feat(xo-server): 5.71.0-0

(keeping it for another branch)

@xen-orchestra/backups-cli/package.json

@@ -7,14 +7,14 @@
   "bugs": "https://github.com/vatesfr/xen-orchestra/issues",
   "dependencies": {
     "@xen-orchestra/backups": "^0.1.1",
-    "@xen-orchestra/fs": "^0.11.1",
+    "@xen-orchestra/fs": "^0.12.0-0",
     "filenamify": "^4.1.0",
     "getopts": "^2.2.5",
     "limit-concurrency-decorator": "^0.4.0",
     "lodash": "^4.17.15",
     "promise-toolbox": "^0.15.0",
     "proper-lockfile": "^4.1.1",
-    "vhd-lib": "^0.8.0"
+    "vhd-lib": "^0.9.0-0"
   },
   "engines": {
     "node": ">=7.10.1"

@xen-orchestra/fs/package.json

@@ -1,7 +1,7 @@
 {
   "private": false,
   "name": "@xen-orchestra/fs",
-  "version": "0.11.1",
+  "version": "0.12.0-0",
   "license": "AGPL-3.0-or-later",
   "description": "The File System for Xen Orchestra backups.",
   "keywords": [],
@@ -36,6 +36,7 @@
     "readable-stream": "^3.0.6",
     "through2": "^4.0.2",
     "tmp": "^0.2.1",
+    "syscall": "^0.2.0",
     "xo-remote-parser": "^0.6.0"
   },
   "devDependencies": {

@xen-orchestra/fs/src/abstract.js

@@ -311,6 +311,28 @@ export default class RemoteHandlerAbstract {
     )
   }
 
+  /**
+   * Copy a range from one file to the other, kernel side, server side or with a reflink if possible.
+   *
+   * Slightly different from the copy_file_range linux system call:
+   *  - offsets are mandatory (because some remote handlers don't have a current pointer for files)
+   *  - flags is fixed to 0
+   *  - will not return until copy is finished.
+   *
+   * @param fdIn read open file descriptor
+   * @param offsetIn either start offset in the source file
+   * @param fdOut write open file descriptor (not append!)
+   * @param offsetOut offset in the target file
+   * @param dataLen how long to copy
+   * @returns {Promise<void>}
+   */
+  async copyFileRange(fdIn, offsetIn, fdOut, offsetOut, dataLen) {
+    // default implementation goes through the network
+    const buffer = Buffer.alloc(dataLen)
+    await this._read(fdIn, buffer, offsetIn)
+    await this._write(fdOut, buffer, offsetOut)
+  }
+
   async readFile(
     file: string,
     { flags = 'r' }: { flags?: string } = {}
@@ -357,27 +379,51 @@ export default class RemoteHandlerAbstract {
   }
 
   async test(): Promise<Object> {
-    const SIZE = 1024 * 1024 * 10
-    const testFileName = normalizePath(`${Date.now()}.test`)
-    const data = await fromCallback(randomBytes, SIZE)
+    const SIZE = 1024 * 1024 * 100
+    const now = Date.now()
+    const testFileName = normalizePath(`${now}.test`)
+    const testFileName2 = normalizePath(`${now}__dup.test`)
+    // get random ASCII for easy debug
+    const data = Buffer.from((await fromCallback(randomBytes, SIZE)).toString('base64'), 'ascii').slice(0, SIZE)
     let step = 'write'
     try {
       const writeStart = process.hrtime()
       await this._outputFile(testFileName, data, { flags: 'wx' })
       const writeDuration = process.hrtime(writeStart)
+      let cloneDuration
+      const fd1 = await this.openFile(testFileName, 'r+')
+      try {
+        const fd2 = await this.openFile(testFileName2, 'wx')
+        try {
+          step = 'duplicate'
+          const cloneStart = process.hrtime()
+          await this.copyFileRange(fd1, 0, fd2, 0, data.byteLength)
+          cloneDuration = process.hrtime(cloneStart)
+          console.log('cloneDuration', cloneDuration)
+        } finally {
+          await this._closeFile(fd2)
+        }
+      } finally {
+        await this._closeFile(fd1)
+      }
 
