Skip to content

File Tunnel v3.5.8

Latest

Choose a tag to compare

@fiddyschmitt fiddyschmitt released this 17 Nov 00:49

Reliability

  • SMB on macOS: fixed tunnels with a Mac end stalling until they went offline. The Mac end could wait for a command that had already arrived, and with both ends on one Mac and a Windows server, macOS lost some of the other end's writes. ft now reads the tunnel file uncached, rather than opening it again on every poll.
  • Linux and macOS: fixed a case where a tunnel could skip a command at the start of a session or after a purge. A share can show a file's new size before its data, and ft kept the zeros it read there rather than reading the file again, so it waited until the tunnel timed out and then went on past the command.
  • macOS on Apple silicon: ft now really turns read-ahead off on SMB and NFS mounts. Its calls to fcntl passed their setting where Apple silicon doesn't look for it, so read-ahead stayed on, and caching was turned off only by chance.
  • File handles: ft now closes each handle it replaces, when it reopens the other end's file after a purge on macOS and 9P, and when a tunnel end starts over after an error, rather than leaving it open until the garbage collector gets to it. On a Mac's SMB mount, a handle closed at the wrong moment could make macOS lose the other end's writes.
  • File names: tunnel files named without a folder, such as ft -r 2.dat -w 1.dat run inside the share, now get the same mount detection as full paths. On a Mac that turns on the SMB and NFS handling, and on Linux the sshfs, VirtualBox shared folder and 9P detection.
  • Dropbox: a server error or a connection timeout when ft gets its access token no longer stops ft with "Could not authenticate with Dropbox". ft tries again up to three times. A wrong app key, app secret or refresh token still fails at once.

Improvements

  • --isolated-io: tunnels are about three times as fast. ft reads ahead rather than opening the other end's file for each part of a command, and purges the files every 8 MB rather than every 1 MB. On Linux, ft picks this mode by itself for sshfs and VirtualBox shared folders.
  • NFS: when the Linux end of a tunnel starts first, the tunnel now comes online within seconds. It used to wait up to 30 s to see the other end's file.
  • Performance: ft spends about half as much CPU polling and flushing. It flushes the tunnel file once per batch rather than three times, and no longer wakes a thousand times a second while a tunnel or a connection is idle. Over SMB, where every flush is a round trip to the server, transfers take about half as long.

Other

  • Android: built on .NET 10.0.15, the runtime the other builds use, in place of 10.0.9.