Add zfs_object_mutex_size module option

Add a zfs_object_mutex_size module option to facilitate resizing the
the per-dataset znode mutex array.  Increasing this value may help
make the deadlock described in #4106 less common, but this is not a
proper fix.  This patch is primarily to aid debugging and analysis.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Issue #4106

diff --git a/include/sys/zfs_vfsops.h b/include/sys/zfs_vfsops.h
index 28407c6f76fe1c725b3a56aff10e6d7bccaa0129..a59114a1a7ea6370f7693247e7a0412d056f7688 100644 (file)
--- a/include/sys/zfs_vfsops.h
+++ b/include/sys/zfs_vfsops.h
@@ -112,7 +112,7 @@ typedef struct zfs_sb {
        uint64_t        z_groupquota_obj;
        uint64_t        z_replay_eof;   /* New end of file - replay only */
        sa_attr_type_t  *z_attr_table;  /* SA attr mapping->id */
-#define        ZFS_OBJ_MTX_SZ  256
+       uint64_t        z_hold_mtx_size; /* znode hold locks size */
        kmutex_t        *z_hold_mtx;    /* znode hold locks */
 } zfs_sb_t;
 

diff --git a/include/sys/zfs_znode.h b/include/sys/zfs_znode.h
index 7e73cf99182e50ea936c9f340324e270ea350233..59ca085efe7a586c947d1e08592916a8f3c14759 100644 (file)
--- a/include/sys/zfs_znode.h
+++ b/include/sys/zfs_znode.h
@@ -273,9 +273,15 @@ typedef struct znode {
 /*
  * Macros for dealing with dmu_buf_hold
  */
-#define        ZFS_OBJ_HASH(obj_num)   ((obj_num) & (ZFS_OBJ_MTX_SZ - 1))
+#define        ZFS_OBJ_MTX_SZ  64
+#define        ZFS_OBJ_MTX_MAX (1024 * 1024)
+
+extern unsigned int zfs_object_mutex_size;
+
+#define        ZFS_OBJ_HASH(zsb, obj_num)      \
+       ((obj_num) & ((zsb->z_hold_mtx_size) - 1))
 #define        ZFS_OBJ_MUTEX(zsb, obj_num)     \
-       (&(zsb)->z_hold_mtx[ZFS_OBJ_HASH(obj_num)])
+       (&(zsb)->z_hold_mtx[ZFS_OBJ_HASH(zsb, obj_num)])
 #define        ZFS_OBJ_HOLD_ENTER(zsb, obj_num) \
        mutex_enter(ZFS_OBJ_MUTEX((zsb), (obj_num)))
 #define        ZFS_OBJ_HOLD_TRYENTER(zsb, obj_num) \

diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c
index fc63ea569ac80c4a41b36b47e31aea9c8fbf6d4f..5e15faca4bf625b7c860a23a096e799662408a07 100644 (file)
--- a/module/zfs/zfs_vfsops.c
+++ b/module/zfs/zfs_vfsops.c
@@ -795,9 +795,11 @@ zfs_sb_create(const char *osname, zfs_mntopts_t *zmo, zfs_sb_t **zsbp)
        rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
        rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
 
-       zsb->z_hold_mtx = vmem_zalloc(sizeof (kmutex_t) * ZFS_OBJ_MTX_SZ,
+       zsb->z_hold_mtx_size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
+           ZFS_OBJ_MTX_MAX);
+       zsb->z_hold_mtx = vmem_zalloc(sizeof (kmutex_t) * zsb->z_hold_mtx_size,
            KM_SLEEP);
-       for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
+       for (i = 0; i != zsb->z_hold_mtx_size; i++)
                mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
 
        *zsbp = zsb;
@@ -807,7 +809,7 @@ out:
        dmu_objset_disown(os, zsb);
        *zsbp = NULL;
 
-       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * ZFS_OBJ_MTX_SZ);
+       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * zsb->z_hold_mtx_size);
        kmem_free(zsb, sizeof (zfs_sb_t));
        return (error);
 }
@@ -909,9 +911,9 @@ zfs_sb_free(zfs_sb_t *zsb)
        rrm_destroy(&zsb->z_teardown_lock);
        rw_destroy(&zsb->z_teardown_inactive_lock);
        rw_destroy(&zsb->z_fuid_lock);
-       for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
+       for (i = 0; i != zsb->z_hold_mtx_size; i++)
                mutex_destroy(&zsb->z_hold_mtx[i]);
-       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * ZFS_OBJ_MTX_SZ);
+       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * zsb->z_hold_mtx_size);
        zfs_mntopts_free(zsb->z_mntopts);
        kmem_free(zsb, sizeof (zfs_sb_t));
 }

diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index bd5f738a72461a7a1c91100073f240b41eaa729f..94e26fb13450834d47acba6386cd7e3ab420914d 100644 (file)
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -95,6 +95,7 @@
 #ifdef _KERNEL
 
 static kmem_cache_t *znode_cache = NULL;
+unsigned int zfs_object_mutex_size = ZFS_OBJ_MTX_SZ;
 
 /*ARGSUSED*/
 static int
@@ -1734,9 +1735,11 @@ zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
        list_create(&zsb->z_all_znodes, sizeof (znode_t),
            offsetof(znode_t, z_link_node));
 
-       zsb->z_hold_mtx = vmem_zalloc(sizeof (kmutex_t) * ZFS_OBJ_MTX_SZ,
+       zsb->z_hold_mtx_size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
+           ZFS_OBJ_MTX_MAX);
+       zsb->z_hold_mtx = vmem_zalloc(sizeof (kmutex_t) * zsb->z_hold_mtx_size,
            KM_SLEEP);
-       for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
+       for (i = 0; i != zsb->z_hold_mtx_size; i++)
                mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
 
        VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
@@ -1757,10 +1760,10 @@ zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
        error = zfs_create_share_dir(zsb, tx);
        ASSERT(error == 0);
 
-       for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
+       for (i = 0; i != zsb->z_hold_mtx_size; i++)
                mutex_destroy(&zsb->z_hold_mtx[i]);
 
-       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * ZFS_OBJ_MTX_SZ);
+       vmem_free(zsb->z_hold_mtx, sizeof (kmutex_t) * zsb->z_hold_mtx_size);
        kmem_free(sb, sizeof (struct super_block));
        kmem_free(zsb, sizeof (zfs_sb_t));
 }
@@ -2023,4 +2026,7 @@ zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(zfs_create_fs);
 EXPORT_SYMBOL(zfs_obj_to_path);
+
+module_param(zfs_object_mutex_size, uint, 0644);
+MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");
 #endif