Repository navigation
Hall of Fame
Lauri Ojansivu edited this page Nov 8, 2026
·
3 revisions
Security issues found in, and fixed in, caddy-gitea-plugin - Gitea inside Caddy, part of Weddy. The same columns as WeKan's Hall of Fame. Report a vulnerability privately as described in WeKan's SECURITY.md; stars follow WeKan's star audit. Findings of automatic scanners have no stars.
The code is upstream Gitea's (gitea/, made by releases/sync-gitea.sh), so a fix is a
plugin patch in dist/gitea/ - here security-hardening - and worth sending upstream
too. Inside Weddy, Gitea shares its process with Caddy and WeKan: a request that
exhausts memory stops all three.
| CVE | Icon | Vulnerability name | Date | Responsible Security Disclosure by | Stars | Process | Vulnerabilities |
|---|---|---|---|---|---|---|---|
| - | 📦🧠🩸 | ContainerTagBleed | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alert #39 (go/uncontrolled-allocation-size) at gitea/models/packages/container/search.go; the other 20 alerts of this kind were checked one by one (below). |
DetailsThe container registry's tag list,/v2/<image>/tags/list?n=..., reserved its result slice with the request's n as the capacity: n=1000000000 asked for gigabytes before reading one row - for a public package without signing in - and inside Weddy that stops Caddy and WeKan with Gitea. The capacity is now at most 100; n still limits the query. Fix: commit. |
| - | 🔢↩️🩸 | WrapBleed | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #32, #33 and #34 (go/incorrect-integer-conversion) at gitea/services/context/api.go, pagination.go and gitea/services/migrations/github.go. |
Detailsint64 totals in the pagination and an int64 issue number in the GitHub migration became int, which wraps on the 32-bit platforms Weddy is built for (linux/386, arm); upstream's own min(total, math.MaxInt) (#33) is no bound there. Each is now bounded by math.MaxInt32 with an explicit comparison, and an issue number outside 1..MaxInt32 is refused. Fix: commit. |
| - | 📝🔑🩸 | AccessLogBleed | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #26 and #27 (go/clear-text-logging) at gitea/services/context/access_log.go. #24 and #25, at the general log wrappers, are not (below). |
DetailsThe access log template was given the request itself, so a token in the query string (?token=, ?access_token=) or an Authorization or Cookie header could be written to the access log with a custom template. It now gets a copy with the token, access_token, private_token and password parameters and the Authorization, Cookie, X-Gitea-OTP and Proxy-Authorization headers replaced by [REDACTED]; the request is unchanged. TestRedactedRequest pins it. Fix: commit. |
| - | 🧠🛡️ | Bounded allocations (not a vulnerability) | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #38, #40-#58 (go/uncontrolled-allocation-size): reviewed, to be dismissed as false positives. |
DetailsEach slice's capacity is a page size that is already bounded before it gets there: the API'slimit by convert.ToCorrectPageSize to MAX_RESPONSE_ITEMS, the LFS locks' limit by LOCKS_PAGING_NUM, the container catalog's n to 100 in its handler, and the web pages' sizes by their settings - or the length of a list already in memory. |
| - | 🌐🛡️ | Escaped output (not a vulnerability) | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #1-#5 (go/html-template-escaping-bypass, go/reflected-xss): reviewed, to be dismissed as false positives. |
Details#1-#4 are Gitea's response writer wrappers (Write, the page renderer), not a place request data is written unescaped; pages are html/template. #5 is the mail preview under /devtest, which Gitea registers only when it is not in production mode (!setting.IsProd). |
| - | 📁🛡️ | Checked paths (not a vulnerability) | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #6-#23 (go/path-injection): reviewed, to be dismissed as false positives. |
DetailsThe file session store builds its path from the session cookie only after isValidSessionID allowed exactly 16 characters of 0-9 and a-z. Git hook names are fixed by Gitea and written by admins; temporary directories and file retries take paths Gitea made itself; restoring a repository dump reads a path an administrator gave on the command line. |
| - | 🗄️↪️⚙️🔒🛡️ | Parameterised queries, a checked redirect, argument lists, a local TLS check (not vulnerabilities) | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #24, #25 (go/clear-text-logging), #28-#31 (go/sql-injection), #35 (go/unvalidated-url-redirection), #36 (go/command-injection) and #37 (go/disabled-certificate-check): reviewed, to be dismissed as false positives or by design. |
Details#24, #25: the general log wrappers, which upstream already marks for CodeQL. #28-#31: xorm builder conditions and constant column lists with? placeholders. #35: the redirect is checked by IsCurrentGiteaSiteURL first. #36: commands are run with an argument list, never a shell. #37: the internal API skips certificate verification only when LOCAL_ROOT_URL is localhost or a loopback address - Gitea talking to itself. |
| - | #️⃣🛡️ | Hashes that are not password storage (not a vulnerability) | 2026-11-08 | GitHub CodeQL and xet7 | - | ProcessCode scanning alerts #60-#63 (go/weak-sensitive-data-hashing), closed the way WeKan closed its #599: by the code. |
Details#60: SHA-1 is what Have I Been Pwned's k-anonymity range API takes - only the first five hex digits leave the server. #61: a general SHA-256 helper. #62, #63: SHA-256 turns SECRET_KEY into a 32-byte AES key; changing it would make every stored secret unreadable. Passwords are stored with Gitea's slow hashes (pbkdf2, argon2, scrypt, bcrypt). The three are now computed through the crypto package's registry, saying what they are, with the same bytes. Fix: commit. |