staging: ramster: ramster-specific changes to zcache/tmem
authorDan Magenheimer <dan.magenheimer@oracle.com>
Wed, 15 Feb 2012 15:54:18 +0000 (07:54 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 15 Feb 2012 17:02:03 +0000 (09:02 -0800)
RAMster implements peer-to-peer transcendent memory, allowing a "cluster"
of kernels to dynamically pool their RAM.

This patch incorporates changes transforming zcache to work with
a remote store.

In tmem.[ch], new "repatriate" (provoke async get) and "localify" (handle
incoming data resulting from an async get) routines combine with a handful
of changes to existing pamops interfaces allow the generic tmem code
to support asynchronous operations.  Also, a new tmem_xhandle struct
groups together key information that must be passed to remote tmem stores.

Zcache-main.c is augmented with a large amount of ramster-specific code
to handle remote operations and "foreign" pages on both ends of the
"remotify" protocol.  New "foreign" pools are auto-created on demand.
A "selfshrinker" thread periodically repatriates remote persistent pages
when local memory conditions allow.  For certain operations, a queue is
necessary to guarantee strict ordering as out-of-order puts/flushes can
cause strange race conditions.  Pampd pointers now either point to local
memory OR describe a remote page; to allow the same 64-bits to describe
either, the LSB is used to differentiate.  Some acrobatics must be performed
to ensure local memory is available to handle a remote persistent get,
or deal with the data directly anyway if the malloc failed.  Lots
of ramster-specific statistics are available via sysfs.

Note: Some debug ifdefs left in for now.
Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/ramster/Kconfig
drivers/staging/ramster/Makefile
drivers/staging/ramster/tmem.c
drivers/staging/ramster/tmem.h
drivers/staging/ramster/zcache-main.c

index 7fabcb2bc80dec25757938693dbedf60d533c53e..acfb974f9ef037f10ea0a92774f8a24069ec7e0e 100644 (file)
@@ -1,13 +1,13 @@
-config ZCACHE
-       tristate "Dynamic compression of swap pages and clean pagecache pages"
-       depends on CLEANCACHE || FRONTSWAP
-       select XVMALLOC
+config RAMSTER
+       bool "Cross-machine RAM capacity sharing, aka peer-to-peer tmem"
+       depends on (CLEANCACHE || FRONTSWAP) && CONFIGFS_FS && !ZCACHE && !XVMALLOC && !HIGHMEM
        select LZO_COMPRESS
        select LZO_DECOMPRESS
        default n
        help
-         Zcache doubles RAM efficiency while providing a significant
-         performance boosts on many workloads.  Zcache uses lzo1x
-         compression and an in-kernel implementation of transcendent
-         memory to store clean page cache pages and swap in RAM,
-         providing a noticeable reduction in disk I/O.
+         RAMster allows RAM on other machines in a cluster to be utilized
+         dynamically and symmetrically instead of swapping to a local swap
+         disk, thus improving performance on memory-constrained workloads
+         while minimizing total RAM across the cluster.  RAMster, like
+         zcache, compresses swap pages into local RAM, but then remotifies
+         the compressed pages to another node in the RAMster cluster.
index 60daa272c204b6191525d1990b4b38937400b94e..bcc13c87f9960f6112676c07f0c96a68079a9c18 100644 (file)
@@ -1,3 +1 @@
-zcache-y       :=      zcache-main.o tmem.o
-
-obj-$(CONFIG_ZCACHE)   +=      zcache.o
+obj-$(CONFIG_RAMSTER)  +=      zcache-main.o tmem.o r2net.o xvmalloc.o cluster/
index 1ca66ea9b28123825a4db9b08d8621da909c8c66..8f2f6892d8d3b11aa37b9052c74b7e45c4f2dee1 100644 (file)
@@ -27,6 +27,7 @@
 #include <linux/list.h>
 #include <linux/spinlock.h>
 #include <linux/atomic.h>
+#include <linux/delay.h>
 
 #include "tmem.h"
 
@@ -316,7 +317,7 @@ static void *tmem_pampd_lookup_in_obj(struct tmem_obj *obj, uint32_t index)
 }
 
 static void *tmem_pampd_replace_in_obj(struct tmem_obj *obj, uint32_t index,
-                                       void *new_pampd)
+                                       void *new_pampd, bool no_free)
 {
        struct tmem_objnode **slot;
        void *ret = NULL;
@@ -325,7 +326,9 @@ static void *tmem_pampd_replace_in_obj(struct tmem_obj *obj, uint32_t index,
        if ((slot != NULL) && (*slot != NULL)) {
                void *old_pampd = *(void **)slot;
                *(void **)slot = new_pampd;
-               (*tmem_pamops.free)(old_pampd, obj->pool, NULL, 0);
+               if (!no_free)
+                       (*tmem_pamops.free)(old_pampd, obj->pool,
+                                               NULL, 0, false);
                ret = new_pampd;
        }
        return ret;
@@ -481,7 +484,7 @@ static void tmem_objnode_node_destroy(struct tmem_obj *obj,
                        if (ht == 1) {
                                obj->pampd_count--;
                                (*tmem_pamops.free)(objnode->slots[i],
-                                               obj->pool, NULL, 0);
+                                               obj->pool, NULL, 0, true);
                                objnode->slots[i] = NULL;
                                continue;
                        }
@@ -498,7 +501,8 @@ static void tmem_pampd_destroy_all_in_obj(struct tmem_obj *obj)
                return;
        if (obj->objnode_tree_height == 0) {
                obj->pampd_count--;
-               (*tmem_pamops.free)(obj->objnode_tree_root, obj->pool, NULL, 0);
+               (*tmem_pamops.free)(obj->objnode_tree_root,
+                                       obj->pool, NULL, 0, true);
        } else {
                tmem_objnode_node_destroy(obj, obj->objnode_tree_root,
                                        obj->objnode_tree_height);
@@ -529,7 +533,7 @@ static void tmem_pampd_destroy_all_in_obj(struct tmem_obj *obj)
  * always flushes for simplicity.
  */
 int tmem_put(struct tmem_pool *pool, struct tmem_oid *oidp, uint32_t index,
-               char *data, size_t size, bool raw, bool ephemeral)
+               char *data, size_t size, bool raw, int ephemeral)
 {
        struct tmem_obj *obj = NULL, *objfound = NULL, *objnew = NULL;
        void *pampd = NULL, *pampd_del = NULL;
@@ -545,7 +549,7 @@ int tmem_put(struct tmem_pool *pool, struct tmem_oid *oidp, uint32_t index,
                        /* if found, is a dup put, flush the old one */
                        pampd_del = tmem_pampd_delete_from_obj(obj, index);
                        BUG_ON(pampd_del != pampd);
-                       (*tmem_pamops.free)(pampd, pool, oidp, index);
+                       (*tmem_pamops.free)(pampd, pool, oidp, index, true);
                        if (obj->pampd_count == 0) {
                                objnew = obj;
                                objfound = NULL;
@@ -576,7 +580,7 @@ delete_and_free:
        (void)tmem_pampd_delete_from_obj(obj, index);
 free:
        if (pampd)
-               (*tmem_pamops.free)(pampd, pool, NULL, 0);
+               (*tmem_pamops.free)(pampd, pool, NULL, 0, true);
        if (objnew) {
                tmem_obj_free(objnew, hb);
                (*tmem_hostops.obj_free)(objnew, pool);
@@ -586,6 +590,65 @@ out:
        return ret;
 }
 
