tmpfs: dynamically register tmpfs pager

Remove OBJT_SWAP_TMPFS. Move tmpfs-specific swap pager bits into
tmpfs_subr.c.

There is no longer any code to directly support tmpfs in sys/vm, most
tmpfs knowledge is shared by non-anon swap object type implementation.
The tmpfs-specific methods are provided by registered tmpfs pager, which
inherits from the swap pager.

Reviewed by:	markj
Tested by:	pho
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D30168

sys/fs/tmpfs/tmpfs.h

@@ -45,6 +45,8 @@
 MALLOC_DECLARE(M_TMPFSNAME);
 #endif
 
+#define	OBJ_TMPFS	OBJ_PAGERPRIV	/* has tmpfs vnode allocated */
+
 /*
  * Internal representation of a tmpfs directory entry.
  */
@@ -514,9 +516,11 @@ tmpfs_update(struct vnode *vp)
 
 size_t tmpfs_mem_avail(void);
 size_t tmpfs_pages_used(struct tmpfs_mount *tmp);
-void tmpfs_subr_init(void);
+int tmpfs_subr_init(void);
 void tmpfs_subr_uninit(void);
 
+extern int tmpfs_pager_type;
+
 /*
  * Macros/functions to convert from generic data structures to tmpfs
  * specific ones.

sys/fs/tmpfs/tmpfs_subr.c

@@ -54,6 +54,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/smr.h>
 #include <sys/stat.h>
 #include <sys/sysctl.h>
+#include <sys/user.h>
 #include <sys/vnode.h>
 #include <sys/vmmeter.h>
 
@@ -79,6 +80,61 @@ MALLOC_DEFINE(M_TMPFSDIR, "tmpfs dir", "tmpfs dirent structure");
 static uma_zone_t tmpfs_node_pool;
 VFS_SMR_DECLARE;
 
+int tmpfs_pager_type = -1;
+
+static vm_object_t
+tmpfs_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
+    vm_ooffset_t offset, struct ucred *cred)
+{
+	vm_object_t object;
+
+	MPASS(handle == NULL);
+	MPASS(offset == 0);
+	object = vm_object_allocate_dyn(tmpfs_pager_type, size,
+	    OBJ_COLORED | OBJ_SWAP);
+	if (!swap_pager_init_object(object, NULL, NULL, size, 0)) {
+		vm_object_deallocate(object);
+		object = NULL;
+	}
+	return (object);
+}
+
+static void
+tmpfs_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp)
+{
+	struct vnode *vp;
+
+	/*
+	 * Tmpfs VREG node, which was reclaimed, has tmpfs_pager_type
+	 * type, but not OBJ_TMPFS flag.  In this case there is no
+	 * v_writecount to adjust.
+	 */
+	if (vp_heldp != NULL)
+		VM_OBJECT_RLOCK(object);
+	else
+		VM_OBJECT_ASSERT_LOCKED(object);
+	if ((object->flags & OBJ_TMPFS) != 0) {
+		vp = object->un_pager.swp.swp_tmpfs;
+		if (vp != NULL) {
+			*vpp = vp;
+			if (vp_heldp != NULL) {
+				vhold(vp);
+				*vp_heldp = true;
+			}
+		}
+	}
+	if (vp_heldp != NULL)
+		VM_OBJECT_RUNLOCK(object);
+}
+
+struct pagerops tmpfs_pager_ops = {
+	.pgo_kvme_type = KVME_TYPE_VNODE,
+	.pgo_alloc = tmpfs_pager_alloc,
+	.pgo_set_writeable_dirty = vm_object_set_writeable_dirty_,
+	.pgo_mightbedirty = vm_object_mightbedirty_,
+	.pgo_getvp = tmpfs_pager_getvp,
+};
+
 static int
 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
 {
@@ -126,18 +182,26 @@ tmpfs_node_fini(void *mem, int size)
 	mtx_destroy(&node->tn_interlock);
 }
 
