Plane's frontend calls getFileMetaDataForUpload() which uses the file-type
library to sniff MIME from magic bytes. For unsniffable formats (plain text,
.json, .csv, etc.) it returns "" — and that empty string was being threaded
through to S3Storage.generate_presigned_post(), signing the presigned URL
with `Content-Type=""`. Browsers can't reliably send an empty Content-Type
header, so the SigV4 signature never matched and R2 returned 403
SignatureDoesNotMatch. UI showed an opaque upload error.
Two-sided fix:
* apps/api/plane/settings/storage.py — default file_type to
"application/octet-stream" when empty/None. The signed URL now always has
a non-empty Content-Type the browser can match.
* packages/services/src/file/helper.ts — generateFileUploadPayload now
prefers the signed Content-Type from upload_data.fields["Content-Type"]
over file.type. The browser must send EXACTLY the signed value, not its
own MIME guess from extension. Belt-and-suspenders defense alongside the
backend default.
Reproduced empirically against R2 with the new keys 2026-05-01: empty
Content-Type signs, then PUT with `Content-Type: text/plain` returns 403
SignatureDoesNotMatch. With this patch, signing "application/octet-stream"
+ sending it back verbatim returns 200.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
== WHY (KEEP THIS — IT'S WHY THE FORK EXISTS) ==
Vanilla Plane's upload flow uses AWS S3 PostObject (presigned POST +
multipart/form-data + signed-policy-document). Cloudflare R2 AND
Backblaze B2 — the two most common self-host S3-compatible backends —
both return HTTP 501 NotImplemented for PostObject. Empirically verified
2026-04-30 against B2 s3.us-west-004.backblazeb2.com from inside Plane's
own prod api container, replicating Plane's exact boto3 call:
PUT against B2: 200 OK
POST against B2: 501 NotImplemented "This API call is not supported."
POST against R2: 501 NotImplemented (failure that started this thread)
The error code is `NotImplemented` (not `SignatureDoesNotMatch` etc),
meaning the server rejects the verb itself — no boto3 config, addressing-
style flag, or signature variant fixes it. Tested both path-style and
virtual-hosted-style URLs against B2; both fail identically for POST.
This patch rewrites the upload flow to use presigned PUT, which is
universally supported (R2, B2, AWS S3 native, MinIO, Wasabi, etc).
== WHAT (FIVE-FILE BACKEND, FIVE-FILE FRONTEND) ==
Backend:
* apps/api/plane/settings/storage.py — S3Storage.generate_presigned_post
now mints a presigned PUT URL via generate_presigned_url(HttpMethod="PUT").
Method name kept for caller compat. Response shape:
{url, method: "PUT", fields: {Content-Type, key}}.
* apps/api/plane/utils/openapi/responses.py — example response updated.
* apps/api/plane/tests/unit/settings/test_storage.py — 2 tests updated to
assert the new boto3 call.
Frontend:
* packages/types/src/file.ts — TFileSignedURLResponse.upload_data adds
optional method?: "PUT" | "POST"; drops AWS POST-form-data fields.
* packages/services/src/file/helper.ts — generateFileUploadPayload now
returns a TFileUploadRequest descriptor (url+method+body+headers) that
dispatches on method. POST branch kept for upstream parity but the
fork backend never emits POST.
* packages/services/src/file/file-upload.service.ts +
apps/web/core/services/file-upload.service.ts — uploadFile signature
changes from (url, FormData, progress?) to (payload, progress?).
* 5 caller sites updated (apps/web/core/services/file.service.ts x3,
issue_attachment.service.ts x1, sites-file.service.ts x1).
== TRADEOFFS ACCEPTED ==
* Lost: signed `content-length-range` enforcement at the storage layer.
Server-side validation in the API view still rejects oversized requests
with 413 before minting the URL, so a determined client could only
over-upload by misreporting size, capped at the bucket's own size limit.
* Different request shape on the wire (PUT with raw binary body vs POST
with multipart form). Externally invisible to users.
== ROLLBACK ==
If this becomes a maintenance nightmare:
git revert <this-commit-sha>
# rebuild + push images, swap compose tags, redeploy
After revert, uploads will only work against backends that implement
PostObject (MinIO, AWS S3 native). R2 and B2 will return 501 again.
== FULL DECISION RECORD ==
binarybeachio repo: docs/features/storage-upload-flow.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: add sync functionality for OAuth providers
- Implemented `check_sync_enabled` method to verify if sync is enabled for Google, GitHub, GitLab, and Gitea.
- Added `sync_user_data` method to update user details, including first name, last name, display name, and avatar.
- Updated configuration variables to include sync options for each provider.
- Integrated sync check into the login/signup process.
* feat: add sync toggle for OAuth providers in configuration forms
* fix: remove default value for sync options in OAuth configuration forms
* chore: delete old avatar and upload a new one
* chore: update class method
* chore: add email nullable
* refactor: streamline sync check for multiple providers and improve avatar deletion logic
* fix: ensure ENABLE_SYNC configurations default to "0" for Gitea, Github, Gitlab, and Google forms
* fix: simplify toggle switch value handling in ControllerSwitch component
---------
Co-authored-by: b-saikrishnakanth <bsaikrishnakanth97@gmail.com>
* feat: add avatar download and upload functionality in authentication adapter
- Implemented `download_and_upload_avatar` method to fetch and store user avatars from OAuth providers.
- Enhanced user data saving process to include avatar handling.
- Updated `S3Storage` class with a new `upload_file` method for direct file uploads to S3.
* feat: enhance avatar download functionality with size limit checks
- Added checks for content length before downloading avatar images to ensure they do not exceed the maximum allowed size.
- Implemented chunked downloading of avatar images to handle large files efficiently.
- Updated the upload process to return None if the upload fails, improving error handling.
* feat: improve avatar filename generation with content type handling
- Refactored avatar download logic to determine file extension based on the content type from the response headers.
- Removed redundant code for extension mapping, ensuring a cleaner implementation.
- Enhanced error handling by returning None for unsupported content types.
* fix: remove authorization header for avatar download
- Updated the avatar download logic to remove the Authorization header when token data is not present, ensuring compatibility with scenarios where authentication is not required.
* feat: add method for avatar download headers
- Introduced `get_avatar_download_headers` method to centralize header management for avatar downloads.
- Updated `download_and_upload_avatar` method to utilize the new header method, improving code clarity and maintainability.
- Update S3Storage to use configurable expiration time
- Default remains 3600 seconds (1 hour) for backward compatibility
- Add comprehensive unit tests with mocked S3 client
- Update .env.example with documentation and examples
