fe9c4f49 keyed the ETag on the SOURCE file's content hash. That is wrong
whenever the response comes from a satellite table, and for attachments
it is wrong in two ways.
Uploading a preview does not change the file's content, so a
source-keyed ETag does not change either — and with `immutable` set,
clients never revalidate and keep the previous render for up to a year.
The exact staleness fe9c4f49 set out to fix, re-entering through the
attachment path.
Worse: a copy inherits the source hash, so an original and a copy have
identical ETags. Give either one a different uploaded preview and they
serve different bytes under one validator, which a shared cache may hand
to either request. That is a collision, not just staleness.
thumbnail_content_id resolves the identity through the same tier
precedence the read path uses: an attached blob's own hash, else a
derived blob's own hash, else the source-keyed form. An ETag naming a
different tier than the one answering is worse than a coarse one, so the
two orders must not drift.
Derived-hash keying is strictly better than source-keying and never
worse. The sidecar and the derived row are written from the same bytes;
where they can diverge — a sidecar re-rendered while the derived row
stays pinned by ON CONFLICT DO NOTHING — source-keying is wrong too,
because the renderer is not part of that key. This is the step 10 change
arriving early, forced by the attachment case; the plan note stands for
the read-order flip itself.
Known gap: a legacy ext-{file_id}.jpg with no file_attached_blobs row
yet falls through to the source-keyed form. No worse than today, and it
resolves when the import backfills.
attached_thumbnail_copy.hurl now asserts ETags, which is why this went
unnoticed: it compared bytes only, and thumbnail_etag_content_keyed
covers content replacement rather than preview upload. A fresh GET
returned the right bytes throughout — the same "healthy locally, broken
for anyone caching" shape as the two bugs before it.
The disk check proves no BYTES are left. This proves no ROWS are, which
fails differently and worse: a stale storage.blobs row with nothing
behind it means a reference was never released, and dedup_gc will skip
it forever because its count never reaches zero. Silent, permanent, and
invisible to a check that only looks at the filesystem.
Zero is the right assertion, not "fewer than before". By this point the
suite has deleted its users, their drives and everything cascading
beneath, and the disk check has already insisted the blob store is
empty. A non-zero registry beside an empty disk is exactly the
divergence the consistency jobs report — caught here first because one
number is easier to read than a findings list.
Degrades to a warning if the endpoint is unavailable rather than
failing, so a build without the admin dedup surface still runs the rest.
The disk checks above prove nothing leaked. These prove the bookkeeping
behind them is honest: every refcount matches what the reference sources
hold, and no row points at bytes that are gone.
End of suite is the only place this is cheap. One database serves every
hurl file, so by here the counters have absorbed every upload, copy,
move, share, trash and purge the suite performed — across both copy
paths, the derived tier and the attached tier. Drift that no individual
test would notice, because each only inspects its own file, surfaces as a
mismatch.
Runs after the GC drain deliberately: mid-sweep state is legitimately
inconsistent — a manifest can sit at zero waiting for the next pass — so
checking earlier would report normal in-flight state as drift.
Zero findings is the assertion. These four tenants are read-only, so
anything they report is a real invariant violation rather than a repair
opportunity. A job missing from the build is skipped with a warning
instead of failing, so this does not break on a feature-gated build.
Unknown job names and unwrapped-vs-wrapped response shapes both degrade
to a visible warning rather than a silent pass: list_job_runs currently
returns a bare array, and the .runs/.items fallbacks exist so a future
wrapping does not quietly turn the whole check into a no-op.
Completes step 9. The PUT wrote `ext-{file_id}.jpg` and nothing else —
keyed by file id, on local disk. No copy path duplicates it and no other
instance can see it, so a copied file lost the preview its owner
uploaded. Silently: the server falls back to rendering one from the
source, or to 204 for a PDF, which has no render path at all. A
user-supplied preview is not derivable from the content, so once lost it
is gone.
The PUT now also records a storage.file_attached_blobs row, which
copy_file_satellites already duplicates, so both copy paths carry it.
