Cause: When a reservation conflict occurred, mpathpersist was not correctly iterating over all of the registration threads, and it was not correctly rolling back the registration of multiple devices
Consequence: mpathpersist could crash on reservation conflicts.
Fix: mpathpersist now correctly iterates over all of the registration threads, and correctly issues the rollback request
Result: mpathpersist no longer crashes when registrations fail with reservation conflicts.
Description of problem:
mpathpersist can fail with a crash that often looks like this:
Program terminated with signal 11, Segmentation fault.
#0 udev_list_entry_delete (entry=0xbad7aad20, entry@entry=0x558bad7aad20)
at src/libudev/libudev-list.c:243
243 free(entry->value);
However the crash could look very different on other systems. The common factor is that variables on the process stack are corrupted. This happens because
mpathpersist creates threads that access variables on the creating thread's
stack. If the creating function fails to wait for the created threads, it
exits while the threads still have access to the stack variables. This means
that they can overwrite stack variables used by later functions.
Version-Release number of selected component (if applicable):
How reproducible:
Reliably with the right setup.
Steps to Reproduce:
run
# /usr/sbin/mpathpersist -o --register --param-sark=<key> -d <dev>
while access to the array is slowed for some reason.
Actual results:
mpathpersist crashes
Expected results:
mpathpersist completes without crashing
Additional info:
To reproduce this, you need a multipath device with at least 3 active paths, but the more paths, the easier it is to hit. Then you need to use sgpersist to add a registration to the last path in the multipath device table. Finally you need to use mpathpersist to register a different key.
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.
https://access.redhat.com/errata/RHBA-2019:2138
Description of problem: mpathpersist can fail with a crash that often looks like this: Program terminated with signal 11, Segmentation fault. #0 udev_list_entry_delete (entry=0xbad7aad20, entry@entry=0x558bad7aad20) at src/libudev/libudev-list.c:243 243 free(entry->value); However the crash could look very different on other systems. The common factor is that variables on the process stack are corrupted. This happens because mpathpersist creates threads that access variables on the creating thread's stack. If the creating function fails to wait for the created threads, it exits while the threads still have access to the stack variables. This means that they can overwrite stack variables used by later functions. Version-Release number of selected component (if applicable): How reproducible: Reliably with the right setup. Steps to Reproduce: run # /usr/sbin/mpathpersist -o --register --param-sark=<key> -d <dev> while access to the array is slowed for some reason. Actual results: mpathpersist crashes Expected results: mpathpersist completes without crashing Additional info: