X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/2a6e0785f8fb675038f867aebc65033dc8a85211..a9a677e1655c461e742e46cc3c239f8605f4fc6b:/services/keepstore/handlers.go diff --git a/services/keepstore/handlers.go b/services/keepstore/handlers.go index f8e4d71452..289dce15a0 100644 --- a/services/keepstore/handlers.go +++ b/services/keepstore/handlers.go @@ -9,6 +9,7 @@ package main import ( "container/list" + "context" "crypto/md5" "encoding/json" "fmt" @@ -20,6 +21,7 @@ import ( "regexp" "runtime" "strconv" + "strings" "sync" "time" ) @@ -53,8 +55,8 @@ func MakeRESTRouter() *mux.Router { // Replace the current trash queue. rest.HandleFunc(`/trash`, TrashHandler).Methods("PUT") - // Undelete moves blocks from trash back into store - rest.HandleFunc(`/undelete/{hash:[0-9a-f]{32}}`, UndeleteHandler).Methods("PUT") + // Untrash moves blocks from trash back into store + rest.HandleFunc(`/untrash/{hash:[0-9a-f]{32}}`, UntrashHandler).Methods("PUT") // Any request which does not match any of these routes gets // 400 Bad Request. @@ -70,30 +72,88 @@ func BadRequestHandler(w http.ResponseWriter, r *http.Request) { // GetBlockHandler is a HandleFunc to address Get block requests. func GetBlockHandler(resp http.ResponseWriter, req *http.Request) { - if enforcePermissions { + ctx, cancel := contextForResponse(context.TODO(), resp) + defer cancel() + + if theConfig.RequireSignatures { locator := req.URL.Path[1:] // strip leading slash - if err := VerifySignature(locator, GetApiToken(req)); err != nil { + if err := VerifySignature(locator, GetAPIToken(req)); err != nil { http.Error(resp, err.Error(), err.(*KeepError).HTTPCode) return } } - block, err := GetBlock(mux.Vars(req)["hash"]) + // TODO: Probe volumes to check whether the block _might_ + // exist. Some volumes/types could support a quick existence + // check without causing other operations to suffer. If all + // volumes support that, and assure us the block definitely + // isn't here, we can return 404 now instead of waiting for a + // buffer. + + buf, err := getBufferWithContext(ctx, bufs, BlockSize) if err != nil { - // This type assertion is safe because the only errors - // GetBlock can return are DiskHashError or NotFoundError. - http.Error(resp, err.Error(), err.(*KeepError).HTTPCode) + http.Error(resp, err.Error(), http.StatusServiceUnavailable) return } - defer bufs.Put(block) + defer bufs.Put(buf) - resp.Header().Set("Content-Length", strconv.Itoa(len(block))) + size, err := GetBlock(ctx, mux.Vars(req)["hash"], buf, resp) + if err != nil { + code := http.StatusInternalServerError + if err, ok := err.(*KeepError); ok { + code = err.HTTPCode + } + http.Error(resp, err.Error(), code) + return + } + + resp.Header().Set("Content-Length", strconv.Itoa(size)) resp.Header().Set("Content-Type", "application/octet-stream") - resp.Write(block) + resp.Write(buf[:size]) +} + +// Return a new context that gets cancelled by resp's CloseNotifier. +func contextForResponse(parent context.Context, resp http.ResponseWriter) (context.Context, context.CancelFunc) { + ctx, cancel := context.WithCancel(parent) + if cn, ok := resp.(http.CloseNotifier); ok { + go func(c <-chan bool) { + select { + case <-c: + theConfig.debugLogf("cancel context") + cancel() + case <-ctx.Done(): + } + }(cn.CloseNotify()) + } + return ctx, cancel +} + +// Get a buffer from the pool -- but give up and return a non-nil +// error if ctx ends before we get a buffer. +func getBufferWithContext(ctx context.Context, bufs *bufferPool, bufSize int) ([]byte, error) { + bufReady := make(chan []byte) + go func() { + bufReady <- bufs.Get(bufSize) + }() + select { + case buf := <-bufReady: + return buf, nil + case <-ctx.Done(): + go func() { + // Even if closeNotifier happened first, we + // need to keep waiting for our buf so we can + // return it to the pool. + bufs.Put(<-bufReady) + }() + return