Failing to run Shadow on Ubuntu in corporate environment #2271
-
|
Hello, I encounter a problem with setting up Shadow, and after several days trying to figure out what the problem could be, I dare to ask here for help. I am in a corporate environment behind a proxy. I have tried to install Shadow on Ubuntu 20.04 and 22.04 in several environments:
Every time, I fail in installing Shadow and running the file transfer demo successfully. On the Bare metal install, when running the tests after building Shadow from sources, most of them fail: When I look at the verbose output for the tests I get the following error most of the time and processes executed in the tests exit with code 134. I have changed my system's limits as mentioned in https://github.com/shadow/shadow/blob/0171ea8282e19bc439add4ad952bf75a98444627/docs/system_configuration.md If you have an idea about why I face those issues, I would really appreciate any help ! Thanks in advance, Antoine |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 11 replies
-
|
Do you have any other information about your system that would help us replicate this? What Linux kernel version are you using (output of
This error causes the managed process to exit, which is why the tests are failing on your baremetal server. I'm not sure why this is failing, but you can try skipping the vdso interception by change the line: shadow/src/lib/shim/patch_vdso.c Lines 178 to 179 in ea80705 to warning("Can't inject %zd byte trampoline into %zd byte symbol '%s'", trampolineSize,
symbol->st_size, vdsoFnName);
return;If you make this change, do the tests pass? |
Beta Was this translation helpful? Give feedback.
-
|
Hello Here is the result of the command on my machine (Ubuntu 22.04 LTS, 5.15.0-39-generic kernel) |
Beta Was this translation helpful? Give feedback.
-
Yes, this seems to be specific to the vdso module, which is part of the kernel. e.g. in the case of Docker I expect the problem to surface or not based on the kernel version of the host machine.
It seems that readelf agrees that the symbol is only 5 bytes large, which is too small for us to inject our trampoline. I'm guessing it must be a relative jump to somewhere else in the vdso, which would only take a few bytes.
I'm guessing that this internal symbol contains the actual code, and is where I'll have to think about if there's a way to shrink our trampoline to <= 5 bytes. Maybe we can inject something that'll raise a signal, and get control in the signal handler. @afressancourt, @stevenengler's workaround of changing the panic to a warning is probably good enough for your purposes. If you're interested in digging further, the next step is probably to inspect the disassembly of these symbols to validate my hypothesis. e.g. |
Beta Was this translation helpful? Give feedback.
-
|
I am interested in helping as much as I can given the effort you make to answer my questions (I guess this is the purpose of open sourcing a project like Shadow ;-)) Here is the information you have asked for: And Regarding the kernel version I run, I have not done anything special with it, I think this is the version you get when you make a clean Ubuntu 22.04 LTS install on a baremetal server. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! Yup, it looks pretty much like what I thought. I think we'll need to develop a smaller trampoline, and start running our CI on an Ubuntu 22.04 base platform. (We currently test a 22.04 Docker container, but from an older base platform and hence an older kernel). I'll add this info to the issue #2273 |
Beta Was this translation helpful? Give feedback.
-
|
This should be fixed now as of 884cccc |
Beta Was this translation helpful? Give feedback.
Do you have any other information about your system that would help us replicate this? What Linux kernel version are you using (output of
uname -a)? Both Ubuntu 20.04 and 22.04 work for me locally and in our CI tests.This error causes the managed process to exit, which is why the tests are failing on your baremetal server. I'm not sure why this is failing, but you can try skipping the vdso interception by change the line:
shadow/src/lib/shim/patch_vdso.c
Lines 178 to 179 in ea80705