mirror of
https://github.com/BobbyWibowo/lolisafe.git
synced 2024-12-13 16:06:21 +00:00
1696 lines
54 KiB
JavaScript
1696 lines
54 KiB
JavaScript
const blake3 = require('blake3')
|
|
const fetch = require('node-fetch')
|
|
const fs = require('fs')
|
|
const path = require('path')
|
|
const randomstring = require('randomstring')
|
|
const searchQuery = require('search-query-parser')
|
|
const paths = require('./pathsController')
|
|
const perms = require('./permissionController')
|
|
const utils = require('./utilsController')
|
|
const ClientError = require('./utils/ClientError')
|
|
const ServerError = require('./utils/ServerError')
|
|
const config = require('./../config')
|
|
const logger = require('./../logger')
|
|
|
|
const self = {
|
|
onHold: new Set(),
|
|
scanHelpers: {}
|
|
}
|
|
|
|
/** Preferences */
|
|
|
|
const fileIdentifierLengthFallback = 32
|
|
const fileIdentifierLengthChangeable = !config.uploads.fileIdentifierLength.force &&
|
|
typeof config.uploads.fileIdentifierLength.min === 'number' &&
|
|
typeof config.uploads.fileIdentifierLength.max === 'number'
|
|
|
|
const maxSize = parseInt(config.uploads.maxSize)
|
|
const maxSizeBytes = maxSize * 1e6
|
|
const urlMaxSize = parseInt(config.uploads.urlMaxSize)
|
|
const urlMaxSizeBytes = urlMaxSize * 1e6
|
|
|
|
const maxFilesPerUpload = 20
|
|
|
|
const chunkedUploads = config.uploads.chunkSize &&
|
|
typeof config.uploads.chunkSize === 'object' &&
|
|
config.uploads.chunkSize.default
|
|
const chunkedUploadsTimeout = config.uploads.chunkSize.timeout || 1800000
|
|
const chunksData = {}
|
|
// Hard-coded min chunk size of 1 MB (e.g. 50 MB = max 50 chunks)
|
|
const maxChunksCount = maxSize
|
|
// Use fs.copyFile() instead of fs.rename() if chunks dir is NOT inside uploads dir
|
|
const chunksCopyFile = !paths.chunks.startsWith(paths.uploads)
|
|
|
|
const extensionsFilter = Array.isArray(config.extensionsFilter) &&
|
|
config.extensionsFilter.length
|
|
const urlExtensionsFilter = Array.isArray(config.uploads.urlExtensionsFilter) &&
|
|
config.uploads.urlExtensionsFilter.length
|
|
|
|
/** Chunks helper class & function **/
|
|
|
|
class ChunksData {
|
|
constructor (uuid, root) {
|
|
this.uuid = uuid
|
|
this.root = root
|
|
this.filename = 'tmp'
|
|
this.chunks = 0
|
|
this.writeStream = null
|
|
this.hashStream = null
|
|
}
|
|
|
|
onTimeout () {
|
|
self.cleanUpChunks(this.uuid)
|
|
}
|
|
|
|
setTimeout (delay) {
|
|
this.clearTimeout()
|
|
this._timeout = setTimeout(this.onTimeout.bind(this), delay)
|
|
}
|
|
|
|
clearTimeout () {
|
|
if (this._timeout) {
|
|
clearTimeout(this._timeout)
|
|
}
|
|
}
|
|
}
|
|
|
|
const initChunks = async uuid => {
|
|
if (chunksData[uuid] === undefined) {
|
|
const root = path.join(paths.chunks, uuid)
|
|
try {
|
|
await paths.access(root)
|
|
} catch (err) {
|
|
// Re-throw error
|
|
if (err && err.code !== 'ENOENT') throw err
|
|
await paths.mkdir(root)
|
|
}
|
|
chunksData[uuid] = new ChunksData(uuid, root)
|
|
}
|
|
chunksData[uuid].setTimeout(chunkedUploadsTimeout)
|
|
return chunksData[uuid]
|
|
}
|
|
|
|
/** Helper functions */
|
|
|
|
self.isExtensionFiltered = extname => {
|
|
// If empty extension needs to be filtered
|
|
if (!extname && config.filterNoExtension) return true
|
|
|
|
// If there are extensions that have to be filtered
|
|
if (extname && extensionsFilter) {
|
|
const match = config.extensionsFilter.includes(extname.toLowerCase())
|
|
const whitelist = config.extensionsFilterMode === 'whitelist'
|
|
if ((!whitelist && match) || (whitelist && !match)) return true
|
|
}
|
|
|
|
return false
|
|
}
|
|
|
|
self.parseFileIdentifierLength = fileLength => {
|
|
if (!config.uploads.fileIdentifierLength) return fileIdentifierLengthFallback
|
|
|
|
const parsed = parseInt(fileLength)
|
|
if (isNaN(parsed) ||
|
|
!fileIdentifierLengthChangeable ||
|
|
parsed < config.uploads.fileIdentifierLength.min ||
|
|
parsed > config.uploads.fileIdentifierLength.max) {
|
|
return config.uploads.fileIdentifierLength.default || fileIdentifierLengthFallback
|
|
} else {
|
|
return parsed
|
|
}
|
|
}
|
|
|
|
self.getUniqueRandomName = async (length, extension) => {
|
|
for (let i = 0; i < utils.idMaxTries; i++) {
|
|
const identifier = randomstring.generate(length)
|
|
const name = identifier + extension
|
|
if (config.uploads.cacheFileIdentifiers) {
|
|
if (utils.idSet.has(identifier)) {
|
|
logger.log(`Identifier ${identifier} is already in use (${i + 1}/${utils.idMaxTries}).`)
|
|
continue
|
|
}
|
|
utils.idSet.add(identifier)
|
|
logger.debug(`Added ${identifier} to identifiers cache`)
|
|
} else if (config.uploads.queryDbForFileCollisions) {
|
|
if (self.onHold.has(identifier)) continue
|
|
|
|
// Put token on-hold (wait for it to be inserted to DB)
|
|
self.onHold.add(identifier)
|
|
|
|
const file = await utils.db.table('files')
|
|
.whereRaw('?? like ?', ['name', `${identifier}.%`])
|
|
.select('id')
|
|
.first()
|
|
if (file) {
|
|
self.onHold.delete(identifier)
|
|
logger.log(`Identifier ${identifier} is already in use (${i + 1}/${utils.idMaxTries}).`)
|
|
continue
|
|
}
|
|
} else {
|
|
try {
|
|
await paths.access(path.join(paths.uploads, name))
|
|
logger.log(`${name} is already in use (${i + 1}/${utils.idMaxTries}).`)
|
|
continue
|
|
} catch (error) {
|
|
// Re-throw non-ENOENT error
|
|
if (error & error.code !== 'ENOENT') throw error
|
|
}
|
|
}
|
|
return name
|
|
}
|
|
|
|
throw new ServerError('Failed to allocate a unique name for the upload. Try again?')
|
|
}
|
|
|
|
self.assertRetentionPeriod = (user, age) => {
|
|
if (!utils.retentions.enabled) return null
|
|
|
|
const group = user ? perms.group(user) : '_'
|
|
if (!group || !utils.retentions.periods[group]) {
|
|
throw new ClientError('You are not eligible for any file retention periods.', { statusCode: 403 })
|
|
}
|
|
|
|
let parsed = parseFloat(age)
|
|
if (Number.isNaN(parsed) || age < 0) {
|
|
parsed = utils.retentions.default[group]
|
|
} else if (!utils.retentions.periods[group].includes(parsed)) {
|
|
throw new ClientError('You are not eligible for the specified file retention period.', { statusCode: 403 })
|
|
}
|
|
|
|
if (!parsed && !utils.retentions.periods[group].includes(0)) {
|
|
throw new ClientError('Permanent uploads are not permitted.', { statusCode: 403 })
|
|
}
|
|
|
|
return parsed
|
|
}
|
|
|
|
self.parseStripTags = stripTags => {
|
|
if (!config.uploads.stripTags) return false
|
|
|
|
if (config.uploads.stripTags.force || stripTags === undefined) {
|
|
return config.uploads.stripTags.default
|
|
}
|
|
|
|
return Boolean(parseInt(stripTags))
|
|
}
|
|
|
|
/** File uploads */
|
|
|
|
self.upload = async (req, res) => {
|
|
let user
|
|
if (config.private === true) {
|
|
user = await utils.authorize(req)
|
|
} else if (req.headers.token) {
|
|
user = await utils.assertUser(req.headers.token)
|
|
}
|
|
|
|
if (config.privateUploadGroup) {
|
|
if (!user || !perms.is(user, config.privateUploadGroup)) {
|
|
throw new ClientError(config.privateUploadCustomResponse || 'Your usergroup is not permitted to upload new files.', { statusCode: 403 })
|
|
}
|
|
}
|
|
|
|
let albumid = parseInt(req.headers.albumid || req.params.albumid)
|
|
if (isNaN(albumid)) albumid = null
|
|
|
|
const age = self.assertRetentionPeriod(user, req.headers.age)
|
|
|
|
// Init empty Request.body
|
|
req.body = {}
|
|
|
|
// Initially try to parse as multipart
|
|
let hasMultipartField = true
|
|
|
|
await req.multipart({
|
|
// https://github.com/mscdex/busboy/tree/v1.6.0#exports
|
|
limits: {
|
|
fileSize: maxSizeBytes,
|
|
// Maximum number of non-file fields.