Best-effort: the sidecar has already succeeded by then and the user can
see their thumbnail, so failing the request would report an error for an
operation that visibly worked.
Read path consults attachments ahead of every content-derived tier: an
uploaded preview is an explicit choice about THIS file and must beat
anything rendered from its content. Cached under the per-file key — a
content key would leak those bytes to every other file sharing the
content, which is the poisoning the file-keyed table exists to prevent.
store_attached_blob is ON CONFLICT DO UPDATE, unlike its derived twin:
re-uploading a preview is a deliberate replacement, where a re-derived
thumbnail is the same bytes again. The superseded blob's reference is
released, or it would be pinned forever with nothing pointing at it.
Deletion goes through a trigger, not a hook. file_id is ON DELETE
CASCADE, and on_file_deleted fires AFTER delete_file — by then the
cascade has run and there is nothing left to enumerate. This matters
most for folder deletion, where PG cascades folders to files to
attachments and Rust never sees the rows at all. storage.decrement_blob_ref
keys off OLD.blob_hash and is otherwise table-agnostic, so it is reused
verbatim rather than transcribed into a second trigger that can drift.
DELETE only: a replacement updates in place and is handled in Rust, so
adding UPDATE would double-decrement.
Extracted read_blob_to_bytes, shared by the attached and derived tiers —
the only difference between them is which table produced the hash.
tests/api/attached_thumbnail_copy.hurl guards it. The file is red and
the uploaded thumbnail is green, so a render could never produce the
uploaded bytes; the pre-upload render is captured first and required to
change, which stops three identical renders from satisfying the
byte-equality. Then both copy paths must serve the upload, and after the
original is purged and GC runs, both copies must still serve it — each
holds its own reference, because the rows are duplicated rather than
shared.
GET /api/files/{id} is the download route — it returned the PNG bytes,
so the jsonpath capture failed on a UTF-8 decode. /{id}/metadata is the
EXIF endpoint and carries no FileDto either. Listing the folder gives
the DTO, and since the folder holds exactly this one file, count == 1
also proves the WebDAV PUT overwrote in place rather than creating a
second file beside it.
Also drops an unused bytes capture and records why the body is not
asserted after the overwrite: the moka tier is keyed on file_id and
invalidated from the spawned task in on_file_updated, so a request
landing first sees the previous bytes under the new ETag. Asserting on
bytes would be a race.
The upload omitted folder_id, which the handler needs to resolve the
file's owner — it answers 500, not a root upload. Every other upload in
the suite passes it; this was the only one that did not, which is why
nothing caught it earlier.
The folder also gives the WebDAV overwrite a deterministic path
(/webdav/hurl-etag-src/<name>) instead of depending on where a
folder-less upload would have landed. Teardown now removes it and
purges it from trash, keeping the shared database clean for the files
that run after.
The thumbnail ETag was "thumb-{file_id}-{size}-{format}", sent with
Cache-Control: public, max-age=31536000, immutable. Replacing a file's
content preserves its id — file_upload_service rebuilds the entity with
parts.id and a new hash, then fires on_file_updated, which deletes and
regenerates the thumbnails — so the server produced a new thumbnail while
still advertising the old ETag. Because `immutable` tells a conforming
browser not to revalidate at all inside the freshness window, clients kept
rendering the previous image for up to a year, unfixably.
Keyed on the content hash the directive becomes honest: a thumbnail is a
pure function of (source bytes, size, format), so that triple identifies
the response. New content yields a new ETag.
The same change fixes the opposite direction. A copy, or any dedup twin,
had a different id and therefore a different ETag, so clients refetched
bytes they already held even though both are served from the same derived
blob. Now identical content agrees on an ETag and revalidates to 304
across files, users and copies.
Both thumbnail endpoints were affected: the REST handler and the
NextCloud preview handler.