+void *tmem_localify_get_pampd(struct tmem_pool *pool, struct tmem_oid *oidp,
+                               uint32_t index, struct tmem_obj **ret_obj,
+                               void **saved_hb)
+{
+       struct tmem_hashbucket *hb;
+       struct tmem_obj *obj = NULL;
+       void *pampd = NULL;
+
+       hb = &pool->hashbucket[tmem_oid_hash(oidp)];
+       spin_lock(&hb->lock);
+       obj = tmem_obj_find(hb, oidp);
+       if (likely(obj != NULL))
+               pampd = tmem_pampd_lookup_in_obj(obj, index);
+       *ret_obj = obj;
+       *saved_hb = (void *)hb;
+       /* note, hashbucket remains locked */
+       return pampd;
+}
+
+void tmem_localify_finish(struct tmem_obj *obj, uint32_t index,
+                         void *pampd, void *saved_hb, bool delete)
+{
+       struct tmem_hashbucket *hb = (struct tmem_hashbucket *)saved_hb;
+
+       BUG_ON(!spin_is_locked(&hb->lock));
+       if (pampd != NULL) {
+               BUG_ON(obj == NULL);
+               (void)tmem_pampd_replace_in_obj(obj, index, pampd, 1);
+       } else if (delete) {
+               BUG_ON(obj == NULL);
+               (void)tmem_pampd_delete_from_obj(obj, index);
+       }
+       spin_unlock(&hb->lock);
+}
+
+static int tmem_repatriate(void **ppampd, struct tmem_hashbucket *hb,
+                               struct tmem_pool *pool, struct tmem_oid *oidp,
+                               uint32_t index, bool free, char *data)
+{
+       void *old_pampd = *ppampd, *new_pampd = NULL;
+       bool intransit = false;
+       int ret = 0;
+
+
+       if (!is_ephemeral(pool))
+               new_pampd = (*tmem_pamops.repatriate_preload)(
+                               old_pampd, pool, oidp, index, &intransit);
+       if (intransit)
+               ret = -EAGAIN;
+       else if (new_pampd != NULL)
+               *ppampd = new_pampd;
+       /* must release the hb->lock else repatriate can't sleep */
+       spin_unlock(&hb->lock);
+       if (!intransit)
+               ret = (*tmem_pamops.repatriate)(old_pampd, new_pampd, pool,
+                                               oidp, index, free, data);
+       return ret;
+}
+
 /*
  * "Get" a page, e.g. if one can be found, copy the tmem page with the
  * matching handle from PAM space to the kernel.  By tmem definition,
@@ -607,14 +670,36 @@ int tmem_get(struct tmem_pool *pool, struct tmem_oid *oidp, uint32_t index,
        int ret = -1;
        struct tmem_hashbucket *hb;
        bool free = (get_and_free == 1) || ((get_and_free == 0) && ephemeral);
-       bool lock_held = false;
+       bool lock_held = 0;
+       void **ppampd;
 
+again:
        hb = &pool->hashbucket[tmem_oid_hash(oidp)];
        spin_lock(&hb->lock);
-       lock_held = true;
+       lock_held = 1;
        obj = tmem_obj_find(hb, oidp);
        if (obj == NULL)
                goto out;
+       ppampd = __tmem_pampd_lookup_in_obj(obj, index);
+       if (ppampd == NULL)
+               goto out;
+       if (tmem_pamops.is_remote(*ppampd)) {
+               ret = tmem_repatriate(ppampd, hb, pool, oidp,
+                                       index, free, data);
+               lock_held = 0; /* note hb->lock has been unlocked */
+               if (ret == -EAGAIN) {
+                       /* rare I think, but should cond_resched()??? */
+                       usleep_range(10, 1000);
+                       goto again;
+               } else if (ret != 0) {
+                       if (ret != -ENOENT)
+                               pr_err("UNTESTED case in tmem_get, ret=%d\n",
+                                               ret);
+                       ret = -1;
+                       goto out;
+               }
+               goto out;
+       }
        if (free)
                pampd = tmem_pampd_delete_from_obj(obj, index);
        else
@@ -628,10 +713,6 @@ int tmem_get(struct tmem_pool *pool, struct tmem_oid *oidp, uint32_t index,
                        obj = NULL;
                }
        }
-       if (tmem_pamops.is_remote(pampd)) {
-               lock_held = false;
-               spin_unlock(&hb->lock);
-       }
        if (free)
                ret = (*tmem_pamops.get_data_and_free)(
                                data, size, raw, pampd, pool, oidp, index);
@@ -668,7 +749,7 @@ int tmem_flush_page(struct tmem_pool *pool,
        pampd = tmem_pampd_delete_from_obj(obj, index);
        if (pampd == NULL)
                goto out;
-       (*tmem_pamops.free)(pampd, pool, oidp, index);
+       (*tmem_pamops.free)(pampd, pool, oidp, index, true);
        if (obj->pampd_count == 0) {
                tmem_obj_free(obj, hb);
                (*tmem_hostops.obj_free)(obj, pool);
@@ -682,8 +763,8 @@ out:
 
 /*
  * If a page in tmem matches the handle, replace the page so that any
- * subsequent "get" gets the new page.  Returns 0 if
- * there was a page to replace, else returns -1.
+ * subsequent "get" gets the new page.  Returns the new page if
+ * there was a page to replace, else returns NULL.
  */
 int tmem_replace(struct tmem_pool *pool, struct tmem_oid *oidp,
                        uint32_t index, void *new_pampd)
@@ -697,7 +778,7 @@ int tmem_replace(struct tmem_pool *pool, struct tmem_oid *oidp,
        obj = tmem_obj_find(hb, oidp);
        if (obj == NULL)
                goto out;
-       new_pampd = tmem_pampd_replace_in_obj(obj, index, new_pampd);
+       new_pampd = tmem_pampd_replace_in_obj(obj, index, new_pampd, 0);
        ret = (*tmem_pamops.replace_in_obj)(new_pampd, obj);
 out:
        spin_unlock(&hb->lock);
index ed147c4b110d41b8f7febbb9b5efcb0ce3666f0e..47f1918c8314d32740ee653d0f9f7ad8a3760cb4 100644 (file)
@@ -9,7 +9,6 @@
 #ifndef _TMEM_H_
 #define _TMEM_H_
 
-#include <linux/types.h>
 #include <linux/highmem.h>
 #include <linux/hash.h>
 #include <linux/atomic.h>
@@ -89,6 +88,31 @@ struct tmem_oid {
        uint64_t oid[3];
 };
 
+struct tmem_xhandle {
+       uint8_t client_id;
+       uint8_t xh_data_cksum;
+       uint16_t xh_data_size;
+       uint16_t pool_id;
+       struct tmem_oid oid;
+       uint32_t index;
+       void *extra;
+};
+
+static inline struct tmem_xhandle tmem_xhandle_fill(uint16_t client_id,
+                                       struct tmem_pool *pool,
+                                       struct tmem_oid *oidp,
+                                       uint32_t index)
+{
+       struct tmem_xhandle xh;
+       xh.client_id = client_id;
+       xh.xh_data_cksum = (uint8_t)-1;
+       xh.xh_data_size = (uint16_t)-1;
+       xh.pool_id = pool->pool_id;
+       xh.oid = *oidp;
+       xh.index = index;
+       return xh;
+}
+
 static inline void tmem_oid_set_invalid(struct tmem_oid *oidp)
 {
        oidp->oid[0] = oidp->oid[1] = oidp->oid[2] = -1UL;
@@ -147,7 +171,11 @@ struct tmem_obj {
        unsigned int objnode_tree_height;
        unsigned long objnode_count;
        long pampd_count;
-       void *extra; /* for private use by pampd implementation */
+       /* for current design of ramster, all pages belonging to
+        * an object reside on the same remotenode and extra is
+        * used to record the number of the remotenode so a
+        * flush-object operation can specify it */
+       void *extra; /* for use by pampd implementation */
        DECL_SENTINEL
 };
 
@@ -174,9 +202,14 @@ struct tmem_pamops {
        int (*get_data_and_free)(char *, size_t *, bool, void *,
                                struct tmem_pool *, struct tmem_oid *,
                                uint32_t);
-       void (*free)(void *, struct tmem_pool *, struct tmem_oid *, uint32_t);
+       void (*free)(void *, struct tmem_pool *,
+                               struct tmem_oid *, uint32_t, bool);
        void (*free_obj)(struct tmem_pool *, struct tmem_obj *);
        bool (*is_remote)(void *);
+       void *(*repatriate_preload)(void *, struct tmem_pool *,
+                                       struct tmem_oid *, uint32_t, bool *);
+       int (*repatriate)(void *, void *, struct tmem_pool *,
+                               struct tmem_oid *, uint32_t, bool, void *);
        void (*new_obj)(struct tmem_obj *);
        int (*replace_in_obj)(void *, struct tmem_obj *);
 };
@@ -193,11 +226,16 @@ extern void tmem_register_hostops(struct tmem_hostops *m);
 
 /* core tmem accessor functions */
 extern int tmem_put(struct tmem_pool *, struct tmem_oid *, uint32_t index,
-                       char *, size_t, bool, bool);
+                       char *, size_t, bool, int);
 extern int tmem_get(struct tmem_pool *, struct tmem_oid *, uint32_t index,
                        char *, size_t *, bool, int);
 extern int tmem_replace(struct tmem_pool *, struct tmem_oid *, uint32_t index,
                        void *);
+extern void *tmem_localify_get_pampd(struct tmem_pool *, struct tmem_oid *,
+                                  uint32_t index, struct tmem_obj **,
+                                  void **);
+extern void tmem_localify_finish(struct tmem_obj *, uint32_t index,
+                                void *, void *, bool);
 extern int tmem_flush_page(struct tmem_pool *, struct tmem_oid *,
                        uint32_t index);
 extern int tmem_flush_object(struct tmem_pool *, struct tmem_oid *);
index 49c8791462f46e4a33f923d2b0219c26bf9ac18a..36d53ed9d71a8a6099ecd4889b918dacda6ebf48 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * zcache.c
  *
- * Copyright (c) 2010,2011, Dan Magenheimer, Oracle Corp.
+ * Copyright (c) 2010-2012, Dan Magenheimer, Oracle Corp.
  * Copyright (c) 2010,2011, Nitin Gupta
  *
  * Zcache provides an in-kernel "host implementation" for transcendent memory
@@ -17,6 +17,9 @@
  *
  * [1] For a definition of page-accessible memory (aka PAM), see:
  *   http://marc.info/?l=linux-mm&m=127811271605009
+ *  RAMSTER TODO:
+ *   - handle remotifying of buddied pages (see zbud_remotify_zbpg)
+ *   - kernel boot params: nocleancache/nofrontswap don't always work?!?
  */
 
