#636 dsi_stream_read: len:0, unexpected EOF remains unfixed

[masc]

https://sourceforge.net/p/netatalk/bugs/633/ remains closed though it has been repeatedly reported not to be fixed. this issue seems critical on 10.12, where time machine backups fail frequently or accessing files with applications using autosave (eg. preview) may disconnect the entire volume almost instantly. Please re-open 633. Replicated with 3.1.10.

[John Gubert]

I compiled netatalk 3.1.10 on Xenial 16.04.1 using this howto:

http://netatalk.sourceforge.net/wiki/index.php/Install_Netatalk_3.1.9_on_Ubuntu_16.04_Xenial

While normal acces to folders works fine, as soon as I switch to time machine = yes, I get errors:

Dec 22 17:17:45 timemachine afpd[469]: Login by me (AFP3.4)
Dec 22 17:17:45 timemachine afpd[469]: AFP logout by me
Dec 22 17:17:45 timemachine afpd[469]: dsi_stream_read: len:0, unexpected EOF
Dec 22 17:17:45 timemachine afpd[469]: afp_over_dsi: client logged out, terminating DSI session
Dec 22 17:17:45 timemachine afpd[469]: pam_unix(netatalk:session): session closed for user me
Dec 22 17:17:45 timemachine dbus[357]: [system] Rejected send message, 2 matched rules; type="method_call", sender=":1.7" (uid=1001 pid=469
Dec 22 17:17:45 timemachine afpd[469]: pam_systemd(netatalk:session): Failed to release session: Access denied
Dec 22 17:17:45 timemachine afpd[469]: AFP statistics: 0.61 KB read, 0.45 KB written

[Bryan Stansell]

Same issue here (running 3.1.10 on OmniOS with ZFS). Exporting photos via the Photos application on Sierra is also provoking the issue for me. Seems like Time Machine backup are fine, however (dunno about restores). This is the pattern I see with the Photos application (this over and over again)...

Dec 27 17:32:44.894723 afpd[7685] {dsi_tcp.c:241} (info:DSI): AFP/TCP session from 192.168.0.134:65143
Dec 27 17:32:44.900555 afpd[7685] {uams_dhx2_passwd.c:265} (info:UAMS): DHX2 login: bryan
Dec 27 17:32:45.888319 afpd[7684] {auth.c:561} (note:AFPDaemon): afp_disconnect: primary reconnect succeeded
Dec 27 17:32:47.983988 afpd[7685] {auth.c:235} (note:AFPDaemon): Login by bryan (AFP3.4)
Dec 27 17:32:47.987623 afpd[7685] {auth.c:536} (note:AFPDaemon): afp_disconnect: trying primary reconnect
Dec 27 17:32:47.988023 afpd[490] {main.c:151} (info:AFPDaemon): child[7684]: done
Dec 27 17:32:47.988264 afpd[490] {server_child.c:256} (note:Default): Reconnect: transfering session to child[7651]
Dec 27 17:32:47.988379 afpd[490] {server_ipc.c:235} (note:AFPDaemon): Reconnect: killing new session child[7685] after transfer
Dec 27 17:32:47.988643 afpd[7651] {afp_dsi.c:188} (note:AFPDaemon): afp_dsi_transfer_session: succesfull primary reconnect
Dec 27 17:32:47.991338 afpd[7651] {afp_dsi.c:614} (note:AFPDaemon): AFP Replay Cache match: id: 367 / cmd: AFP_GETEXTATTR
Dec 27 17:32:48.057814 afpd[7651] {dsi_stream.c:504} (error:DSI): dsi_stream_read: len:0, unexpected EOF
Dec 27 17:32:48.057964 afpd[7651] {dsi_stream.c:504} (error:DSI): dsi_stream_read: len:0, unexpected EOF
Dec 27 17:32:48.058005 afpd[7651] {dsi_stream.c:259} (note:DSI): dsi_disconnect: entering disconnected state

[masc]

I can confirm the time machine errors are (presumably) not caused by this, but rather slow io and thus broken pipes. The issue persists for the other mentioned use case(s) though.

[Bryan Stansell]

I was hunting around again for references to this issue and found the xattr.patch that was reposted here: https://sourceforge.net/p/netatalk/mailman/message/35446938/

I applied that to my 3.1.10 build and things are running smoothly now. No issues detected. Posting here so others know as well. Doesn't look like that patch made it into the source repo, but, given it's fixing issues, hopefully it will get adopted.

[Adonis]

Also getting this issue. This issue is corrupting my Photos library and Time machine backups to an AFS share. I am unable to apply the patch given that i install netatalk via the freebsd port. What has to happen so that this makes it to the repo?

[Adonis]

Well.. The patch is applied to the repo. It was applied back in october 2016. Issue is that in FreeBSD's case for example the netatalk3 port is building off of 3.1.10 which does not include these changes. The changes related to this patch seem to be coming as part of 3.1.11 which has not been release currently. So if your looking to get this patch build off HEAD which does have this patch included.

[Thorsten Hirsch]

The error still ocurrs in 3.1.12.

[Daniel Markstedt]

Just for the record: This exact issue affects 2.2.7 as well. A sure way to reproduce this is to use an EXTREMELY fast classic Mac, e.g. the Basilisk II emulator running System 7 on a modern PC. Then copy a folder containing a lot of small files from the emulated Mac to the shared volume. A good test case is Realmz 7.1.2 (uncompressed.)

[Daniel Markstedt]

This is tracked in GitHub now https://github.com/Netatalk/netatalk/issues/1631