Git LFS support v2 (#122)

* Import github.com/git-lfs/lfs-test-server as lfs module base

Imported commit is 3968aac269a77b73924649b9412ae03f7ccd3198

Removed:

Dockerfile CONTRIBUTING.md mgmt* script/ vendor/ kvlogger.go
.dockerignore .gitignore README.md

* Remove config, add JWT support from github.com/mgit-at/lfs-test-server

Imported commit f0cdcc5a01599c5a955dc1bbf683bb4acecdba83

* Add LFS settings

* Add LFS meta object model

* Add LFS routes and initialization

* Import github.com/dgrijalva/jwt-go into vendor/

* Adapt LFS module: handlers, routing, meta store

* Move LFS routes to /user/repo/info/lfs/*

* Add request header checks to LFS BatchHandler / PostHandler

* Implement LFS basic authentication

* Rework JWT secret generation / load

* Implement LFS SSH token authentication with JWT

Specification: https://github.com/github/git-lfs/tree/master/docs/api

* Integrate LFS settings into install process

* Remove LFS objects when repository is deleted

Only removes objects from content store when deleted repo is the only
referencing repository

* Make LFS module stateless

Fixes bug where LFS would not work after installation without
restarting Gitea

* Change 500 'Internal Server Error' to 400 'Bad Request'

* Change sql query to xorm call

* Remove unneeded type from LFS module

* Change internal imports to code.gitea.io/gitea/

* Add Gitea authors copyright

* Change basic auth realm to "gitea-lfs"

* Add unique indexes to LFS model

* Use xorm count function in LFS check on repository delete

* Return io.ReadCloser from content store and close after usage

* Add LFS info to runWeb()

* Export LFS content store base path

* LFS file download from UI

* Work around git-lfs client issue with unauthenticated requests

Returning a dummy Authorization header for unauthenticated requests
lets git-lfs client skip asking for auth credentials
See: git-lfs/git-lfs#1088

* Fix unauthenticated UI downloads from public repositories

* Authentication check order, Finish LFS file view logic

* Ignore LFS hooks if installed for current OS user

Fixes Gitea UI actions for repositories tracking LFS files.
Checks for minimum needed git version by parsing the semantic version
string.

* Hide LFS metafile diff from commit view, marking as binary

* Show LFS notice if file in commit view is tracked

* Add notbefore/nbf JWT claim

* Correct lint suggestions - comments for structs and functions

- Add comments to LFS model
- Function comment for GetRandomBytesAsBase64
- LFS server function comments and lint variable suggestion

* Move secret generation code out of conditional

Ensures no LFS code may run with an empty secret

* Do not hand out JWT tokens if LFS server support is disabled

Author: fabian-z

cmd/serve.go

@@ -7,6 +7,7 @@ package cmd
 
 import (
 	"crypto/tls"
+	"encoding/json"
 	"fmt"
 	"os"
 	"os/exec"
@@ -21,12 +22,14 @@ import (
 	"code.gitea.io/gitea/modules/log"
 	"code.gitea.io/gitea/modules/setting"
 	"github.com/Unknwon/com"
+	"github.com/dgrijalva/jwt-go"
 	gouuid "github.com/satori/go.uuid"
 	"github.com/urfave/cli"
 )
 
 const (
-	accessDenied = "Repository does not exist or you do not have access"
+	accessDenied        = "Repository does not exist or you do not have access"
+	lfsAuthenticateVerb = "git-lfs-authenticate"
 )
 
 // CmdServ represents the available serv sub-command.
@@ -73,6 +76,7 @@ var (
 		"git-upload-pack":    models.AccessModeRead,
 		"git-upload-archive": models.AccessModeRead,
 		"git-receive-pack":   models.AccessModeWrite,
+		lfsAuthenticateVerb:  models.AccessModeNone,
 	}
 )
 
@@ -161,6 +165,21 @@ func runServ(c *cli.Context) error {
 	}
 
 	verb, args := parseCmd(cmd)
+
+	var lfsVerb string
+	if verb == lfsAuthenticateVerb {
+
+		if !setting.LFS.StartServer {
+			fail("Unknown git command", "LFS authentication request over SSH denied, LFS support is disabled")
+		}
+
+		if strings.Contains(args, " ") {
+			argsSplit := strings.SplitN(args, " ", 2)
+			args = strings.TrimSpace(argsSplit[0])
+			lfsVerb = strings.TrimSpace(argsSplit[1])
+		}
+	}
+
 	repoPath := strings.ToLower(strings.Trim(args, "'"))
 	rr := strings.SplitN(repoPath, "/", 2)
 	if len(rr) != 2 {
@@ -196,6 +215,14 @@ func runServ(c *cli.Context) error {
 		fail("Unknown git command", "Unknown git command %s", verb)
 	}
 
+	if verb == lfsAuthenticateVerb {
+		if lfsVerb == "upload" {
+			requestedMode = models.AccessModeWrite
+		} else {
+			requestedMode = models.AccessModeRead
+		}
+	}
+
 	// Prohibit push to mirror repositories.
 	if requestedMode > models.AccessModeRead && repo.IsMirror {
 		fail("mirror repository is read-only", "")
@@ -261,6 +288,41 @@ func runServ(c *cli.Context) error {
 		}
 	}
 