 #include <linux/module.h>
 #include <linux/atomic.h>
 #include <linux/math64.h>
 #include "tmem.h"
+#include "zcache.h"
+#include "ramster.h"
+#include "cluster/tcp.h"
 
-#include "../zram/xvmalloc.h" /* if built in drivers/staging */
+#include "xvmalloc.h"  /* temporary until change to zsmalloc */
+
+#define        RAMSTER_TESTING
 
 #if (!defined(CONFIG_CLEANCACHE) && !defined(CONFIG_FRONTSWAP))
-#error "zcache is useless without CONFIG_CLEANCACHE or CONFIG_FRONTSWAP"
+#error "ramster is useless without CONFIG_CLEANCACHE or CONFIG_FRONTSWAP"
 #endif
 #ifdef CONFIG_CLEANCACHE
 #include <linux/cleancache.h>
 #include <linux/frontswap.h>
 #endif
 
+enum ramster_remotify_op {
+       RAMSTER_REMOTIFY_EPH_PUT,
+       RAMSTER_REMOTIFY_PERS_PUT,
+       RAMSTER_REMOTIFY_FLUSH_PAGE,
+       RAMSTER_REMOTIFY_FLUSH_OBJ,
+       RAMSTER_INTRANSIT_PERS
+};
+
+struct ramster_remotify_hdr {
+       enum ramster_remotify_op op;
+       struct list_head list;
+};
+
+#define ZBH_SENTINEL  0x43214321
+#define ZBPG_SENTINEL  0xdeadbeef
+
+#define ZBUD_MAX_BUDS 2
+
+struct zbud_hdr {
+       struct ramster_remotify_hdr rem_op;
+       uint16_t client_id;
+       uint16_t pool_id;
+       struct tmem_oid oid;
+       uint32_t index;
+       uint16_t size; /* compressed size in bytes, zero means unused */
+       DECL_SENTINEL
+};
+
+#define ZVH_SENTINEL  0x43214321
+static const int zv_max_page_size = (PAGE_SIZE / 8) * 7;
+
+struct zv_hdr {
+       struct ramster_remotify_hdr rem_op;
+       uint16_t client_id;
+       uint16_t pool_id;
+       struct tmem_oid oid;
+       uint32_t index;
+       DECL_SENTINEL
+};
+
+struct flushlist_node {
+       struct ramster_remotify_hdr rem_op;
+       struct tmem_xhandle xh;
+};
+
+union {
+       struct ramster_remotify_hdr rem_op;
+       struct zv_hdr zv;
+       struct zbud_hdr zbud;
+       struct flushlist_node flist;
+} remotify_list_node;
+
+static LIST_HEAD(zcache_rem_op_list);
+static DEFINE_SPINLOCK(zcache_rem_op_list_lock);
+
 #if 0
 /* this is more aggressive but may cause other problems? */
 #define ZCACHE_GFP_MASK        (GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN)
@@ -98,20 +161,6 @@ static inline bool is_local_client(struct zcache_client *cli)
  * read or written unless the zbpg's lock is held.
  */
 
-#define ZBH_SENTINEL  0x43214321
-#define ZBPG_SENTINEL  0xdeadbeef
-
-#define ZBUD_MAX_BUDS 2
-
-struct zbud_hdr {
-       uint16_t client_id;
-       uint16_t pool_id;
-       struct tmem_oid oid;
-       uint32_t index;
-       uint16_t size; /* compressed size in bytes, zero means unused */
-       DECL_SENTINEL
-};
-
 struct zbud_page {
        struct list_head bud_list;
        spinlock_t lock;
@@ -141,20 +190,43 @@ static unsigned long zcache_zbud_buddied_count;
 /* protects the buddied list and all unbuddied lists */
 static DEFINE_SPINLOCK(zbud_budlists_spinlock);
 
-static LIST_HEAD(zbpg_unused_list);
-static unsigned long zcache_zbpg_unused_list_count;
-
-/* protects the unused page list */
-static DEFINE_SPINLOCK(zbpg_unused_list_spinlock);
-
 static atomic_t zcache_zbud_curr_raw_pages;
 static atomic_t zcache_zbud_curr_zpages;
 static unsigned long zcache_zbud_curr_zbytes;
 static unsigned long zcache_zbud_cumul_zpages;
 static unsigned long zcache_zbud_cumul_zbytes;
 static unsigned long zcache_compress_poor;
+static unsigned long zcache_policy_percent_exceeded;
 static unsigned long zcache_mean_compress_poor;
 
+/*
+ * RAMster counters
+ * - Remote pages are pages with a local pampd but the data is remote
+ * - Foreign pages are pages stored locally but belonging to another node
+ */
+static atomic_t ramster_remote_pers_pages = ATOMIC_INIT(0);
+static unsigned long ramster_pers_remotify_enable;
+static unsigned long ramster_eph_remotify_enable;
+static unsigned long ramster_eph_pages_remoted;
+static unsigned long ramster_eph_pages_remote_failed;
+static unsigned long ramster_pers_pages_remoted;
+static unsigned long ramster_pers_pages_remote_failed;
+static unsigned long ramster_pers_pages_remote_nomem;
+static unsigned long ramster_remote_objects_flushed;
+static unsigned long ramster_remote_object_flushes_failed;
+static unsigned long ramster_remote_pages_flushed;
+static unsigned long ramster_remote_page_flushes_failed;
+static unsigned long ramster_remote_eph_pages_succ_get;
+static unsigned long ramster_remote_pers_pages_succ_get;
+static unsigned long ramster_remote_eph_pages_unsucc_get;
+static unsigned long ramster_remote_pers_pages_unsucc_get;
+static atomic_t ramster_curr_flnode_count = ATOMIC_INIT(0);
+static unsigned long ramster_curr_flnode_count_max;
+static atomic_t ramster_foreign_eph_pampd_count = ATOMIC_INIT(0);
+static unsigned long ramster_foreign_eph_pampd_count_max;
+static atomic_t ramster_foreign_pers_pampd_count = ATOMIC_INIT(0);
+static unsigned long ramster_foreign_pers_pampd_count_max;
+
 /* forward references */
 static void *zcache_get_free_page(void);
 static void zcache_free_page(void *p);
@@ -210,6 +282,29 @@ static char *zbud_data(struct zbud_hdr *zh, unsigned size)
        return p;
 }
 
