diff options
| author | Brian Behlendorf <[email protected]> | 2016-03-17 15:32:33 -0700 |
|---|---|---|
| committer | Brian Behlendorf <[email protected]> | 2016-03-21 10:20:02 -0700 |
| commit | 505d9655c9e759a145b4812c654c51c34dbe3425 (patch) | |
| tree | 7ad9dfd11ce94dd597a72b61c6d1563e55f6208b /cmd | |
| parent | fc0c72b1670156622af26562030e84bd56d4cdfb (diff) | |
Fix zdb -e and zhack thread_init()
This issue was caused by calling `thread_init()` and `thread_fini()`
multiple times resulting in `kthread_key` being invalid. To resolve
the issue the explicit calls to `thread_init()` and `thread_fini()`
required by the `zpool` command have been moved in to the command.
Consumers such as `zdb` and `zhack` perform the same initialized
through `kernel_init()` and `kernel_fini()`.
Resolving this issue allows multiple additional test cases to
be enabled.
Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Signed-off-by: Chunwei Chen <[email protected]>
Signed-off-by: Tim Chase <[email protected]>
Closes #4331
Diffstat (limited to 'cmd')
| -rw-r--r-- | cmd/zpool/zpool_main.c | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/cmd/zpool/zpool_main.c b/cmd/zpool/zpool_main.c index b74dc219a..c9b600650 100644 --- a/cmd/zpool/zpool_main.c +++ b/cmd/zpool/zpool_main.c @@ -2331,7 +2331,15 @@ zpool_do_import(int argc, char **argv) idata.cachefile = cachefile; idata.scan = do_scan; + /* + * Under Linux the zpool_find_import_impl() function leverages the + * taskq implementation to parallelize device scanning. It is + * therefore necessary to initialize this functionality for the + * duration of the zpool_search_import() function. + */ + thread_init(); pools = zpool_search_import(g_zfs, &idata); + thread_fini(); if (pools != NULL && idata.exists && (argc == 1 || strcmp(argv[0], argv[1]) == 0)) { |