nil, ErrClientDisconnect + } } // PutBlockHandler is a HandleFunc to address Put block requests. func PutBlockHandler(resp http.ResponseWriter, req *http.Request) { + ctx, cancel := contextForResponse(context.TODO(), resp) + defer cancel() + hash := mux.Vars(req)["hash"] // Detect as many error conditions as possible before reading @@ -115,29 +175,37 @@ func PutBlockHandler(resp http.ResponseWriter, req *http.Request) { return } - buf := bufs.Get(int(req.ContentLength)) - _, err := io.ReadFull(req.Body, buf) + buf, err := getBufferWithContext(ctx, bufs, int(req.ContentLength)) + if err != nil { + http.Error(resp, err.Error(), http.StatusServiceUnavailable) + return + } + + _, err = io.ReadFull(req.Body, buf) if err != nil { http.Error(resp, err.Error(), 500) bufs.Put(buf) return } - replication, err := PutBlock(buf, hash) + replication, err := PutBlock(ctx, buf, hash) bufs.Put(buf) if err != nil { - ke := err.(*KeepError) - http.Error(resp, ke.Error(), ke.HTTPCode) + code := http.StatusInternalServerError + if err, ok := err.(*KeepError); ok { + code = err.HTTPCode + } + http.Error(resp, err.Error(), code) return } // Success; add a size hint, sign the locator if possible, and // return it to the client. returnHash := fmt.Sprintf("%s+%d", hash, req.ContentLength) - apiToken := GetApiToken(req) - if PermissionSecret != nil && apiToken != "" { - expiry := time.Now().Add(blobSignatureTTL) + apiToken := GetAPIToken(req) + if theConfig.blobSigningKey != nil && apiToken != "" { + expiry := time.Now().Add(theConfig.BlobSignatureTTL.Duration()) returnHash = SignLocator(returnHash, apiToken, expiry) } resp.Header().Set("X-Keep-Replicas-Stored", strconv.Itoa(replication)) @@ -147,7 +215,7 @@ func PutBlockHandler(resp http.ResponseWriter, req *http.Request) { // IndexHandler is a HandleFunc to address /index and /index/{prefix} requests. func IndexHandler(resp http.ResponseWriter, req *http.Request) { // Reject unauthorized requests. - if !IsDataManagerToken(GetApiToken(req)) { + if !IsSystemAuth(GetAPIToken(req)) { http.Error(resp, UnauthorizedError.Error(), UnauthorizedError.HTTPCode) return } @@ -279,13 +347,13 @@ func DeleteHandler(resp http.ResponseWriter, req *http.Request) { hash := mux.Vars(req)["hash"] // Confirm that this user is an admin and has a token with unlimited scope. - var tok = GetApiToken(req) + var tok = GetAPIToken(req) if tok == "" || !CanDelete(tok) { http.Error(resp, PermissionError.Error(), PermissionError.HTTPCode) return } - if neverDelete { + if !theConfig.EnableDelete { http.Error(resp, MethodDisabledError.Error(), MethodDisabledError.HTTPCode) return } @@ -370,7 +438,7 @@ type PullRequest struct { // PullHandler processes "PUT /pull" requests for the data manager. func PullHandler(resp http.ResponseWriter, req *http.Request) { // Reject unauthorized requests. - if !IsDataManagerToken(GetApiToken(req)) { + if !IsSystemAuth(GetAPIToken(req)) { http.Error(resp, UnauthorizedError.Error(), UnauthorizedError.HTTPCode) return } @@ -406,7 +474,7 @@ type TrashRequest struct { // TrashHandler processes /trash requests. func TrashHandler(resp http.ResponseWriter, req *http.Request) { // Reject unauthorized requests. - if !IsDataManagerToken(GetApiToken(req)) { + if !IsSystemAuth(GetAPIToken(req)) { http.Error(resp, UnauthorizedError.Error(), UnauthorizedError.HTTPCode) return } @@ -433,37 +501,53 @@ func TrashHandler(resp http.ResponseWriter, req *http.Request) { trashq.ReplaceQueue(tlist) } -// UndeleteHandler processes "PUT /undelete/{hash:[0-9a-f]{32}}" requests for the data manager. -func UndeleteHandler(resp http.ResponseWriter, req *http.Request) { +// UntrashHandler processes "PUT /untrash/{hash:[0-9a-f]{32}}" requests for the data manager. +func UntrashHandler(resp http.ResponseWriter, req *http.Request) { // Reject unauthorized requests. - if !IsDataManagerToken(GetApiToken(req)) { + if !IsSystemAuth(GetAPIToken(req)) { http.Error(resp, UnauthorizedError.Error(), UnauthorizedError.HTTPCode) return } hash := mux.Vars(req)["hash"] - successResp := "Untrashed on volume: " - var st int + + if len(KeepVM.AllWritable()) == 0 { + http.Error(resp, "No writable volumes", http.StatusNotFound) + return + } + + var untrashedOn, failedOn []string + var numNotFound int for _, vol := range KeepVM.AllWritable() { - if err := vol.Untrash(hash); err == nil { - log.Printf("Untrashed %v on volume %v", hash, vol.String()) - st = http.StatusOK - successResp += vol.String() - break + err := vol.Untrash(hash) + + if os.IsNotExist(err) { + numNotFound++ + } else if err != nil { + log.Printf("Error untrashing %v on volume %v", hash, vol.String()) + failedOn = append(failedOn, vol.String()) } else { - log.Printf("Error untrashing %v on volume %v: %v", hash, vol.String(), err) - st = 500 + log.Printf("Untrashed %v on volume %v", hash, vol.String()) + untrashedOn = append(untrashedOn, vol.String()) } } - if st == http.StatusOK { - resp.Write([]byte(successResp)) + if numNotFound == len(KeepVM.AllWritable()) { + http.Error(resp, "Block not found on any of the writable volumes", http.StatusNotFound) + return } - resp.WriteHeader(st) + if len(failedOn) == len(KeepVM.AllWritable()) { + http.Error(resp, "Failed to untrash on all writable volumes", http.StatusInternalServerError) + } else { + respBody := "Successfully untrashed on: " + strings.Join(untrashedOn, ",") + if len(failedOn) > 0 { + respBody += "; Failed to untrash on: " + strings.Join(failedOn, ",") + } + resp.Write([]byte(respBody)) + } } -// ============================== // GetBlock and PutBlock implement lower-level code for handling // blocks by rooting through volumes connected to the local machine. // Once the handler has determined that system policy permits the @@ -474,24 +558,26 @@ func UndeleteHandler(resp http.ResponseWriter, req *http.Request) { // should be the only part of the code that cares about which volume a // block is stored on, so it should be responsible for figuring out // which volume to check for fetching blocks, storing blocks, etc. -// ============================== -// GetBlock fetches and returns the block identified by "hash". -// -// On success, GetBlock returns a byte slice with the block data, and -// a nil error. +// GetBlock fetches the block identified by "hash" into the provided +// buf, and returns the data size. // // If the block cannot be found on any volume, returns NotFoundError. // // If the block found does not have the correct MD5 hash, returns // DiskHashError. // -func GetBlock(hash string) ([]byte, error) { +func GetBlock(ctx context.Context, hash string, buf []byte, resp http.ResponseWriter) (int, error) { // Attempt to read the requested hash from a keep volume. errorToCaller := NotFoundError for _, vol := range KeepVM.AllReadable() { - buf, err := vol.Get(hash) + size, err := vol.Get(ctx, hash, buf) + select { + case <-ctx.Done(): + return 0, ErrClientDisconnect + default: + } if err != nil { // IsNotExist is an expected error and may be // ignored. All other errors are logged. In @@ -505,28 +591,27 @@ func GetBlock(hash string) ([]byte, error) { } // Check the file checksum. // - filehash := fmt.Sprintf("%x", md5.Sum(buf)) + filehash := fmt.Sprintf("%x", md5.Sum(buf[:size])) if filehash != hash { // TODO: Try harder to tell a sysadmin about // this. log.Printf("%s: checksum mismatch for request %s (actual %s)", vol, hash, filehash) errorToCaller = DiskHashError - bufs.Put(buf) continue } if errorToCaller == DiskHashError { log.Printf("%s: checksum mismatch for request %s but a good copy was found on another volume and returned", vol, hash) } - return buf, nil + return size, nil } - return nil, errorToCaller + return 0, errorToCaller } // PutBlock Stores the BLOCK (identified by the content id HASH) in