       step = 'read'
       const readStart = process.hrtime()
       const read = await this._readFile(testFileName, { flags: 'r' })
       const readDuration = process.hrtime(readStart)
-
       if (!data.equals(read)) {
         throw new Error('output and input did not match')
       }
+
+      const read2 = await this._readFile(testFileName2, { flags: 'r' })
+      if (!data.equals(read2)) {
+        throw new Error('duplicated and input did not match')
+      }
       return {
         success: true,
         writeRate: computeRate(writeDuration, SIZE),
         readRate: computeRate(readDuration, SIZE),
+        cloneDuration: computeRate(cloneDuration, SIZE),
       }
     } catch (error) {
       return {
@@ -388,6 +434,7 @@ export default class RemoteHandlerAbstract {
       }
     } finally {
       ignoreErrors.call(this._unlink(testFileName))
+      ignoreErrors.call(this._unlink(testFileName2))
     }
   }
 
@@ -428,7 +475,7 @@ export default class RemoteHandlerAbstract {
   // Methods that can be called by private methods to avoid parallel limit on public methods
 
   async __closeFile(fd: FileDescriptor): Promise<void> {
-    await timeout.call(this._closeFile(fd.fd), this._timeout)
+    await timeout.call(this._closeFile(fd), this._timeout)
   }
 
   async __mkdir(dir: string): Promise<void> {

@xen-orchestra/fs/src/local.js

@@ -1,10 +1,39 @@
 import df from '@sindresorhus/df'
 import fs from 'fs-extra'
 import { fromEvent } from 'promise-toolbox'
+import { Syscall6 } from 'syscall'
 
 import RemoteHandlerAbstract from './abstract'
 
+/**
+ * @returns the number of byte effectively copied, needs to be called in a loop!
+ * @throws Error if the syscall returned -1
+ */
+function copyFileRangeSyscall(fdIn, offsetIn, fdOut, offsetOut, dataLen, flags = 0) {
+  // we are stuck on linux x86_64 because of int64 representation and syscall numbers
+  function wrapOffset(offsetIn) {
+    if (offsetIn == null)
+      return 0
+    const offsetInBuffer = new Uint32Array(2)
+    new DataView(offsetInBuffer.buffer).setBigUint64(0, BigInt(offsetIn), true)
+    return offsetInBuffer
+  }
+
+  // https://man7.org/linux/man-pages/man2/copy_file_range.2.html
+  const SYS_copy_file_range = 326
+  const [copied, _, errno] = Syscall6(SYS_copy_file_range, fdIn, wrapOffset(offsetIn), fdOut, wrapOffset(offsetOut), dataLen, flags)
+  if (copied === -1) {
+    throw new Error('Error no ' + errno)
+  }
+  return copied
+}
+
 export default class LocalHandler extends RemoteHandlerAbstract {
+  constructor(remote: any, options: Object = {}) {
+    super(remote, options)
+    this._canFallocate = true
+  }
+
   get type() {
     return 'file'
   }
@@ -18,7 +47,7 @@ export default class LocalHandler extends RemoteHandlerAbstract {
   }
 
   async _closeFile(fd) {
-    return fs.close(fd)
+    return fs.close(fd.fd)
   }
 
   async _createReadStream(file, options) {
@@ -81,6 +110,26 @@ export default class LocalHandler extends RemoteHandlerAbstract {
     return fs.open(this._getFilePath(path), flags)
   }
 
+  /**
+   * Slightly different from the linux system call:
+   *  - offsets are mandatory (because some remote handlers don't have a current pointer for files)
+   *  - flags is fixed to 0
+   *  - will not return until copy is finished.
+   *
+   * @param fdIn read open file descriptor
+   * @param offsetIn either start offset in the source file
+   * @param fdOut write open file descriptor (not append!)
+   * @param offsetOut offset in the target file
+   * @param dataLen how long to copy
+   * @returns {Promise<void>}
+   */
+  async copyFileRange(fdIn, offsetIn, fdOut, offsetOut, dataLen) {
+    let copied = 0
+    do {
+      copied += await copyFileRangeSyscall(fdIn.fd, offsetIn + copied, fdOut.fd, offsetOut + copied, dataLen - copied)
+    } while (dataLen - copied > 0)
+  }
+
   async _read(file, buffer, position) {
     const needsClose = typeof file === 'string'
     file = needsClose ? await fs.open(this._getFilePath(file), 'r') : file.fd

