Submissions are exported twice

[conorsch]

Recently during Workstation QA, I was surprised to see that files exported to a physical USB stick were exported twice from the client. After tinkering for a bit with package versions, it appears that on my system the latest nightly securedrop-export_0.2.4-dev-20210315-061408 reliably doubles exports, whereas the prod version 0.2.4 does not.

Here's an example of the timestamps generated:

$ find /media/conor/Transfer/ | sort
/media/conor/Transfer/
/media/conor/Transfer/lost+found
/media/conor/Transfer/sd-export-20210315-161039
/media/conor/Transfer/sd-export-20210315-161039/export_data
/media/conor/Transfer/sd-export-20210315-161039/export_data/surf.mp4
/media/conor/Transfer/sd-export-20210315-161048
/media/conor/Transfer/sd-export-20210315-161048/export_data
/media/conor/Transfer/sd-export-20210315-161048/export_data/surf.mp4
/media/conor/Transfer/sd-export-20210315-161439
/media/conor/Transfer/sd-export-20210315-161439/export_data
/media/conor/Transfer/sd-export-20210315-161439/export_data/wikipedia.jpg
/media/conor/Transfer/sd-export-20210315-161455
/media/conor/Transfer/sd-export-20210315-161455/export_data
/media/conor/Transfer/sd-export-20210315-161455/export_data/wikipedia.jpg
/media/conor/Transfer/sd-export-20210315-161939
/media/conor/Transfer/sd-export-20210315-161939/export_data
/media/conor/Transfer/sd-export-20210315-161939/export_data/surf.mp4

There were three (3) export operations performed:

• surf.mp4 on latest nightly, was doubled
• wikipedia.jpg on latest nightly, was doubled
• surf.mp4 on 0.2.4 prod, was not doubled

[eloquence]

I've not been able to reproduce with the same nightly version yet -- I only get one copy per export operation. If you can reliably reproduce, it would be helpful to have exact steps, as the order of operations (when does sd-devices VM start, when is USB plugged in, when removed) and the system toast notifications displayed could be significant.

[emkll]

I can reproduce on the latest nightly (20210316) as well, regardless of the order of operations (plug in before sd-devices VM is started, plug in before the "insert usb drive" prompt)

[eloquence]

Still no "luck" reproducing it here. @creviera mentioned that it may be related to the USB device or formatting; I've only tested with the same export device so far. It is formatted as follows:

Or in lsblk language:

sdb                                             8:16   1  29.7G  0 disk  
└─sdb1                                          8:17   1  29.7G  0 part  
  └─luks-0dba726c-fdff-4f35-aede-2dfac81c0230 253:1    0  29.7G  0 crypt /media/erik/Backup

The device itself is a USB-based Flash Card reader, not a traditional USB stick. I'll test with the same device using the latest nightlies with the changes in freedomofpress/securedrop-export#67 to test for any regression on this device. For folks who've seen the issue, it would be useful to compare the above output or other debugging info.

[eloquence]

OK, I think we can discount the device/formatting hypothesis. After updating (including the nightlies with freedomofpress/securedrop-export#67) I've now seen the issue a couple of times with the exact same device. However, I'm still not able to reliably reproduce it -- I restarted sys-usb and the issue went away on the next couple of exports.

Note that my last update also pulled in dom0 updates, so it's not necessarily related to any change in the nightlies, especially since Conor/Mickael reported seeing the issue with older nightlies.

[sssoleileraaa]

Based on logs, it looks like the continue_button from the GUI is sending a clicked signal twice when exporting sometimes. So I'm switching my focus to the client code instead of changes that were made here: freedomofpress/securedrop-export#67

[sssoleileraaa]

I'm able to repro every time by pressing "Enter" instead of clicking the "Continue" button, which tells you that our QKeyEvent handler isn't handling the even properly. I'll have a fix shortly and would appreciate some extra eyes on testing in case there's another way to repro the issue.