cloudron-box/src/storage/s3.js

'use strict';

exports = module.exports = {
    getBackupRootPath,
    getProviderStatus,
    getAvailableSize,

    upload,
    exists,
    download,
    copy,

    listDir,

    remove,
    removeDir,

    remount,

    testConfig,
    removePrivateFields,
    injectPrivateFields,

    // Used to mock AWS
    _mockInject: mockInject,
    _mockRestore: mockRestore,
    _chunk: chunk
};

// https://github.com/aws/aws-sdk-js/issues/4354
require('aws-sdk/lib/maintenance_mode_message').suppress = true;

const assert = require('assert'),
    async = require('async'),
    AwsSdk = require('aws-sdk'),
    BoxError = require('../boxerror.js'),
    constants = require('../constants.js'),
    debug = require('debug')('box:storage/s3'),
    https = require('https'),
    path = require('path'),
    Readable = require('stream').Readable,
    safe = require('safetydance'),
    util = require('util'),
    _ = require('underscore');

let aws = AwsSdk;

// test only
let originalAWS;
function mockInject(mock) {
    originalAWS = aws;
    aws = mock;
}

function mockRestore() {
    aws = originalAWS;
}

function S3_NOT_FOUND(error) {
    return error.code === 'NoSuchKey' || error.code === 'NotFound' || error.code === 'ENOENT';
}

function getS3Config(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');

    const credentials = {
        signatureVersion: apiConfig.signatureVersion || 'v4',
        s3ForcePathStyle: false, // Use vhost style instead of path style - https://forums.aws.amazon.com/ann.jspa?annID=6776
        accessKeyId: apiConfig.accessKeyId,
        secretAccessKey: apiConfig.secretAccessKey,
        region: apiConfig.region || 'us-east-1',
        maxRetries: 10,
        retryDelayOptions: {
            customBackoff: (/* retryCount, error */) => 20000 // constant backoff - https://docs.aws.amazon.com/AWSJavaScriptSDK/latest/AWS/Config.html#retryDelayOptions-property
        },
        httpOptions: {
            connectTimeout: 60000, // https://github.com/aws/aws-sdk-js/pull/1446
            timeout: 0 // https://github.com/aws/aws-sdk-js/issues/1704 (allow unlimited time for chunk upload)
        }
    };

    if (apiConfig.endpoint) credentials.endpoint = apiConfig.endpoint;

    if (apiConfig.s3ForcePathStyle === true) credentials.s3ForcePathStyle = true;

    // s3 endpoint names come from the SDK
    const isHttps = (credentials.endpoint && credentials.endpoint.startsWith('https://')) || apiConfig.provider === 's3';
    if (isHttps) { // only set agent for https calls. otherwise, it crashes
        if (apiConfig.acceptSelfSignedCerts || apiConfig.bucket.includes('.')) {
            credentials.httpOptions.agent = new https.Agent({ rejectUnauthorized: false });
        }
    }

    return credentials;
}

// storage api
function getBackupRootPath(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');

    return apiConfig.prefix;
}

async function getProviderStatus(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');

    return { state: 'active' };
}

async function getAvailableSize(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');

    return Number.POSITIVE_INFINITY;
}

function upload(apiConfig, backupFilePath, sourceStream, callback) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof backupFilePath, 'string');
    assert.strictEqual(typeof sourceStream, 'object');
    assert.strictEqual(typeof callback, 'function');

    const credentials = getS3Config(apiConfig);

    const params = {
        Bucket: apiConfig.bucket,
        Key: backupFilePath,
        Body: sourceStream
    };

    const s3 = new aws.S3(credentials);

    // s3.upload automatically does a multi-part upload. we set queueSize to 3 to reduce memory usage
    // uploader will buffer at most queueSize * partSize bytes into memory at any given time.
    // scaleway only supports 1000 parts per object (https://www.scaleway.com/en/docs/s3-multipart-upload/)
    // s3: https://docs.aws.amazon.com/AmazonS3/latest/dev/qfacts.html (max 10k parts and no size limit on the last part!)
    const partSize = apiConfig.limits?.uploadPartSize || (apiConfig.provider === 'scaleway-objectstorage' ? 100 * 1024 * 1024 : 10 * 1024 * 1024);

    s3.upload(params, { partSize, queueSize: 3 }, function (error, data) {
        if (error) {
            debug(`upload: [${backupFilePath}] s3 upload error. %o`, error);
            return callback(new BoxError(BoxError.EXTERNAL_ERROR, `Error uploading ${backupFilePath}. Message: ${error.message} HTTP Code: ${error.code}`));
        }

