.gitbay/wiki/Threat-Model.org

  1#+title: gitbay threat model
  2
  3What the forge trusts, what it refuses to do, and where the boundaries
  4are. This is the reference for security review; it complements the audit
  5log and hardening notes in [[Admin]]. Diagrams, data flows, a controls
  6matrix and the open gaps are in the [[file:Architecture/00-Overview.org][Architecture]] pages.
  7
  8* What gitbay never does
  9
 10- *Execute repository content.* Git object contents are never run. Hooks
 11  are gitbay's own binary, invoked by git; they compute facts and ask the
 12  daemon over a unix socket. Repo files are only ever read.
 13- *Hold a signing key.* There is no server-side signing key. "Verified"
 14  means a signature made by a key the *user* registered — the server
 15  never vouches for a commit it did not receive already signed. Merge
 16  commits the server creates are honestly =unsigned=.
 17- *Serve repository HTML on its own origin as active content.* Raw file
 18  serving is =text/plain= with =nosniff=. Rendered markdown/org is
 19  sanitized (bluemonday) and served under a CSP that forbids scripts.
 20- *Put secrets in argv, URLs, or logs, with one documented exception.*
 21  Import and mirror credentials, registration invites, and API tokens
 22  travel on stdin or in request bodies, never as command arguments
 23  (visible in =/proc=) or query strings. The one exception is the
 24  emailed login link, =/login?token=...=: single-use, 15-minute expiry,
 25  and the response that consumes it carries =Cache-Control: no-store= so
 26  no intermediary keeps a copy. An operator running gitbay behind a
 27  reverse proxy should configure that proxy to strip the query string
 28  from its own access logs. Tokens are stored only as SHA-256 hashes.
 29- *Name a mail recipient in the log.* A queued mail is logged by its
 30  queue row id, never by address, and the relay's own error is redacted
 31  before it is logged because a rejection usually quotes the address it
 32  rejected. The unredacted error and the address stay on the row, which
 33  an instance admin reads on =/admin=: the database holds who, the log
 34  holds which. One rule for every mail type — notifications, dependency
 35  reports and login links all drain the same queue (krz/gitbay#173).
 36- *Confirm the existence of private repositories.* Every surface answers
 37  "not found" identically for a private repo and a nonexistent one — web
 38  pages, git transport, control commands, release asset downloads.
 39
 40* Trust boundaries
 41
 42- *SSH public key = identity.* The SSH username is ignored; the presented
 43  key's fingerprint resolves to an account. Key uniqueness is global.
 44- *Revocation is immediate.* Every exec and every git transport session
 45  re-reads its key. Removing a key, removing a deploy key, disabling or
 46  deleting an account closes the connections the affected keys opened:
 47  a git transport is killed with its children, and a push killed before
 48  its pre-receive hook answers moves no ref.
 49  An LFS transfer token names the key that obtained it and is refused
 50  from the moment that key is. A control command already
 51  inside its database write finishes it; its output is lost. A
 52  revocation made by =gitbayd admin= on the host, another process, is
 53  found within 15 seconds. In =ssh.mode = "system"= each exec is its
 54  own process: the next exec is refused, one already running is not cut.
 55- *Per-instance trust.* Email verification and key registration are local
 56  to an instance and never transfer. Account migration re-registers keys
 57  and re-verifies emails on the target by design.
 58- *The control plane is one command registry*, fully usable from stock
 59  OpenSSH and fronted unchanged by the JSON API and the web. No command
 60  belongs to one surface (#234): what a caller may do is the account's
 61  rights narrowed by its credential's scope, decided in one place, so a
 62  bearer token is worth exactly its scope and no more, and a token or
 63  SSH key with an expiry cannot create a credential that outlives it.
 64  A browser session can create one, or grant access, only within 15
 65  minutes of signing in (=control.ReauthWindow=, #297). Git transport
 66  never runs over the API.
 67- *Anonymous surfaces* — HTTPS clone of public repos, =git://= where
 68  enabled, the read-only web UI — carry no credentials and expose only
 69  public data. HTTP push is refused via a pkt-line =ERR=, never a 401.