+static void zbud_copy_from_pampd(char *data, size_t *size, struct zbud_hdr *zh)
+{
+       struct zbud_page *zbpg;
+       char *p;
+       unsigned budnum;
+
+       ASSERT_SENTINEL(zh, ZBH);
+       budnum = zbud_budnum(zh);
+       zbpg = container_of(zh, struct zbud_page, buddy[budnum]);
+       spin_lock(&zbpg->lock);
+       BUG_ON(zh->size > *size);
+       p = (char *)zbpg;
+       if (budnum == 0)
+               p += ((sizeof(struct zbud_page) + CHUNK_SIZE - 1) &
+                                                       CHUNK_MASK);
+       else if (budnum == 1)
+               p += PAGE_SIZE - ((zh->size + CHUNK_SIZE - 1) & CHUNK_MASK);
+       /* client should be filled in by caller */
+       memcpy(data, p, zh->size);
+       *size = zh->size;
+       spin_unlock(&zbpg->lock);
+}
+
 /*
  * zbud raw page management
  */
@@ -218,38 +313,17 @@ static struct zbud_page *zbud_alloc_raw_page(void)
 {
        struct zbud_page *zbpg = NULL;
        struct zbud_hdr *zh0, *zh1;
-       bool recycled = 0;
-
-       /* if any pages on the zbpg list, use one */
-       spin_lock(&zbpg_unused_list_spinlock);
-       if (!list_empty(&zbpg_unused_list)) {
-               zbpg = list_first_entry(&zbpg_unused_list,
-                               struct zbud_page, bud_list);
-               list_del_init(&zbpg->bud_list);
-               zcache_zbpg_unused_list_count--;
-               recycled = 1;
-       }
-       spin_unlock(&zbpg_unused_list_spinlock);
-       if (zbpg == NULL)
-               /* none on zbpg list, try to get a kernel page */
                zbpg = zcache_get_free_page();
        if (likely(zbpg != NULL)) {
                INIT_LIST_HEAD(&zbpg->bud_list);
                zh0 = &zbpg->buddy[0]; zh1 = &zbpg->buddy[1];
                spin_lock_init(&zbpg->lock);
-               if (recycled) {
-                       ASSERT_INVERTED_SENTINEL(zbpg, ZBPG);
-                       SET_SENTINEL(zbpg, ZBPG);
-                       BUG_ON(zh0->size != 0 || tmem_oid_valid(&zh0->oid));
-                       BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid));
-               } else {
-                       atomic_inc(&zcache_zbud_curr_raw_pages);
-                       INIT_LIST_HEAD(&zbpg->bud_list);
-                       SET_SENTINEL(zbpg, ZBPG);
-                       zh0->size = 0; zh1->size = 0;
-                       tmem_oid_set_invalid(&zh0->oid);
-                       tmem_oid_set_invalid(&zh1->oid);
-               }
+               atomic_inc(&zcache_zbud_curr_raw_pages);
+               INIT_LIST_HEAD(&zbpg->bud_list);
+               SET_SENTINEL(zbpg, ZBPG);
+               zh0->size = 0; zh1->size = 0;
+               tmem_oid_set_invalid(&zh0->oid);
+               tmem_oid_set_invalid(&zh1->oid);
        }
        return zbpg;
 }
@@ -265,10 +339,8 @@ static void zbud_free_raw_page(struct zbud_page *zbpg)
        BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid));
        INVERT_SENTINEL(zbpg, ZBPG);
        spin_unlock(&zbpg->lock);
-       spin_lock(&zbpg_unused_list_spinlock);
-       list_add(&zbpg->bud_list, &zbpg_unused_list);
-       zcache_zbpg_unused_list_count++;
-       spin_unlock(&zbpg_unused_list_spinlock);
+       atomic_dec(&zcache_zbud_curr_raw_pages);
+       zcache_free_page(zbpg);
 }
 
 /*
@@ -299,6 +371,10 @@ static void zbud_free_and_delist(struct zbud_hdr *zh)
        struct zbud_page *zbpg =
                container_of(zh, struct zbud_page, buddy[budnum]);
 
+       /* FIXME, should be BUG_ON, pool destruction path doesn't disable
+        * interrupts tmem_destroy_pool()->tmem_pampd_destroy_all_in_obj()->
+        * tmem_objnode_node_destroy()-> zcache_pampd_free() */
+       WARN_ON(!irqs_disabled());
        spin_lock(&zbpg->lock);
        if (list_empty(&zbpg->bud_list)) {
                /* ignore zombie page... see zbud_evict_pages() */
@@ -358,8 +434,8 @@ static struct zbud_hdr *zbud_create(uint16_t client_id, uint16_t pool_id,
        if (unlikely(zbpg == NULL))
                goto out;
        /* ok, have a page, now compress the data before taking locks */
-       spin_lock(&zbpg->lock);
        spin_lock(&zbud_budlists_spinlock);
+       spin_lock(&zbpg->lock);
        list_add_tail(&zbpg->bud_list, &zbud_unbuddied[nchunks].list);
        zbud_unbuddied[nchunks].count++;
        zh = &zbpg->buddy[0];
@@ -389,12 +465,10 @@ init_zh:
        zh->oid = *oid;
        zh->pool_id = pool_id;
        zh->client_id = client_id;
-       /* can wait to copy the data until the list locks are dropped */
-       spin_unlock(&zbud_budlists_spinlock);
-
        to = zbud_data(zh, size);
        memcpy(to, cdata, size);
        spin_unlock(&zbpg->lock);
+       spin_unlock(&zbud_budlists_spinlock);
        zbud_cumul_chunk_counts[nchunks]++;
        atomic_inc(&zcache_zbud_curr_zpages);
        zcache_zbud_cumul_zpages++;
@@ -458,9 +532,9 @@ static void zbud_evict_zbpg(struct zbud_page *zbpg)
        uint32_t index[ZBUD_MAX_BUDS];
        struct tmem_oid oid[ZBUD_MAX_BUDS];
        struct tmem_pool *pool;
+       unsigned long flags;
 
        ASSERT_SPINLOCK(&zbpg->lock);
-       BUG_ON(!list_empty(&zbpg->bud_list));
        for (i = 0, j = 0; i < ZBUD_MAX_BUDS; i++) {
                zh = &zbpg->buddy[i];
                if (zh->size) {
@@ -469,20 +543,18 @@ static void zbud_evict_zbpg(struct zbud_page *zbpg)
                        oid[j] = zh->oid;
                        index[j] = zh->index;
                        j++;
-                       zbud_free(zh);
                }
        }
        spin_unlock(&zbpg->lock);
        for (i = 0; i < j; i++) {
                pool = zcache_get_pool_by_id(client_id[i], pool_id[i]);
-               if (pool != NULL) {
-                       tmem_flush_page(pool, &oid[i], index[i]);
-                       zcache_put_pool(pool);
-               }
+               BUG_ON(pool == NULL);
+               local_irq_save(flags);
+               /* these flushes should dispose of any local storage */
+               tmem_flush_page(pool, &oid[i], index[i]);
+               local_irq_restore(flags);
+               zcache_put_pool(pool);
        }
-       ASSERT_SENTINEL(zbpg, ZBPG);
-       spin_lock(&zbpg->lock);
-       zbud_free_raw_page(zbpg);
 }
 
 /*
@@ -496,26 +568,8 @@ static void zbud_evict_zbpg(struct zbud_page *zbpg)
 static void zbud_evict_pages(int nr)
 {
        struct zbud_page *zbpg;
-       int i;
+       int i, newly_unused_pages = 0;
 
-       /* first try freeing any pages on unused list */
-retry_unused_list:
-       spin_lock_bh(&zbpg_unused_list_spinlock);
-       if (!list_empty(&zbpg_unused_list)) {
-               /* can't walk list here, since it may change when unlocked */
-               zbpg = list_first_entry(&zbpg_unused_list,
-                               struct zbud_page, bud_list);
-               list_del_init(&zbpg->bud_list);
-               zcache_zbpg_unused_list_count--;
-               atomic_dec(&zcache_zbud_curr_raw_pages);
-               spin_unlock_bh(&zbpg_unused_list_spinlock);
-               zcache_free_page(zbpg);
-               zcache_evicted_raw_pages++;
-               if (--nr <= 0)
-                       goto out;
-               goto retry_unused_list;
-       }
-       spin_unlock_bh(&zbpg_unused_list_spinlock);
 