        debug(`Uploaded ${backupFilePath} with partSize ${partSize}: ${JSON.stringify(data)}`);

        callback(null);
    });
}

async function exists(apiConfig, backupFilePath) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof backupFilePath, 'string');

    const credentials = getS3Config(apiConfig);

    const s3 = new aws.S3(_.omit(credentials, 'retryDelayOptions', 'maxRetries'));

    if (!backupFilePath.endsWith('/')) { // check for file
        const params = {
            Bucket: apiConfig.bucket,
            Key: backupFilePath
        };

        const [error] = await safe(s3.headObject(params).promise());
        if (!Object.keys(this.httpResponse.headers).some(h => h.startsWith('x-amz'))) throw new BoxError(BoxError.EXTERNAL_ERROR, 'not a s3 endpoint');
        if (error && S3_NOT_FOUND(error)) return false;
        if (error) throw new BoxError(BoxError.EXTERNAL_ERROR, `Error headObject ${backupFilePath}. Message: ${error.message} HTTP Code: ${error.code}`);

        return true;
    } else { // list dir contents
        const listParams = {
            Bucket: apiConfig.bucket,
            Prefix: backupFilePath,
            MaxKeys: 1
        };

        const [error, listData] = await safe(s3.listObjects(listParams).promise());
        if (error) throw new BoxError(BoxError.EXTERNAL_ERROR, `Error listing objects ${backupFilePath}. Message: ${error.message} HTTP Code: ${error.code}`);

        return listData.Contents.length !== 0;
    }
}

// Download the object in small parts. By downloading small parts, we reduce the chance of sporadic network errors when downloading large objects
// We can retry each part individually, but we haven't had the need for this yet
class S3MultipartDownloadStream extends Readable {
    constructor (s3, params, options) {
        super(options);
        this._s3 = s3;
        this._params = params;
        this._readSize = 0;
        this._fileSize = -1;
        this._path = params.Bucket + '/' + params.Key;

        this._blockSize = options.blockSize || 64 * 1048576; // MB
    }

    _done() {
        this._readSize = 0;
        this.push(null); // EOF
    }

    _handleError(error) {
        if (S3_NOT_FOUND(error)) {
            this.destroy(new BoxError(BoxError.NOT_FOUND, `Backup not found: ${this._path}`));
        } else {
            debug(`download: ${this._path} s3 stream error. %o`, error);
            this.destroy(new BoxError(BoxError.EXTERNAL_ERROR, `Error multipartDownload ${this._path}. Message: ${error.message} HTTP Code: ${error.code}`));
        }
    }

    _downloadRange(offset, length) {
        const params = Object.assign({}, this._params);
        const lastPos = offset + length - 1;
        const range = `bytes=${offset}-${lastPos}`;
        params['Range'] = range;

        this._s3.getObject(params, (error, data) => {
            if (error) return this._handleError(error);

            const length = parseInt(data.ContentLength, 10);

            if (length > 0) {
                this._readSize += length;
                this.push(data.Body);
            } else {
                this._done();
            }
        });
    }

    _nextDownload() {
        let len = 0;
        if (this._readSize + this._blockSize < this._fileSize) {
            len = this._blockSize;
        } else {
            len = this._fileSize - this._readSize;
        }
        this._downloadRange(this._readSize, len);
    }

    _fetchSize() {
        this._s3.headObject(this._params, (error, data) => {
            if (error) return this._handleError(error);

            const length = parseInt(data.ContentLength, 10);

            if (length > 0) {
                this._fileSize = length;
                this._nextDownload();
            } else {
                this._done();
            }
        });
    }

    _read() {
        if (this._readSize === this._fileSize) return this._done();
        if (this._readSize === 0) return this._fetchSize();

        this._nextDownload();
    }
}

function download(apiConfig, backupFilePath, callback) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof backupFilePath, 'string');
    assert.strictEqual(typeof callback, 'function');

    const credentials = getS3Config(apiConfig);

    const params = {
        Bucket: apiConfig.bucket,
        Key: backupFilePath
    };

    const s3 = new aws.S3(credentials);

    const multipartDownloadStream = new S3MultipartDownloadStream(s3, params, { blockSize: 64 * 1024 * 1024 });
    return callback(null, multipartDownloadStream);
}

function listDir(apiConfig, dir, batchSize, iteratorCallback, callback) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof dir, 'string');
    assert.strictEqual(typeof batchSize, 'number');
    assert.strictEqual(typeof iteratorCallback, 'function');
    assert.strictEqual(typeof callback, 'function');

    const credentials = getS3Config(apiConfig);

