prevent 500 error on cyclic space hierarchies

Reaster commented

2026-07-23 12:40:59 +00:00

Contributor

Requesting GET /_matrix/client/v1/rooms/{roomId}/hierarchy on a space whose graph contains a cycle — for example a space that
includes itself as a child, or two spaces that reference each other — fails with:

500 Internal Server Error
{"errcode":"M_UNKNOWN","error":"Space hierarchy is unreasonably large"}

This makes the space impossible to open in clients such as Element.

get_room_hierarchy_for_user in src/service/rooms/summary/mod.rs traverses the space tree depth-first using a stack of layers
(queue), but the only deduplication set it keeps is inaccessible_children. There is no visited-set for accessible rooms, so when
the graph contains a cycle, the same room is fetched and its children pushed again on every revisit. Each revisit adds a new
layer, so the traversal depth grows without bound.

Two factors then turn this into a 500:

The client route (src/api/client/space.rs) only applies a depth limit when the client explicitly sends max_depth. Element does
not, so max_depth stays None and the depth guard never fires.
The traversal eventually trips the hardcoded queue.len() > 50 backstop, which returns a bare-string error that maps to
M_UNKNOWN / HTTP 500.

Additionally, even when max_depth is provided, a cycle currently produces duplicate room entries in the response, which violates
the spec: MSC2946 requires each room to appear at most once.

Changes

src/service/rooms/summary/mod.rs: add a visited set, seeded with the root room, and skip any room already present in the
response. This terminates cycles naturally and guarantees room uniqueness as required by the spec.
src/api/client/space.rs: default max_depth to MAX_MAX_DEPTH (10) when the client omits it, so the traversal is always
depth-bounded.
The queue.len() > 50 backstop is kept for genuinely enormous hierarchies, but now returns 413 M_TOO_LARGE instead of a 500,
since it is a property of the request target rather than a server fault.
Added a changelog entry (changelog.d/space-hierarchy-cycle.bugfix.md).

The federation hierarchy endpoint (/_matrix/federation/v1/hierarchy) is unaffected: it only summarizes a single level and does
not use this traversal.

Requesting GET /_matrix/client/v1/rooms/{roomId}/hierarchy on a space whose graph contains a cycle — for example a space that includes itself as a child, or two spaces that reference each other — fails with: 500 Internal Server Error {"errcode":"M_UNKNOWN","error":"Space hierarchy is unreasonably large"} This makes the space impossible to open in clients such as Element. get_room_hierarchy_for_user in src/service/rooms/summary/mod.rs traverses the space tree depth-first using a stack of layers (queue), but the only deduplication set it keeps is inaccessible_children. There is no visited-set for accessible rooms, so when the graph contains a cycle, the same room is fetched and its children pushed again on every revisit. Each revisit adds a new layer, so the traversal depth grows without bound. Two factors then turn this into a 500: - The client route (src/api/client/space.rs) only applies a depth limit when the client explicitly sends max_depth. Element does not, so max_depth stays None and the depth guard never fires. - The traversal eventually trips the hardcoded queue.len() > 50 backstop, which returns a bare-string error that maps to M_UNKNOWN / HTTP 500. Additionally, even when max_depth is provided, a cycle currently produces duplicate room entries in the response, which violates the spec: MSC2946 requires each room to appear at most once. Changes - src/service/rooms/summary/mod.rs: add a visited set, seeded with the root room, and skip any room already present in the response. This terminates cycles naturally and guarantees room uniqueness as required by the spec. - src/api/client/space.rs: default max_depth to MAX_MAX_DEPTH (10) when the client omits it, so the traversal is always depth-bounded. - The queue.len() > 50 backstop is kept for genuinely enormous hierarchies, but now returns 413 M_TOO_LARGE instead of a 500, since it is a property of the request target rather than a server fault. - Added a changelog entry (changelog.d/space-hierarchy-cycle.bugfix.md). The federation hierarchy endpoint (/_matrix/federation/v1/hierarchy) is unaffected: it only summarizes a single level and does not use this traversal.