        /* now try freeing unbuddied pages, starting with least space avail */
        for (i = 0; i < MAX_CHUNK; i++) {
@@ -528,15 +582,15 @@ retry_unbud_list_i:
                list_for_each_entry(zbpg, &zbud_unbuddied[i].list, bud_list) {
                        if (unlikely(!spin_trylock(&zbpg->lock)))
                                continue;
-                       list_del_init(&zbpg->bud_list);
                        zbud_unbuddied[i].count--;
                        spin_unlock(&zbud_budlists_spinlock);
                        zcache_evicted_unbuddied_pages++;
                        /* want budlists unlocked when doing zbpg eviction */
                        zbud_evict_zbpg(zbpg);
+                       newly_unused_pages++;
                        local_bh_enable();
                        if (--nr <= 0)
-                               goto out;
+                               goto evict_unused;
                        goto retry_unbud_list_i;
                }
                spin_unlock_bh(&zbud_budlists_spinlock);
@@ -547,27 +601,413 @@ retry_bud_list:
        spin_lock_bh(&zbud_budlists_spinlock);
        if (list_empty(&zbud_buddied_list)) {
                spin_unlock_bh(&zbud_budlists_spinlock);
-               goto out;
+               goto evict_unused;
        }
        list_for_each_entry(zbpg, &zbud_buddied_list, bud_list) {
                if (unlikely(!spin_trylock(&zbpg->lock)))
                        continue;
-               list_del_init(&zbpg->bud_list);
                zcache_zbud_buddied_count--;
                spin_unlock(&zbud_budlists_spinlock);
                zcache_evicted_buddied_pages++;
                /* want budlists unlocked when doing zbpg eviction */
                zbud_evict_zbpg(zbpg);
+               newly_unused_pages++;
                local_bh_enable();
                if (--nr <= 0)
-                       goto out;
+                       goto evict_unused;
                goto retry_bud_list;
        }
        spin_unlock_bh(&zbud_budlists_spinlock);
+
+evict_unused:
+       return;
+}
+
+static DEFINE_PER_CPU(unsigned char *, zcache_remoteputmem);
+
+static int zbud_remotify_zbud(struct tmem_xhandle *xh, char *data,
+                               size_t size)
+{
+       struct tmem_pool *pool;
+       int i, remotenode, ret = -1;
+       unsigned char cksum, *p;
+       unsigned long flags;
+
+       for (p = data, cksum = 0, i = 0; i < size; i++)
+               cksum += *p;
+       ret = ramster_remote_put(xh, data, size, true, &remotenode);
+       if (ret == 0) {
+               /* data was successfully remoted so change the local version
+                * to point to the remote node where it landed */
+               pool = zcache_get_pool_by_id(LOCAL_CLIENT, xh->pool_id);
+               BUG_ON(pool == NULL);
+               local_irq_save(flags);
+               /* tmem_replace will also free up any local space */
+               (void)tmem_replace(pool, &xh->oid, xh->index,
+                       pampd_make_remote(remotenode, size, cksum));
+               local_irq_restore(flags);
+               zcache_put_pool(pool);
+               ramster_eph_pages_remoted++;
+               ret = 0;
+       } else
+               ramster_eph_pages_remote_failed++;
+       return ret;
+}
+
+static int zbud_remotify_zbpg(struct zbud_page *zbpg)
+{
+       struct zbud_hdr *zh1, *zh2 = NULL;
+       struct tmem_xhandle xh1, xh2 = { 0 };
+       char *data1 = NULL, *data2 = NULL;
+       size_t size1 = 0, size2 = 0;
+       int ret = 0;
+       unsigned char *tmpmem = __get_cpu_var(zcache_remoteputmem);
+
+       ASSERT_SPINLOCK(&zbpg->lock);
+       if (zbpg->buddy[0].size == 0)
+               zh1 = &zbpg->buddy[1];
+       else if (zbpg->buddy[1].size == 0)
+               zh1 = &zbpg->buddy[0];
+       else {
+               zh1 = &zbpg->buddy[0];
+               zh2 = &zbpg->buddy[1];
+       }
+       /* don't remotify pages that are already remotified */
+       if (zh1->client_id != LOCAL_CLIENT)
+               zh1 = NULL;
+       if ((zh2 != NULL) && (zh2->client_id != LOCAL_CLIENT))
+               zh2 = NULL;
+
+       /* copy the data and metadata so can release lock */
+       if (zh1 != NULL) {
+               xh1.client_id = zh1->client_id;
+               xh1.pool_id = zh1->pool_id;
+               xh1.oid = zh1->oid;
+               xh1.index = zh1->index;
+               size1 = zh1->size;
+               data1 = zbud_data(zh1, size1);
+               memcpy(tmpmem, zbud_data(zh1, size1), size1);
+               data1 = tmpmem;
+               tmpmem += size1;
+       }
+       if (zh2 != NULL) {
+               xh2.client_id = zh2->client_id;
+               xh2.pool_id = zh2->pool_id;
+               xh2.oid = zh2->oid;
+               xh2.index = zh2->index;
+               size2 = zh2->size;
+               memcpy(tmpmem, zbud_data(zh2, size2), size2);
+               data2 = tmpmem;
+       }
+       spin_unlock(&zbpg->lock);
+       preempt_enable();
+
+       /* OK, no locks held anymore, remotify one or both zbuds */
+       if (zh1 != NULL)
+               ret = zbud_remotify_zbud(&xh1, data1, size1);
+       if (zh2 != NULL)
+               ret |= zbud_remotify_zbud(&xh2, data2, size2);
+       return ret;
+}
+
+void zbud_remotify_pages(int nr)
+{
+       struct zbud_page *zbpg;
+       int i, ret;
+
+       /*
+        * for now just try remotifying unbuddied pages, starting with
+        * least space avail
+        */
+       for (i = 0; i < MAX_CHUNK; i++) {
+retry_unbud_list_i:
+               preempt_disable();  /* enable in zbud_remotify_zbpg */
+               spin_lock_bh(&zbud_budlists_spinlock);
+               if (list_empty(&zbud_unbuddied[i].list)) {
+                       spin_unlock_bh(&zbud_budlists_spinlock);
+                       preempt_enable();
+                       continue; /* next i in for loop */
+               }
+               list_for_each_entry(zbpg, &zbud_unbuddied[i].list, bud_list) {
+                       if (unlikely(!spin_trylock(&zbpg->lock)))
+                               continue; /* next list_for_each_entry */
+                       zbud_unbuddied[i].count--;
+                       /* want budlists unlocked when doing zbpg remotify */
+                       spin_unlock_bh(&zbud_budlists_spinlock);
+                       ret = zbud_remotify_zbpg(zbpg);
+                       /* preemption is re-enabled in zbud_remotify_zbpg */
+                       if (ret == 0) {
+                               if (--nr <= 0)
+                                       goto out;
+                               goto retry_unbud_list_i;
+                       }
+                       /* if fail to remotify any page, quit */
+                       pr_err("TESTING zbud_remotify_pages failed on page,"
+                               " trying to re-add\n");
+                       spin_lock_bh(&zbud_budlists_spinlock);
+                       spin_lock(&zbpg->lock);
+                       list_add_tail(&zbpg->bud_list, &zbud_unbuddied[i].list);
+                       zbud_unbuddied[i].count++;
+                       spin_unlock(&zbpg->lock);
+                       spin_unlock_bh(&zbud_budlists_spinlock);
+                       pr_err("TESTING zbud_remotify_pages failed on page,"
+                               " finished re-add\n");
+                       goto out;
+               }
+               spin_unlock_bh(&zbud_budlists_spinlock);
+               preempt_enable();
+       }
+
+next_buddied_zbpg:
+       preempt_disable();  /* enable in zbud_remotify_zbpg */
+       spin_lock_bh(&zbud_budlists_spinlock);
+       if (list_empty(&zbud_buddied_list))
+               goto unlock_out;
+       list_for_each_entry(zbpg, &zbud_buddied_list, bud_list) {
+               if (unlikely(!spin_trylock(&zbpg->lock)))
+                       continue; /* next list_for_each_entry */
+               zcache_zbud_buddied_count--;
+               /* want budlists unlocked when doing zbpg remotify */
+               spin_unlock_bh(&zbud_budlists_spinlock);
+               ret = zbud_remotify_zbpg(zbpg);
+               /* preemption is re-enabled in zbud_remotify_zbpg */
+               if (ret == 0) {
+                       if (--nr <= 0)
+                               goto out;
+                       goto next_buddied_zbpg;
+               }
+               /* if fail to remotify any page, quit */
+               pr_err("TESTING zbud_remotify_pages failed on BUDDIED page,"
+                       " trying to re-add\n");
+               spin_lock_bh(&zbud_budlists_spinlock);
+               spin_lock(&zbpg->lock);
+               list_add_tail(&zbpg->bud_list, &zbud_buddied_list);
+               zcache_zbud_buddied_count++;
+               spin_unlock(&zbpg->lock);
+               spin_unlock_bh(&zbud_budlists_spinlock);
+               pr_err("TESTING zbud_remotify_pages failed on BUDDIED page,"
+                       " finished re-add\n");
+               goto out;
+       }
+unlock_out:
+       spin_unlock_bh(&zbud_budlists_spinlock);
+       preempt_enable();
 out:
        return;
 }
 