    const s3 = new aws.S3(credentials);
    const listParams = {
        Bucket: apiConfig.bucket,
        Prefix: dir,
        MaxKeys: batchSize
    };

    let done = false;

    async.whilst((testDone) => testDone(null, !done), function listAndDownload(whilstCallback) {
        s3.listObjects(listParams, function (error, listData) {
            if (error) return whilstCallback(new BoxError(BoxError.EXTERNAL_ERROR, `Error listing objects in ${dir}. Message: ${error.message} HTTP Code: ${error.code}`));

            if (listData.Contents.length === 0) { done = true; return whilstCallback(); }

            const entries = listData.Contents.map(function (c) { return { fullPath: c.Key, size: c.Size }; });

            iteratorCallback(entries, function (error) {
                if (error) return whilstCallback(error);

                if (!listData.IsTruncated) { done = true; return whilstCallback(); }

                listParams.Marker = listData.Contents[listData.Contents.length - 1].Key; // NextMarker is returned only with delimiter

                whilstCallback();
            });
        });
    }, callback);
}

// https://github.com/aws/aws-sdk-js/blob/2b6bcbdec1f274fe931640c1b61ece999aae7a19/lib/util.js#L41
// https://github.com/GeorgePhillips/node-s3-url-encode/blob/master/index.js
// See aws-sdk-js/issues/1302
function encodeCopySource(bucket, path) {
    // AWS percent-encodes some extra non-standard characters in a URI
    const output = encodeURI(path).replace(/[+!"#$@&'()*+,:;=?@]/g, function(ch) {
        return '%' + ch.charCodeAt(0).toString(16).toUpperCase();
    });

    // the slash at the beginning is optional
    return `/${bucket}/${output}`;
}

async function copy(apiConfig, oldFilePath, newFilePath, progressCallback) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof oldFilePath, 'string');
    assert.strictEqual(typeof newFilePath, 'string');
    assert.strictEqual(typeof progressCallback, 'function');

    function copyFile(entry, iteratorCallback) {
        const credentials = getS3Config(apiConfig);

        const s3 = new aws.S3(credentials);
        const relativePath = path.relative(oldFilePath, entry.fullPath);

        function done(error) {
            if (error) debug(`copy: s3 copy error when copying ${entry.fullPath}: ${error}`);

            if (error && S3_NOT_FOUND(error)) return iteratorCallback(new BoxError(BoxError.NOT_FOUND, `Old backup not found: ${entry.fullPath}`));
            if (error) return iteratorCallback(new BoxError(BoxError.EXTERNAL_ERROR, `Error copying ${entry.fullPath} (${entry.size} bytes): ${error.code || ''} ${error}`));

            iteratorCallback(null);
        }

        const copyParams = {
            Bucket: apiConfig.bucket,
            Key: path.join(newFilePath, relativePath)
        };

        // S3 copyObject has a file size limit of 5GB so if we have larger files, we do a multipart copy
        // Exoscale and B2 take too long to copy 5GB
        const largeFileLimit = (apiConfig.provider === 'exoscale-sos' || apiConfig.provider === 'backblaze-b2' || apiConfig.provider === 'digitalocean-spaces') ? 1024 * 1024 * 1024 : 5 * 1024 * 1024 * 1024;

        if (entry.size < largeFileLimit) {
            progressCallback({ message: `Copying ${relativePath || oldFilePath}` });

            copyParams.CopySource = encodeCopySource(apiConfig.bucket, entry.fullPath);
            s3.copyObject(copyParams, done).on('retry', function (response) {
                progressCallback({ message: `Retrying (${response.retryCount+1}) copy of ${relativePath || oldFilePath}. Error: ${response.error} ${response.httpResponse.statusCode}` });
                // on DO, we get a random 408. these are not retried by the SDK
                if (response.error) response.error.retryable = true; // https://github.com/aws/aws-sdk-js/issues/412
            });

            return;
        }

        progressCallback({ message: `Copying (multipart) ${relativePath || oldFilePath}` });

        s3.createMultipartUpload(copyParams, function (error, multipart) {
            if (error) return done(error);