Cost is one PK lookup ahead of the 304 decision, where the id-keyed
version needed none — paid for by no longer serving stale images. It is
partly recovered: both handlers already resolved the same hash further
down for the render path, and that second lookup is now gone, so the
cache-miss path is unchanged and only the 304 path pays. The resolved
hash is also handed to get_cached_thumbnail instead of None, saving the
service its own lookup.
No new disclosure: content_hash is already on FileDto and returned by
GET /api/files/{id}.
Tests: thumbnail_etag_content_keyed.hurl covers invalidation — overwrite
in place via WebDAV PUT, assert the ETag changed, assert a client holding
the stale one gets 200 rather than 304. derived_blob_copy.hurl gains the
sharing direction: a copy answers with the SAME ETag and revalidates to
304, which is the one externally observable consequence of content-keying
and was not previously testable.
The byte-identity assertions were documented as proving that a copy
shares the original's content_derived_blobs row. They prove no such
thing: rendering is deterministic in the source bytes and the variant,
so a copy that re-rendered from scratch returns identical bytes. The
copy is in fact a moka hit — that cache is keyed on
(source_hash, size, format), which the copy shares — so it never
reaches the derived tier here at all.
Nor is there an assertion that would fix it. Duplication is impossible
by construction: the PK is (source_hash, kind, variant), a copy carries
the same source_hash, and store_derived_blob is ON CONFLICT DO NOTHING.
The schema enforces the property, so no runtime behaviour can violate
it and there is nothing to catch.
Same limitation narrows step 11: it proves the SOURCE content survived
GC, not the derived blob — a reaped derived blob is re-rendered
transparently from the live source.
What the file does prove is unchanged and is the part that was broken:
both copy paths take a real blob reference (ref_count 1 -> 2 -> 3), and
purging the original does not destroy the copies. No assertions changed.
Step 8 of docs/plan/derived-blobs.md. "What follows a file on copy" was
written twice — the copy_file CTE and storage.copy_folder_tree — and had
already drifted: the tree path bumped storage.blobs only, missing
manifests, which was silent data loss on any multi-chunk file. Fixing it
meant writing the same logic a second time. Step 9 adds a file-keyed
satellite table, which would mean a third and fourth.
Two SQL functions:
storage.add_blob_references(TEXT[]) — the manifest-first reference
contract for SQL callers, returning hashes that matched no registry
row. Set-based so the tree path keeps its single-statement cost; a
per-row helper would have made a 10k-file copy 10k calls.
storage.copy_file_satellites(UUID[], UUID[]) — dead properties plus
the blob reference. The body is the copy-semantics declaration: what
is absent (comments, favorites, content-keyed derived rows) is listed
with its reason, so the taxonomy is executable rather than documented
elsewhere and drifting.
Both copy paths now call it. The single-file path becomes a real
transaction, which also fixes the reference being best-effort: a failed
add_reference used to log a warning and leave a copy holding no
reference at all — the exact shape that gets its content reaped. It
cannot be a CTE arm, because data-modifying CTEs share one snapshot and
the function must read the row the INSERT just wrote.
Verified against a scratch PG with all migrations applied: multi-chunk
manifest 1→2, single-chunk alias bumped at manifest level only (the
NOT EXISTS guard), chunks behind a manifest untouched, dead properties
duplicated, length mismatch rejected, repeats counted.
tests/api/derived_blob_copy.hurl covers it end-to-end and answers the
question the copy raises: content_derived_blobs is NOT copied. A copy
carries the same blob_hash, so it resolves the same derived row — the
test asserts byte-identical thumbnails from both copy paths, then
deletes the original, runs GC, and requires both copies to still serve.
That last step only passes if the references are real.
One dedup_gc pass cannot fully drain now that thumbnails are derived
blobs. Reaping a source releases the references its derived artifacts
hold (each content_derived_blobs row pins a manifest), and those
releases happen mid-sweep — the derived chunks are stamped orphaned as
the pass is already walking past them, because
remove_manifest_reference deliberately does not unlink, to avoid racing
a concurrent upload re-referencing the same chunk. They are collectible
only on the NEXT sweep, which is why the check saw 15 leftover blobs.