+/* the "flush list" asynchronously collects pages to remotely flush */
+#define FLUSH_ENTIRE_OBJECT ((uint32_t)-1)
+static void ramster_flnode_free(struct flushlist_node *,
+                               struct tmem_pool *);
+
+static void zcache_remote_flush_page(struct flushlist_node *flnode)
+{
+       struct tmem_xhandle *xh;
+       int remotenode, ret;
+
+       preempt_disable();
+       xh = &flnode->xh;
+       remotenode = flnode->xh.client_id;
+       ret = ramster_remote_flush(xh, remotenode);
+       if (ret >= 0)
+               ramster_remote_pages_flushed++;
+       else
+               ramster_remote_page_flushes_failed++;
+       preempt_enable_no_resched();
+       ramster_flnode_free(flnode, NULL);
+}
+
+static void zcache_remote_flush_object(struct flushlist_node *flnode)
+{
+       struct tmem_xhandle *xh;
+       int remotenode, ret;
+
+       preempt_disable();
+       xh = &flnode->xh;
+       remotenode = flnode->xh.client_id;
+       ret = ramster_remote_flush_object(xh, remotenode);
+       if (ret >= 0)
+               ramster_remote_objects_flushed++;
+       else
+               ramster_remote_object_flushes_failed++;
+       preempt_enable_no_resched();
+       ramster_flnode_free(flnode, NULL);
+}
+
+static void zcache_remote_eph_put(struct zbud_hdr *zbud)
+{
+       /* FIXME */
+}
+
+static void zcache_remote_pers_put(struct zv_hdr *zv)
+{
+       struct tmem_xhandle xh;
+       uint16_t size;
+       bool ephemeral;
+       int remotenode, ret = -1;
+       char *data;
+       struct tmem_pool *pool;
+       unsigned long flags;
+       unsigned char cksum;
+       char *p;
+       int i;
+       unsigned char *tmpmem = __get_cpu_var(zcache_remoteputmem);
+
+       ASSERT_SENTINEL(zv, ZVH);
+       BUG_ON(zv->client_id != LOCAL_CLIENT);
+       local_bh_disable();
+       xh.client_id = zv->client_id;
+       xh.pool_id = zv->pool_id;
+       xh.oid = zv->oid;
+       xh.index = zv->index;
+       size = xv_get_object_size(zv) - sizeof(*zv);
+       BUG_ON(size == 0 || size > zv_max_page_size);
+       data = (char *)zv + sizeof(*zv);
+       for (p = data, cksum = 0, i = 0; i < size; i++)
+               cksum += *p;
+       memcpy(tmpmem, data, size);
+       data = tmpmem;
+       pool = zcache_get_pool_by_id(zv->client_id, zv->pool_id);
+       ephemeral = is_ephemeral(pool);
+       zcache_put_pool(pool);
+       /* now OK to release lock set in caller */
+       spin_unlock(&zcache_rem_op_list_lock);
+       local_bh_enable();
+       preempt_disable();
+       ret = ramster_remote_put(&xh, data, size, ephemeral, &remotenode);
+       preempt_enable_no_resched();
+       if (ret != 0) {
+               /*
+                * This is some form of a memory leak... if the remote put
+                * fails, there will never be another attempt to remotify
+                * this page.  But since we've dropped the zv pointer,
+                * the page may have been freed or the data replaced
+                * so we can't just "put it back" in the remote op list.
+                * Even if we could, not sure where to put it in the list
+                * because there may be flushes that must be strictly
+                * ordered vs the put.  So leave this as a FIXME for now.
+                * But count them so we know if it becomes a problem.
+                */
+               ramster_pers_pages_remote_failed++;
+               goto out;
+       } else
+               atomic_inc(&ramster_remote_pers_pages);
+       ramster_pers_pages_remoted++;
+       /*
+        * data was successfully remoted so change the local version to
+        * point to the remote node where it landed
+        */
+       local_bh_disable();
+       pool = zcache_get_pool_by_id(LOCAL_CLIENT, xh.pool_id);
+       local_irq_save(flags);
+       (void)tmem_replace(pool, &xh.oid, xh.index,
+                       pampd_make_remote(remotenode, size, cksum));
+       local_irq_restore(flags);
+       zcache_put_pool(pool);
+       local_bh_enable();
+out:
+       return;
+}
+
+static void zcache_do_remotify_ops(int nr)
+{
+       struct ramster_remotify_hdr *rem_op;
+       union remotify_list_node *u;
+
+       while (1) {
+               if (!nr)
+                       goto out;
+               spin_lock(&zcache_rem_op_list_lock);
+               if (list_empty(&zcache_rem_op_list)) {
+                       spin_unlock(&zcache_rem_op_list_lock);
+                       goto out;
+               }
+               rem_op = list_first_entry(&zcache_rem_op_list,
+                               struct ramster_remotify_hdr, list);
+               list_del_init(&rem_op->list);
+               if (rem_op->op != RAMSTER_REMOTIFY_PERS_PUT)
+                       spin_unlock(&zcache_rem_op_list_lock);
+               u = (union remotify_list_node *)rem_op;
+               switch (rem_op->op) {
+               case RAMSTER_REMOTIFY_EPH_PUT:
+BUG();
+                       zcache_remote_eph_put((struct zbud_hdr *)rem_op);
+                       break;
+               case RAMSTER_REMOTIFY_PERS_PUT:
+                       zcache_remote_pers_put((struct zv_hdr *)rem_op);
+                       break;
+               case RAMSTER_REMOTIFY_FLUSH_PAGE:
+                       zcache_remote_flush_page((struct flushlist_node *)u);
+                       break;
+               case RAMSTER_REMOTIFY_FLUSH_OBJ:
+                       zcache_remote_flush_object((struct flushlist_node *)u);
+                       break;
+               default:
+                       BUG();
+               }
+       }
+out:
+       return;
+}
+
+/*
+ * Communicate interface revision with userspace
+ */
+#include "cluster/ramster_nodemanager.h"
+static unsigned long ramster_interface_revision  = R2NM_API_VERSION;
+
+/*
+ * For now, just push over a few pages every few seconds to
+ * ensure that it basically works
+ */
+static struct workqueue_struct *ramster_remotify_workqueue;
+static void ramster_remotify_process(struct work_struct *work);
+static DECLARE_DELAYED_WORK(ramster_remotify_worker,
+               ramster_remotify_process);
+
+static void ramster_remotify_queue_delayed_work(unsigned long delay)
+{
+       if (!queue_delayed_work(ramster_remotify_workqueue,
+                               &ramster_remotify_worker, delay))
+               pr_err("ramster_remotify: bad workqueue\n");
+}
+
+
+static int use_frontswap;
+static int use_cleancache;
+static int ramster_remote_target_nodenum = -1;
+static void ramster_remotify_process(struct work_struct *work)
+{
+       static bool remotify_in_progress;
+
+       BUG_ON(irqs_disabled());
+       if (remotify_in_progress)
+               ramster_remotify_queue_delayed_work(HZ);
+       else if (ramster_remote_target_nodenum != -1) {
+               remotify_in_progress = true;
+#ifdef CONFIG_CLEANCACHE
+       if (use_cleancache && ramster_eph_remotify_enable)
+               zbud_remotify_pages(5000); /* FIXME is this a good number? */
+#endif
+#ifdef CONFIG_FRONTSWAP
+       if (use_frontswap && ramster_pers_remotify_enable)
+               zcache_do_remotify_ops(500); /* FIXME is this a good number? */
+#endif
+               remotify_in_progress = false;
+               ramster_remotify_queue_delayed_work(HZ);
+       }
+}
+
+static void ramster_remotify_init(void)
+{
+       unsigned long n = 60UL;
+       ramster_remotify_workqueue =
+               create_singlethread_workqueue("ramster_remotify");
+       ramster_remotify_queue_delayed_work(n * HZ);
+}
+
+
 static void zbud_init(void)
 {
        int i;
@@ -631,15 +1071,6 @@ static int zbud_show_cumul_chunk_counts(char *buf)
  * necessary for decompression) immediately preceding the compressed data.
  */
 
-#define ZVH_SENTINEL  0x43214321
-
-struct zv_hdr {
-       uint32_t pool_id;
-       struct tmem_oid oid;
-       uint32_t index;
-       DECL_SENTINEL
-};
-
 /* rudimentary policy limits */
 /* total number of persistent pages may not exceed this percentage */
 static unsigned int zv_page_count_policy_percent = 75;
@@ -655,10 +1086,11 @@ static unsigned int zv_max_zsize = (PAGE_SIZE / 8) * 7;
  */
 static unsigned int zv_max_mean_zsize = (PAGE_SIZE / 8) * 5;
 
-static unsigned long zv_curr_dist_counts[NCHUNKS];
-static unsigned long zv_cumul_dist_counts[NCHUNKS];
+static atomic_t zv_curr_dist_counts[NCHUNKS];
+static atomic_t zv_cumul_dist_counts[NCHUNKS];
 
-static struct zv_hdr *zv_create(struct xv_pool *xvpool, uint32_t pool_id,
+
+static struct zv_hdr *zv_create(struct zcache_client *cli, uint32_t pool_id,
                                struct tmem_oid *oid, uint32_t index,
                                void *cdata, unsigned clen)
 {
@@ -671,23 +1103,61 @@ static struct zv_hdr *zv_create(struct xv_pool *xvpool, uint32_t pool_id,
 