|
|
// Dropzone.js will add 6 extra fields for chunked uploads.
|
|
// We don't use them for anything else.
|
|
fields: 6,
|
|
// Maximum number of file fields.
|
|
// Chunked uploads still need to provide ONLY 1 file field.
|
|
// Otherwise, only one of the files will end up being properly stored,
|
|
// and that will also be as a chunk.
|
|
files: maxFilesPerUpload
|
|
}
|
|
}, async field => {
|
|
hasMultipartField = true
|
|
|
|
// Keep non-files fields in Request.body
|
|
// Since fields get processed in sequence depending on the order at which they were defined,
|
|
// chunked uploads data must be set before the files[] field which contain the actual file
|
|
if (field.truncated) {
|
|
// Re-map Dropzone chunked uploads keys so people can manually use the API without prepending 'dz'
|
|
let name = field.name
|
|
if (name.startsWith('dz')) {
|
|
name = name.replace(/^dz/, '')
|
|
}
|
|
|
|
req.body[name] = field.value
|
|
return
|
|
}
|
|
|
|
// Process files immediately and push into Request.files array
|
|
if (field.file) {
|
|
// Init Request.files array if not previously set
|
|
if (req.files === undefined) {
|
|
req.files = []
|
|
}
|
|
|
|
// Push immediately as we will only be adding props into the file object down the line
|
|
const file = {
|
|
albumid,
|
|
age,
|
|
mimetype: field.mime_type,
|
|
isChunk: req.body.uuid !== undefined &&
|
|
req.body.chunkindex !== undefined
|
|
}
|
|
req.files.push(file)
|
|
|
|
if (file.isChunk) {
|
|
if (!chunkedUploads) {
|
|
throw new ClientError('Chunked uploads are disabled at the moment.')
|
|
} else if (req.files.length > 1) {
|
|
throw new ClientError('Chunked uploads may only be uploaded 1 chunk at a time.')
|
|
}
|
|
}
|
|
|
|
// NOTE: Since busboy@1, filenames no longer get automatically parsed as UTF-8, so we force it here
|
|
file.originalname = field.file.name &&
|
|
Buffer.from(field.file.name, 'latin1').toString('utf8')
|
|
|
|
file.extname = utils.extname(file.originalname)
|
|
if (self.isExtensionFiltered(file.extname)) {
|
|
throw new ClientError(`${file.extname ? `${file.extname.substr(1).toUpperCase()} files` : 'Files with no extension'} are not permitted.`)
|
|
}
|
|
|
|
let destination
|
|
if (file.isChunk) {
|
|
// Calling this will also reset this chunked uploads' timeout
|
|
file.chunksData = await initChunks(req.body.uuid)
|
|
file.filename = file.chunksData.filename
|
|
destination = file.chunksData.root
|
|
} else {
|
|
const length = self.parseFileIdentifierLength(req.headers.filelength)
|
|
file.filename = await self.getUniqueRandomName(length, file.extname)
|
|
destination = paths.uploads
|
|
}
|
|
|
|
file.path = path.join(destination, file.filename)
|
|
|
|
// Write the file into disk, and supply required props into file object
|
|
await new Promise((resolve, reject) => {
|
|
// "weighted" resolve function, to be able to "await" multiple callbacks
|
|
const REQUIRED_WEIGHT = 2
|
|
let _weight = 0
|
|
const _resolve = (props = {}, weight = 2) => {
|
|
if (file.error) return
|
|
Object.assign(file, props)
|
|
_weight += weight
|
|
if (_weight >= REQUIRED_WEIGHT) {
|
|
return resolve()
|
|
}
|
|
}
|
|
|
|
const readStream = field.file.stream
|
|
let writeStream
|
|
let hashStream
|
|
let scanStream
|
|
|
|
const _reject = error => {
|
|
file.error = true
|
|
if (writeStream && !writeStream.closed) {
|
|
writeStream.end()
|
|
}
|
|
if (hashStream.hash.hash) {
|
|
hashStream.dispose()
|
|
}
|
|
return reject(error)
|
|
}
|
|
|
|
if (file.isChunk) {
|
|
if (!file.chunksData.writeStream) {
|
|
file.chunksData.writeStream = fs.createWriteStream(file.path, { flags: 'a' })
|
|
file.chunksData.writeStream.on('error', _reject)
|
|
}
|
|
if (!file.chunksData.hashStream) {
|
|
file.chunksData.hashStream = blake3.createHash()
|
|
}
|
|
|
|
writeStream = file.chunksData.writeStream
|
|
hashStream = file.chunksData.hashStream
|
|
} else {
|
|
writeStream = fs.createWriteStream(file.path)
|
|
writeStream.on('error', _reject)
|
|
hashStream = blake3.createHash()
|
|
|
|
if (utils.scan.passthrough &&
|
|
!self.scanHelpers.assertUserBypass(req._user, file.filename) &&
|
|
!self.scanHelpers.assertFileBypass({ filename: file.filename })) {
|
|
scanStream = utils.scan.instance.passthrough()
|
|
}
|
|
}
|
|
|
|
readStream.on('error', _reject)
|
|
readStream.on('data', d => {
|
|
// .dispose() will destroy this internal component,
|
|
// so use it as an indicator of whether the hashStream has been .dispose()'d
|
|
if (hashStream.hash.hash) {
|
|
hashStream.update(d)
|
|
}
|
|
})
|
|
|
|
if (file.isChunk) {
|
|
readStream.on('end', () => _resolve())
|
|
readStream.pipe(writeStream, { end: false })
|
|
} else {
|
|
// Callback's weight is 1 when passthrough scanning is enabled,
|
|
// so that the Promise will be resolved only after
|
|
// both writeStream and scanStream finish
|
|
writeStream.on('finish', () => _resolve({
|
|
size: writeStream.bytesWritten,
|
|
hash: hashStream.hash.hash ? hashStream.digest('hex') : null
|
|
}, scanStream ? 1 : 2))
|
|
|
|
if (scanStream) {
|
|
logger.debug(`[ClamAV]: ${file.filename}: Passthrough scanning\u2026`)
|
|
scanStream.on('error', _reject)
|
|
scanStream.on('scan-complete', scan => _resolve({
|
|
scan
|
|
}, 1))
|
|
readStream
|
|
.pipe(scanStream)
|
|
.pipe(writeStream)
|
|
} else {
|
|
readStream
|
|
.pipe(writeStream)
|
|
}
|
|
}
|
|
})
|
|
|
|
if (config.filterEmptyFile && file.size === 0) {
|
|
throw new ClientError('Empty files are not allowed.')
|
|
}
|
|
}
|
|
}).catch(error => {
|
|
// Unlink temp files (do not wait)
|
|
if (Array.isArray(req.files)) {
|
|
Promise.all(req.files.map(async file => {
|
|
if (file.filename) {
|
|
return utils.unlinkFile(file.filename).catch(logger.error)
|
|
}
|
|
}))
|
|
}
|
|
|
|
// res.multipart() itself may throw string errors
|
|
if (typeof error === 'string') {
|
|
throw new ClientError(error)
|
|
} else {
|
|
throw error
|
|
}
|
|
})
|
|
|
|
// If Request connection dropped during multiform upload
|
|
if (req._readableState.closed) {
|
|
return
|
|
}
|
|
|
|
if (hasMultipartField) {
|
|
return self.actuallyUpload(req, res, user)
|
|
} else {
|
|
// Parse POST body
|
|
req.body = await req.json()
|
|
return self.actuallyUploadUrls(req, res, user, { albumid, age })
|
|
}
|
|
}
|
|
|
|
self.actuallyUpload = async (req, res, user, data = {}) => {
|
|
if (!req.files || !req.files.length) {
|
|
throw new ClientError('No files.')