+	//LFS token authentication
+
+	if verb == lfsAuthenticateVerb {
+
+		url := fmt.Sprintf("%s%s/%s.git/info/lfs", setting.AppURL, repoUser.Name, repo.Name)
+
+		now := time.Now()
+		token := jwt.NewWithClaims(jwt.SigningMethodHS256, jwt.MapClaims{
+			"repo": repo.ID,
+			"op":   lfsVerb,
+			"exp":  now.Add(5 * time.Minute).Unix(),
+			"nbf":  now.Unix(),
+		})
+
+		// Sign and get the complete encoded token as a string using the secret
+		tokenString, err := token.SignedString(setting.LFS.JWTSecretBytes)
+		if err != nil {
+			fail("Internal error", "Failed to sign JWT token: %v", err)
+		}
+
+		tokenAuthentication := &models.LFSTokenResponse{
+			Header: make(map[string]string),
+			Href:   url,
+		}
+		tokenAuthentication.Header["Authorization"] = fmt.Sprintf("Bearer %s", tokenString)
+
+		enc := json.NewEncoder(os.Stdout)
+		err = enc.Encode(tokenAuthentication)
+		if err != nil {
+			fail("Internal error", "Failed to encode LFS json response: %v", err)
+		}
+
+		return nil
+	}
+
 	uuid := gouuid.NewV4().String()
 	os.Setenv("GITEA_UUID", uuid)
 	// Keep the old env variable name for backward compability

cmd/web.go

@@ -17,6 +17,7 @@ import (
 	"code.gitea.io/gitea/models"
 	"code.gitea.io/gitea/modules/auth"
 	"code.gitea.io/gitea/modules/context"
+	"code.gitea.io/gitea/modules/lfs"
 	"code.gitea.io/gitea/modules/log"
 	"code.gitea.io/gitea/modules/options"
 	"code.gitea.io/gitea/modules/public"
@@ -29,6 +30,7 @@ import (
 	"code.gitea.io/gitea/routers/org"
 	"code.gitea.io/gitea/routers/repo"
 	"code.gitea.io/gitea/routers/user"
+
 	"github.com/go-macaron/binding"
 	"github.com/go-macaron/cache"
 	"github.com/go-macaron/captcha"
@@ -564,6 +566,12 @@ func runWeb(ctx *cli.Context) error {
 		}, ignSignIn, context.RepoAssignment(true), context.RepoRef())
 
 		m.Group("/:reponame", func() {
+			m.Group("/info/lfs", func() {
+				m.Post("/objects/batch", lfs.BatchHandler)
+				m.Get("/objects/:oid/:filename", lfs.ObjectOidHandler)
+				m.Any("/objects/:oid", lfs.ObjectOidHandler)
+				m.Post("/objects", lfs.PostHandler)
+			}, ignSignInAndCsrf)
 			m.Any("/*", ignSignInAndCsrf, repo.HTTP)
 			m.Head("/tasks/trigger", repo.TriggerTask)
 		})
@@ -600,6 +608,10 @@ func runWeb(ctx *cli.Context) error {
 	}
 	log.Info("Listen: %v://%s%s", setting.Protocol, listenAddr, setting.AppSubURL)
 
+	if setting.LFS.StartServer {
+		log.Info("LFS server enabled")
+	}
+
 	var err error
 	switch setting.Protocol {
 	case setting.HTTP:

models/git_diff.go

@@ -200,6 +200,7 @@ type DiffFile struct {
 	IsCreated          bool
 	IsDeleted          bool
 	IsBin              bool
+	IsLFSFile          bool
 	IsRenamed          bool
 	IsSubmodule        bool
 	Sections           []*DiffSection
@@ -245,6 +246,7 @@ func ParsePatch(maxLines, maxLineCharacteres, maxFiles int, reader io.Reader) (*
 		leftLine, rightLine int
 		lineCount           int
 		curFileLinesCount   int
+		curFileLFSPrefix    bool
 	)
 
 	input := bufio.NewReader(reader)
@@ -268,6 +270,28 @@ func ParsePatch(maxLines, maxLineCharacteres, maxFiles int, reader io.Reader) (*
 			continue
 		}
 
+		trimLine := strings.Trim(line, "+- ")
+
+		if trimLine == LFSMetaFileIdentifier {
+			curFileLFSPrefix = true
+		}
+
+		if curFileLFSPrefix && strings.HasPrefix(trimLine, LFSMetaFileOidPrefix) {
+			oid := strings.TrimPrefix(trimLine, LFSMetaFileOidPrefix)
+
+			if len(oid) == 64 {
+				m := &LFSMetaObject{Oid: oid}
+				count, err := x.Count(m)
+
+				if err == nil && count > 0 {
+					curFile.IsBin = true
+					curFile.IsLFSFile = true
+					curSection.Lines = nil
+					break
+				}
+			}
+		}
+
 		curFileLinesCount++
 		lineCount++
 
