staging: lustre: obdclass: serialize lu_site purge

Umount process relies on lu_site_purge(-1) to purge all
objects before umount, however, if there happen to have a
cache shrinker which calls lu_site_purge(nr) in parallel,
some objects may still being freed by cache shrinker even
after the lu_site_purge(-1) called by umount done.

This can be simply fixed by serializing purge threads,
since it doesn't make any sense to have them in parallel.

Signed-off-by: Niu Yawei <yawei.niu@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5331
Reviewed-on: http://review.whamcloud.com/11099
Reviewed-by: Lai Siyao <lai.siyao@intel.com>
Reviewed-by: Mike Pershin <mike.pershin@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/staging/lustre/lustre/include/lu_object.h b/drivers/staging/lustre/lustre/include/lu_object.h
index 502bc41658c6381a909ba14819b4a05bde13208d..fe40b428346d396411bfab36e92c0e2f3b1d80ea 100644 (file)
--- a/drivers/staging/lustre/lustre/include/lu_object.h
+++ b/drivers/staging/lustre/lustre/include/lu_object.h
@@ -622,6 +622,11 @@ struct lu_site {
        struct list_head                ls_ld_linkage;
        spinlock_t              ls_ld_lock;
 
+       /**
+        * Lock to serialize site purge.
+        */
+       struct mutex            ls_purge_mutex;
+
        /**
         * lu_site stats
         */

diff --git a/drivers/staging/lustre/lustre/obdclass/lu_object.c b/drivers/staging/lustre/lustre/obdclass/lu_object.c
index 9d1c96b8ab7355c38846eb260270735315f9f551..b6fd9af93f04bf60e29a92e0429d335a173d6a99 100644 (file)
--- a/drivers/staging/lustre/lustre/obdclass/lu_object.c
+++ b/drivers/staging/lustre/lustre/obdclass/lu_object.c
@@ -354,6 +354,11 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
        start = s->ls_purge_start;
        bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1;
  again:
+       /*
+        * It doesn't make any sense to make purge threads parallel, that can
+        * only bring troubles to us. See LU-5331.
+        */
+       mutex_lock(&s->ls_purge_mutex);
        did_sth = 0;
        cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
                if (i < start)
@@ -399,6 +404,7 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
                if (nr == 0)
                        break;
        }
+       mutex_unlock(&s->ls_purge_mutex);
 
        if (nr != 0 && did_sth && start != 0) {
                start = 0; /* restart from the first bucket */
@@ -983,6 +989,7 @@ int lu_site_init(struct lu_site *s, struct lu_device *top)
        char name[16];
 
        memset(s, 0, sizeof(*s));
+       mutex_init(&s->ls_purge_mutex);
        snprintf(name, 16, "lu_site_%s", top->ld_type->ldt_name);
        for (bits = lu_htable_order(top); bits >= LU_SITE_BITS_MIN; bits--) {
                s->ls_obj_hash = cfs_hash_create(name, bits, bits,