|
|
}
|
|
|
|
// If chunked uploads is enabled and the uploaded file is a chunk, then just say that it was a success
|
|
const uuid = req.body.uuid
|
|
if (chunkedUploads && chunksData[uuid] !== undefined) {
|
|
req.files.forEach(file => {
|
|
chunksData[uuid].chunks++
|
|
})
|
|
return res.json({ success: true })
|
|
}
|
|
|
|
const filesData = req.files
|
|
|
|
if (utils.scan.instance) {
|
|
let scanResult
|
|
if (utils.scan.passthrough) {
|
|
scanResult = await self.assertPassthroughScans(req, user, filesData)
|
|
} else {
|
|
scanResult = await self.scanFiles(req, user, filesData)
|
|
}
|
|
if (scanResult) {
|
|
throw new ClientError(scanResult)
|
|
}
|
|
}
|
|
|
|
await self.stripTags(req, filesData)
|
|
|
|
const result = await self.storeFilesToDb(req, res, user, filesData)
|
|
return self.sendUploadResponse(req, res, user, result)
|
|
}
|
|
|
|
/** URL uploads */
|
|
|
|
self.actuallyUploadUrls = async (req, res, user, data = {}) => {
|
|
if (!config.uploads.urlMaxSize) {
|
|
throw new ClientError('Upload by URLs is disabled at the moment.', { statusCode: 403 })
|
|
}
|
|
|
|
const urls = req.body.urls
|
|
if (!urls || !(urls instanceof Array)) {
|
|
throw new ClientError('Missing "urls" property (array).')
|
|
}
|
|
|
|
if (urls.length > maxFilesPerUpload) {
|
|
throw new ClientError(`Maximum ${maxFilesPerUpload} URLs at a time.`)
|
|
}
|
|
|
|
const assertSize = (size, isContentLength = false) => {
|
|
if (config.filterEmptyFile && size === 0) {
|
|
throw new ClientError('Empty files are not allowed.')
|
|
} else if (size > urlMaxSizeBytes) {
|
|
if (isContentLength) {
|
|
throw new ClientError(`File too large. Content-Length header reports file is bigger than ${urlMaxSize} MB.`)
|
|
} else {
|
|
throw new ClientError(`File too large. File is bigger than ${urlMaxSize} MB.`)
|
|
}
|
|
}
|
|
}
|
|
|
|
const filesData = []
|
|
|
|
await Promise.all(urls.map(async url => {
|
|
// Push immediately as we will only be adding props into the file object down the line
|
|
const file = {
|
|
url,
|
|
albumid: data.albumid,
|
|
age: data.age
|
|
}
|
|
filesData.push(file)
|
|
|
|
file.originalname = path.basename(url).split(/[?#]/)[0]
|
|
file.extname = utils.extname(file.originalname)
|
|
|
|
// Extensions filter
|
|
let filtered = false
|
|
if (urlExtensionsFilter && ['blacklist', 'whitelist'].includes(config.uploads.urlExtensionsFilterMode)) {
|
|
const match = config.uploads.urlExtensionsFilter.includes(file.extname.toLowerCase())
|
|
const whitelist = config.uploads.urlExtensionsFilterMode === 'whitelist'
|
|
filtered = ((!whitelist && match) || (whitelist && !match))
|
|
} else {
|
|
filtered = self.isExtensionFiltered(file.extname)
|
|
}
|
|
|
|
if (filtered) {
|
|
throw new ClientError(`${file.extname ? `${file.extname.substr(1).toUpperCase()} files` : 'Files with no extension'} are not permitted.`)
|
|
}
|
|
|
|
if (config.uploads.urlProxy) {
|
|
url = config.uploads.urlProxy
|
|
.replace(/{url}/g, encodeURIComponent(url))
|
|
.replace(/{url-noprot}/g, encodeURIComponent(url.replace(/^https?:\/\//, '')))
|
|
}
|
|
|
|
// Try to determine size early via Content-Length header,
|
|
// but continue anyway if it isn't a valid number
|
|
try {
|
|
const head = await fetch(url, { method: 'HEAD', size: urlMaxSizeBytes })
|
|
if (head.status === 200) {
|
|
const contentLength = parseInt(head.headers.get('content-length'))
|
|
if (!Number.isNaN(contentLength)) {
|
|
assertSize(contentLength, true)
|
|
}
|
|
}
|
|
} catch (ex) {
|
|
// Re-throw only if ClientError, otherwise ignore
|
|
if (ex instanceof ClientError) {
|
|
throw ex
|
|
}
|
|
}
|
|
|
|
const length = self.parseFileIdentifierLength(req.headers.filelength)
|
|
file.filename = await self.getUniqueRandomName(length, file.extname)
|
|
file.path = path.join(paths.uploads, file.filename)
|
|
|
|
let writeStream
|
|
let hashStream
|
|
return Promise.resolve().then(async () => {
|
|
writeStream = fs.createWriteStream(file.path)
|
|
hashStream = blake3.createHash()
|
|
|
|
// Limit max response body size with maximum allowed size
|
|
const fetchFile = await fetch(url, { method: 'GET', size: urlMaxSizeBytes })
|
|
.then(res => new Promise((resolve, reject) => {
|
|
if (res.status === 200) {
|
|
writeStream.on('error', reject)
|
|
res.body.on('error', reject)
|
|
res.body.on('data', d => hashStream.update(d))
|
|
|
|
res.body.pipe(writeStream)
|
|
writeStream.on('finish', () => resolve(res))
|
|
} else {
|
|
resolve(res)
|
|
}
|
|
}))
|
|
|
|
if (fetchFile.status !== 200) {
|
|
throw new ServerError(`${fetchFile.status} ${fetchFile.statusText}`)
|
|
}
|
|
|
|
// Re-test size via actual bytes written to physical file
|
|
assertSize(writeStream.bytesWritten)
|
|
|
|
// Finalize file props
|
|
const contentType = fetchFile.headers.get('content-type')
|
|
file.mimetype = contentType ? contentType.split(';')[0] : 'application/octet-stream'
|
|
file.size = writeStream.bytesWritten
|
|
file.hash = hashStream.digest('hex')
|
|
}).catch(err => {
|
|
// Dispose of unfinished write & hasher streams
|
|
if (writeStream && !writeStream.destroyed) {
|
|
writeStream.destroy()
|
|
}
|
|
try {
|
|
if (hashStream) {
|
|
hashStream.dispose()
|
|
}
|
|
} catch (_) {}
|
|
|
|
// Re-throw errors
|
|
throw err
|
|
})
|
|
})).catch(async error => {
|
|
// Unlink temp files (do not wait)
|
|
if (filesData.length) {
|
|
Promise.all(filesData.map(async file =>
|
|
utils.unlinkFile(file.filename).catch(logger.error)
|
|
))
|
|
}
|
|
|
|
// Re-throw suppressed errors as ClientError, otherwise as-is
|
|
const errorString = error.toString()
|
|
const suppress = [
|
|
/ over limit:/
|
|
]
|
|
if (suppress.some(t => t.test(errorString))) {
|
|
throw new ClientError(errorString)
|
|
} else {
|
|
throw error
|
|
}
|
|
})
|
|
|
|
if (utils.scan.instance) {
|
|
const scanResult = await self.scanFiles(req, user, filesData)
|
|
if (scanResult) throw new ClientError(scanResult)
|
|
}
|
|
|
|
const result = await self.storeFilesToDb(req, res, user, filesData)
|
|
return self.sendUploadResponse(req, res, user, result)
|
|
}
|
|
|
|
/** Chunk uploads */
|
|
|
|
self.finishChunks = async (req, res) => {
|
|
if (!chunkedUploads) {
|
|
throw new ClientError('Chunked upload is disabled.', { statusCode: 403 })
|
|
}
|
|
|
|
let user
|
|
if (config.private === true) {
|
|
user = await utils.authorize(req)
|
|
if (!user) return
|
|
} else if (req.headers.token) {
|
|
user = await utils.assertUser(req.headers.token)
|
|
}
|
|
|
|
// Parse POST body
|
|
req.body = await req.json()
|
|
|
|
const files = req.body.files
|
|
if (!Array.isArray(files) || !files.length) {
|
|
throw new ClientError('Bad request.')