Reaster added 1 commit

2026-07-23 12:40:59 +00:00

deduplicate rooms in hierarchy traversal

Documentation / Build and Deploy Documentation (pull_request) Has been skipped

Details

Auto Labeler / Apply labels based on changed files (pull_request_target) Successful in 3s

Details

Checks / Changelog / Check changelog is added (pull_request_target) Successful in 6s

Details

Checks / Prek / Check changed files (pull_request) Successful in 5s

Details

Checks / Prek / Pre-commit & Formatting (pull_request) Successful in 3m2s

Details

Checks / Prek / Clippy and Cargo Tests (pull_request) Failing after 6m37s

Details

965d861365

forgejo-actions added the

Changelog

Added

label

2026-07-23 12:41:05 +00:00

Reaster commented

2026-07-23 12:44:07 +00:00

Author

Contributor

my deadass really just went "now what happend if i try to put the same space into itself?" and pouf space bricked wonderfully haha

ginger commented

2026-07-23 13:04:52 +00:00

Owner

🤨 I was under the impression that it was both legal and expected for a space to appear multiple times in the hierarchy output if it exists multiple times in the room graph

Reaster commented

2026-07-23 13:08:45 +00:00

Author

Contributor

i don't know how it could possibly work to have the same space twice in a same space three without causing any recursion error when showing this three

Reaster commented

2026-07-23 13:10:49 +00:00

Author

Contributor

i think theres no problem to see the same space appear multiple time in a same space "group" at different "branch" but if it happend more than one on the same "branch" (aka: at one point space Y contain space Y under itself) theres no way to not have a recursion issue

ginger commented

2026-07-23 13:33:41 +00:00

Owner

I think this is an acceptable change for now to at least keep recursive space hierarchies from becoming unrecoverable

ginger commented

2026-07-23 13:34:19 +00:00

Owner

This currently fails clippy, I'll approve it once that's fixed

Reaster added 1 commit

2026-07-23 13:55:34 +00:00

make clippy happy again

Documentation / Build and Deploy Documentation (pull_request) Has been skipped

Details

Checks / Changelog / Check changelog is added (pull_request_target) Successful in 6s

Details

Checks / Prek / Check changed files (pull_request) Successful in 5s

Details

Checks / Prek / Pre-commit & Formatting (pull_request) Successful in 1m20s

Details

Checks / Prek / Clippy and Cargo Tests (pull_request) Successful in 8m51s

Details

e1726d8733

Reaster commented

2026-07-23 14:08:48 +00:00

Author

Contributor

done, it was just a one-liner issue

Reaster force-pushed self-contained-room from e1726d8733

Documentation / Build and Deploy Documentation (pull_request) Has been skipped

Details

Checks / Changelog / Check changelog is added (pull_request_target) Successful in 6s

Details

Checks / Prek / Check changed files (pull_request) Successful in 5s

Details

Checks / Prek / Pre-commit & Formatting (pull_request) Successful in 1m20s

Details

Checks / Prek / Clippy and Cargo Tests (pull_request) Successful in 8m51s

Details

to fc0b960622

Documentation / Build and Deploy Documentation (pull_request) Has been skipped

Details

Checks / Prek / Check changed files (pull_request) Successful in 5s

Details

Checks / Prek / Pre-commit & Formatting (pull_request) Successful in 56s

Details

Update flake hashes / update-flake-hashes (pull_request) Successful in 1m10s

Details

Checks / Prek / Clippy and Cargo Tests (pull_request) Successful in 8m35s

Details

Checks / Changelog / Check changelog is added (pull_request_target) Successful in 7s

Details

2026-07-23 15:07:25 +00:00

Compare

ginger merged commit a838d59cfc into main

2026-07-25 13:55:30 +00:00

ginger added this to the 26.7.0 milestone

2026-07-25 13:55:34 +00:00

ginger added the

Bug

Matrix/Client

labels

2026-07-25 13:55:39 +00:00

Rows
Columns

prevent 500 error on cyclic space hierarchies #2036