826846 - Infinite Crash Loop of Doom on the Homescreen if I reinstall a packaged app that already exists - phone homescreen becomes entirely unusable, even after a power down and power up

Status: VERIFIED FIXED

(Firefox OS Graveyard :: General, defect)

Description

[Jason Smith [:jsmith]]

Attachment: Logcat

Build: B2G 18 1/3/2013
Device: Unagi

Steps:

1. Install a packaged app from http://mozqa.com/webapi-permissions-tests/ ("Packaged App Test Case 2" was the one I used)
2. After installation completes, reinstall that packaged app

Expected:

The app should be successfully installed.

Actual:

The homescreen crashes. Then, after going past the crash reporter dialog, the homecsreen crashes again. And again. And it goes into an infinite crash loop of doom.

This looks bad:

01-04 21:59:06.773: I/Gecko(926): [Child 926] ###!!! ABORT: [PRemoteOpenFileChild] abort()ing as a result: file PRemoteOpenFileChild.cpp, line 361

Comment 1

[Jason Smith [:jsmith]]

This is probably the worst crash I've ever seen. The homescreen goes into an infinite crash loop making it unusable.

Comment 2

[Jason Smith [:jsmith]]

https://crash-stats.mozilla.com/report/index/bp-ff6611dd-5a68-46ea-941c-878732130104

Comment 3

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Yup, this looking bad.

Would be interesting to understand both why we are crashing here, since apps shouldn't be able to crash, and why this is an error situation at all.

Comment 4

[Jason Smith [:jsmith]]

Attachment: Submitted Crash Reports Zip File

Here's a dump of every crash report that was submitted in the infinite crash loop.

Looks like it's consistently these two stacks that I'm seeing:

https://crash-stats.mozilla.com/report/index/bp-f90b9616-3887-4080-86ab-6c5be2130104

https://crash-stats.mozilla.com/report/index/bp-f81bf21d-ea00-4a95-9045-8e2172130104

Comment 5

[Jason Smith [:jsmith]]

Btw - if you power off the phone and restart it, you'll still get the crashers after you go back to the homescreen.

So basically the phone is borked at this point.

Comment 6

[Jason Smith [:jsmith]]

So I re-flashed a fresh build and tried the repro steps again - it's consistently 100% reproducible.

Comment 7

[Jason Smith [:jsmith]]

(In reply to Jason Smith [:jsmith] from comment #6)
> So I re-flashed a fresh build and tried the repro steps again - it's
> consistently 100% reproducible.

And I just got it to reproduce on a different phone too.

Comment 8

[Marcia Knous [:marcia]]

Adding the two signatures so it gets picked up in crash stats.

Comment 9

[Gregor Wagner [:gwagner]]

Maybe a regression from bug 815523

Comment 10

[Jason Smith [:jsmith]]

Apparently this also reproduces after installing two large packaged apps and viewing it the settings apps perms UI.

Comment 11

[Johnny Stenback (:jst)]

Blocking and over to jduell to investigate whether this has something to do with bug 815523.

Comment 12

[Jason Duell]

Attachment: v1: return NS_FILE_NOT_FOUND instead of crashing when file is missing

What was happening:  The HomeScreen app was getting launched, and tried to open the application.zip for an application, but that file no longer exists (which is going to be a separate bug: looks like "install twice" == remove application.zip).  

IPDL kills the child process if you try to send fd==-1 as a file descriptor, so work around it by using a separate IPDL message.  Now we return NS_FILE_NOT_FOUND instead of crashing.  This gets rid of the thundering herd of dying HomeScreen apps that renders the phone useless.

Comment 13

[Jason Duell]

https://hg.mozilla.org/integration/mozilla-inbound/rev/2c0352886e95
https://hg.mozilla.org/releases/mozilla-b2g18/rev/c68fd32f4336

Seemingly valid rumor has it that I should mark B2G blockers FIXED as soon as they're on inbound/b2g18, and don't need to wait for m-c.

Comment 14

[Jason Duell]

I've verified that the crash on the phone is fixed, and gets replaced by bug 827502 :)

Comment 15

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/2c0352886e95

Comment 16

[Jason Smith [:jsmith]]

Verified then per comment 14