|
|
}
|
|
|
|
return self.actuallyFinishChunks(req, res, user, files)
|
|
.catch(error => {
|
|
// Unlink temp files (do not wait)
|
|
Promise.all(files.map(async file => {
|
|
if (file.uuid && chunksData[file.uuid]) {
|
|
return self.cleanUpChunks(file.uuid).catch(logger.error)
|
|
}
|
|
}))
|
|
// Re-throw errors
|
|
throw error
|
|
})
|
|
}
|
|
|
|
self.actuallyFinishChunks = async (req, res, user, files) => {
|
|
const filesData = []
|
|
await Promise.all(files.map(async file => {
|
|
if (!file.uuid || typeof chunksData[file.uuid] === 'undefined') {
|
|
throw new ClientError('Invalid file UUID, or chunks data had already timed out. Try again?')
|
|
}
|
|
|
|
// Suspend timeout
|
|
// If the chunk errors out there, it will be immediately cleaned up anyway
|
|
chunksData[file.uuid].clearTimeout()
|
|
|
|
// Conclude write and hasher streams
|
|
chunksData[file.uuid].writeStream.end()
|
|
const hash = chunksData[file.uuid].hashStream.digest('hex')
|
|
|
|
if (chunksData[file.uuid].chunks < 2 || chunksData[file.uuid].chunks > maxChunksCount) {
|
|
throw new ClientError('Invalid chunks count.')
|
|
}
|
|
|
|
file.extname = typeof file.original === 'string' ? utils.extname(file.original) : ''
|
|
if (self.isExtensionFiltered(file.extname)) {
|
|
throw new ClientError(`${file.extname ? `${file.extname.substr(1).toUpperCase()} files` : 'Files with no extension'} are not permitted.`)
|
|
}
|
|
|
|
file.age = self.assertRetentionPeriod(user, file.age)
|
|
|
|
file.size = chunksData[file.uuid].writeStream.bytesWritten
|
|
if (config.filterEmptyFile && file.size === 0) {
|
|
throw new ClientError('Empty files are not allowed.')
|
|
} else if (file.size > maxSizeBytes) {
|
|
throw new ClientError(`File too large. Chunks are bigger than ${maxSize} MB.`)
|
|
}
|
|
|
|
// Double-check file size
|
|
const tmpfile = path.join(chunksData[file.uuid].root, chunksData[file.uuid].filename)
|
|
const lstat = await paths.lstat(tmpfile)
|
|
if (lstat.size !== file.size) {
|
|
throw new ClientError(`File size mismatched (${lstat.size} vs. ${file.size}).`)
|
|
}
|
|
|
|
// Generate name
|
|
const length = self.parseFileIdentifierLength(file.filelength)
|
|
const name = await self.getUniqueRandomName(length, file.extname)
|
|
|
|
// Move tmp file to final destination
|
|
// For fs.copyFile(), tmpfile will eventually be unlinked by self.cleanUpChunks()
|
|
const destination = path.join(paths.uploads, name)
|
|
if (chunksCopyFile) {
|
|
await paths.copyFile(tmpfile, destination)
|
|
} else {
|
|
await paths.rename(tmpfile, destination)
|
|
}
|
|
|
|
// Continue even when encountering errors
|
|
await self.cleanUpChunks(file.uuid).catch(logger.error)
|
|
|
|
let albumid = parseInt(file.albumid)
|
|
if (isNaN(albumid)) albumid = null
|
|
|
|
filesData.push({
|
|
filename: name,
|
|
originalname: file.original || '',
|
|
extname: file.extname,
|
|
mimetype: file.type || '',
|
|
path: destination,
|
|
size: file.size,
|
|
hash,
|
|
albumid,
|
|
age: file.age
|
|
})
|
|
}))
|
|
|
|
if (utils.scan.instance) {
|
|
const scanResult = await self.scanFiles(req, user, filesData)
|
|
if (scanResult) throw new ClientError(scanResult)
|
|
}
|
|
|
|
await self.stripTags(req, filesData)
|
|
|
|
const result = await self.storeFilesToDb(req, res, user, filesData)
|
|
return self.sendUploadResponse(req, res, user, result)
|
|
}
|
|
|
|
self.cleanUpChunks = async uuid => {
|
|
// Dispose of unfinished write & hasher streams
|
|
if (chunksData[uuid].writeStream && !chunksData[uuid].writeStream.destroyed) {
|
|
chunksData[uuid].writeStream.destroy()
|
|
}
|
|
try {
|
|
if (chunksData[uuid].hashStream) {
|
|
chunksData[uuid].hashStream.dispose()
|
|
}
|
|
} catch (_) {}
|
|
|
|
// Remove tmp file
|
|
await paths.unlink(path.join(chunksData[uuid].root, chunksData[uuid].filename))
|
|
.catch(error => {
|
|
// Re-throw non-ENOENT error
|
|
if (error.code !== 'ENOENT') logger.error(error)
|
|
})
|
|
|
|
// Remove UUID dir
|
|
await paths.rmdir(chunksData[uuid].root)
|
|
|
|
// Delete cached chunks data
|
|
delete chunksData[uuid]
|
|
}
|
|
|
|
/** Virus scanning (ClamAV) */
|
|
|
|
self.scanHelpers.assertUserBypass = (user, filenames) => {
|
|
if (!user || !utils.scan.groupBypass) return false
|
|
if (!Array.isArray(filenames)) filenames = [filenames]
|
|
logger.debug(`[ClamAV]: ${filenames.join(', ')}: Skipped, uploaded by ${user.username} (${utils.scan.groupBypass})`)
|
|
return perms.is(user, utils.scan.groupBypass)
|
|
}
|
|
|
|
self.scanHelpers.assertFileBypass = data => {
|
|
if (typeof data !== 'object' || !data.filename) return false
|
|
|
|
const extname = data.extname || utils.extname(data.filename)
|
|
if (utils.scan.whitelistExtensions && utils.scan.whitelistExtensions.includes(extname)) {
|
|
logger.debug(`[ClamAV]: ${data.filename}: Skipped, extension whitelisted`)
|
|
return true
|
|
}
|
|
|
|
if (utils.scan.maxSize && Number.isFinite(data.size) && data.size > utils.scan.maxSize) {
|
|
logger.debug(`[ClamAV]: ${data.filename}: Skipped, size ${data.size} > ${utils.scan.maxSize}`)
|
|
return true
|
|
}
|
|
|
|
return false
|
|
}
|
|
|
|
self.assertPassthroughScans = async (req, user, filesData) => {
|
|
const foundThreats = []
|
|
const unableToScan = []
|
|
|
|
for (const file of filesData) {
|
|
if (file.scan) {
|
|
if (file.scan.isInfected) {
|
|
logger.log(`[ClamAV]: ${file.filename}: ${file.scan.viruses.join(', ')}`)
|
|
foundThreats.push(...file.scan.viruses)
|
|
} else if (file.scan.isInfected === null) {
|
|
logger.log(`[ClamAV]: ${file.filename}: Unable to scan`)
|
|
unableToScan.push(file.filename)
|
|
} else {
|
|
logger.debug(`[ClamAV]: ${file.filename}: File is clean`)
|
|
}
|
|
}
|
|
}
|
|
|
|
let result = ''
|
|
if (foundThreats.length) {
|
|
const more = foundThreats.length > 1
|
|
result = `Threat${more ? 's' : ''} detected: ${foundThreats[0]}${more ? ', and more' : ''}.`
|
|
} else if (unableToScan.length) {
|
|
const more = unableToScan.length > 1
|
|
result = `Unable to scan: ${unableToScan[0]}${more ? ', and more' : ''}.`
|
|
}
|
|
|
|
if (result) {
|
|
// Unlink temp files (do not wait)
|
|
Promise.all(filesData.map(async file =>
|
|
utils.unlinkFile(file.filename).catch(logger.error)
|
|
))
|
|
}
|
|
|
|
return result
|
|
}
|
|
|
|
self.scanFiles = async (req, user, filesData) => {
|
|
const filenames = filesData.map(file => file.filename)
|
|
if (self.scanHelpers.assertUserBypass(user, filenames)) {
|
|
return false
|
|
}
|
|
|
|
const foundThreats = []
|
|
const unableToScan = []
|
|
const result = await Promise.all(filesData.map(async file => {
|
|
if (self.scanHelpers.assertFileBypass({
|
|
filename: file.filename,
|
|
extname: file.extname,
|
|
size: file.size
|
|
})) return
|
|
|
|
logger.debug(`[ClamAV]: ${file.filename}: Scanning\u2026`)
|
|
const response = await utils.scan.instance.isInfected(file.path)
|
|
if (response.isInfected) {
|
|
logger.log(`[ClamAV]: ${file.filename}: ${response.viruses.join(', ')}`)
|
|
foundThreats.push(...response.viruses)
|
|
} else if (response.isInfected === null) {
|
|
logger.log(`[ClamAV]: ${file.filename}: Unable to scan`)
|
|
unableToScan.push(file.filename)
|
|
} else {
|
|
logger.debug(`[ClamAV]: ${file.filename}: File is clean`)
|
|
}
|
|
})).then(() => {
|
|
if (foundThreats.length) {
|
|
const more = foundThreats.length > 1
|
|
return `Threat${more ? 's' : ''} detected: ${foundThreats[0]}${more ? ', and more' : ''}.`
|
|
} else if (unableToScan.length) {
|
|
const more = unableToScan.length > 1
|
|
return `Unable to scan: ${unableToScan[0]}${more ? ', and more' : ''}.`
|
|
}
|
|
}).catch(error => {
|
|
logger.error(`[ClamAV]: ${filenames.join(', ')}: ${error.toString()}`)
|
|
return 'An unexpected error occurred with ClamAV, please contact the site owner.'
