Introduce new X-OC-ETag header to fix some sync issues.

[guruz]

The syncing client uses the ETag header for its syncing algorithm.

There is however a difference in semantics of what the sync client needs and what the HTTP RFC (and servers) implement.

The syncing client uses the ETag to uniquely identify (changed) file contents.
The HTTP servers do the same, however compressing responses (Content-Encoding) also changes or removes the ETag.
(The sync client doesn't want that!)

This can be seen for example here:
https://httpd.apache.org/docs/trunk/mod/mod_deflate.html#deflatealteretag
http://lxr.nginx.org/source/src/http/modules/ngx_http_gzip_filter_module.c#0309
pagespeed modules (lost the link)

So a plain webserver works fine with the sync client, but as soon as users configure gzip on their servers, it might break.

A solution would be for the server to additionally(!) send a X-OC-ETag header with the "real" oC file ETag.

I will try to do a pull request. Probably code change in lib/private/connector/sabre/file.php in get() if no one has a better idea

(This should be done for oC6 and oC7)

CC @ogoffart @PVince81 @dragotin

Source issues:
owncloud/client#1291
owncloud-archive/documentation#344
owncloud/client#1097
owncloud/client#1306

[PVince81]

Sounds good.
Until now I believed that my ownCloud server runnining on openSUSE + Apache 2 had gzip enabled by default. But what you wrote above seem to imply that it doesn't, else I'd have experienced that etag issue already.

CC @karlitschek @craigpg @icewind1991

[blizzz]

@PVince81 see the first link – Apache's behaviour can be configured so it keeps, modifies or removes it. Unchanged was default until Apache 2.4.

[ghost]

Hi,

some so called "security guides" are also recommending to set:

Header unset ETag
FileETag None

in the apache config which removes the used ETag header completely.

Source: https://forum.owncloud.org/viewtopic.php?f=14&t=16800&start=10

[PVince81]

@craigpg @karlitschek please evaluate the priority. Thanks.

[PVince81]

Seems the priority was set to 7.0.1

[DeepDiver1975]

Regarding the naming of the header - we shall not prefix it with 'X-' - this is an deprecated convention.
Simply OC-ETag is valid

[PVince81]

Reopening this, setting to 8.0.1

[PVince81]

(the original fix was reverted here: #13932)

[nsymms]

The ETag issue has been opened / closed / fixed / reverted so many times I've lost track.

I'm still having the same problem with server v8.0.2 using nginx. The v1.7.1 windows client continues to receive a tiny text file over and over again because "No E-Tag received from server."

What's going to happen with this? Since this particular issue (9005) is still open, is the issue still active? Or did someone just forget to close this issue?

I can post logs and other info if needed.

[ghost]

@nsymms
Try to disable gzip in your NginX config: owncloud-archive/documentation#922

AFAIK this "X-OC-ETag" will be added in OC 8.1 so thats probably the reason why this issue here is still open.

[nsymms]

Thanks. gzip disabled, error disappeared. I was concerned because I saw PVince set it to target 8.0.1. Now I see that it is an 8.1 fix.

On a side note, I am using OC inside a docker container, so I had to disable gzip on both the container's nginx and the outside nginx proxy (just for the OC server).

[ghost]

@nsymms
Great, thanks for the feedback. Hope my documentation change linked above will be merged soon.

Really wonder why no one had found the initial problem which i had already reported nearly a year ago in owncloud-archive/documentation#344

[PVince81]

I merged it too quickly... sorry. Revert PR here #15218

The problem is described here: #15168 (comment)

[PVince81]

Fix here: #15227
This one is much simpler and copies the Etag header in Sabre's "afterMethod"

[PVince81]

Fixed through #15227