Loops until a pass reclaims nothing rather than hardcoding two. Two is
correct only while the derivation graph is one level deep — a thumbnail
is derived from a file, nothing is derived from a thumbnail. That is a
property of the data, not an invariant the code enforces, so a fixed
count would silently under-drain the day transcodes-of-thumbnails or
E2E-wrapped derivatives exist, and the failure would surface as a
confusing leftover-file assertion rather than the design change it is.
Bounded at 3 with a warning if it does not settle.
Sleeps between passes. The JobRegistry serialises runs of the same job,
so a back-to-back trigger risks rejection as already-running — which
returns 0 reaped and would exit the loop early, declaring success with
blobs still on disk. A false pass is worse than a slow one. It also
gives the previous pass's detached unlink tasks (spawned by
on_blob_deleted, awaited by nothing) time to land.
Deliberately NOT fixed in production code: derived chunks land inside
the 1-hour orphan grace, so a second immediate sweep would collect
nothing there and the next scheduled run picks them up. A fixpoint loop
in garbage_collect would be dead code outside force=true, which is only
this test.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The 'broken pipe' error under set -euo pipefail was masking the real
answer. Capture status + content-type + first-char in one curl, dump
first 200 bytes of body on failure so we can see what the server
actually served.
link are checking that email matches, +email alias are normalize into email
if email is already used on another account, link is not possible
not usurpation risk as the IDP is choosen by the admin
this make OIDC compliant with the invariant binding (issuer and subject)
admin can now rename their provider without breaking
clarifing federation_kind: report the kind of federation wired not the allowed login method
hybryd login method are still allowed
- POST /api/auth/opaque/login/ke1 (public) — takes {userIdentifier, startLoginRequest}, resolves the identifier via the same @-dispatch as legacy login (AuthApplicationService::lookup_user_for_login, factored out), fetches envelope, runs ServerLogin::start (real branch for known users, dummy branch for anti-enum on unknown/unregistered), stashes state under a random exchange_id in the moka cache, returns {exchangeId, loginResponse}.
- POST /api/auth/opaque/login/ke3 (public) — atomic take from the cache FIRST (anti-enum + anti-replay), then decodes the payload, runs ServerLogin::finish, stamps opaque_migrated_at (Phase 3 signal), and mints a session via the new AuthApplicationService::mint_session_for_authenticated_user helper — returns the same AuthResponseDto shape as legacy login so the SPA has one downstream handler.
- Session mint factored: mint_session_for_authenticated_user(User) extracted from login() so both the legacy password path and OPAQUE KE3 converge through one implementation.
- OpaqueRepositoryPort::mark_migrated with COALESCE-preserving idempotent stamp of opaque_migrated_at.
- opaque-setup CLI + Dockerfile wiring already shipped (Step 0 hygiene).
- Routing fix: sub-prefix split (/api/auth/opaque/register vs /api/auth/opaque/login) — axum composes middleware between sibling nests at the same prefix, which was cross-applying auth+CSRF to my public login routes. Distinct prefixes side-step that cleanly. Documented in both main.rs and the router builder doc.
- Rate-limit sharing: login KE1/KE3 layered with the SAME login_limiter instance as legacy POST /api/auth/login, so an attacker can't halve the per-IP budget by spraying both endpoints.
Anti-enum + anti-replay hardening in KE3: take runs BEFORE payload parse so:
- Unknown / expired / already-consumed exchange_id → 401 InvalidCredentials (same shape as wrong-passphrase, no payload-shape leak)
- Consumed handle can't be re-used to spam parse attempts
OPAQUE (RFC 9807) implementation (using `opaque-ke` crate)
with opaque authentfication, server will never receive the password (in the auth=password mode)
this is a must have to create trust with users to permit end to end encryption in the future
(we cannot know if user use the same password/passphrase for his asymetric key or his oxicloud auth,
this is why server must never have the password)
pass1: prepare server
Edouard Vanbelle