallelcluster/pam_ssh_key_generator\.log}
-        ignore_missing true
-      end
+    delete_lines "Ensure /etc/profile is not configured to call SSH key generation script" do
+      path "/etc/profile"
+      line ssh_key_generator_profile_config_line
+    end
+  end
+
+  # Ensure pam.d based SSH generation is removed
+  ssh_key_generator_pam_config_line = "session    optional     pam_exec.so log=/var/log/parallelcluster/pam_ssh_key_generator.log #{generate_ssh_key_path}"
+  pam_services = %w(sudo su sshd)
+  pam_config_dir = "/etc/pam.d"
+  pam_services.each do |pam_service|
+    pam_config_file = "#{pam_config_dir}/#{pam_service}"
+    delete_lines "Ensure PAM service #{pam_service} is not configured to call SSH key generation script" do
+      path pam_config_file
+      pattern %r{session\s+optional\s+pam_exec\.so\s+log=/var/log/parallelcluster/pam_ssh_key_generator\.log}
+      ignore_missing true
     end
   end
 else

cookbooks/aws-parallelcluster-config/templates/default/directory_service/generate_ssh_key.sh.erb

@@ -3,16 +3,11 @@ set -ex
 env
 
 # Root does not need SSH key generation
-[ ${PAM_USER} == "root" ] && exit 0
-
-# The home directory for every user must be determined in the most generic way because
-# we should not assume that every user has its home directory in /home/$USER.
-user_home_dir="$(getent passwd ${PAM_USER} | cut -d ':' -f 6)"
-[ ! -d "${user_home_dir}" ] && echo "ERROR Cannot create SSH key for user ${PAM_USER} if its home directory is not found" && exit 1
+[ ${USER} == "root" ] && exit 0
 
 # Skip SSH key creation if the SSH has been already configured for the user.
 # We assume that SSH has been already configured if the directory .ssh already exists in the user home.
-user_ssh_dir="${user_home_dir}/.ssh"
+user_ssh_dir="${HOME}/.ssh"
 [ -d "${user_ssh_dir}" ] && exit 0
 
 mkdir -m 0700 "${user_ssh_dir}"
@@ -22,4 +17,4 @@ cat "${user_ssh_dir}/id_rsa.pub" >> "${user_ssh_dir}/authorized_keys"
 chmod 0600 "${user_ssh_dir}/authorized_keys"
 ssh-keyscan <%=  node['hostname'] %> > "${user_ssh_dir}/known_hosts"
 chmod 0600 "${user_ssh_dir}/known_hosts"
-chown ${PAM_USER}:$(id -g ${PAM_USER}) -R "${user_ssh_dir}"
+chown ${USER}:$(id -g ${USER}) -R "${user_ssh_dir}"