Keep. // -// PutBlock(block, hash) +// PutBlock(ctx, block, hash) // Stores the BLOCK (identified by the content id HASH) in Keep. // // The MD5 checksum of the block must be identical to the content id HASH. @@ -551,7 +636,7 @@ func GetBlock(hash string) ([]byte, error) { // all writes failed). The text of the error message should // provide as much detail as possible. // -func PutBlock(block []byte, hash string) (int, error) { +func PutBlock(ctx context.Context, block []byte, hash string) (int, error) { // Check that BLOCK's checksum matches HASH. blockhash := fmt.Sprintf("%x", md5.Sum(block)) if blockhash != hash { @@ -562,16 +647,21 @@ func PutBlock(block []byte, hash string) (int, error) { // If we already have this data, it's intact on disk, and we // can update its timestamp, return success. If we have // different data with the same hash, return failure. - if n, err := CompareAndTouch(hash, block); err == nil || err == CollisionError { + if n, err := CompareAndTouch(ctx, hash, block); err == nil || err == CollisionError { return n, err + } else if ctx.Err() != nil { + return 0, ErrClientDisconnect } // Choose a Keep volume to write to. // If this volume fails, try all of the volumes in order. if vol := KeepVM.NextWritable(); vol != nil { - if err := vol.Put(hash, block); err == nil { + if err := vol.Put(ctx, hash, block); err == nil { return vol.Replication(), nil // success! } + if ctx.Err() != nil { + return 0, ErrClientDisconnect + } } writables := KeepVM.AllWritable() @@ -582,7 +672,10 @@ func PutBlock(block []byte, hash string) (int, error) { allFull := true for _, vol := range writables { - err := vol.Put(hash, block) + err := vol.Put(ctx, hash, block) + if ctx.Err() != nil { + return 0, ErrClientDisconnect + } if err == nil { return vol.Replication(), nil // success! } @@ -608,10 +701,13 @@ func PutBlock(block []byte, hash string) (int, error) { // the relevant block's modification time in order to protect it from // premature garbage collection. Otherwise, it returns a non-nil // error. -func CompareAndTouch(hash string, buf []byte) (int, error) { +func CompareAndTouch(ctx context.Context, hash string, buf []byte) (int, error) { var bestErr error = NotFoundError for _, vol := range KeepVM.AllWritable() { - if err := vol.Compare(hash, buf); err == CollisionError { + err := vol.Compare(ctx, hash, buf) + if ctx.Err() != nil { + return 0, ctx.Err() + } else if err == CollisionError { // Stop if we have a block with same hash but // different content. (It will be impossible // to tell which one is wanted if we have @@ -653,10 +749,10 @@ func IsValidLocator(loc string) bool { var authRe = regexp.MustCompile(`^OAuth2\s+(.*)`) -// GetApiToken returns the OAuth2 token from the Authorization +// GetAPIToken returns the OAuth2 token from the Authorization // header of a HTTP request, or an empty string if no matching // token is found. -func GetApiToken(req *http.Request) string { +func GetAPIToken(req *http.Request) string { if auth, ok := req.Header["Authorization"]; ok { if match := authRe.FindStringSubmatch(auth[0]); match != nil { return match[1] @@ -685,7 +781,7 @@ func CanDelete(apiToken string) bool { } // Blocks may be deleted only when Keep has been configured with a // data manager. - if IsDataManagerToken(apiToken) { + if IsSystemAuth(apiToken) { return true } // TODO(twp): look up apiToken with the API server @@ -694,8 +790,8 @@ func CanDelete(apiToken string) bool { return false } -// IsDataManagerToken returns true if apiToken represents the data -// manager's token. -func IsDataManagerToken(apiToken string) bool { - return dataManagerToken != "" && apiToken == dataManagerToken +// IsSystemAuth returns true if the given token is allowed to perform +// system level actions like deleting data. +func IsSystemAuth(token string) bool { + return token != "" && token == theConfig.systemAuthToken }