@@ -354,6 +378,7 @@ func ParsePatch(maxLines, maxLineCharacteres, maxFiles int, reader io.Reader) (*
 				break
 			}
 			curFileLinesCount = 0
+			curFileLFSPrefix = false
 
 			// Check file diff type and is submodule.
 			for {

models/lfs.go

@@ -0,0 +1,122 @@
+package models
+
+import (
+	"errors"
+	"github.com/go-xorm/xorm"
+	"time"
+)
+
+// LFSMetaObject stores metadata for LFS tracked files.
+type LFSMetaObject struct {
+	ID           int64     `xorm:"pk autoincr"`
+	Oid          string    `xorm:"UNIQUE(s) INDEX NOT NULL"`
+	Size         int64     `xorm:"NOT NULL"`
+	RepositoryID int64     `xorm:"UNIQUE(s) INDEX NOT NULL"`
+	Existing     bool      `xorm:"-"`
+	Created      time.Time `xorm:"-"`
+	CreatedUnix  int64
+}
+
+// LFSTokenResponse defines the JSON structure in which the JWT token is stored.
+// This structure is fetched via SSH and passed by the Git LFS client to the server
+// endpoint for authorization.
+type LFSTokenResponse struct {
+	Header map[string]string `json:"header"`
+	Href   string            `json:"href"`
+}
+
+var (
+	// ErrLFSObjectNotExist is returned from lfs models functions in order
+	// to differentiate between database and missing object errors.
+	ErrLFSObjectNotExist = errors.New("LFS Meta object does not exist")
+)
+
+const (
+	// LFSMetaFileIdentifier is the string appearing at the first line of LFS pointer files.
+	// https://github.com/git-lfs/git-lfs/blob/master/docs/spec.md
+	LFSMetaFileIdentifier = "version https://git-lfs.github.com/spec/v1"
+
+	// LFSMetaFileOidPrefix appears in LFS pointer files on a line before the sha256 hash.
+	LFSMetaFileOidPrefix = "oid sha256:"
+)
+
+// NewLFSMetaObject stores a given populated LFSMetaObject structure in the database
+// if it is not already present.
+func NewLFSMetaObject(m *LFSMetaObject) (*LFSMetaObject, error) {
+	var err error
+
+	has, err := x.Get(m)
+	if err != nil {
+		return nil, err
+	}
+
+	if has {
+		m.Existing = true
+		return m, nil
+	}
+
+	sess := x.NewSession()
+	defer sessionRelease(sess)
+	if err = sess.Begin(); err != nil {
+		return nil, err
+	}
+
+	if _, err = sess.Insert(m); err != nil {
+		return nil, err
+	}
+
+	return m, sess.Commit()
+}
+
+// GetLFSMetaObjectByOid selects a LFSMetaObject entry from database by its OID.
+// It may return ErrLFSObjectNotExist or a database error. If the error is nil,
+// the returned pointer is a valid LFSMetaObject.
+func GetLFSMetaObjectByOid(oid string) (*LFSMetaObject, error) {
+	if len(oid) == 0 {
+		return nil, ErrLFSObjectNotExist
+	}
+
+	m := &LFSMetaObject{Oid: oid}
+	has, err := x.Get(m)
+	if err != nil {
+		return nil, err
+	} else if !has {
+		return nil, ErrLFSObjectNotExist
+	}
+	return m, nil
+}
+
+// RemoveLFSMetaObjectByOid removes a LFSMetaObject entry from database by its OID.
+// It may return ErrLFSObjectNotExist or a database error.
+func RemoveLFSMetaObjectByOid(oid string) error {
+	if len(oid) == 0 {
+		return ErrLFSObjectNotExist
+	}
+
+	sess := x.NewSession()
+	defer sessionRelease(sess)
+	if err := sess.Begin(); err != nil {
+		return err
+	}
+
+	m := &LFSMetaObject{Oid: oid}
+
+	if _, err := sess.Delete(m); err != nil {
+		return err
+	}
+
+	return sess.Commit()
+}
+
+// BeforeInsert sets the time at which the LFSMetaObject was created.
+func (m *LFSMetaObject) BeforeInsert() {
+	m.CreatedUnix = time.Now().Unix()
+}
+
+// AfterSet stores the LFSMetaObject creation time in the database as local time.
+func (m *LFSMetaObject) AfterSet(colName string, _ xorm.Cell) {
+	switch colName {
+	case "created_unix":
+		m.Created = time.Unix(m.CreatedUnix, 0).Local()
+	}
+}

models/models.go

@@ -79,7 +79,7 @@ func init() {
 		new(Mirror), new(Release), new(LoginSource), new(Webhook),
 		new(UpdateTask), new(HookTask),
 		new(Team), new(OrgUser), new(TeamUser), new(TeamRepo),
-		new(Notice), new(EmailAddress))
+		new(Notice), new(EmailAddress), new(LFSMetaObject))
 
 	gonicNames := []string{"SSL", "UID"}
 	for _, name := range gonicNames {