@xen-orchestra/fs/src/s3.js

@@ -138,14 +138,21 @@ export default class S3Handler extends RemoteHandlerAbstract {
       file = file.fd
     }
     const uploadParams = this._createParams(file)
-    const fileSize = +(await this._s3.headObject(uploadParams).promise())
-      .ContentLength
+    let fileSize
+    try {
+      fileSize = +(await this._s3.headObject(uploadParams).promise()).ContentLength
+    } catch (e) {
+      if (e.code === 'NotFound') {
+        fileSize = 0
+      } else {
+        throw e
+      }
+    }
     if (fileSize < MIN_PART_SIZE) {
       const resultBuffer = Buffer.alloc(
         Math.max(fileSize, position + buffer.length)
       )
-      const fileContent = (await this._s3.getObject(uploadParams).promise())
-        .Body
+      const fileContent = fileSize ? (await this._s3.getObject(uploadParams).promise()).Body : Buffer.alloc(0)
       fileContent.copy(resultBuffer)
       buffer.copy(resultBuffer, position)
       await this._s3

@xen-orchestra/fs/src/smb-mount.js

@@ -2,6 +2,7 @@ import { parse } from 'xo-remote-parser'
 
 import MountHandler from './_mount'
 import normalizePath from './_normalizePath'
+import { fromEvent } from "promise-toolbox"
 
 export default class SmbMountHandler extends MountHandler {
   constructor(remote, opts) {
@@ -22,4 +23,21 @@ export default class SmbMountHandler extends MountHandler {
   get type() {
     return 'smb'
   }
+
+  // nraynaud: in some circumstances, renaming the file triggers a bug where we can't re-open it afterwards in SMB2
+  // SMB linux client Linux xoa 4.19.0-12-amd64 #1 SMP Debian 4.19.152-1 (2020-10-18) x86_64 GNU/Linux
+  // server Windows 10 Family Edition 1909 (v18363.1139)
+  async _outputStream(input, path, { checksum }) {
+    const output = await this.createOutputStream(path, { checksum })
+    try {
+      input.pipe(output)
+      await fromEvent(output, 'finish')
+      await output.checksumWritten
+      // $FlowFixMe
+      await input.task
+    } catch (error) {
+      await this.unlink(path, { checksum })
+      throw error
+    }
+  }
 }

CHANGELOG.unreleased.md

@@ -7,6 +7,8 @@
 
 > Users must be able to say: “Nice enhancement, I'm eager to test it”
 
+- [backup] improve merge speed after backup when using SMB3.1.1 or NFS4.2 (PR [#5331](https://github.com/vatesfr/xen-orchestra/pull/5331))
+
 ### Bug fixes
 
 > Users must be able to say: “I had this issue, happy to know it's fixed”
@@ -27,3 +29,8 @@
 > - major: if the change breaks compatibility
 >
 > In case of conflict, the highest (lowest in previous list) `$version` wins.
+
+- @xen-orchestra/fs minor
+- vhd-lib minor
+- xo-server minor
+- xo-web minor

package.json

@@ -40,7 +40,6 @@
   "jest": {
     "collectCoverage": true,
     "moduleNameMapper": {
-      "^.": "./src",
       "^(@vates/[^/]+)": "$1/src",
       "^(@xen-orchestra/[^/]+)": "$1/src",
       "^(value-matcher)": "$1/src",

packages/vhd-cli/package.json

@@ -28,12 +28,12 @@
     "node": ">=8.10"
   },
   "dependencies": {
-    "@xen-orchestra/fs": "^0.11.1",
+    "@xen-orchestra/fs": "^0.12.0-0",
     "cli-progress": "^3.1.0",
     "exec-promise": "^0.7.0",
     "getopts": "^2.2.3",
     "struct-fu": "^1.2.0",
-    "vhd-lib": "^0.8.0"
+    "vhd-lib": "^0.9.0-0"
   },
   "devDependencies": {
     "@babel/cli": "^7.0.0",

packages/vhd-lib/merge.integ.spec.js

@@ -11,7 +11,7 @@ import { pFromCallback } from 'promise-toolbox'
 import { pipeline } from 'readable-stream'
 import { randomBytes } from 'crypto'
 
