Skip to content

Commit

Permalink
Use a dedicated taskq for vdev_file
Browse files Browse the repository at this point in the history
Originally, vdev_file used system_taskq. This would cause a deadlock,
especially on system with few CPUs. The reason is that the prefetcher
threads, which are on system_taskq, will sometimes be blocked waiting
for I/O to finish. If the prefetcher threads consume all the tasks in
system_taskq, the I/O cannot be served and thus results in a deadlock.

We fix this by creating a dedicated vdev_file_taskq for vdev_file I/O.

Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2270
  • Loading branch information
tuxoko authored and behlendorf committed May 14, 2014
1 parent 2c33b91 commit bc25c93
Show file tree
Hide file tree
Showing 3 changed files with 24 additions and 1 deletion.
3 changes: 3 additions & 0 deletions include/sys/vdev_file.h
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,9 @@ typedef struct vdev_file {
vnode_t *vf_vnode;
} vdev_file_t;

extern void vdev_file_init(void);
extern void vdev_file_fini(void);

#ifdef __cplusplus
}
#endif
Expand Down
3 changes: 3 additions & 0 deletions module/zfs/spa_misc.c
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include <sys/zap.h>
#include <sys/zil.h>
#include <sys/vdev_impl.h>
#include <sys/vdev_file.h>
#include <sys/metaslab.h>
#include <sys/uberblock_impl.h>
#include <sys/txg.h>
Expand Down Expand Up @@ -1660,6 +1661,7 @@ spa_init(int mode)
dmu_init();
zil_init();
vdev_cache_stat_init();
vdev_file_init();
zfs_prop_init();
zpool_prop_init();
zpool_feature_init();
Expand All @@ -1674,6 +1676,7 @@ spa_fini(void)

spa_evict_all();

vdev_file_fini();
vdev_cache_stat_fini();
zil_fini();
dmu_fini();
Expand Down
19 changes: 18 additions & 1 deletion module/zfs/vdev_file.c
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,8 @@
* Virtual device vector for files.
*/

static taskq_t *vdev_file_taskq;

static void
vdev_file_hold(vdev_t *vd)
{
Expand Down Expand Up @@ -184,7 +186,7 @@ vdev_file_io_start(zio_t *zio)
return (ZIO_PIPELINE_CONTINUE);
}

VERIFY3U(taskq_dispatch(system_taskq, vdev_file_io_strategy, zio,
VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
TQ_PUSHPAGE), !=, 0);

return (ZIO_PIPELINE_STOP);
Expand All @@ -209,6 +211,21 @@ vdev_ops_t vdev_file_ops = {
B_TRUE /* leaf vdev */
};

void
vdev_file_init(void)
{
vdev_file_taskq = taskq_create("vdev_file_taskq", 100, minclsyspri,
max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);

VERIFY(vdev_file_taskq);
}

void
vdev_file_fini(void)
{
taskq_destroy(vdev_file_taskq);
}

/*
* From userland we access disks just like files.
*/
Expand Down

0 comments on commit bc25c93

Please sign in to comment.