 70
 71* Attacker-controlled parsers
 72
 73Every parser that eats bytes from a pusher, a key registrant, or an
 74anonymous client has a fuzz target and must never panic:
 75
 76- =internal/protocol= — the SSH command tokenizer (fuzzed against argv
 77  round-tripping).
 78- =internal/gitd= — the =git://= pkt-line reader.
 79- =internal/sig= — the commit parser, the SSHSIG armor decoder and blob
 80  parser, and the OpenPGP armored-key reader.
 81
 82Run =deploy/audit.sh= to exercise them plus =go vet= and =govulncheck=.
 83
 84* Secret handling
 85
 86Tokens (web sessions, login links, email verification, API bearer tokens,
 87deploy/CI) are random 256-bit values. Only their SHA-256 hash is stored,
 88and verification is a database index lookup on that hash — the secret
 89itself is never compared in Go, so there is no timing oracle to exploit.
 90Webhook payloads are signed outbound with HMAC-SHA256; the forge verifies
 91no inbound HMAC.
 92
 93* Network-facing request forgery
 94
 95Webhook delivery, GitHub-history import =--api-base=, mirror remotes
 96and =repo import --from=, which make the *server* open an outbound
 97connection to a user-supplied address, pass the same SSRF guard: the
 98scheme must be http/https and, unless =webhooks.allow_local= is set,
 99the resolved address must not be loopback, private, shared
100(100.64.0.0/10), link-local, or multicast. The webhook dialer re-checks
101at connect time, and mirror sync, =repo import= and =repo
102import-issues= resolve and check immediately before running git and pin
103it to the checked addresses (=internal/gitpin=), so a DNS answer that
104changes after validation still cannot reach private space;
105=import-issues= holds its API client to the API host's checked
106addresses the same way, with no proxy taken from the environment.
107Redirects are never followed.
108=repo import= refuses =git://=, which cannot be pinned. Mirror sync
109and =repo import= refuse a host written as a bare number or in
110hex/octal (=0x7f.1=, =2130706433=, =127.1=) rather than dotted
111decimal, since that form resolves differently across parsers; =repo
112mirror add= refuses it when the mirror is saved, and =repo
113import-issues= refuses it in =--api-base=.
114
115* Rendering pushed markup
116
117Rendered markup is attacker-controlled: a README, a wiki page and a
118profile's about text are all whatever someone pushed or typed. The risk
119is not only what the output contains but what the *parser* is willing to
120go and fetch — the filesystem counterpart of the SSRF guard above.
121
122Org is rendered by go-org, whose default configuration resolves
123=#+INCLUDE:= and =#+SETUPFILE:= targets with =os.ReadFile=. Both are
124refused outright (=orgConfig()= in =internal/httpd=): the file is never
125opened and the keyword stays the inert text it already was, so the rest
126of the document renders normally. There is no safe subset to allow
127instead — an absolute path skips go-org's relative-path join, a relative
128one resolves against the daemon's working directory, and the content
129came from a git object rather than a checkout, so there is no directory
130to scope a read to. Markdown is goldmark, which has no include
131mechanism. go-org's parse warnings are discarded rather than logged, so
132pushed content cannot write to the server's log.
133
134Org output is then sanitized (bluemonday UGC policy) because go-org
135passes raw HTML through — export blocks and inline export snippets —
136while goldmark drops it and needs no pass. The policy admits chroma's
137short token classes and nothing else.
138
139* Web responses
140
141Every response carries =Content-Security-Policy= (no scripts, no plugins,
142no embedding; inline styles allowed for chroma and label chips; images
143from any origin so external README images render), =X-Frame-Options:
144DENY=, =X-Content-Type-Options: nosniff=, =Referrer-Policy: no-referrer=,
145and =Strict-Transport-Security= when TLS is on. The UI needs no
146JavaScript, so =script-src 'none'= costs nothing.
147
148* The CI runner
149
150=gitbay-runner= is the one component that executes repository content.
151gitbayd never does: it reads =.gitbay/ci.yml= and queues a build, and a
152runner, polling over SSH, clones the commit and runs its steps.