|
|
})
|
|
|
|
if (result) {
|
|
// Unlink temp files (do not wait)
|
|
Promise.all(filesData.map(async file =>
|
|
utils.unlinkFile(file.filename).catch(logger.error)
|
|
))
|
|
}
|
|
|
|
return result
|
|
}
|
|
|
|
/** Strip tags (EXIF, etc.) */
|
|
|
|
self.stripTags = async (req, filesData) => {
|
|
if (!self.parseStripTags(req.headers.striptags)) return
|
|
|
|
try {
|
|
await Promise.all(filesData.map(async file =>
|
|
utils.stripTags(file.filename, file.extname)
|
|
))
|
|
} catch (error) {
|
|
// Unlink temp files (do not wait)
|
|
Promise.all(filesData.map(async file =>
|
|
utils.unlinkFile(file.filename).catch(logger.error)
|
|
))
|
|
|
|
// Re-throw error
|
|
throw error
|
|
}
|
|
}
|
|
|
|
/** Database functions */
|
|
|
|
self.storeFilesToDb = async (req, res, user, filesData) => {
|
|
const files = []
|
|
const exists = []
|
|
const albumids = []
|
|
|
|
await Promise.all(filesData.map(async file => {
|
|
// Check if the file exists by checking its hash and size
|
|
const dbFile = await utils.db.table('files')
|
|
.where(function () {
|
|
if (user === undefined) {
|
|
this.whereNull('userid')
|
|
} else {
|
|
this.where('userid', user.id)
|
|
}
|
|
})
|
|
.where({
|
|
hash: file.hash,
|
|
size: String(file.size)
|
|
})
|
|
// Select expirydate to display expiration date of existing files as well
|
|
.select('name', 'expirydate')
|
|
.first()
|
|
|
|
if (dbFile) {
|
|
// Continue even when encountering errors
|
|
await utils.unlinkFile(file.filename).catch(logger.error)
|
|
logger.debug(`Unlinked ${file.filename} since a duplicate named ${dbFile.name} exists`)
|
|
|
|
// If on /nojs route, append original name reported by client,
|
|
// instead of the actual original name from database
|
|
if (req.path === '/nojs') {
|
|
dbFile.original = file.originalname
|
|
}
|
|
|
|
exists.push(dbFile)
|
|
return
|
|
}
|
|
|
|
const timestamp = Math.floor(Date.now() / 1000)
|
|
const data = {
|
|
name: file.filename,
|
|
original: file.originalname,
|
|
type: file.mimetype,
|
|
size: String(file.size),
|
|
hash: file.hash,
|
|
// Only disable if explicitly set to false in config
|
|
ip: config.uploads.storeIP !== false ? req.ip : null,
|
|
timestamp
|
|
}
|
|
|
|
if (user) {
|
|
data.userid = user.id
|
|
data.albumid = file.albumid
|
|
if (data.albumid !== null && !albumids.includes(data.albumid)) {
|
|
albumids.push(data.albumid)
|
|
}
|
|
}
|
|
|
|
if (file.age) {
|
|
data.expirydate = data.timestamp + (file.age * 3600) // Hours to seconds
|
|
}
|
|
|
|
files.push(data)
|
|
|
|
// Generate thumbs, but do not wait
|
|
if (utils.mayGenerateThumb(file.extname)) {
|
|
utils.generateThumbs(file.filename, file.extname, true).catch(logger.error)
|
|
}
|
|
}))
|
|
|
|
if (files.length) {
|
|
let authorizedIds = []
|
|
if (albumids.length) {
|
|
authorizedIds = await utils.db.table('albums')
|
|
.where({ userid: user.id })
|
|
.whereIn('id', albumids)
|
|
.select('id')
|
|
.then(rows => rows.map(row => row.id))
|
|
|
|
// Remove albumid if user do not own the album
|
|
for (const file of files) {
|
|
if (file.albumid !== null && !authorizedIds.includes(file.albumid)) {
|
|
file.albumid = null
|
|
}
|
|
}
|
|
}
|
|
|
|
// Insert new files to DB
|
|
await utils.db.table('files').insert(files)
|
|
utils.invalidateStatsCache('uploads')
|
|
|
|
if (config.uploads.queryDbForFileCollisions) {
|
|
for (const file of files) {
|
|
const extname = utils.extname(file.name)
|
|
const identifier = file.name.slice(0, -(extname.length))
|
|
self.onHold.delete(identifier)
|
|
}
|
|
}
|
|
|
|
// Update albums' timestamp
|
|
if (authorizedIds.length) {
|
|
await utils.db.table('albums')
|
|
.whereIn('id', authorizedIds)
|
|
.update('editedAt', Math.floor(Date.now() / 1000))
|
|
utils.deleteStoredAlbumRenders(authorizedIds)
|
|
}
|
|
}
|
|
|
|
return [...files, ...exists]
|
|
}
|
|
|
|
/** Final response */
|
|
|
|
self.sendUploadResponse = async (req, res, user, result) => {
|
|
// Send response
|
|
return res.json({
|
|
success: true,
|
|
files: result.map(file => {
|
|
const map = {
|
|
name: file.name,
|
|
url: `${utils.conf.domain ? `${utils.conf.domain}/` : ''}${file.name}`
|
|
}
|
|
|
|
// If a temporary upload, add expiry date
|
|
if (file.expirydate) {
|
|
map.expirydate = file.expirydate
|
|
}
|
|
|
|
// If on /nojs route, add original name
|
|
if (req.path === '/nojs') {
|
|
map.original = file.original
|
|
}
|
|
|
|
// If uploaded by user, add delete URL (intended for ShareX and its derivatives)
|
|
// Homepage uploader will not use this (use dashboard instead)
|
|
if (user) {
|
|
map.deleteUrl = `${utils.conf.homeDomain}/file/${file.name}?delete`
|
|
}
|
|
|
|
return map
|
|
})
|
|
})
|
|
}
|
|
|
|
/** Delete uploads */
|
|
|
|
self.delete = async (req, res) => {
|
|
// Parse POST body and re-map for .bulkDelete()
|
|
// Original API used by lolisafe v3's frontend
|
|
// Meanwhile this fork's frontend uses .bulkDelete() straight away
|
|
req.body = await req.json()
|
|
.then(obj => {
|
|
const id = parseInt(obj.id)
|
|
return {
|
|
field: 'id',
|
|
values: isNaN(id) ? undefined : [id]
|
|
}
|
|
})
|
|
return self.bulkDelete(req, res)
|
|
}
|
|
|
|
self.bulkDelete = async (req, res) => {
|
|
const user = await utils.authorize(req)
|
|
|
|
// Parse POST body, if required
|
|
req.body = req.body || await req.json()
|
|
|
|
const field = req.body.field || 'id'
|
|
const values = req.body.values
|
|
|
|
if (!Array.isArray(values) || !values.length) {
|
|
throw new ClientError('No array of files specified.')