            // Exoscale (96M) was suggested by exoscale. 1GB - rather random size for others
            const chunkSize = apiConfig.provider === 'exoscale-sos' ? 96 * 1024 * 1024 : 1024 * 1024 * 1024;
            const uploadId = multipart.UploadId;
            let uploadedParts = [], ranges = [];

            let cur = 0;
            while (cur + chunkSize < entry.size) {
                ranges.push({ startBytes: cur, endBytes: cur + chunkSize - 1 });
                cur += chunkSize;
            }
            ranges.push({ startBytes: cur, endBytes: entry.size-1 });

            async.eachOfLimit(ranges, 3, function copyChunk(range, index, iteratorDone) {
                const partCopyParams = {
                    Bucket: apiConfig.bucket,
                    Key: path.join(newFilePath, relativePath),
                    CopySource: encodeCopySource(apiConfig.bucket, entry.fullPath), // See aws-sdk-js/issues/1302
                    CopySourceRange: 'bytes=' + range.startBytes + '-' + range.endBytes,
                    PartNumber: index+1,
                    UploadId: uploadId
                };

                progressCallback({ message: `Copying part ${partCopyParams.PartNumber} - ${partCopyParams.CopySource} ${partCopyParams.CopySourceRange}` });

                s3.uploadPartCopy(partCopyParams, function (error, part) {
                    if (error) return iteratorDone(error);

                    progressCallback({ message: `Copying part ${partCopyParams.PartNumber} - Etag: ${part.CopyPartResult.ETag}` });

                    if (!part.CopyPartResult.ETag) return iteratorDone(new Error('Multi-part copy is broken or not implemented by the S3 storage provider'));

                    uploadedParts[index] = { ETag: part.CopyPartResult.ETag, PartNumber: partCopyParams.PartNumber };

                    iteratorDone();
                }).on('retry', function (response) {
                    progressCallback({ message: `Retrying (${response.retryCount+1}) multipart copy of ${relativePath || oldFilePath}. Error: ${response.error} ${response.httpResponse.statusCode}` });
                });
            }, function chunksCopied(error) {
                if (error) { // we must still recommend the user to set a AbortIncompleteMultipartUpload lifecycle rule
                    const abortParams = {
                        Bucket: apiConfig.bucket,
                        Key: path.join(newFilePath, relativePath),
                        UploadId: uploadId
                    };
                    progressCallback({ message: `Aborting multipart copy of ${relativePath || oldFilePath}` });
                    return s3.abortMultipartUpload(abortParams, () => done(error)); // ignore any abort errors
                }

                const completeMultipartParams = {
                    Bucket: apiConfig.bucket,
                    Key: path.join(newFilePath, relativePath),
                    MultipartUpload: { Parts: uploadedParts },
                    UploadId: uploadId
                };

                progressCallback({ message: `Finishing multipart copy - ${completeMultipartParams.Key}` });

                s3.completeMultipartUpload(completeMultipartParams, done);
            });
        });
    }

    let total = 0;
    const concurrency = apiConfig.limits?.copyConcurrency || (apiConfig.provider === 's3' ? 500 : 10);
    progressCallback({ message: `Copying with concurrency of ${concurrency}` });

    const listDirAsync = util.promisify(listDir);

    const [copyError] = await safe(listDirAsync(apiConfig, oldFilePath, 1000, function listDirIterator(entries, done) {
        total += entries.length;

        progressCallback({ message: `Copying files from ${total-entries.length}-${total}` });

        async.eachLimit(entries, concurrency, copyFile, done);
    }));

    progressCallback({ message: `Copied ${total} files with error: ${copyError}` });
    if (copyError) throw copyError;
}

async function remove(apiConfig, filename) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof filename, 'string');

    const credentials = getS3Config(apiConfig);

    const s3 = new aws.S3(credentials);

    const deleteParams = {
        Bucket: apiConfig.bucket,
        Delete: {
            Objects: [{ Key: filename }]
        }
    };

    // deleteObjects does not return error if key is not found
    const [error] = await safe(s3.deleteObjects(deleteParams).promise());
    if (error) throw new BoxError(BoxError.EXTERNAL_ERROR, `Unable to remove ${deleteParams.Key}. error: ${error.message}`);
}

function chunk(array, size) {
    assert(Array.isArray(array));
    assert.strictEqual(typeof size, 'number');

    const length = array.length;
    if (!length) return [];
    let index = 0, resIndex = 0, result = Array(Math.ceil(length / size));

    for (; index < length; index += size) {
        result[resIndex++] = array.slice(index, index+size);
    }

    return result;
}

async function removeDir(apiConfig, pathPrefix, progressCallback) {
    assert.strictEqual(typeof apiConfig, 'object');
    assert.strictEqual(typeof pathPrefix, 'string');
    assert.strictEqual(typeof progressCallback, 'function');


    const credentials = getS3Config(apiConfig);
    const s3 = new aws.S3(credentials);
    const listDirAsync = util.promisify(listDir);

    let total = 0;

    await listDirAsync(apiConfig, pathPrefix, 1000, function listDirIterator(entries, done) {
        total += entries.length;

        const chunkSize = apiConfig.limits?.deleteConcurrency || (apiConfig.provider !== 'digitalocean-spaces' ? 1000 : 100); // throttle objects in each request
        const chunks = chunk(entries, chunkSize);

        async.eachSeries(chunks, async function deleteFiles(objects) {
            const deleteParams = {
                Bucket: apiConfig.bucket,
                Delete: {
                    Objects: objects.map(function (o) { return { Key: o.fullPath }; })
                }
            };

            progressCallback({ message: `Removing ${objects.length} files from ${objects[0].fullPath} to ${objects[objects.length-1].fullPath}` });

