Teddy Hogeborn kirjoitti 17.02.2018 klo 15:59: > Edward Shornock <ed.shorn...@gmail.com> writes: > >> On 10.02.2018 23:24, Debian Bug Tracking System wrote: >>> * Fix "fails with "LeakSanitizer has encountered a fatal error"" >>> by fixing memory leak in plugin-runner (Closes: #886595) >> For what it's worth I still see an occasional "LeakSanitizer" crash >> with 1.7.18-1. I don't have any packages installed with *plymouth* in >> the name. >> >> Please let me know what I can do to help troubleshoot this. > You could probably *avoid* the problem entirely by recompiling the > program without "-fsanitize=leak". Of course, that only hides the > actual memory leak, but it avoids the problem it causes. Yes, this does work to avoid the problem. I built the package with -fsanitize=leak removed from Makefile and I had 10 successful boots with all lines commented out in plugin-runner-conf.
Then I reverted to the most recent mandos-client package in Debian (1.7.18-1) and tried what was suggested below: > To help *finding* the memory leak, I would suggest the following: > Disable all the plugins you don't use; put the following into > /etc/mandos/plugin-runner.conf: > > --disable=askpass-fifo > --disable=plymouth > --disable=splashy > --disable=usplash > > (Make sure to rebuild the initramfs image with "update-initramfs -k all > -u" after changing that file.) After doing this, do you still see > "LeakSanitizer" output when booting? I configured the following and rebuilt the initramfs: --disable=askpass-fifo --disable=plymouth --disable=splashy --disable=usplash With 10 boots I saw the leak sanitizer output all 10 times. Then I added --disable=password-prompt and rebuilt the initramfs. With 20 more boots I saw the leak sanitizer output each time. I logged onto the mandos server and disabled the client host and rebooted the system which sees the leak sanitizer output. In mandos-monitor (on the server) I could see the client connecting but of course it didn't receive the key since it was disabled. Once I re-enabled the client host with mandos-ctl, I saw the leak sanitizer output on the client again. After I used "rescue media" to get back in on the client system I changed --disable=password-prompt to --disable=mandos-client and rebuilt the initramfs. Once I entered the passphrase, the drives were unlocked and the system booted. I booted 10 times with this configuration and I did not see the leak sanitizer any of those times. > If you do, try to disable the "password-prompt" and/or the > "mandos-client" plugins as well. Note: disabling the "mandos-client" > plugin will necessitate the manual typing of the password on boot, and > disabling "password-prompt" plugin will rely on the Mandos client for > getting the password. You could also disable them *both*, and the > plugin-runner will fall back to asking for the password itself on the > console. I am wondering which, if any, of these disablings will make > the LeakSanitizer errors go away on your system. That would help to > narrow the problem down significantly. I have only been able to reproduce the problem when the mandos-client plugin is enabled along with a package which was built with "-fsanitize=leak" enabled.
signature.asc
Description: OpenPGP digital signature