|
|
}
|
|
|
|
const failed = await utils.bulkDeleteFromDb(field, values, user)
|
|
|
|
return res.json({ success: true, failed })
|
|
}
|
|
|
|
/** List uploads */
|
|
|
|
self.list = async (req, res) => {
|
|
const user = await utils.authorize(req)
|
|
|
|
const all = req.headers.all === '1'
|
|
const filters = req.headers.filters
|
|
const minoffset = Number(req.headers.minoffset) || 0
|
|
const ismoderator = perms.is(user, 'moderator')
|
|
if (all && !ismoderator) return res.status(403).end()
|
|
|
|
const basedomain = utils.conf.domain
|
|
|
|
// Thresholds for regular users
|
|
const MAX_WILDCARDS_IN_KEY = 2
|
|
const MAX_TEXT_QUERIES = 3 // non-keyed keywords
|
|
const MAX_SORT_KEYS = 1
|
|
const MAX_IS_KEYS = 1
|
|
|
|
const filterObj = {
|
|
uploaders: [],
|
|
excludeUploaders: [],
|
|
queries: {
|
|
exclude: {}
|
|
},
|
|
typeIs: [
|
|
'image',
|
|
'video',
|
|
'audio'
|
|
],
|
|
flags: {}
|
|
}
|
|
|
|
const sortObj = {
|
|
// Cast columns to specific type if they are stored differently
|
|
casts: {
|
|
size: 'integer'
|
|
},
|
|
// Columns mapping
|
|
maps: {
|
|
date: 'timestamp',
|
|
expiry: 'expirydate',
|
|
originalname: 'original'
|
|
},
|
|
// Columns with which to use SQLite's NULLS LAST option
|
|
nullsLast: [
|
|
'userid',
|
|
'expirydate',
|
|
'ip'
|
|
],
|
|
parsed: []
|
|
}
|
|
|
|
// Parse glob wildcards into SQL wildcards
|
|
function sqlLikeParser (pattern) {
|
|
// Escape SQL operators
|
|
const escaped = pattern
|
|
.replace(/(?<!\\)%/g, '\\%')
|
|
.replace(/(?<!\\)_/g, '\\_')
|
|
|
|
// Look for any glob operators
|
|
const match = pattern.match(/(?<!\\)(\*|\?)/g)
|
|
if (match && match.length) {
|
|
return {
|
|
count: match.length,
|
|
// Replace glob operators with their SQL equivalents
|
|
escaped: escaped
|
|
.replace(/(?<!\\)\*/g, '%')
|
|
.replace(/(?<!\\)\?/g, '_')
|
|
}
|
|
} else {
|
|
return {
|
|
count: 0,
|
|
// Assume partial match
|
|
escaped: `%${escaped}%`
|
|
}
|
|
}
|
|
}
|
|
|
|
if (filters) {
|
|
const keywords = []
|
|
|
|
if (req.params.id === undefined) keywords.push('albumid')
|
|
|
|
// Only allow filtering by 'ip' and 'user' keys when listing all uploads
|
|
if (all) keywords.push('ip', 'user')
|
|
|
|
const ranges = [
|
|
'date',
|
|
'expiry'
|
|
]
|
|
|
|
keywords.push('is', 'sort', 'orderby')
|
|
filterObj.queries = searchQuery.parse(filters, {
|
|
keywords,
|
|
ranges,
|
|
tokenize: true,
|
|
alwaysArray: true,
|
|
offsets: false
|
|
})
|
|
|
|
// Accept orderby as alternative for sort
|
|
if (filterObj.queries.orderby) {
|
|
if (!filterObj.queries.sort) filterObj.queries.sort = []
|
|
filterObj.queries.sort.push(...filterObj.queries.orderby)
|
|
delete filterObj.queries.orderby
|
|
}
|
|
|
|
// For some reason, single value won't be in Array even with 'alwaysArray' option
|
|
if (typeof filterObj.queries.exclude.text === 'string') {
|
|
filterObj.queries.exclude.text = [filterObj.queries.exclude.text]
|
|
}
|
|
|
|
// Text (non-keyed keywords) queries
|
|
let textQueries = 0
|
|
if (filterObj.queries.text) textQueries += filterObj.queries.text.length
|
|
if (filterObj.queries.exclude.text) textQueries += filterObj.queries.exclude.text.length
|
|
|
|
// Regular user threshold check
|
|
if (!ismoderator && textQueries > MAX_TEXT_QUERIES) {
|
|
throw new ClientError(`Users are only allowed to use ${MAX_TEXT_QUERIES} non-keyed keyword${MAX_TEXT_QUERIES === 1 ? '' : 's'} at a time.`)
|
|
}
|
|
|
|
if (filterObj.queries.text) {
|
|
for (let i = 0; i < filterObj.queries.text.length; i++) {
|
|
const result = sqlLikeParser(filterObj.queries.text[i])
|
|
if (!ismoderator && result.count > MAX_WILDCARDS_IN_KEY) {
|
|
throw new ClientError(`Users are only allowed to use ${MAX_WILDCARDS_IN_KEY} wildcard${MAX_WILDCARDS_IN_KEY === 1 ? '' : 's'} per key.`)
|
|
}
|
|
filterObj.queries.text[i] = result.escaped
|
|
}
|
|
}
|
|
|
|
if (filterObj.queries.exclude.text) {
|
|
for (let i = 0; i < filterObj.queries.exclude.text.length; i++) {
|
|
const result = sqlLikeParser(filterObj.queries.exclude.text[i])
|
|
if (!ismoderator && result.count > MAX_WILDCARDS_IN_KEY) {
|
|
throw new ClientError(`Users are only allowed to use ${MAX_WILDCARDS_IN_KEY} wildcard${MAX_WILDCARDS_IN_KEY === 1 ? '' : 's'} per key.`)
|
|
}
|
|
filterObj.queries.exclude.text[i] = result.escaped
|
|
}
|
|
}
|
|
|
|
for (const key of keywords) {
|
|
let queryIndex = -1
|
|
let excludeIndex = -1
|
|
|
|
// Make sure keyword arrays only contain unique values
|
|
if (filterObj.queries[key]) {
|
|
filterObj.queries[key] = filterObj.queries[key].filter((v, i, a) => a.indexOf(v) === i)
|
|
queryIndex = filterObj.queries[key].indexOf('-')
|
|
}
|
|
if (filterObj.queries.exclude[key]) {
|
|
filterObj.queries.exclude[key] = filterObj.queries.exclude[key].filter((v, i, a) => a.indexOf(v) === i)
|
|
excludeIndex = filterObj.queries.exclude[key].indexOf('-')
|
|
}
|
|
|
|
// Flag to match NULL values
|
|
const inQuery = queryIndex !== -1
|
|
const inExclude = excludeIndex !== -1
|
|
if (inQuery || inExclude) {
|
|
// Prioritize exclude keys when both types found
|
|
filterObj.flags[`${key}Null`] = inExclude ? false : inQuery
|
|
if (inQuery) {
|
|
if (filterObj.queries[key].length === 1) {
|
|
// Delete key to avoid unexpected behavior
|
|
delete filterObj.queries[key]
|
|
} else {
|
|
filterObj.queries[key].splice(queryIndex, 1)
|
|
}
|
|
}
|
|
if (inExclude) {
|
|
if (filterObj.queries.exclude[key].length === 1) {
|
|
// Delete key to avoid unexpected behavior
|
|
delete filterObj.queries.exclude[key]
|
|
} else {
|
|
filterObj.queries.exclude[key].splice(excludeIndex, 1)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
const parseDate = (date, minoffset, resetMs) => {
|
|
// [YYYY][/MM][/DD] [HH][:MM][:SS]
|
|
// e.g. 2020/01/01 00:00:00, 2018/01/01 06, 2019/11, 12:34:00
|
|
const match = date.match(/^(\d{4})?(\/\d{2})?(\/\d{2})?\s?(\d{2})?(:\d{2})?(:\d{2})?$/)
|
|
|
|
if (match) {
|
|
let offset = 0
|
|
if (minoffset !== undefined) {
|
|
offset = 60000 * (utils.timezoneOffset - minoffset)
|
|
}
|
|
|
|
const dateObj = new Date(Date.now() + offset)
|
|
|
|
if (match[1] !== undefined) {
|
|
dateObj.setFullYear(Number(match[1]), // full year
|
|
match[2] !== undefined ? (Number(match[2].slice(1)) - 1) : 0, // month, zero-based
|
|
match[3] !== undefined ? Number(match[3].slice(1)) : 1) // date
|
|
}
|
|
|
|
if (match[4] !== undefined) {
|
|
dateObj.setHours(Number(match[4]), // hours
|
|
match[5] !== undefined ? Number(match[5].slice(1)) : 0, // minutes
|
|
match[6] !== undefined ? Number(match[6].slice(1)) : 0) // seconds
|
|