-import Vhd, { chainVhd, createSyntheticStream, mergeVhd as vhdMerge } from './'
+import Vhd, { chainVhd, createSyntheticStream, mergeVhd as vhdMerge } from './src/index'
 
 import { SECTOR_SIZE } from './src/_constants'
 

packages/vhd-lib/package.json

@@ -1,7 +1,7 @@
 {
   "private": false,
   "name": "vhd-lib",
-  "version": "0.8.0",
+  "version": "0.9.0-0",
   "license": "AGPL-3.0-or-later",
   "description": "Primitives for VHD file handling",
   "keywords": [],
@@ -36,7 +36,7 @@
     "@babel/core": "^7.0.0",
     "@babel/preset-env": "^7.0.0",
     "@babel/preset-flow": "^7.0.0",
-    "@xen-orchestra/fs": "^0.11.1",
+    "@xen-orchestra/fs": "^0.12.0-0",
     "babel-plugin-lodash": "^3.3.2",
     "cross-env": "^7.0.2",
     "execa": "^4.0.2",

packages/vhd-lib/src/merge.js

@@ -17,6 +17,8 @@ export default concurrency(2)(async function merge(
   childPath,
   { onProgress = noop } = {}
 ) {
+  // merges blocks
+  let mergedDataSize = 0
   const parentFd = await parentHandler.openFile(parentPath, 'r+')
   try {
     const parentVhd = new Vhd(parentHandler, parentFd)
@@ -68,14 +70,12 @@ export default concurrency(2)(async function merge(
 
       onProgress({ total: nBlocks, done: 0 })
 
-      // merges blocks
-      let mergedDataSize = 0
       for (let i = 0, block = firstBlock; i < nBlocks; ++i, ++block) {
         while (!childVhd.containsBlock(block)) {
           ++block
         }
 
-        mergedDataSize += await parentVhd.coalesceBlock(childVhd, block)
+        mergedDataSize += await parentVhd.coalesceBlock(childVhd, block,childFd, parentFd )
         onProgress({
           total: nBlocks,
           done: i + 1,
@@ -94,12 +94,11 @@ export default concurrency(2)(async function merge(
       // necessary to update values and to recreate the footer after block
       // creation
       await parentVhd.writeFooter()
-
-      return mergedDataSize
     } finally {
       await childHandler.closeFile(childFd)
     }
   } finally {
     await parentHandler.closeFile(parentFd)
   }
+  return mergedDataSize
 })

packages/vhd-lib/src/vhd.js

@@ -199,30 +199,35 @@ export default class Vhd {
   }
 
   // return the first sector (bitmap) of a block
-  _getBatEntry(block) {
-    const i = block * 4
+  _getBatEntry(blockId) {
+    const i = blockId * 4
     const { blockTable } = this
     return i < blockTable.length ? blockTable.readUInt32BE(i) : BLOCK_UNUSED
   }
 
-  _readBlock(blockId, onlyBitmap = false) {
+  // returns actual byt offset in the file or null
+  _getBlockOffsetBytes(blockId) {
     const blockAddr = this._getBatEntry(blockId)
-    if (blockAddr === BLOCK_UNUSED) {
+    return blockAddr === BLOCK_UNUSED ? null : sectorsToBytes(blockAddr)
+  }
+
+  _readBlock(blockId, onlyBitmap = false) {
+    const blockAddr = this._getBlockOffsetBytes(blockId)
+    if (blockAddr === null) {
       throw new Error(`no such block ${blockId}`)
     }
 
-    return this._read(
-      sectorsToBytes(blockAddr),
+    return this._read(blockAddr,
       onlyBitmap ? this.bitmapSize : this.fullBlockSize
     ).then(buf =>
       onlyBitmap
         ? { id: blockId, bitmap: buf }
         : {
-            id: blockId,
-            bitmap: buf.slice(0, this.bitmapSize),
-            data: buf.slice(this.bitmapSize),
-            buffer: buf,
-          }
+          id: blockId,
+          bitmap: buf.slice(0, this.bitmapSize),
+          data: buf.slice(this.bitmapSize),
+          buffer: buf,
+        }
     )
   }
 