153
154- *What the runner holds.* A key of scope =runner=, which the dispatcher
155  confines to =runner next=, =runner log= and =runner done= and to
156  read-only git, and which claims, logs and finishes builds only for
157  the repositories it is attached to (=repo runner add=). A step that
158  reads the key off the disk gets exactly that: it cannot administer
159  the instance, push, read a repository the runner's account cannot, or
160  touch another repository's builds. An admin key still works for the
161  runner protocol so an operator can rotate at their own pace; a runner
162  host should not hold one. Untrusted builds are skipped unless the
163  runner asks with =-untrusted=, so a runner on a user's machine never
164  executes a stranger's branch by default.
165- *What a build sees.* The commit, the =GITBAY_*= variables and the
166  repository's secrets — unless the head came from another repository.
167  A merge request from a fork is built in the target as untrusted, with
168  no secrets, so a stranger's branch cannot read the target's deploy
169  credentials. The step environment is *constructed*, not inherited: a
170  build gets =PATH=, =HOME=, =LANG=, =CI=, its own variables and its
171  secrets, and nothing the operator set on the service. =HOME= is a build
172  home under the runner's =-workdir=, not the runner's own home, so a
173  build cannot read the =.netrc=, =.npmrc= or =.gitconfig= where tools
174  keep credentials. A trusted build's home belongs to its repository
175  and persists, so caches survive; an untrusted build's home is new,
176  empty and removed when the build ends, so nothing a fork's build
177  writes is read by a later build (krz/gitbay#255). The claim names a
178  build's trust explicitly, and a runner that finds no trust flag treats
179  the build as untrusted.
180- *Where it runs.* Steps run in a rootless podman container, one per
181  job, with the workspace bind mounted and nothing else. The clone
182  happens outside it with the runner's key, so the container never sees
183  =GIT_SSH_COMMAND=, the key, or the runner's environment. =-isolation
184  none= runs steps on the host as before, for an instance where every
185  repository is trusted; there is no automatic fallback to it — a runner
186  configured for podman that cannot find one refuses to start, because
187  dropping isolation silently is worse than a stopped runner. The
188  systemd drop-in still adds =ProtectSystem=full= and the cgroup
189  protections, and =-repos= still limits a runner to named
190  repositories. =ProtectKernelTunables= is *off*: it overmounts =/proc=
191  in the unit's namespace and the kernel then refuses a proc mount in
192  any child user namespace, which every rootless container needs; the
193  container masks the same paths for the build itself.
194  =NoNewPrivileges= is *off*: rootless podman sets up its
195  namespace with the setuid =newuidmap=, which that flag blocks, so the
196  choice is between it and containers at all. Containers are the stronger
197  boundary — the flag constrained a process that was already running
198  arbitrary repository code, and under podman that code no longer runs in
199  the runner's process context. Under =-isolation none= there is no
200  container and the flag should be on.
201- *Images are provisioned by the operator, not fetched by a build.* The
202  runner passes =--pull=never=, so =image:= chooses among what the host
203  already has rather than naming anything on the internet. On an
204  instance with open registration that is the difference between a
205  curated set and arbitrary code from a registry nobody vetted.
206- *What a build can reach.* Outbound internet, trusted or not: a fork's
207  merge request to a Go repository has to fetch its modules. On the
208  runner's host, the rules below limit it to the forge's public ports
209  22, 80 and 443. Under pasta a build's container holds the host's own
210  public address, and pasta translates =169.254.1.2= (its
211  =--map-guest-addr=) to that address. A runner that polls the daemon
212  over loopback starts its containers with =--network
213  pasta:--no-map-gw=, which is podman's default stated explicitly, and
214  gives them =GITBAY_SSH= at =169.254.1.2=, with the forge's port when
215  it is not 22. The runner's source address is =127.0.0.1=. An nftables
216  table (=deploy/gitbay-runner-egress.nft=) rejects every connection
217  the runner's user makes to the host's own addresses except
218  =127.0.0.1:22=, DNS on loopback, and 22, 80 and 443 on the public
219  address: the operator's sshd on 2222 and anything bound to loopback
220  are closed to it. Under rootless podman a build's connections are
221  made by pasta as the runner's user, so the table cannot tell a build
222  from its runner and leaves =127.0.0.1:22= open; that no build reaches
223  the host's loopback is measured from inside a build by runbook R3.