}
|
|
|
|
if (resetMs) {
|
|
dateObj.setMilliseconds(0)
|
|
}
|
|
|
|
// Calculate timezone differences
|
|
return new Date(dateObj.getTime() - offset)
|
|
} else {
|
|
return null
|
|
}
|
|
}
|
|
|
|
// Parse dates to timestamps
|
|
for (const range of ranges) {
|
|
if (filterObj.queries[range]) {
|
|
if (filterObj.queries[range].from) {
|
|
const parsed = parseDate(filterObj.queries[range].from, minoffset, true)
|
|
filterObj.queries[range].from = parsed ? Math.floor(parsed / 1000) : null
|
|
}
|
|
if (filterObj.queries[range].to) {
|
|
const parsed = parseDate(filterObj.queries[range].to, minoffset, true)
|
|
filterObj.queries[range].to = parsed ? Math.ceil(parsed / 1000) : null
|
|
}
|
|
}
|
|
}
|
|
|
|
// Query users table for user IDs
|
|
if (filterObj.queries.user || filterObj.queries.exclude.user) {
|
|
const usernames = []
|
|
if (filterObj.queries.user) {
|
|
usernames.push(...filterObj.queries.user)
|
|
}
|
|
if (filterObj.queries.exclude.user) {
|
|
usernames.push(...filterObj.queries.exclude.user)
|
|
}
|
|
|
|
const uploaders = await utils.db.table('users')
|
|
.whereIn('username', usernames)
|
|
.select('id', 'username')
|
|
|
|
// If no matches, or mismatched results
|
|
if (!uploaders || (uploaders.length !== usernames.length)) {
|
|
const notFound = usernames.filter(username => {
|
|
return !uploaders.find(uploader => uploader.username === username)
|
|
})
|
|
if (notFound) {
|
|
throw new ClientError(`User${notFound.length === 1 ? '' : 's'} not found: ${notFound.join(', ')}.`)
|
|
}
|
|
}
|
|
|
|
for (const uploader of uploaders) {
|
|
if (filterObj.queries.user && filterObj.queries.user.includes(uploader.username)) {
|
|
filterObj.uploaders.push(uploader)
|
|
} else {
|
|
filterObj.excludeUploaders.push(uploader)
|
|
}
|
|
}
|
|
|
|
// Delete keys to avoid unexpected behavior
|
|
delete filterObj.queries.user
|
|
delete filterObj.queries.exclude.user
|
|
}
|
|
|
|
// Parse sort keys
|
|
if (filterObj.queries.sort) {
|
|
const allowed = [
|
|
'expirydate',
|
|
'id',
|
|
'name',
|
|
'original',
|
|
'size',
|
|
'timestamp'
|
|
]
|
|
|
|
// Only allow sorting by 'albumid' when not listing album's uploads
|
|
if (req.params.id === undefined) allowed.push('albumid')
|
|
|
|
// Only allow sorting by 'ip' and 'userid' columns when listing all uploads
|
|
if (all) allowed.push('ip', 'userid')
|
|
|
|
for (const obQuery of filterObj.queries.sort) {
|
|
const tmp = obQuery.toLowerCase().split(':')
|
|
const column = sortObj.maps[tmp[0]] || tmp[0]
|
|
|
|
if (!allowed.includes(column)) {
|
|
// Alert users if using disallowed/missing columns
|
|
throw new ClientError(`Column "${column}" cannot be used for sorting.\n\nTry the following instead:\n${allowed.join(', ')}`)
|
|
}
|
|
|
|
sortObj.parsed.push({
|
|
column,
|
|
order: (tmp[1] && /^d/.test(tmp[1])) ? 'desc' : 'asc',
|
|
clause: sortObj.nullsLast.includes(column) ? 'nulls last' : '',
|
|
cast: sortObj.casts[column] || null
|
|
})
|
|
}
|
|
|
|
// Regular user threshold check
|
|
if (!ismoderator && sortObj.parsed.length > MAX_SORT_KEYS) {
|
|
throw new ClientError(`Users are only allowed to use ${MAX_SORT_KEYS} sort key${MAX_SORT_KEYS === 1 ? '' : 's'} at a time.`)
|
|
}
|
|
|
|
// Delete key to avoid unexpected behavior
|
|
delete filterObj.queries.sort
|
|
}
|
|
|
|
// Parse is keys
|
|
let isKeys = 0
|
|
let isLast
|
|
if (filterObj.queries.is || filterObj.queries.exclude.is) {
|
|
for (const type of filterObj.typeIs) {
|
|
const inQuery = filterObj.queries.is && filterObj.queries.is.includes(type)
|
|
const inExclude = filterObj.queries.exclude.is && filterObj.queries.exclude.is.includes(type)
|
|
|
|
// Prioritize exclude keys when both types found
|
|
if (inQuery || inExclude) {
|
|
filterObj.flags[`is${type}`] = inExclude ? false : inQuery
|
|
if (isLast !== undefined && isLast !== filterObj.flags[`is${type}`]) {
|
|
throw new ClientError('Cannot mix inclusion and exclusion type-is keys.')
|
|
}
|
|
isKeys++
|
|
isLast = filterObj.flags[`is${type}`]
|
|
}
|
|
}
|
|
|
|
// Delete keys to avoid unexpected behavior
|
|
delete filterObj.queries.is
|
|
delete filterObj.queries.exclude.is
|
|
}
|
|
|
|
// Regular user threshold check
|
|
if (!ismoderator && isKeys > MAX_IS_KEYS) {
|
|
throw new ClientError(`Users are only allowed to use ${MAX_IS_KEYS} type-is key${MAX_IS_KEYS === 1 ? '' : 's'} at a time.`)
|
|
}
|
|
}
|
|
|
|
function filter () {
|
|
// If listing all uploads
|
|
if (all) {
|
|
this.where(function () {
|
|
// Filter uploads matching any of the supplied 'user' keys and/or NULL flag
|
|
// Prioritze exclude keys when both types found
|
|
this.orWhere(function () {
|
|
if (filterObj.excludeUploaders.length) {
|
|
this.whereNotIn('userid', filterObj.excludeUploaders.map(v => v.id))
|
|
} else if (filterObj.uploaders.length) {
|
|
this.orWhereIn('userid', filterObj.uploaders.map(v => v.id))
|
|
}
|
|
// Such overbearing logic for NULL values, smh...
|
|
if ((filterObj.excludeUploaders.length && filterObj.flags.userNull !== false) ||
|
|
(filterObj.uploaders.length && filterObj.flags.userNull) ||
|
|
(!filterObj.excludeUploaders.length && !filterObj.uploaders.length && filterObj.flags.userNull)) {
|
|
this.orWhereNull('userid')
|
|
} else if (filterObj.flags.userNull === false) {
|
|
this.whereNotNull('userid')
|
|
}
|
|
})
|
|
|
|
// Filter uploads matching any of the supplied 'ip' keys and/or NULL flag
|
|
// Same prioritization logic as above
|
|
this.orWhere(function () {
|
|
if (filterObj.queries.exclude.ip) {
|
|
this.whereNotIn('ip', filterObj.queries.exclude.ip)
|
|
} else if (filterObj.queries.ip) {
|
|
this.orWhereIn('ip', filterObj.queries.ip)
|
|
}
|
|
// ...
|
|
if ((filterObj.queries.exclude.ip && filterObj.flags.ipNull !== false) ||
|
|
(filterObj.queries.ip && filterObj.flags.ipNull) ||
|
|
(!filterObj.queries.exclude.ip && !filterObj.queries.ip && filterObj.flags.ipNull)) {
|
|
this.orWhereNull('ip')
|
|
} else if (filterObj.flags.ipNull === false) {
|
|
this.whereNotNull('ip')
|
|
}
|
|
})
|
|
})
|
|
} else {
|
|
// If not listing all uploads, list user's uploads
|
|
this.where('userid', user.id)
|
|
}
|
|
|
|
// Then, refine using any of the supplied 'albumid' keys and/or NULL flag
|
|
// Same prioritization logic as 'userid' and 'ip' above
|
|
if (req.params.id === undefined) {
|
|
this.andWhere(function () {
|
|
if (filterObj.queries.exclude.albumid) {
|
|
this.whereNotIn('albumid', filterObj.queries.exclude.albumid)
|
|
} else if (filterObj.queries.albumid) {
|
|
this.orWhereIn('albumid', filterObj.queries.albumid)
|
|
}
|
|
// ...
