Segmentation fault whenever dialogue appears for more than ~1 second

Victor Engmark said:

PS: I believe this has been happening at least since September 2017, when I started using Thunar https://github.com/l0b0/root/commit/1574fed43478d9dd26d9f6afa6564a97b1710946.

Andre Miranda @andreldm said:

Unfortunately I can't reproduce here with Thunar from master and 1.6.14, I have used touch {0..1000} and all three view modes (Icons, Detailed and Compact). I know you opened the bug against 1.6.14, but could please confirm you're using this version? A couple of previous versions were crashing on simple file operations. Besides that, which distro are you running?

Victor Engmark said:

I'm using Arch Linux. Details:

$ thunar --version | head -1 Thunar 1.6.14 (Xfce 4.12)

$ uname --kernel-name --kernel-release --kernel-version --machine --processor --hardware-platform --operating-system Linux 4.15.9-1-ARCH #1 (closed) SMP PREEMPT Sun Mar 11 17:54:33 UTC 2018 x86_64 unknown unknown GNU/Linux

$ cat ~/.config/xfce4/xfconf/xfce-perchannel-xml/thunar.xml

`` >>>

Alexander Schwinn @alexxcons said:

I just took a try with thunar 1.6.14 as well

used your xfconf configuration touch {0..299} and later touch {0..1000}

However I cannot reproduce the bug.

It looks like something else is making the difference.

Can you reproduce the bug when thunar is started from command line, and when you quit any possible running thunar-daemon ? thunar -q; thunar

Possibly you installed some extra thunar-plugins ?

Maybe a different filesystem ? I tested on ext4.

Could you generate and attach the coredump + the thunar binary used ?

Alexander Schwinn @alexxcons said:

When using touch {0..2000}, a spinner appears after the dialog for me and my CPU load goes max for some seconds ... possibly a performance issue.

Can you reproduce the bug on other systems ?

Victor Engmark said:

I did reproduce it from the command line to see if there were any useful messages. This is the entire output (the warning pops up every time I change directory during normal use):

$ thunar -q; thunar

(thunar:5815): Gdk-WARNING **: gdk_window_set_icon_list: icons too large Segmentation fault (core dumped)

I never used Thunar as a daemon.

No plugins are installed:

$ pacman --query | grep thunar thunar 1.6.14-1

The crash is reproducible on tmpfs and encrypted EXT4 LVM:

$ df -h /tmp /home/$USER Filesystem Size Used Avail Use% Mounted on tmpfs 7.9G 2.0M 7.9G 1% /tmp /dev/mapper/vg-home 111G 83G 23G 79% /home

$ mount | grep /home /dev/mapper/vg-home on /home type ext4 (rw,relatime,data=ordered)

I'm able to reproduce with the commands I specified on both my laptop and desktop machine (including stopping any running Thunar instances beforehand). It might be relevant that even moving files is incredibly slow in Thunar. Moving the 300 files in the test takes about 3-4 seconds within the same tmpfs on both machines, which is kind of ludicrous.

My setup is mostly automated:

Base install: https://l0b0.wordpress.com/2016/10/07/full-disk-encryption-with-arch-linux-footnotes/
Software installed: https://github.com/l0b0/root/
User configuration: https://github.com/l0b0/tilde/

Victor Engmark said:

Do you have somewhere I can put a 300 MB core dump?

Victor Engmark uploaded an attachment:

Attachment 7641, "Thunar binary":
thunar

Alexander Schwinn @alexxcons said:

Thanks for the info ! Possibly the disk encryption makes a difference .. have to test that. On a unencrypted ssd the movement of 300 files is something like 1-2 sec. for me ... did not compare with other file managers so far.

For the coredump, maybe this one would be an option: https://www.transfernow.net/ ?

Victor Engmark said:

Thanks! Core dump: https://www.transfernow.net/4aj3d4w2haab

Andre Miranda @andreldm submitted a patch:

I was finally able to reproduce the bug after many attempts. However, I found an easier way to force the crash: keep switching between the target folder and its parent, Thunar will certainly crash.

These are the warnings messages it normally emits before crashing (not in this order): (thunar:4043): GLib-GObject-WARNING **: invalid cast from 'GtkCssImageBuiltin' to 'ThunarFolder' (thunar:4244): GLib-GObject-WARNING **: invalid cast from 'GtkCssTransition' to 'ThunarFolder' (thunar:4170): GLib-GObject-WARNING **: invalid uninstantiatable type '(null)' in cast to 'ThunarFolder' (thunar:4170): thunar-WARNING **: Content type loading failed for 370: Error when getting information for file “/tmp/tmp.KJj2GZa3j3/foo/370”: No such file or directory

And this is the backtrace (sometimes it segfaults at line 424):