            // deleteObjects does not return error if key is not found
            const [error] = await safe(s3.deleteObjects(deleteParams).promise());
            if (error) {
                progressCallback({ message: `Unable to remove ${deleteParams.Key} ${error.message || error.code}` });
                throw new BoxError(BoxError.EXTERNAL_ERROR, `Unable to remove ${deleteParams.Key}. error: ${error.message}`);
            }
        }, done);
    });

    progressCallback({ message: `Removed ${total} files` });
}

async function remount(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');
}

async function testConfig(apiConfig) {
    assert.strictEqual(typeof apiConfig, 'object');

    if (typeof apiConfig.accessKeyId !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'accessKeyId must be a string');
    if (typeof apiConfig.secretAccessKey !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'secretAccessKey must be a string');

    if (typeof apiConfig.bucket !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'bucket must be a string');
    // the node module seems to incorrectly accept bucket name with '/'
    if (apiConfig.bucket.includes('/')) throw new BoxError(BoxError.BAD_FIELD, 'bucket name cannot contain "/"');

    // names must be lowercase and start with a letter or number. can contain dashes
    if (apiConfig.bucket.includes('_') || apiConfig.bucket.match(/[A-Z]/)) throw new BoxError(BoxError.BAD_FIELD, 'bucket name cannot contain "_" or capitals');

    if (typeof apiConfig.prefix !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'prefix must be a string');
    if ('signatureVersion' in apiConfig && typeof apiConfig.signatureVersion !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'signatureVersion must be a string');
    if ('endpoint' in apiConfig && typeof apiConfig.endpoint !== 'string') throw new BoxError(BoxError.BAD_FIELD, 'endpoint must be a string');

    if ('acceptSelfSignedCerts' in apiConfig && typeof apiConfig.acceptSelfSignedCerts !== 'boolean') throw new BoxError(BoxError.BAD_FIELD, 'acceptSelfSignedCerts must be a boolean');
    if ('s3ForcePathStyle' in apiConfig && typeof apiConfig.s3ForcePathStyle !== 'boolean') throw new BoxError(BoxError.BAD_FIELD, 's3ForcePathStyle must be a boolean');

    // attempt to upload and delete a file with new credentials
    const credentials = getS3Config(apiConfig);

    const putParams = {
        Bucket: apiConfig.bucket,
        Key: path.join(apiConfig.prefix, 'snapshot/cloudron-testfile'),
        Body: 'testcontent'
    };

    const s3 = new aws.S3(_.omit(credentials, 'retryDelayOptions', 'maxRetries'));
    const [putError] = await safe(s3.putObject(putParams).promise());
    if (putError) throw new BoxError(BoxError.EXTERNAL_ERROR, `Error put object cloudron-testfile. Message: ${putError.message} HTTP Code: ${putError.code}`);

    const listParams = {
        Bucket: apiConfig.bucket,
        Prefix: path.join(apiConfig.prefix, 'snapshot'),
        MaxKeys: 1
    };

    const [listError] = await safe(s3.listObjects(listParams).promise());
    if (listError) throw new BoxError(BoxError.EXTERNAL_ERROR, `Error listing objects. Message: ${listError.message} HTTP Code: ${listError.code}`);

    const delParams = {
        Bucket: apiConfig.bucket,
        Key: path.join(apiConfig.prefix, 'snapshot/cloudron-testfile')
    };

    const [delError] = await safe(s3.deleteObject(delParams).promise());
    if (delError) throw new BoxError(BoxError.EXTERNAL_ERROR, `Error del object cloudron-testfile. Message: ${delError.message} HTTP Code: ${delError.code}`);
}

function removePrivateFields(apiConfig) {
    apiConfig.secretAccessKey = constants.SECRET_PLACEHOLDER;
    return apiConfig;
}

function injectPrivateFields(newConfig, currentConfig) {
    if (newConfig.secretAccessKey === constants.SECRET_PLACEHOLDER) newConfig.secretAccessKey = currentConfig.secretAccessKey;
}