-void
+int
 tmpfs_subr_init(void)
 {
+	tmpfs_pager_type = vm_pager_alloc_dyn_type(&tmpfs_pager_ops,
+	    OBJT_SWAP);
+	if (tmpfs_pager_type == -1)
+		return (EINVAL);
 	tmpfs_node_pool = uma_zcreate("TMPFS node",
 	    sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
 	    tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
 	VFS_SMR_ZONE_SET(tmpfs_node_pool);
+	return (0);
 }
 
 void
 tmpfs_subr_uninit(void)
 {
+	if (tmpfs_pager_type != -1)
+		vm_pager_free_dyn_type(tmpfs_pager_type);
+	tmpfs_pager_type = -1;
 	uma_zdestroy(tmpfs_node_pool);
 }
 
@@ -364,7 +428,7 @@ tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
 
 	case VREG:
 		obj = nnode->tn_reg.tn_aobj =
-		    vm_pager_allocate(OBJT_SWAP_TMPFS, NULL, 0,
+		    vm_pager_allocate(tmpfs_pager_type, NULL, 0,
 			VM_PROT_DEFAULT, 0,
 			NULL /* XXXKIB - tmpfs needs swap reservation */);
 		/* OBJ_TMPFS is set together with the setting of vp->v_object */
@@ -1588,7 +1652,7 @@ tmpfs_check_mtime(struct vnode *vp)
 	if (vp->v_type != VREG)
 		return;
 	obj = vp->v_object;
-	KASSERT(obj->type == OBJT_SWAP_TMPFS &&
+	KASSERT(obj->type == tmpfs_pager_type &&
 	    (obj->flags & (OBJ_SWAP | OBJ_TMPFS)) ==
 	    (OBJ_SWAP | OBJ_TMPFS), ("non-tmpfs obj"));
 	/* unlocked read */

sys/fs/tmpfs/tmpfs_vfsops.c

@@ -103,8 +103,8 @@ static const char *tmpfs_updateopts[] = {
  * Handle updates of time from writes to mmaped regions, if allowed.
  * Use MNT_VNODE_FOREACH_ALL instead of MNT_VNODE_FOREACH_LAZY, since
  * unmap of the tmpfs-backed vnode does not call vinactive(), due to
- * vm object type is OBJT_SWAP.  If lazy, only handle delayed update
+ * vm object type is basically OBJT_SWAP.  If lazy, only handle
- * of mtime due to the writes to mapped files.
+ * delayed update of mtime due to the writes to mapped files.
  */
 static void
 tmpfs_update_mtime(struct mount *mp, bool lazy)
@@ -120,7 +120,7 @@ tmpfs_update_mtime(struct mount *mp, bool lazy)
 			continue;
 		}
 		obj = vp->v_object;
-		MPASS(obj->type == OBJT_SWAP_TMPFS);
+		MPASS(obj->type == tmpfs_pager_type);
 		MPASS((obj->flags & OBJ_TMPFS) != 0);
 
 		/*
@@ -225,7 +225,7 @@ tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
 			    (entry->max_protection & VM_PROT_WRITE) == 0)
 				continue;
 			object = entry->object.vm_object;
-			if (object == NULL || object->type != OBJT_SWAP_TMPFS)
+			if (object == NULL || object->type != tmpfs_pager_type)
 				continue;
 			/*
 			 * No need to dig into shadow chain, mapping
@@ -661,7 +661,11 @@ tmpfs_sync(struct mount *mp, int waitfor)
 static int
 tmpfs_init(struct vfsconf *conf)
 {
-	tmpfs_subr_init();
+	int res;
+
+	res = tmpfs_subr_init();
+	if (res != 0)
+		return (res);
 	memcpy(&tmpfs_fnops, &vnops, sizeof(struct fileops));
 	tmpfs_fnops.fo_close = tmpfs_fo_close;
 	return (0);

sys/fs/tmpfs/tmpfs_vnops.c

@@ -623,7 +623,7 @@ tmpfs_read_pgcache(struct vop_read_pgcache_args *v)
 	if (object == NULL)
 		goto out_smr;
 