        BUG_ON(!irqs_disabled());
        BUG_ON(chunks >= NCHUNKS);
-       ret = xv_malloc(xvpool, alloc_size,
+       ret = xv_malloc(cli->xvpool, clen + sizeof(struct zv_hdr),
                        &page, &offset, ZCACHE_GFP_MASK);
        if (unlikely(ret))
                goto out;
-       zv_curr_dist_counts[chunks]++;
-       zv_cumul_dist_counts[chunks]++;
+       atomic_inc(&zv_curr_dist_counts[chunks]);
+       atomic_inc(&zv_cumul_dist_counts[chunks]);
        zv = kmap_atomic(page, KM_USER0) + offset;
        zv->index = index;
        zv->oid = *oid;
        zv->pool_id = pool_id;
        SET_SENTINEL(zv, ZVH);
+       INIT_LIST_HEAD(&zv->rem_op.list);
+       zv->client_id = get_client_id_from_client(cli);
+       zv->rem_op.op = RAMSTER_REMOTIFY_PERS_PUT;
+       if (zv->client_id == LOCAL_CLIENT) {
+               spin_lock(&zcache_rem_op_list_lock);
+               list_add_tail(&zv->rem_op.list, &zcache_rem_op_list);
+               spin_unlock(&zcache_rem_op_list_lock);
+       }
        memcpy((char *)zv + sizeof(struct zv_hdr), cdata, clen);
        kunmap_atomic(zv, KM_USER0);
 out:
        return zv;
 }
 
+/* similar to zv_create, but just reserve space, no data yet */
+static struct zv_hdr *zv_alloc(struct tmem_pool *pool,
+                               struct tmem_oid *oid, uint32_t index,
+                               unsigned clen)
+{
+       struct zcache_client *cli = pool->client;
+       struct page *page;
+       struct zv_hdr *zv = NULL;
+       uint32_t offset;
+       int ret;
+
+       BUG_ON(!irqs_disabled());
+       BUG_ON(!is_local_client(pool->client));
+       ret = xv_malloc(cli->xvpool, clen + sizeof(struct zv_hdr),
+                       &page, &offset, ZCACHE_GFP_MASK);
+       if (unlikely(ret))
+               goto out;
+       zv = kmap_atomic(page, KM_USER0) + offset;
+       SET_SENTINEL(zv, ZVH);
+       INIT_LIST_HEAD(&zv->rem_op.list);
+       zv->client_id = LOCAL_CLIENT;
+       zv->rem_op.op = RAMSTER_INTRANSIT_PERS;
+       zv->index = index;
+       zv->oid = *oid;
+       zv->pool_id = pool->pool_id;
+       kunmap_atomic(zv, KM_USER0);
+out:
+       return zv;
+}
+
 static void zv_free(struct xv_pool *xvpool, struct zv_hdr *zv)
 {
        unsigned long flags;
@@ -698,10 +1168,15 @@ static void zv_free(struct xv_pool *xvpool, struct zv_hdr *zv)
 
        ASSERT_SENTINEL(zv, ZVH);
        BUG_ON(chunks >= NCHUNKS);
-       zv_curr_dist_counts[chunks]--;
+       atomic_dec(&zv_curr_dist_counts[chunks]);
        size -= sizeof(*zv);
+       spin_lock(&zcache_rem_op_list_lock);
+       size = xv_get_object_size(zv) - sizeof(*zv);
        BUG_ON(size == 0);
        INVERT_SENTINEL(zv, ZVH);
+       if (!list_empty(&zv->rem_op.list))
+               list_del_init(&zv->rem_op.list);
+       spin_unlock(&zcache_rem_op_list_lock);
        page = virt_to_page(zv);
        offset = (unsigned long)zv & ~PAGE_MASK;
        local_irq_save(flags);
@@ -727,6 +1202,29 @@ static void zv_decompress(struct page *page, struct zv_hdr *zv)
        BUG_ON(clen != PAGE_SIZE);
 }
 
+static void zv_copy_from_pampd(char *data, size_t *bufsize, struct zv_hdr *zv)
+{
+       unsigned size;
+
+       ASSERT_SENTINEL(zv, ZVH);
+       size = xv_get_object_size(zv) - sizeof(*zv);
+       BUG_ON(size == 0 || size > zv_max_page_size);
+       BUG_ON(size > *bufsize);
+       memcpy(data, (char *)zv + sizeof(*zv), size);
+       *bufsize = size;
+}
+
+static void zv_copy_to_pampd(struct zv_hdr *zv, char *data, size_t size)
+{
+       unsigned zv_size;
+
+       ASSERT_SENTINEL(zv, ZVH);
+       zv_size = xv_get_object_size(zv) - sizeof(*zv);
+       BUG_ON(zv_size != size);
+       BUG_ON(zv_size == 0 || zv_size > zv_max_page_size);
+       memcpy((char *)zv + sizeof(*zv), data, size);
+}
+
 #ifdef CONFIG_SYSFS
 /*
  * show a distribution of compression stats for zv pages.
@@ -738,7 +1236,7 @@ static int zv_curr_dist_counts_show(char *buf)
        char *p = buf;
 
        for (i = 0; i < NCHUNKS; i++) {
-               n = zv_curr_dist_counts[i];
+               n = atomic_read(&zv_curr_dist_counts[i]);
                p += sprintf(p, "%lu ", n);
                chunks += n;
                sum_total_chunks += i * n;
@@ -754,7 +1252,7 @@ static int zv_cumul_dist_counts_show(char *buf)
        char *p = buf;
 
        for (i = 0; i < NCHUNKS; i++) {
-               n = zv_cumul_dist_counts[i];
+               n = atomic_read(&zv_cumul_dist_counts[i]);
                p += sprintf(p, "%lu ", n);
                chunks += n;
                sum_total_chunks += i * n;
@@ -787,7 +1285,7 @@ static ssize_t zv_max_zsize_store(struct kobject *kobj,
        if (!capable(CAP_SYS_ADMIN))
                return -EPERM;
 
-       err = strict_strtoul(buf, 10, &val);
+       err = kstrtoul(buf, 10, &val);
        if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7))
                return -EINVAL;
        zv_max_zsize = val;
@@ -819,7 +1317,7 @@ static ssize_t zv_max_mean_zsize_store(struct kobject *kobj,
        if (!capable(CAP_SYS_ADMIN))
                return -EPERM;
 
-       err = strict_strtoul(buf, 10, &val);
+       err = kstrtoul(buf, 10, &val);
        if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7))
                return -EINVAL;
        zv_max_mean_zsize = val;
@@ -853,7 +1351,7 @@ static ssize_t zv_page_count_policy_percent_store(struct kobject *kobj,
        if (!capable(CAP_SYS_ADMIN))
                return -EPERM;
 
-       err = strict_strtoul(buf, 10, &val);
+       err = kstrtoul(buf, 10, &val);
        if (err || (val == 0) || (val > 150))
                return -EINVAL;
        zv_page_count_policy_percent = val;
@@ -890,6 +1388,7 @@ static unsigned long zcache_flush_found;
 static unsigned long zcache_flobj_total;
 static unsigned long zcache_flobj_found;
 static unsigned long zcache_failed_eph_puts;
+static unsigned long zcache_nonactive_puts;
 static unsigned long zcache_failed_pers_puts;
 
 /*
@@ -970,6 +1469,7 @@ static unsigned long zcache_put_to_flush;
  */
 static struct kmem_cache *zcache_objnode_cache;
 static struct kmem_cache *zcache_obj_cache;
+static struct kmem_cache *ramster_flnode_cache;
 static atomic_t zcache_curr_obj_count = ATOMIC_INIT(0);
 static unsigned long zcache_curr_obj_count_max;
 static atomic_t zcache_curr_objnode_count = ATOMIC_INIT(0);
@@ -985,6 +1485,7 @@ struct zcache_preload {
        struct tmem_obj *obj;
        int nr;
        struct tmem_objnode *objnodes[OBJNODE_TREE_MAX_PATH];
+       struct flushlist_node *flnode;
 };
 static DEFINE_PER_CPU(struct zcache_preload, zcache_preloads) = { 0, };
 
@@ -993,6 +1494,7 @@ static int zcache_do_preload(struct tmem_pool *pool)
        struct zcache_preload *kp;
        struct tmem_objnode *objnode;
        struct tmem_obj *obj;
+       struct flushlist_node *flnode;
        void *page;
        int ret = -ENOMEM;
 
@@ -1023,27 +1525,61 @@ static int zcache_do_preload(struct tmem_pool *pool)
                zcache_failed_alloc++;
                goto out;
        }
-       page = (void *)__get_free_page(ZCACHE_GFP_MASK);
-       if (unlikely(page == NULL)) {
-               zcache_failed_get_free_pages++;
-               kmem_cache_free(zcache_obj_cache, obj);
+       flnode = kmem_cache_alloc(ramster_flnode_cache, ZCACHE_GFP_MASK);
+       if (unlikely(flnode == NULL)) {
+               zcache_failed_alloc++;
                goto out;
        }
+       if (is_ephemeral(pool)) {
+               page = (void *)__get_free_page(ZCACHE_GFP_MASK);
+               if (unlikely(page == NULL)) {
+                       zcache_failed_get_free_pages++;
+                       kmem_cache_free(zcache_obj_cache, obj);
+                       kmem_cache_free(ramster_flnode_cache, flnode);
+                       goto out;
+               }
+       }
        preempt_disable();
        kp = &__get_cpu_var(zcache_preloads);
        if (kp->obj == NULL)
                kp->obj = obj;
        else
                kmem_cache_free(zcache_obj_cache, obj);
-       if (kp->page == NULL)
-               kp->page = page;
+       if (kp->flnode == NULL)
+               kp->flnode = flnode;
        else
-               free_page((unsigned long)page);
+               kmem_cache_free(ramster_flnode_cache, flnode);
+       if (is_ephemeral(pool)) {
+               if (kp->page == NULL)
+                       kp->page = page;
+               else
+                       free_page((unsigned long)page);
+       }
        ret = 0;
 out:
        return ret;
 }
 