@@ -307,15 +312,18 @@ export default class Vhd {
     return this._write(blockTable.slice(i, i + 4), this.header.tableOffset + i)
   }
 
-  // Allocate a new uninitialized block in the BAT
-  async _createBlock(blockId) {
-    assert.strictEqual(this._getBatEntry(blockId), BLOCK_UNUSED)
-
+  // Make a new empty block at vhd end.
+  // Update block allocation table in context and in file.
+  async _createBlock(blockId, fullBlock = Buffer.alloc(this.fullBlockSize)) {
     const blockAddr = Math.ceil(this._getEndOfData() / SECTOR_SIZE)
 
     debug(`create block ${blockId} at ${blockAddr}`)
 
-    await this._setBatEntry(blockId, blockAddr)
+    await Promise.all([
+      // Write an empty block and addr in vhd file.
+      this._write(fullBlock, sectorsToBytes(blockAddr)),
+      this._setBatEntry(blockId, blockAddr),
+    ])
 
     return blockAddr
   }
@@ -338,12 +346,13 @@ export default class Vhd {
     await this._write(bitmap, sectorsToBytes(blockAddr))
   }
 
-  async _writeEntireBlock(block) {
-    let blockAddr = this._getBatEntry(block.id)
+  async _getAddressOrAllocate(blockId) {
+    const blockAddr = this._getBlockOffsetBytes(blockId)
+    return blockAddr === null ? await this._createBlock(blockId) : blockAddr
+  }
 
-    if (blockAddr === BLOCK_UNUSED) {
-      blockAddr = await this._createBlock(block.id)
-    }
+  async _writeEntireBlock(block) {
+    const blockAddr = this._getAddressOrAllocate(block.id)
     await this._write(block.buffer, sectorsToBytes(blockAddr))
   }
 
@@ -377,15 +386,18 @@ export default class Vhd {
     )
   }
 
-  async coalesceBlock(child, blockId) {
-    const block = await child._readBlock(blockId)
-    const { bitmap, data } = block
+  async coalesceBlock(child, blockId, childFd, parentFd) {
+    const childBlockAddress = child._getBlockOffsetBytes(blockId)
+    const bitmap = (await child._readBlock(blockId, true)).bitmap
 
     debug(`coalesceBlock block=${blockId}`)
 
     // For each sector of block data...
     const { sectorsPerBlock } = child
+    // lazily loaded
     let parentBitmap = null
+    // lazily loaded
+    let childBlock = null
     for (let i = 0; i < sectorsPerBlock; i++) {
       // If no changes on one sector, skip.
       if (!mapTestBit(bitmap, i)) {
@@ -403,19 +415,22 @@ export default class Vhd {
 
       const isFullBlock = i === 0 && endSector === sectorsPerBlock
       if (isFullBlock) {
-        await this._writeEntireBlock(block)
+        await this._handler.copyFileRange(childFd, childBlockAddress, parentFd, await this._getAddressOrAllocate(blockId), this.fullBlockSize)
       } else {
         if (parentBitmap === null) {
           parentBitmap = (await this._readBlock(blockId, true)).bitmap
         }
-        await this._writeBlockSectors(block, i, endSector, parentBitmap)
+        if (childBlock === null) {
+          childBlock = await child._readBlock(blockId)
+        }
+        await this._writeBlockSectors(childBlock, i, endSector, parentBitmap)
       }
 
       i = endSector
     }
 
     // Return the merged data size
-    return data.length
+    return this.fullBlockSize - this.bitmapSize
   }
 
   // Write a context footer. (At the end and beginning of a vhd file.)
@@ -481,7 +496,7 @@ export default class Vhd {
       )
       const endInBuffer = Math.min(
         ((currentBlock + 1) * this.sectorsPerBlock - offsetSectors) *
-          SECTOR_SIZE,
+        SECTOR_SIZE,
         buffer.length
       )
       let inputBuffer

packages/xen-api/examples/package.json

@@ -8,6 +8,6 @@
     "promise-toolbox": "^0.13.0",
     "readable-stream": "^3.1.1",
     "throttle": "^1.0.3",
-    "vhd-lib": "^0.8.0"
+    "vhd-lib": "^0.9.0-0"
   }
 }

packages/xo-server/package.json

@@ -1,7 +1,7 @@
 {
   "private": true,
   "name": "xo-server",
-  "version": "5.70.0",
+  "version": "5.71.0-0",
   "license": "AGPL-3.0-or-later",
   "description": "Server part of Xen-Orchestra",
   "keywords": [
@@ -38,7 +38,7 @@
     "@xen-orchestra/cron": "^1.0.6",
     "@xen-orchestra/defined": "^0.0.0",
     "@xen-orchestra/emit-async": "^0.0.0",
-    "@xen-orchestra/fs": "^0.11.1",
+    "@xen-orchestra/fs": "^0.12.0-0",
     "@xen-orchestra/log": "^0.2.0",
     "@xen-orchestra/mixin": "^0.0.0",
     "@xen-orchestra/self-signed": "^0.1.0",
@@ -130,7 +130,7 @@
     "unzipper": "^0.10.5",
     "uuid": "^3.0.1",
     "value-matcher": "^0.2.0",
-    "vhd-lib": "^0.8.0",
+    "vhd-lib": "^0.9.0-0",
     "ws": "^7.1.2",
     "xdg-basedir": "^4.0.0",
     "xen-api": "^0.29.0",

packages/xo-vmdk-to-vhd/package.json

@@ -30,7 +30,7 @@
     "lodash": "^4.17.15",
     "pako": "^1.0.11",
     "promise-toolbox": "^0.15.0",
-    "vhd-lib": "^0.8.0",
+    "vhd-lib": "^0.9.0-0",
     "xml2js": "^0.4.23"
   },
   "devDependencies": {

yarn.lock

@@ -3974,7 +3974,7 @@ bind-property-descriptor@^1.0.0:
   dependencies:
     lodash "^4.17.4"
 
-bindings@^1.5.0:
+bindings@^1.3.0, bindings@^1.5.0:
   version "1.5.0"
   resolved "https://registry.yarnpkg.com/bindings/-/bindings-1.5.0.tgz#10353c9e945334bc0511a6d90b38fbc7c9c504df"
   integrity sha512-p2q/t/mhvuOj/UeLlV6566GD/guowlr0hHxClI0W9m7MWYkL1F0hLo+0Aexs9HSPCtR1SXQ0TD3MMKrXZajbiQ==
@@ -12784,7 +12784,7 @@ node-gyp-build@~4.1.0:
   resolved "https://registry.yarnpkg.com/node-gyp-build/-/node-gyp-build-4.1.1.tgz#d7270b5d86717068d114cc57fff352f96d745feb"
   integrity sha512-dSq1xmcPDKPZ2EED2S6zw/b9NKsqzXRE6dVr8TVQnI3FJOTteUMuqF3Qqs6LZg+mLGYJWqQzMbIjMtJqTv87nQ==
 
-node-gyp@^3.8.0:
+node-gyp@^3.7.0, node-gyp@^3.8.0:
   version "3.8.0"
   resolved "https://registry.yarnpkg.com/node-gyp/-/node-gyp-3.8.0.tgz#540304261c330e80d0d5edce253a68cb3964218c"
   integrity sha512-3g8lYefrRRzvGeSowdJKAKyks8oUpLEd/DyPV4eMhVlhJ0aNaZqIrNUIPuEWWTAoPqyFkfGrM67MC69baqn6vA==
@@ -17313,6 +17313,14 @@ syntax-error@^1.1.1:
   dependencies:
     acorn-node "^1.2.0"
 
+syscall@^0.2.0:
+  version "0.2.0"
+  resolved "https://registry.yarnpkg.com/syscall/-/syscall-0.2.0.tgz#9308898495dfb5c062ea7a60c46f81f29f532ac4"
+  integrity sha512-MLlgaLAMbOGKUVlqsLVYnJ4dBZmeE1nza4BVgVgGUr2dPV17tgR79JUPIUybX/EqGm1jywsXSXUPXpNbIXDVCw==
+  dependencies:
+    bindings "^1.3.0"
+    node-gyp "^3.7.0"
+
 table@^5.2.3:
   version "5.4.6"
   resolved "https://registry.yarnpkg.com/table/-/table-5.4.6.tgz#1292d19500ce3f86053b05f0e8e7e4a3bb21079e"