-	MPASS(object->type == OBJT_SWAP_TMPFS);
+	MPASS(object->type == tmpfs_pager_type);
 	MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) ==
 	    OBJ_SWAP);
 	if (!VN_IS_DOOMED(vp)) {

sys/vm/swap_pager.c

@@ -419,9 +419,6 @@ static uma_zone_t swpctrie_zone;
 static vm_object_t
 		swap_pager_alloc(void *handle, vm_ooffset_t size,
 		    vm_prot_t prot, vm_ooffset_t offset, struct ucred *);
-static vm_object_t
-		swap_tmpfs_pager_alloc(void *handle, vm_ooffset_t size,
-		    vm_prot_t prot, vm_ooffset_t offset, struct ucred *);
 static void	swap_pager_dealloc(vm_object_t object);
 static int	swap_pager_getpages(vm_object_t, vm_page_t *, int, int *,
     int *);
@@ -437,8 +434,6 @@ static void	swap_pager_update_writecount(vm_object_t object,
     vm_offset_t start, vm_offset_t end);
 static void	swap_pager_release_writecount(vm_object_t object,
     vm_offset_t start, vm_offset_t end);
-static void	swap_tmpfs_pager_getvp(vm_object_t object, struct vnode **vpp,
-    bool *vp_heldp);
 static void	swap_pager_freespace(vm_object_t object, vm_pindex_t start,
     vm_size_t size);
 
@@ -457,23 +452,6 @@ const struct pagerops swappagerops = {
 	.pgo_freespace = swap_pager_freespace,
 };
 
-const struct pagerops swaptmpfspagerops = {
-	.pgo_kvme_type = KVME_TYPE_VNODE,
-	.pgo_alloc =	swap_tmpfs_pager_alloc,
-	.pgo_dealloc =	swap_pager_dealloc,
-	.pgo_getpages =	swap_pager_getpages,
-	.pgo_getpages_async = swap_pager_getpages_async,
-	.pgo_putpages =	swap_pager_putpages,
-	.pgo_haspage =	swap_pager_haspage,
-	.pgo_pageunswapped = swap_pager_unswapped,
-	.pgo_update_writecount = swap_pager_update_writecount,
-	.pgo_release_writecount = swap_pager_release_writecount,
-	.pgo_set_writeable_dirty = vm_object_set_writeable_dirty_,
-	.pgo_mightbedirty = vm_object_mightbedirty_,
-	.pgo_getvp = swap_tmpfs_pager_getvp,
-	.pgo_freespace = swap_pager_freespace,
-};
-
 /*
  * swap_*() routines are externally accessible.  swp_*() routines are
  * internal.
@@ -681,18 +659,31 @@ swap_pager_swap_init(void)
 		    "reduce kern.maxswzone.\n");
 }
 
+bool
+swap_pager_init_object(vm_object_t object, void *handle, struct ucred *cred,
+    vm_ooffset_t size, vm_ooffset_t offset)
+{
+	if (cred != NULL) {
+		if (!swap_reserve_by_cred(size, cred))
+			return (false);
+		crhold(cred);
+	}
+
+	object->un_pager.swp.writemappings = 0;
+	object->handle = handle;
+	if (cred != NULL) {
+		object->cred = cred;
+		object->charge = size;
+	}
+	return (true);
+}
+
 static vm_object_t
 swap_pager_alloc_init(objtype_t otype, void *handle, struct ucred *cred,
     vm_ooffset_t size, vm_ooffset_t offset)
 {
 	vm_object_t object;
 
-	if (cred != NULL) {
-		if (!swap_reserve_by_cred(size, cred))
-			return (NULL);
-		crhold(cred);
-	}
-
 	/*
 	 * The un_pager.swp.swp_blks trie is initialized by
 	 * vm_object_allocate() to ensure the correct order of
@@ -701,11 +692,9 @@ swap_pager_alloc_init(objtype_t otype, void *handle, struct ucred *cred,
 	object = vm_object_allocate(otype, OFF_TO_IDX(offset +
 	    PAGE_MASK + size));
 
-	object->un_pager.swp.writemappings = 0;
+	if (!swap_pager_init_object(object, handle, cred, size, offset)) {
-	object->handle = handle;
+		vm_object_deallocate(object);
-	if (cred != NULL) {
+		return (NULL);
-		object->cred = cred;
-		object->charge = size;
 	}
 	return (object);
 }
@@ -752,18 +741,6 @@ swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
 	return (object);
 }
 
