@@ -0,0 +1,260 @@ |
| 1 | +// SPDX-License-Identifier: AGPL-3.0-or-later |
| 2 | + |
| 3 | +package githttp |
| 4 | + |
| 5 | +import ( |
| 6 | + "io" |
| 7 | + "net/http" |
| 8 | + "os" |
| 9 | + "strconv" |
| 10 | + "strings" |
| 11 | + |
| 12 | + "github.com/go-chi/chi/v5" |
| 13 | + "github.com/jackc/pgx/v5/pgtype" |
| 14 | + |
| 15 | + "github.com/tenseleyFlow/shithub/internal/git/protocol" |
| 16 | + reposdb "github.com/tenseleyFlow/shithub/internal/repos/sqlc" |
| 17 | + "github.com/tenseleyFlow/shithub/internal/web/middleware" |
| 18 | +) |
| 19 | + |
| 20 | +// MountSmartHTTP registers the smart-HTTP routes on r. Caller is |
| 21 | +// responsible for placing this group ahead of any conflicting routes |
| 22 | +// (specifically, register before the /{owner}/{repo} two-segment route |
| 23 | +// from S11 — `.git` URLs would otherwise be eaten by it). |
| 24 | +// |
| 25 | +// Caller is also responsible for stripping CSRF, response-compression, |
| 26 | +// and request-timeout middleware from this group; see web/server.go. |
| 27 | +func (h *Handlers) MountSmartHTTP(r chi.Router) { |
| 28 | + r.Get("/{owner}/{repo}.git/info/refs", h.infoRefs) |
| 29 | + r.Post("/{owner}/{repo}.git/git-upload-pack", h.uploadPack) |
| 30 | + r.Post("/{owner}/{repo}.git/git-receive-pack", h.receivePack) |
| 31 | +} |
| 32 | + |
| 33 | +// infoRefs handles GET /{owner}/{repo}.git/info/refs?service=git-... . |
| 34 | +// |
| 35 | +// Per the smart-HTTP protocol the response body has shape: |
| 36 | +// |
| 37 | +// 001e# service=git-<svc>\n |
| 38 | +// 0000 |
| 39 | +// <advertise-refs output from git> |
| 40 | +// |
| 41 | +// We stream the trailing advertise-refs body straight from the |
| 42 | +// subprocess so a huge ref set doesn't buffer in memory. |
| 43 | +func (h *Handlers) infoRefs(w http.ResponseWriter, r *http.Request) { |
| 44 | + svc, ok := serviceFromQuery(r.URL.Query().Get("service")) |
| 45 | + if !ok { |
| 46 | + http.Error(w, "service query parameter required", http.StatusBadRequest) |
| 47 | + return |
| 48 | + } |
| 49 | + |
| 50 | + row, allow := h.authorizeForService(w, r, svc) |
| 51 | + if !allow { |
| 52 | + return |
| 53 | + } |
| 54 | + |
| 55 | + gitDir, err := h.d.RepoFS.RepoPath(chi.URLParam(r, "owner"), row.Name) |
| 56 | + if err != nil { |
| 57 | + h.d.Logger.ErrorContext(r.Context(), "githttp: path", "error", err) |
| 58 | + http.Error(w, "not found", http.StatusNotFound) |
| 59 | + return |
| 60 | + } |
| 61 | + |
| 62 | + w.Header().Set("Content-Type", "application/x-"+string(svc)+"-advertisement") |
| 63 | + setNoCacheHeaders(w) |
| 64 | + |
| 65 | + if err := protocol.WriteServiceAdvertisement(w, string(svc)); err != nil { |
| 66 | + h.d.Logger.ErrorContext(r.Context(), "githttp: pkt", "error", err) |
| 67 | + return |
| 68 | + } |
| 69 | + cmd := protocol.Cmd(r.Context(), svc, gitDir, true, nil) |
| 70 | + cmd.Stdout = w |
| 71 | + stderr := protocol.DrainStderr(cmd) |
| 72 | + if err := cmd.Run(); err != nil { |
| 73 | + h.d.Logger.ErrorContext(r.Context(), "githttp: info/refs", |
| 74 | + "error", err, "service", svc, "stderr", string(stderr())) |
| 75 | + } |
| 76 | +} |
| 77 | + |
| 78 | +// uploadPack handles POST /{owner}/{repo}.git/git-upload-pack. |
| 79 | +// Streams: req.Body → git-upload-pack stdin; git-upload-pack stdout → w. |
| 80 | +func (h *Handlers) uploadPack(w http.ResponseWriter, r *http.Request) { |
| 81 | + h.runPack(w, r, protocol.UploadPack) |
| 82 | +} |
| 83 | + |
| 84 | +// receivePack handles POST /{owner}/{repo}.git/git-receive-pack. |
| 85 | +// Same streaming shape as uploadPack but with the SHITHUB_* env vars |
| 86 | +// set so the post-receive hook (S14) can identify the actor. |
| 87 | +func (h *Handlers) receivePack(w http.ResponseWriter, r *http.Request) { |
| 88 | + h.runPack(w, r, protocol.ReceivePack) |
| 89 | +} |
| 90 | + |
| 91 | +// runPack is the shared body for both POST endpoints. |
