Signal-desktop: Spell-check unavailable when /tmp mounted as noexec

[pmocek]

• [x] I have searched open and closed issues for duplicates

---

Bug description

Upon new installation of Signal Desktop for Linux, importing Signal data exported from Chrome app fails upon attempt to select directory from which to import.

Steps to reproduce

• Install Signal Desktop Apt package
• Execute signal-desktop
• Click "SET UP WITH CHROME APP EXPORT"
• Click "Choose directory"

Actual result:
Nothing happens in GUI.
Following message is logged to standard output:

"TypeError: Cannot read property 'getDirectoryForImport' of undefined\n at child.onImport (file:///opt/Signal/resources/app.asar/js/views/import_view.js:102:21)\n at HTMLDivElement.dispatch (file:///opt/Signal/resources/app.asar/js/components.js:4409:9)\n at HTMLDivElement.elemData.handle (file:///opt/Signal/resources/app.asar/js/components.js:4095:28)"

Expected result:
File/directory selection dialog appears

Screenshots

Platform info

Operating System: GNU/Linux (4.10.0-32-generic #36~16.04.1-Ubuntu SMP)
Browser: n/a

Signal version: 1.0.34

Link to debug log

https://gist.github.com/pmocek/20ac06d9a44947243d1e46733b0e5a17

[scottnonnenberg]

It looks like basic things were unable to start up in the app, because nodeSetImmediate is not available. Do you happen to have you /tmp mounted as noexec? That's what caused this in https://github.com/WhisperSystems/Signal-Desktop/issues/1618

[pmocek]

Yes, of course, I have /tmp mounted with noexec:

$ fgrep '/tmp' /proc/mounts 
/dev/sda5 /tmp btrfs rw,noexec,noatime,ssd,space_cache,subvolid=273,subvol=/@tmp 0 0
/dev/sda5 /var/tmp btrfs rw,noexec,noatime,ssd,space_cache,subvolid=276,subvol=/@var_tmp 0 0

[pmocek]

/tmp is not really a place for executable files on Linux.

[unDocUMeantIt]

mounting /tmp noexec is also explained in the Securing Debian Manual.

[scottnonnenberg]

(the feature is supporting /tmp mounted as noexec)

[unDocUMeantIt]

is /tmp used in exec mode constantly, or is this just blocking the import dialog for the time being? the thing is, since i exported previous conversations from the chrome extension, i can no longer use it, so right now i'm stuck with neither desktop solution until this is resolved. like i'm locked out.

[scottnonnenberg]

You can go back to using the chrome dialog with these steps: go to chrome://extensions in your browser, find Signal, then click the background.html link, then enter this text in the Console tab:

window.storage.put('migrationEnabled', false);
window.storage.put('migrationState', 0);
window.storage.put('migrationEverCompleted', false);

Wait a bit, then:

window.location.reload();

[pmocek]

By changing this from a bug report to a feature request, you imply that it is working as designed, and thus that the design assumes that a Linux user of Signal Desktop has not implemented a once- if not currently-recommended-by-Debian-among-others system hardening measure. Is that documented anywhere that developers, much less users, could find it?

[scottnonnenberg]

You raise a good point. Bug it is. I liked that other tag since it also indicated that it could potentially be a substantial amount of work. :0/

[unDocUMeantIt]

You can go back to using the chrome dialog with these steps: go to chrome://extensions in your browser, find Signal, then click the background.html link, then enter this text in the Console tab:

thanks, got it back again.
(hint to others: the background link becomes available when developer options are enabled).

[scottnonnenberg]

Some reference material.

The core error, from https://github.com/WhisperSystems/Signal-Desktop/issues/1618:

/opt/Signal/resources/electron.asar/renderer/init.js:155 Error: /tmp/.org.chromium.Chromium.Zl3FPz: failed to map segment from shared object
    at process.module.(anonymous function) [as dlopen] (ELECTRON_ASAR.js:173:20)
    at Object.Module._extensions..node (module.js:598:18)
    at Object.module.(anonymous function) [as .node] (ELECTRON_ASAR.js:187:18)
    at Module.load (module.js:488:32)
    at tryModuleLoad (module.js:447:12)
    at Function.Module._load (module.js:439:3)
    at Module.require (module.js:498:17)
    at require (internal/module.js:20:19)
    at Object.<anonymous> (/opt/Signal/resources/app.asar/node_modules/spellchecker/lib/spellchecker.js:3:16)
    at Object.<anonymous> (/opt/Signal/resources/app.asar/node_modules/spellchecker/lib/spellchecker.js:107:3)

Brave has this same problem: https://github.com/brave/browser-laptop/issues/900

And finally, this PR may make the application useable (though without spell-check) on Linux with /tmp mounted as noexec: https://github.com/WhisperSystems/Signal-Desktop/pull/1734

[unDocUMeantIt]

And finally, this PR may make the application useable (though without spell-check) on Linux with /tmp mounted as noexec: #1734

yes, i just tested v1.0.37 and can confirm the import dialog does open now. thanks!

[balping]

Yes, it works now, this issue should be closed

[scottnonnenberg]

I'm going to keep it open, because there's still something wrong when /tmp is mounted noexec - spellcheck is disabled.

[Unifex]

Just came across this issue. The suggestion from @scottnonnenberg on 3 Nov 2017 to go down the chrome://extensions route resolved my case.

Short version; This is still an issue but the mentioned comment above is a working temporary fix.

In my case this was just after installing the latest .deb. I set things up, saw that my contacts had imported, saw the backup option and took that. I created and used a new dot dir for the backup. It completed and gave me a success window with no buttons. Restarting multiple times did not resolve it. Started searching and found this post.

[scottnonnenberg-signal]

@Unifex Are you sure you're commenting on the right bug? How is the behavior you are seeing related to this? I'd recommend that you open a new bug and include your debug log as directed in the standard issue template.

[jtracey]

I believe this was fixed by 52e594901f6f30269e7bbe45b94b68eb9f7419f7

[kenpowers-signal]

That would make sense since the spell checking implementation has been entirely replaced. I'm going to close this issue for now but feel free to re-open it if the issue persists with the new spellcheck implementation.