-static vm_object_t
-swap_tmpfs_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
-    vm_ooffset_t offset, struct ucred *cred)
-{
-	vm_object_t object;
-
-	MPASS(handle == NULL);
-	object = swap_pager_alloc_init(OBJT_SWAP_TMPFS, handle, cred,
-	    size, offset);
-	return (object);
-}
-
 /*
  * SWAP_PAGER_DEALLOC() -	remove swap metadata from object
  *
@@ -3168,31 +3145,3 @@ swap_pager_release_writecount(vm_object_t object, vm_offset_t start,
 	object->un_pager.swp.writemappings -= (vm_ooffset_t)end - start;
 	VM_OBJECT_WUNLOCK(object);
 }
-
-static void
-swap_tmpfs_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp)
-{
-	struct vnode *vp;
-
-	/*
-	 * Tmpfs VREG node, which was reclaimed, has OBJT_SWAP_TMPFS
-	 * type, but not OBJ_TMPFS flag.  In this case there is no
-	 * v_writecount to adjust.
-	 */
-	if (vp_heldp != NULL)
-		VM_OBJECT_RLOCK(object);
-	else
-		VM_OBJECT_ASSERT_LOCKED(object);
-	if ((object->flags & OBJ_TMPFS) != 0) {
-		vp = object->un_pager.swp.swp_tmpfs;
-		if (vp != NULL) {
-			*vpp = vp;
-			if (vp_heldp != NULL) {
-				vhold(vp);
-				*vp_heldp = true;
-			}
-		}
-	}
-	if (vp_heldp != NULL)
-		VM_OBJECT_RUNLOCK(object);
-}

sys/vm/swap_pager.h

@@ -82,6 +82,7 @@ int swap_pager_reserve(vm_object_t, vm_pindex_t, vm_size_t);
 void swap_pager_status(int *total, int *used);
 u_long swap_pager_swapped_pages(vm_object_t object);
 void swapoff_all(void);
-
+bool swap_pager_init_object(vm_object_t object, void *handle,
+    struct ucred *cred, vm_ooffset_t size, vm_ooffset_t offset);
 #endif				/* _KERNEL */
 #endif				/* _VM_SWAP_PAGER_H_ */

sys/vm/vm.h

@@ -97,7 +97,6 @@ enum obj_type {
 	OBJT_DEAD,
 	OBJT_SG,
 	OBJT_MGTDEVICE,
-	OBJT_SWAP_TMPFS,
 	OBJT_FIRST_DYN,
 };
 typedef u_char objtype_t;

sys/vm/vm_object.c

@@ -413,7 +413,6 @@ vm_object_allocate(objtype_t type, vm_pindex_t size)
 		flags = OBJ_COLORED;
 		break;
 	case OBJT_SWAP:
-	case OBJT_SWAP_TMPFS:
 		flags = OBJ_COLORED | OBJ_SWAP;
 		break;
 	case OBJT_DEVICE:

sys/vm/vm_object.h

@@ -207,7 +207,7 @@ struct vm_object {
 #define	OBJ_COLORED	0x1000		/* pg_color is defined */
 #define	OBJ_ONEMAPPING	0x2000		/* One USE (a single, non-forked) mapping flag */
 #define	OBJ_SHADOWLIST	0x4000		/* Object is on the shadow list. */
-#define	OBJ_TMPFS	0x8000		/* has tmpfs vnode allocated */
+#define	OBJ_PAGERPRIV	0x8000		/* Pager private */
 
 /*
  * Helpers to perform conversion between vm_object page indexes and offsets.

sys/vm/vm_pager.c

@@ -174,7 +174,6 @@ const struct pagerops *pagertab[16] __read_mostly = {
 	[OBJT_DEAD] =		&deadpagerops,
 	[OBJT_SG] = 		&sgpagerops,
 	[OBJT_MGTDEVICE] = 	&mgtdevicepagerops,
-	[OBJT_SWAP_TMPFS] =	&swaptmpfspagerops,
 };
 static struct mtx pagertab_lock;