#0  0x000055555559f2a9 in thunar_folder_content_type_loader_idle (data=<optimized out>) at thunar-folder.c:429
#1  0x00007ffff4c80ca6 in g_main_context_dispatch () at /usr/lib/libglib-2.0.so.0
#2  0x00007ffff4c81081 in  () at /usr/lib/libglib-2.0.so.0
#3  0x00007ffff4c8110e in g_main_context_iteration () at /usr/lib/libglib-2.0.so.0
#4  0x00007ffff524696e in g_application_run () at /usr/lib/libgio-2.0.so.0
#5  0x000055555557ac8d in main (argc=2, argv=0x7fffffffe008) at main.c:165

My suspicion is that we're facing a race condition, where multiple threads are moving files, meanwhile Thunar is trying to update its view and then it uses an invalid reference to a folder.

I have prepared a patch that prevents crashes, but it acts on the problem effects, not on the cause, so CRITICAL messages are emitted. I just ask that you (Victor), if possible, could test it so we are sure to be on the right track.

Patch 7642, "patch":
test.patch

Alexander Schwinn @alexxcons said:

Did not get any more info out of the coredump, anyhow thanks for sending it ! Offtopic, for next time: If you gzip the coredump it will shrink down to 3,8 MB

Victor Engmark said:

Thank you Andre, that seems to have fixed it!

Build instructions for reproducibility and other testers:

The repo doesn't have build instructions. I found a build command in the thunar-git AUR package: ./autogen.sh --prefix=/usr --sysconfdir=/etc --libexecdir=/usr/lib --localstatedir=/var --disable-static --enable-gio-unix --enable-dbus --enable-startup-notification --enable-gudev --enable-exif --enable-pcre --enable-gtk-doc --disable-debug
libxfce4ui-2 needs to be at least version 4.13.2 to build master.
There is no 1.6.15 tag on master. Found "thunar-1.6.15" it on the xfce-4.12 branch.

GitBot @GitBot said:

Andre Miranda referenced this bugreport in commit 2f4a4dec

Avoid segfault on file operations (Bug #14281)

https://git.xfce.org/xfce/thunar/commit?id=2f4a4dec0c4dde4db737e125247a9787edf5eded

GitBot @GitBot said:

Andre Miranda referenced this bugreport in commit e1246248

Avoid segfault on file operations (Bug #14281)

https://git.xfce.org/xfce/thunar/commit?id=e124624880a470aa6ad807368b555cfbd39447ac

GitBot @GitBot said:

Andre Miranda referenced this bugreport in commit a754e7f9

Avoid segfault on file operations (Bug #14281)

https://git.xfce.org/xfce/thunar/commit?id=a754e7f95ea6080d72c6344670d14ab111062bf8

Andre Miranda @andreldm said:

@Victor, sorry for the delay, I haven't had the time to check this again, for now the stopgap will do. Let's keep this bug open until a proper fix is released. And thanks for reporting and testing!

added 1. Crash / Freeze 5. File Operations Patch labels

assigned to @SKefalidis

@alexxcons something similar happened to me today (crash when copying/pasting 150 files), I don't know if it is the same problem since it isn't reproducible if I use gdb. I'll investigate it further.

FYI last evening I got four segfaults with thunar master while copy files between external drives (ext4 and ntfs):

#0  0x0000563000ad9954 in thunar_view_reload (view=0x5630017315e0, reload_info=0) at thunar-view.c:267
#1  0x00007f75dc197d8f in g_closure_invoke () at /usr/lib/libgobject-2.0.so.0
#2  0x00007f75dc1b3588 in  () at /usr/lib/libgobject-2.0.so.0
#3  0x00007f75dc1b4c49 in g_signal_emit_valist () at /usr/lib/libgobject-2.0.so.0
#4  0x00007f75dc1b51a0 in g_signal_emit () at /usr/lib/libgobject-2.0.so.0
#5  0x0000563000a7aa2c in thunar_clipboard_manager_targets_received (clipboard=<optimized out>, selection_data=0x7fffd1153550, user_data=<optimized out>) at thunar-clipboard-manager.c:400

Thanks for info! Planned to release 4.17.5 in a few day .. but meh, lets first fix that :F

Any chance some commit from master sneaked into 4.16.9 causing the segfault seen by @andreldm? I've seen a couple of crashes after upgrading to 4.16.9 and have also received a downstream report about the thunar_view_reload segfault. For me it happened when copying a few tens of ~400MB files to a USB drive, but I have been unable to repro intentionally since then (~4 days ago).

(Edit: Bit weird that we're commenting on a 3 year old bug, but whatever works. :D)

@alexxcons just a random guess, could this be related to what was fixed in a5ec0d35?

e20ac87c is looking a bit suspicious I think.

Let's continue in #653 (closed)? ️

Segmentation fault whenever dialogue appears for more than ~1 second

Submitted by Victor Engmark

Description

Designs

Child items 0

Activity