No description
- C 92.4%
- Roff 4.6%
- Makefile 1.7%
- Python 1.1%
- Shell 0.2%
When plock_ownership mode is disabled, a plock operation read from
the kernel can be sent to all nodes immediately, without looking up
the local resource struct. This is faster, and it avoids a bug in
cases where multiple threads in a process submit concurrent fcntl
lock ops on a file.
This bug occurs because create/free of the resource struct should
only be done in the context of ordered receive_plock cpg messages.
plock unlock results were being incorrectly returned based on the
non-existence of a struct resource checked in process_plocks (outside
of the ordered message context.) In this case, the existence of the
resource struct was used to imply a lock state ("no locks"), which is
invalid to conclude outside of the message context. The bug would
only appear in the unusual scenario where multiple threads in a
process are running concurrent fcntl lock operations on a file.
This fix does not address the problem when plock_ownership is enabled,
so warnings are added for the plock_ownership setting.
|
||
|---|---|---|
| dlm_controld | ||
| dlm_sand | ||
| dlm_tool | ||
| fence | ||
| include | ||
| init | ||
| libdlm | ||
| python | ||
| .gitignore | ||
| dlm.spec.in | ||
| Makefile | ||
| README.license | ||
| README.rst | ||
| ``See https://pagure.io/dlm`` | ``See dlm_controld(8) at dlm.git/dlm_controld/dlm_controld.8`` | ``See dlm.conf(5) at dlm.git/dlm_controld/dlm.conf.5``