Home |
Licence |
FAQ |
Docs |
Download |
Keys |
Links
Mirrors |
Updates |
Feedback |
Changes |
Wishlist |
Team
In the SSH file transfer protocol (SFTP), there is the possibility of the server returning less data than we asked for in response to a read request (SSH_FXP_READ).
Currently, the SFTP implementation in PSFTP/PSCP will barf if this happens other than at EOF ("error while reading: received a short buffer from FXP_READ, but not at EOF"). The filexfer draft at the time seemed to indicate that this was an exceptional condition that only applied to "special" files. However, on closer examination, its "guarantee" of no short reads appears not to be reliable.
We've had one report of this causing trouble with an OpenVMS server that likes to return ZIP files in 512-byte chunks due to the nature of the underlying filesystem (whereas we ask for 4k chunks).
Since PSFTP/PSCP usually have several requests outstanding at any one time, dealing with this properly requires a strategy for dealing with fragmentation and reassembly. (But this shouldn't be too bad as we already deal with out-of-order responses.) We may also want to attempt to adapt to the server's apparent preferred block size, although we should be prepared for this adaptation to fail.
All this should probably be implemented in as paranoid a manner as possible to avoid the server being able to cause unbounded buffering at our end.
Audit trail for this semi-bug.