Skip to content

fix(envd): cap PostInit request body at 1 MiB with MaxBytesReader - #3564

Open
AdaAibaby wants to merge 1 commit into
e2b-dev:mainfrom
AdaAibaby:fix/envd-init-body-size-limit
Open

fix(envd): cap PostInit request body at 1 MiB with MaxBytesReader#3564
AdaAibaby wants to merge 1 commit into
e2b-dev:mainfrom
AdaAibaby:fix/envd-init-body-size-limit

Conversation

@AdaAibaby

Copy link
Copy Markdown
Contributor

Problem

Closes #3563

PostInit read the full request body with no size cap:

body, err := io.ReadAll(r.Body)  // no MaxBytesReader
defer memguard.WipeBytes(body)

Any process inside the VM that can reach the envd HTTP port can send an arbitrarily large body, allocating heap memory until envd is OOM-killed. The auth check happens after the full body is read, so no credentials are required.

envd is the sandbox control plane — OOM-killing it orphans all user processes, breaks graceful teardown (slot release, cleanup callbacks), and can corrupt pause/resume state.

Secondary issue: memguard.WipeBytes(body) is intended to scrub the access token from heap memory after use. If body is a multi-hundred-MiB allocation, Go's GC may have already paged parts of it to disk or the allocator may have moved the backing array before the deferred wipe runs, reducing the security guarantee.

Fix

Wrap r.Body with http.MaxBytesReader(w, r.Body, 1<<20) before io.ReadAll. 1 MiB is a generous upper bound — the largest legitimate /init payload carries EnvVars and CaBundle (multiple PEM certs) but no bulk data.

Oversized requests now return 413 Request Entity Too Large via a dedicated *http.MaxBytesError branch; all other read errors keep the existing 400 path.

Changes

Single file, packages/envd/internal/api/init.go (+17/-2 lines):

Change Detail
r.Body = http.MaxBytesReader(w, r.Body, 1<<20) Cap before io.ReadAll
errors.As(err, &maxErr) branch Return 413 on oversized body
Existing 400 path preserved All other read errors unchanged

No import changes needed (errors, io, net/http already imported).

Testing

  • Normal orchestrator /init payloads are well under 1 MiB — no behaviour change for legitimate callers.
  • A request with a body > 1 MiB now gets 413 and envd memory is unaffected.

PostInit read the request body with an unbounded io.ReadAll. Any process
inside the VM that can reach the envd HTTP port could send an arbitrarily
large body, allocating heap memory until envd is OOM-killed. envd is the
sandbox control plane (file I/O, process management, cgroup control,
live-upgrade handover); an OOM kill leaves user processes orphaned and
breaks graceful sandbox teardown.

The largest legitimate /init payload contains EnvVars and a CaBundle but
no bulk data; 1 MiB is a generous upper bound no real orchestrator payload
approaches. Oversized requests now return 413 instead of silently consuming
memory.

The cap also preserves the intent of memguard.WipeBytes: capping the body
ensures the secret-bearing allocation is small enough to wipe reliably
without racing a GC page-out on a huge heap allocation.

Fixes: e2b-dev#3563

@leonmeijer leonmeijer left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The new 1 MiB limit rejects an envd init payload that is still valid under the public sandbox contract: spec/openapi.yml does not cap envVars, while the orchestrator serializes the complete env map and CA bundle into POST /init and treats a 413 as a fatal create or resume failure. Add an equivalent admission limit at the public API boundary or choose a bound that is already guaranteed there so a successfully admitted sandbox cannot fail only at envd initialization.

This envd behavior change also leaves packages/envd/pkg/version.go at 0.6.13. Without the required version bump, X-Envd-Version and compatibility or rollout logic cannot distinguish nodes that enforce the new body limit.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

bug(envd): PostInit reads request body with io.ReadAll and no size limit, enabling OOM via oversized payload

3 participants