shinyauthr

Lifecycle: stable R-CMD-check Codecov test coverage

shinyauthr is an R package providing module functions that can be used to add an authentication layer to your shiny apps.

Installation

remotes::install_github("paulc91/shinyauthr")

Run example apps

Code for example apps using various UI frameworks can be found in inst/shiny-examples. You can launch 3 example apps with the runExample function.

# login with user1 pass1 or user2 pass2
runExample("basic")
runExample("shinydashboard")
runExample("navbarPage")

Usage

The package provides 2 module functions each with a UI and server element:

Note: the server modules use shiny’s new (version >= 1.5.0) shiny::moduleServer method as opposed to the shiny::callModule method used by the now deprecated shinyauthr::login and shinyauthr::logout functions. These functions will remain in the package for backwards compatibility but it is recommended you migrate to the new server functions. This will require some adjustments to the module server function calling method used in your app. For details on how to migrate see the ‘Migrating from callModule to moduleServer’ section of Modularizing Shiny app code.

Below is a minimal reproducible example of how to use the authentication modules in a shiny app. Note that this package invisibly calls shinyjs::useShinyjs() internally and there is no need for you to do so yourself (although there is no harm if you do).

library(shiny)

# dataframe that holds usernames, passwords and other user data
user_base <- tibble::tibble(
  user = c("user1", "user2"),
  password = c("pass1", "pass2"),
  permissions = c("admin", "standard"),
  name = c("User One", "User Two")
)

ui <- fluidPage(
  # add logout button UI
  div(class = "pull-right", shinyauthr::logoutUI(id = "logout")),
  # add login panel UI function
  shinyauthr::loginUI(id = "login"),
  # setup table output to show user info after login
  tableOutput("user_table")
)

server <- function(input, output, session) {

  # call login module supplying data frame, 
  # user and password cols and reactive trigger
  credentials <- shinyauthr::loginServer(
    id = "login",
    data = user_base,
    user_col = user,
    pwd_col = password,
    log_out = reactive(logout_init())
  )

  # call the logout module with reactive trigger to hide/show
  logout_init <- shinyauthr::logoutServer(
    id = "logout",
    active = reactive(credentials()$user_auth)
  )

  output$user_table <- renderTable({
    # use req to only render results when credentials()$user_auth is TRUE
    req(credentials()$user_auth)
    credentials()$info
  })
}

shinyApp(ui = ui, server = server)

Details

When the login module is called, it returns a reactive list containing 2 elements:

The initial values of these variables are FALSE and NULL respectively. However, given a data frame or tibble containing user names, passwords and other user data (optional), the login module will assign a user_auth value of TRUE if the user supplies a matching user name and password. The value of info then becomes the row of data associated with that user which can be used in the main app to control content based on user permission variables etc.

The logout button will only show when user_auth is TRUE. Clicking the button will reset user_auth back to FALSE which will hide the button and show the login panel again.

You can set the code in your server functions to only run after a successful login through use of the req() function inside all reactives, renders and observers. In the example above, using req(credentials()$user_auth) inside the renderTable function ensures the table showing the returned user information is only rendered when user_auth is TRUE.

Most authentication systems use browser cookies to avoid returning users having to re-enter their user name and password every time they return to the app. shinyauthr provides a method for cookie-based automatic login, but you must create your own functions to save and load session info into a database with persistent data storage.

The first required function must accept two parameters user and session. The first of these is the user name for log in. The second is a randomly generated string that identifies the session. The app asks the user’s web browser to save this session id as a cookie.

The second required function is called without parameters and must return a data.frame of valid user and session ids. If the user’s web browser sends your app a cookie which appears in the session column, then the corresponding user is automatically logged in.

Pass these functions to the login module via shinyauthr::loginServer(...) as the cookie_setter and cookie_getter parameters. A minimal example, using RSQLite as a local database to write and store user session data, is below.

library(shiny)
library(dplyr)
library(lubridate)
library(DBI)
library(RSQLite)

# connect to, or setup and connect to local SQLite db
if (file.exists("my_db_file")) {
  db <- dbConnect(SQLite(), "my_db_file")
} else {
  db <- dbConnect(SQLite(), "my_db_file")
  dbCreateTable(db, "sessionids", c(user = "TEXT", sessionid = "TEXT", login_time = "TEXT"))
}

# a user who has not visited the app for this many days
# will be asked to login with user name and password again
cookie_expiry <- 7 # Days until session expires

# This function must accept two parameters: user and sessionid. It will be called whenever the user
# successfully logs in with a password.  This function saves to your database.