| 92 | +func (h *Handlers) runPack(w http.ResponseWriter, r *http.Request, svc protocol.Service) { |
| 93 | + row, allow := h.authorizeForService(w, r, svc) |
| 94 | + if !allow { |
| 95 | + return |
| 96 | + } |
| 97 | + |
| 98 | + gitDir, err := h.d.RepoFS.RepoPath(chi.URLParam(r, "owner"), row.Name) |
| 99 | + if err != nil { |
| 100 | + h.d.Logger.ErrorContext(r.Context(), "githttp: path", "error", err) |
| 101 | + http.Error(w, "not found", http.StatusNotFound) |
| 102 | + return |
| 103 | + } |
| 104 | + |
| 105 | + // Hard cap on the request body. A 2 GiB+1 push reads up to the cap |
| 106 | + // from req.Body and then errors; we surface the read error from |
| 107 | + // the subprocess as a 413 IF we haven't started writing yet. |
| 108 | + body := http.MaxBytesReader(w, r.Body, h.d.MaxPushBytes) |
| 109 | + defer func() { _ = body.Close() }() |
| 110 | + |
| 111 | + w.Header().Set("Content-Type", "application/x-"+string(svc)+"-result") |
| 112 | + setNoCacheHeaders(w) |
| 113 | + |
| 114 | + var env []string |
| 115 | + if svc == protocol.ReceivePack { |
| 116 | + env = h.hookEnv(r, row) |
| 117 | + } |
| 118 | + cmd := protocol.Cmd(r.Context(), svc, gitDir, false, env) |
| 119 | + cmd.Stdin = body |
| 120 | + cmd.Stdout = w |
| 121 | + stderr := protocol.DrainStderr(cmd) |
| 122 | + if err := cmd.Run(); err != nil { |
| 123 | + // At this point we may have already written headers + bytes; |
| 124 | + // surfacing 413/500 cleanly is best-effort. |
| 125 | + h.d.Logger.ErrorContext(r.Context(), "githttp: pack", |
| 126 | + "error", err, "service", svc, "stderr", string(stderr())) |
| 127 | + } |
| 128 | +} |
| 129 | + |
| 130 | +// authorizeForService resolves the repo + checks visibility/permission. |
| 131 | +// Returns the repo row + allow=true on success. On any failure writes |
| 132 | +// the appropriate response (404, 401, 403, 410) and returns allow=false. |
| 133 | +func (h *Handlers) authorizeForService(w http.ResponseWriter, r *http.Request, svc protocol.Service) (reposdb.Repo, bool) { |
| 134 | + ownerName := chi.URLParam(r, "owner") |
| 135 | + repoName := strings.TrimSuffix(chi.URLParam(r, "repo"), ".git") |
| 136 | + // chi already strips .git for the URL pattern but we defensively |
| 137 | + // trim again — the route is /{owner}/{repo}.git/info/refs, where |
| 138 | + // chi captures `{repo}` WITHOUT the `.git` suffix. |
| 139 | + |
| 140 | + owner, err := h.uq.GetUserByUsername(r.Context(), h.d.Pool, ownerName) |
| 141 | + if err != nil { |
| 142 | + http.Error(w, "not found", http.StatusNotFound) |
| 143 | + return reposdb.Repo{}, false |
| 144 | + } |
| 145 | + row, err := h.rq.GetRepoByOwnerUserAndName(r.Context(), h.d.Pool, reposdb.GetRepoByOwnerUserAndNameParams{ |
| 146 | + OwnerUserID: pgtype.Int8{Int64: owner.ID, Valid: true}, |
| 147 | + Name: repoName, |
| 148 | + }) |
| 149 | + if err != nil { |
| 150 | + http.Error(w, "not found", http.StatusNotFound) |
| 151 | + return reposdb.Repo{}, false |
| 152 | + } |
| 153 | + |
| 154 | + auth, authErr := h.resolveBasicAuth(r.Context(), r.Header.Get("Authorization")) |
| 155 | + requireAuth := svc == protocol.ReceivePack || row.Visibility == reposdb.RepoVisibilityPrivate |
| 156 | + if authErr != nil || (requireAuth && auth.Anonymous) { |
| 157 | + writeChallenge(w) |
| 158 | + return reposdb.Repo{}, false |
| 159 | + } |
| 160 | + |
| 161 | + // Permission check — inline; S15 replaces this with the policy package. |
| 162 | + // V1: only the owner can read a private repo or write to any repo. |
| 163 | + if !auth.Anonymous && auth.UserID != owner.ID { |
| 164 | + http.Error(w, "forbidden", http.StatusForbidden) |
| 165 | + return reposdb.Repo{}, false |
| 166 | + } |
| 167 | + if svc == protocol.ReceivePack { |
| 168 | + if row.IsArchived { |
| 169 | + writeGitErrorMessage(w, http.StatusForbidden, |
| 170 | + "repository is archived; pushes are disabled") |
| 171 | + return reposdb.Repo{}, false |
| 172 | + } |
| 173 | + if row.DeletedAt.Valid { |
| 174 | + http.Error(w, "gone", http.StatusGone) |