|
|
if ((filterObj.queries.exclude.albumid && filterObj.flags.albumidNull !== false) ||
|
|
(filterObj.queries.albumid && filterObj.flags.albumidNull) ||
|
|
(!filterObj.queries.exclude.albumid && !filterObj.queries.albumid && filterObj.flags.albumidNull)) {
|
|
this.orWhereNull('albumid')
|
|
} else if (filterObj.flags.albumidNull === false) {
|
|
this.whereNotNull('albumid')
|
|
}
|
|
})
|
|
} else if (!all) {
|
|
// If not listing all uploads, list uploads from user's album
|
|
this.andWhere('albumid', req.params.id)
|
|
}
|
|
|
|
// Then, refine using the supplied 'date' ranges
|
|
this.andWhere(function () {
|
|
if (!filterObj.queries.date || (!filterObj.queries.date.from && !filterObj.queries.date.to)) return
|
|
if (typeof filterObj.queries.date.from === 'number') {
|
|
if (typeof filterObj.queries.date.to === 'number') {
|
|
this.andWhereBetween('timestamp', [filterObj.queries.date.from, filterObj.queries.date.to])
|
|
} else {
|
|
this.andWhere('timestamp', '>=', filterObj.queries.date.from)
|
|
}
|
|
} else {
|
|
this.andWhere('timestamp', '<=', filterObj.queries.date.to)
|
|
}
|
|
})
|
|
|
|
// Then, refine using the supplied 'expiry' ranges
|
|
this.andWhere(function () {
|
|
if (!filterObj.queries.expiry || (!filterObj.queries.expiry.from && !filterObj.queries.expiry.to)) return
|
|
if (typeof filterObj.queries.expiry.from === 'number') {
|
|
if (typeof filterObj.queries.expiry.to === 'number') {
|
|
this.andWhereBetween('expirydate', [filterObj.queries.expiry.from, filterObj.queries.expiry.to])
|
|
} else {
|
|
this.andWhere('expirydate', '>=', filterObj.queries.expiry.from)
|
|
}
|
|
} else {
|
|
this.andWhere('expirydate', '<=', filterObj.queries.expiry.to)
|
|
}
|
|
})
|
|
|
|
// Then, refine using type-is flags
|
|
this.andWhere(function () {
|
|
for (const type of filterObj.typeIs) {
|
|
let func
|
|
let operator
|
|
if (filterObj.flags[`is${type}`] === true) {
|
|
func = 'orWhere'
|
|
operator = 'like'
|
|
} else if (filterObj.flags[`is${type}`] === false) {
|
|
func = 'andWhere'
|
|
operator = 'not like'
|
|
}
|
|
|
|
if (func) {
|
|
for (const pattern of utils[`${type}Exts`].map(ext => `%${ext}`)) {
|
|
this[func]('name', operator, pattern)
|
|
}
|
|
}
|
|
}
|
|
})
|
|
|
|
// Then, refine using the supplied keywords against their file names
|
|
this.andWhere(function () {
|
|
if (!filterObj.queries.text) return
|
|
for (const pattern of filterObj.queries.text) {
|
|
this.orWhereRaw('?? like ? escape ?', ['name', pattern, '\\'])
|
|
this.orWhereRaw('?? like ? escape ?', ['original', pattern, '\\'])
|
|
}
|
|
})
|
|
|
|
// Finally, refine using the supplied exclusions against their file names
|
|
this.andWhere(function () {
|
|
if (!filterObj.queries.exclude.text) return
|
|
for (const pattern of filterObj.queries.exclude.text) {
|
|
this.andWhereRaw('?? not like ? escape ?', ['name', pattern, '\\'])
|
|
this.andWhereRaw('?? not like ? escape ?', ['original', pattern, '\\'])
|
|
}
|
|
})
|
|
}
|
|
|
|
// Query uploads count for pagination
|
|
const count = await utils.db.table('files')
|
|
.where(filter)
|
|
.count('id as count')
|
|
.then(rows => rows[0].count)
|
|
if (!count) {
|
|
return res.json({ success: true, files: [], count })
|
|
}
|
|
|
|
let offset = Number(req.params.page)
|
|
if (isNaN(offset)) offset = 0
|
|
else if (offset < 0) offset = Math.max(0, Math.ceil(count / 25) + offset)
|
|
|
|
const columns = ['id', 'name', 'original', 'userid', 'size', 'timestamp']
|
|
if (utils.retentions.enabled) columns.push('expirydate')
|
|
if (!all ||
|
|
filterObj.queries.albumid ||
|
|
filterObj.queries.exclude.albumid ||
|
|
filterObj.flags.albumidNull !== undefined) columns.push('albumid')
|
|
|
|
// Only select IPs if we are listing all uploads
|
|
if (all) columns.push('ip')
|
|
|
|
// Build raw query for order by (sorting) operation
|
|
let orderByRaw
|
|
if (sortObj.parsed.length) {
|
|
orderByRaw = sortObj.parsed.map(sort => {
|
|
// Use Knex.raw() to sanitize user inputs
|
|
if (sort.cast) {
|
|
return utils.db.raw(`cast (?? as ${sort.cast}) ${sort.order} ${sort.clause}`.trim(), sort.column)
|
|
} else {
|
|
return utils.db.raw(`?? ${sort.order} ${sort.clause}`.trim(), sort.column)
|
|
}
|
|
}).join(', ')
|
|
} else {
|
|
orderByRaw = '`id` desc'
|
|
}
|
|
|
|
const files = await utils.db.table('files')
|
|
.where(filter)
|
|
.orderByRaw(orderByRaw)
|
|
.limit(25)
|
|
.offset(25 * offset)
|
|
.select(columns)
|
|
|
|
if (!files.length) {
|
|
return res.json({ success: true, files, count, basedomain })
|
|
}
|
|
|
|
for (const file of files) {
|
|
file.extname = utils.extname(file.name)
|
|
if (utils.mayGenerateThumb(file.extname)) {
|
|
file.thumb = `thumbs/${file.name.slice(0, -file.extname.length)}.png`
|
|
}
|
|
}
|
|
|
|
// If we queried albumid, query album names
|
|
let albums = {}
|
|
if (columns.includes('albumid')) {
|
|
const albumids = files
|
|
.map(file => file.albumid)
|
|
.filter((v, i, a) => {
|
|
return v !== null && v !== undefined && v !== '' && a.indexOf(v) === i
|
|
})
|
|
albums = await utils.db.table('albums')
|
|
.whereIn('id', albumids)
|
|
.where('enabled', 1)
|
|
.select('id', 'name')
|
|
.then(rows => {
|
|
// Build Object indexed by their IDs
|
|
const obj = {}
|
|
for (const row of rows) {
|
|
obj[row.id] = row.name
|
|
}
|
|
return obj
|
|
})
|
|
}
|
|
|
|
// If we are not listing all uploads, send response
|
|
if (!all) {
|
|
return res.json({ success: true, files, count, albums, basedomain })
|
|
}
|
|
|
|
// Otherwise proceed to querying usernames
|
|
let usersTable = filterObj.uploaders
|
|
if (!usersTable.length) {
|
|
const userids = files
|
|
.map(file => file.userid)
|
|
.filter((v, i, a) => {
|
|
return v !== null && v !== undefined && v !== '' && a.indexOf(v) === i
|
|
})
|
|
|
|
// If there are no uploads attached to a registered user, send response
|
|
if (!userids.length) {
|
|
return res.json({ success: true, files, count, albums, basedomain })
|
|
}
|
|
|
|
// Query usernames of user IDs from currently selected files
|
|
usersTable = await utils.db.table('users')
|
|
.whereIn('id', userids)
|
|
.select('id', 'username')
|
|
}
|
|
|
|
const users = {}
|
|
for (const user of usersTable) {
|
|
users[user.id] = user.username
|
|
}
|
|
|
|
return res.json({ success: true, files, count, users, albums, basedomain })
|
|
}
|
|
|
|
/** Get file info */
|
|
|
|
self.get = async (req, res) => {
|
|
const user = await utils.authorize(req)
|
|
const ismoderator = perms.is(user, 'moderator')
|
|
|
|
const identifier = req.params.identifier
|
|
if (identifier === undefined) {
|
|
throw new ClientError('No identifier provided.')
|
|
}
|
|
|
|
const file = await utils.db.table('files')
|
|
.where('name', identifier)
|
|
.where(function () {
|
|
if (!ismoderator) {
|
|
this.where('userid', user.id)
|
|
}
|
|
})
|
|
.first()
|
|
|
|
if (!file) {
|
|
throw new ClientError('File not found.', { statusCode: 404 })
|
|
}
|
|
|
|
return res.json({ success: true, file })
|
|
}
|
|
|
|
module.exports = self
|