224  The runner does not start without the table. The SSH auth limiter
225  counts failures per source address and, once an address is over the
226  limit, refuses every key from it until the window passes, the
227  runner's included; with registration open or by invite an unknown
228  key never counts (krz/gitbay#260). Under =-isolation none= a build
229  runs on the host and shares its loopback; the table still applies,
230  since it runs as the same user.
231
232Under =-isolation none=, anything a step can do as the runner's user a
233pushed =ci.yml= can do. Under podman a step is confined to its
234container, the bind-mounted workspace and its build home: a trusted
235build's cache is read only by later trusted builds of the same
236repository, and an untrusted build's home is discarded with it. Treat
237the runner host as executing untrusted code all the same: keep it off
238the daemon's host where the database lives, or scope it to repositories
239whose writers you trust. gitbay.org does the latter — its runner builds
240only the repositories the operator names.
241
242* What has not been audited
243
244The 2026-09 sweep (krz/gitbay#149) read this repository against the
245claims above. Its two findings — review verdicts from accounts without
246write access deciding merge gates (#147), and control commands having no
247write rate limit while the API had one (#148) — are fixed, as is #173,
248found afterwards in the same milestone. What it did not reach, and why,
249is recorded here rather than in a closed issue:
250
251- *The host.* systemd sandboxing, sshd on 2222, the firewall,
252  unattended-upgrades, fail2ban, disk and file modes, restic append-only
253  credentials. Reading production configuration is a separate exercise
254  from auditing the source, and needs doing against bay1.
255- *The supply chain beyond =govulncheck=.* No review of what the
256  dependency set is, who maintains it, or what a compromised release of
257  any of it would reach.
258- *The runner host as an execution environment.* Nobody has tried to
259  escape what is there. #144 covers the missing isolation.
260- *Timing and traffic analysis.* Token comparison is a hash index lookup
261  by design, but nothing has been measured.
262- *Denial of service by resource exhaustion* beyond rate. Concurrent
263  clones, fetches and web archives are bounded by the pack limit
264  (#262); pushes are not, beyond =max_pack_bytes= on each one. Nor are
265  pathological diffs, deep histories, or zip bombs in LFS.
266
267A sweep is a point in time. This section says what a reader should not
268assume has been checked.
269
270* Residual risks, accepted
271
272- External images in rendered READMEs and profile about text load from
273  their origin (no image proxy), which the author can use as a tracking
274  pixel against a viewer. A profile is the wider surface of the two: it
275  is linked from every commit and issue its owner touches. Documented;
276  proxying is future work.
277- Backups snapshot the database first, and repository deletes and
278  moves wait out a full backup. Each repository's refs are archived
279  before its objects, so every archived ref finds the objects it
280  reaches, unless git's own automatic gc after a push repacks during
281  the walk: the archive can then miss objects, and =--verify= reports
282  it. A push during a backup may be missing from the archive, or
283  present as objects no archived ref names, and a repository's refs may
284  be newer than the database snapshot (see [[Admin]]).
285- The audit log lives in the database the daemon writes, so anyone with
286  the daemon user's access can change it. The hash chain is unkeyed:
287  whoever can write the database can edit a row and recompute every
288  later hash. =gitbayd admin audit verify= catches an edited or removed
289  row only when the later hashes were not recomputed, and never catches
290  removing the newest rows or writing new rows under their freed ids.
291  Comparing verify's last id and hash with the daemon's journal copy is
292  the check for any change; rows written outside the daemon (=gitbayd
293  shell= under =ssh.mode = "system"=, host =gitbayd admin= commands)
294  have no journal copy.
295- A global signature-verification epoch over-invalidates the cache on any
296  trust-input change. Correct, not a leak; a performance tradeoff.
297- A build's secrets are environment variables inside its container, so
298  they are visible to =podman inspect= as the runner's user — the same
299  user that already holds them in memory. They reach podman through a
300  0600 env file rather than argv, since =/proc= is world-readable. A
301  value with a newline in it — a private key — cannot go in that file;
302  it is named on podman's command line with =--env NAME= and valued in
303  the runner-owned podman process's environment, so it never touches
304  argv either.