| 175 | + return reposdb.Repo{}, false |
| 176 | + } |
| 177 | + } |
| 178 | + return row, true |
| 179 | +} |
| 180 | + |
| 181 | +// hookEnv assembles the SHITHUB_* env vars to thread through git- |
| 182 | +// receive-pack into S14's hooks. |
| 183 | +func (h *Handlers) hookEnv(r *http.Request, row reposdb.Repo) []string { |
| 184 | + auth, _ := h.resolveBasicAuth(r.Context(), r.Header.Get("Authorization")) |
| 185 | + owner, err := h.uq.GetUserByID(r.Context(), h.d.Pool, ownerIDFromRow(row)) |
| 186 | + ownerName := "" |
| 187 | + if err == nil { |
| 188 | + ownerName = owner.Username |
| 189 | + } |
| 190 | + return []string{ |
| 191 | + "SHITHUB_USER_ID=" + strconv.FormatInt(auth.UserID, 10), |
| 192 | + "SHITHUB_USERNAME=" + auth.Username, |
| 193 | + "SHITHUB_REPO_ID=" + strconv.FormatInt(row.ID, 10), |
| 194 | + "SHITHUB_REPO_FULL_NAME=" + ownerName + "/" + row.Name, |
| 195 | + "SHITHUB_PROTOCOL=http", |
| 196 | + "SHITHUB_REMOTE_IP=" + clientIP(r), |
| 197 | + "SHITHUB_REQUEST_ID=" + middleware.RequestIDFromContext(r.Context()), |
| 198 | + // PATH must be inherited so the subprocess can find git's helpers. |
| 199 | + "PATH=" + os.Getenv("PATH"), |
| 200 | + } |
| 201 | +} |
| 202 | + |
| 203 | +// ownerIDFromRow extracts the user-owner ID; orgs come in S31. Until |
| 204 | +// then we trust the XOR check in the migration. |
| 205 | +func ownerIDFromRow(row reposdb.Repo) int64 { |
| 206 | + if row.OwnerUserID.Valid { |
| 207 | + return row.OwnerUserID.Int64 |
| 208 | + } |
| 209 | + return 0 |
| 210 | +} |
| 211 | + |
| 212 | +// serviceFromQuery maps the ?service=... value to our typed enum. |
| 213 | +func serviceFromQuery(s string) (protocol.Service, bool) { |
| 214 | + switch s { |
| 215 | + case string(protocol.UploadPack): |
| 216 | + return protocol.UploadPack, true |
| 217 | + case string(protocol.ReceivePack): |
| 218 | + return protocol.ReceivePack, true |
| 219 | + } |
| 220 | + return "", false |
| 221 | +} |
| 222 | + |
| 223 | +// writeChallenge emits a 401 with the canonical Basic challenge. |
| 224 | +func writeChallenge(w http.ResponseWriter) { |
| 225 | + w.Header().Set("WWW-Authenticate", `Basic realm="shithub"`) |
| 226 | + w.Header().Set("Content-Type", "text/plain; charset=utf-8") |
| 227 | + w.WriteHeader(http.StatusUnauthorized) |
| 228 | + _, _ = w.Write([]byte("authentication required\n")) |
| 229 | +} |
| 230 | + |
| 231 | +// writeGitErrorMessage writes a friendly message in the body. For |
| 232 | +// non-streamed responses (we haven't started writing the pack stream |
| 233 | +// yet) this surfaces in `git push`'s stderr verbatim. |
| 234 | +func writeGitErrorMessage(w http.ResponseWriter, status int, msg string) { |
| 235 | + w.Header().Set("Content-Type", "text/plain; charset=utf-8") |
| 236 | + w.WriteHeader(status) |
| 237 | + _, _ = io.WriteString(w, msg+"\n") |
| 238 | +} |
| 239 | + |
| 240 | +// setNoCacheHeaders matches what the canonical git http-backend emits |
| 241 | +// — git endpoints are uncacheable both because the stream is dynamic |
| 242 | +// and because intermediaries shouldn't second-guess us. |
| 243 | +func setNoCacheHeaders(w http.ResponseWriter) { |
| 244 | + w.Header().Set("Expires", "Fri, 01 Jan 1980 00:00:00 GMT") |
| 245 | + w.Header().Set("Pragma", "no-cache") |
| 246 | + w.Header().Set("Cache-Control", "no-cache, max-age=0, must-revalidate") |
| 247 | +} |
| 248 | + |
| 249 | +// clientIP returns the request's source address. RealIP middleware in |
| 250 | +// the global stack populates X-Real-IP-style headers; we read those |
| 251 | +// first, then fall back to RemoteAddr's host part. |
| 252 | +func clientIP(r *http.Request) string { |
| 253 | + if ip := middleware.RealIPFromContext(r.Context(), r); ip != "" { |
| 254 | + return ip |
| 255 | + } |
| 256 | + if i := strings.LastIndex(r.RemoteAddr, ":"); i > 0 { |
| 257 | + return r.RemoteAddr[:i] |
| 258 | + } |
| 259 | + return r.RemoteAddr |
| 260 | +} |