add_sessionid_to_db <- function(user, sessionid, conn = db) {
  tibble(user = user, sessionid = sessionid, login_time = as.character(now())) %>%
    dbWriteTable(conn, "sessionids", ., append = TRUE)
}

# This function must return a data.frame with columns user and sessionid  Other columns are also okay
# and will be made available to the app after log in as columns in credentials()$user_auth

get_sessionids_from_db <- function(conn = db, expiry = cookie_expiry) {
  dbReadTable(conn, "sessionids") %>%
    mutate(login_time = ymd_hms(login_time)) %>%
    as_tibble() %>%
    filter(login_time > now() - days(expiry))
}

# dataframe that holds usernames, passwords and other user data
user_base <- tibble::tibble(
  user = c("user1", "user2"),
  password = c("pass1", "pass2"),
  permissions = c("admin", "standard"),
  name = c("User One", "User Two")
)

ui <- fluidPage(
  # add logout button UI
  div(class = "pull-right", shinyauthr::logoutUI(id = "logout")),
  # add login panel UI function
  shinyauthr::loginUI(id = "login", cookie_expiry = cookie_expiry),
  # setup table output to show user info after login
  tableOutput("user_table")
)

server <- function(input, output, session) {

  # call the logout module with reactive trigger to hide/show
  logout_init <- shinyauthr::logoutServer(
    id = "logout",
    active = reactive(credentials()$user_auth)
  )

  # call login module supplying data frame, user and password cols
  # and reactive trigger
  credentials <- shinyauthr::loginServer(
    id = "login",
    data = user_base,
    user_col = user,
    pwd_col = password,
    cookie_logins = TRUE,
    sessionid_col = sessionid,
    cookie_getter = get_sessionids_from_db,
    cookie_setter = add_sessionid_to_db,
    log_out = reactive(logout_init())
  )

  # pulls out the user information returned from login module
  user_data <- reactive({
    credentials()$info
  })

  output$user_table <- renderTable({
    # use req to only render results when credentials()$user_auth is TRUE
    req(credentials()$user_auth)
    user_data() %>%
      mutate(across(starts_with("login_time"), as.character))
  })
}

shinyApp(ui = ui, server = server)

Hashing Passwords with sodium

If you are hosting your user passwords on the internet, it is a good idea to first encrypt them with a hashing algorithm. You can use the sodium package to do this. Sodium uses a slow hashing algorithm that is specifically designed to protect stored passwords from brute-force attacks. More on this here. You then tell the shinyauthr::loginServer module that your passwords have been hashed by sodium and shinyauthr will then decrypt when login is requested. Your plain text passwords must be a character vector, not factors, when hashing for this to work as shiny inputs are passed as character strings.

For example, a sample user base like the following can be incorporated for use with shinyauthr:

# create a user base then hash passwords with sodium
# then save to an rds file in app directory
library(sodium)

user_base <- tibble::tibble(
  user = c("user1", "user2"),
  password = purrr::map_chr(c("pass1", "pass2"), sodium::password_store),
  permissions = c("admin", "standard"),
  name = c("User One", "User Two")
)

saveRDS(user_base, "user_base.rds")
# in your app code, read in the user base rds file
user_base <- readRDS("user_base.rds")
# then when calling the module set sodium_hashed = TRUE
credentials <- shinyauthr::loginServer(
  id = "login",
  data = user_base,
  user_col = user,
  pwd_col = password,
  sodium_hashed = TRUE,
  log_out = reactive(logout_init())
)

Credits

shinyauthr originally borrowed some code from treysp’s shiny_password template with the goal of making implementation simpler for end users and allowing the login/logout UIs to fit easily into any UI framework, including shinydashboard.

Thanks to Michael Dewar for his contribution of cookie-based authentication. Some code was borrowed from calligross’s Shiny Cookie Based Authentication Example and from an earlier PR from aqualogy.

Disclaimer

I’m not a security professional so cannot guarantee this authentication procedure to be foolproof. It is ultimately the shiny app developer’s responsibility not to expose any sensitive content to the client without the necessary login criteria being met.

I would welcome any feedback on any potential vulnerabilities in the process. I know that apps hosted on a server without an SSL certificate could be open to interception of user names and passwords submitted by a user. As such I would not recommend the use of shinyauthr without a HTTPS connection.

For apps intended for use within commercial organisations, I would recommend one of RStudio’s commercial shiny hosting options, or shinyproxy, both of which have built in authentication options.

However, I hope that having an easy-to-implement open-source shiny authentication option like this will prove useful when alternative options are not feasible.

Paul Campbell