+static int ramster_do_preload_flnode_only(struct tmem_pool *pool)
+{
+       struct zcache_preload *kp;
+       struct flushlist_node *flnode;
+       int ret = -ENOMEM;
+
+       BUG_ON(!irqs_disabled());
+       if (unlikely(ramster_flnode_cache == NULL))
+               BUG();
+       kp = &__get_cpu_var(zcache_preloads);
+       flnode = kmem_cache_alloc(ramster_flnode_cache, GFP_ATOMIC);
+       if (unlikely(flnode == NULL) && kp->flnode == NULL)
+               BUG();  /* FIXME handle more gracefully, but how??? */
+       else if (kp->flnode == NULL)
+               kp->flnode = flnode;
+       else
+               kmem_cache_free(ramster_flnode_cache, flnode);
+       return ret;
+}
+
 static void *zcache_get_free_page(void)
 {
        struct zcache_preload *kp;
@@ -1116,6 +1652,30 @@ static void zcache_obj_free(struct tmem_obj *obj, struct tmem_pool *pool)
        kmem_cache_free(zcache_obj_cache, obj);
 }
 
+static struct flushlist_node *ramster_flnode_alloc(struct tmem_pool *pool)
+{
+       struct flushlist_node *flnode = NULL;
+       struct zcache_preload *kp;
+       int count;
+
+       kp = &__get_cpu_var(zcache_preloads);
+       flnode = kp->flnode;
+       BUG_ON(flnode == NULL);
+       kp->flnode = NULL;
+       count = atomic_inc_return(&ramster_curr_flnode_count);
+       if (count > ramster_curr_flnode_count_max)
+               ramster_curr_flnode_count_max = count;
+       return flnode;
+}
+
+static void ramster_flnode_free(struct flushlist_node *flnode,
+                               struct tmem_pool *pool)
+{
+       atomic_dec(&ramster_curr_flnode_count);
+       BUG_ON(atomic_read(&ramster_curr_flnode_count) < 0);
+       kmem_cache_free(ramster_flnode_cache, flnode);
+}
+
 static struct tmem_hostops zcache_hostops = {
        .obj_alloc = zcache_obj_alloc,
        .obj_free = zcache_obj_free,
@@ -1127,6 +1687,14 @@ static struct tmem_hostops zcache_hostops = {
  * zcache implementations for PAM page descriptor ops
  */
 
+
+static inline void dec_and_check(atomic_t *pvar)
+{
+       atomic_dec(pvar);
+       /* later when all accounting is fixed, make this a BUG */
+       WARN_ON_ONCE(atomic_read(pvar) < 0);
+}
+
 static atomic_t zcache_curr_eph_pampd_count = ATOMIC_INIT(0);
 static unsigned long zcache_curr_eph_pampd_count_max;
 static atomic_t zcache_curr_pers_pampd_count = ATOMIC_INIT(0);
@@ -1135,22 +1703,20 @@ static unsigned long zcache_curr_pers_pampd_count_max;
 /* forward reference */
 static int zcache_compress(struct page *from, void **out_va, size_t *out_len);
 
-static void *zcache_pampd_create(char *data, size_t size, bool raw, int eph,
+static int zcache_pampd_eph_create(char *data, size_t size, bool raw,
                                struct tmem_pool *pool, struct tmem_oid *oid,
-                                uint32_t index)
+                               uint32_t index, void **pampd)
 {
-       void *pampd = NULL, *cdata;
-       size_t clen;
-       int ret;
-       unsigned long count;
-       struct page *page = (struct page *)(data);
+       int ret = -1;
+       void *cdata = data;
+       size_t clen = size;
        struct zcache_client *cli = pool->client;
        uint16_t client_id = get_client_id_from_client(cli);
-       unsigned long zv_mean_zsize;
-       unsigned long curr_pers_pampd_count;
-       u64 total_zsize;
+       struct page *page = NULL;
+       unsigned long count;
 
-       if (eph) {
+       if (!raw) {
+               page = virt_to_page(data);
                ret = zcache_compress(page, &cdata, &clen);
                if (ret == 0)
                        goto out;
@@ -1158,46 +1724,114 @@ static void *zcache_pampd_create(char *data, size_t size, bool raw, int eph,
                        zcache_compress_poor++;
                        goto out;
                }
-               pampd = (void *)zbud_create(client_id, pool->pool_id, oid,
-                                               index, page, cdata, clen);
-               if (pampd != NULL) {
-                       count = atomic_inc_return(&zcache_curr_eph_pampd_count);
-                       if (count > zcache_curr_eph_pampd_count_max)
-                               zcache_curr_eph_pampd_count_max = count;
-               }
-       } else {
-               curr_pers_pampd_count =
-                       atomic_read(&zcache_curr_pers_pampd_count);
-               if (curr_pers_pampd_count >
-                   (zv_page_count_policy_percent * totalram_pages) / 100)
-                       goto out;
-               ret = zcache_compress(page, &cdata, &clen);
-               if (ret == 0)
-                       goto out;
-               /* reject if compression is too poor */
-               if (clen > zv_max_zsize) {
-                       zcache_compress_poor++;
+       }
+       *pampd = (void *)zbud_create(client_id, pool->pool_id, oid,
+                                       index, page, cdata, clen);
+       if (*pampd == NULL) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       ret = 0;
+       count = atomic_inc_return(&zcache_curr_eph_pampd_count);
+       if (count > zcache_curr_eph_pampd_count_max)
+               zcache_curr_eph_pampd_count_max = count;
+       if (client_id != LOCAL_CLIENT) {
+               count = atomic_inc_return(&ramster_foreign_eph_pampd_count);
+               if (count > ramster_foreign_eph_pampd_count_max)
+                       ramster_foreign_eph_pampd_count_max = count;
+       }
+out:
+       return ret;
+}
+
+static int zcache_pampd_pers_create(char *data, size_t size, bool raw,
+                               struct tmem_pool *pool, struct tmem_oid *oid,
+                               uint32_t index, void **pampd)
+{
+       int ret = -1;
+       void *cdata = data;
+       size_t clen = size;
+       struct zcache_client *cli = pool->client;
+       struct page *page;
+       unsigned long count;
+       unsigned long zv_mean_zsize;
+       struct zv_hdr *zv;
+       long curr_pers_pampd_count;
+       u64 total_zsize;
+#ifdef RAMSTER_TESTING
+       static bool pampd_neg_warned;
+#endif
+
+       curr_pers_pampd_count = atomic_read(&zcache_curr_pers_pampd_count) -
+                       atomic_read(&ramster_remote_pers_pages);
+#ifdef RAMSTER_TESTING
+       /* should always be positive, but warn if accounting is off */
+       if (!pampd_neg_warned) {
+               pr_warn("ramster: bad accounting for curr_pers_pampd_count\n");
+               pampd_neg_warned = true;
+       }
+#endif
+       if (curr_pers_pampd_count >
+                   (zv_page_count_policy_percent * totalram_pages) / 100) {
+               zcache_policy_percent_exceeded++;
+               goto out;
+       }
+       if (raw)
+               goto ok_to_create;
+       page = virt_to_page(data);
+       if (zcache_compress(page, &cdata, &clen) == 0)
+               goto out;
+       /* reject if compression is too poor */
+       if (clen > zv_max_zsize) {
+               zcache_compress_poor++;
+               goto out;
+       }
+       /* reject if mean compression is too poor */
+       if ((clen > zv_max_mean_zsize) && (curr_pers_pampd_count > 0)) {
+               total_zsize = xv_get_total_size_bytes(cli->xvpool);
+               zv_mean_zsize = div_u64(total_zsize, curr_pers_pampd_count);
+               if (zv_mean_zsize > zv_max_mean_zsize) {
+                       zcache_mean_compress_poor++;
                        goto out;
                }
-               /* reject if mean compression is too poor */
-               if ((clen > zv_max_mean_zsize) && (curr_pers_pampd_count > 0)) {
-                       total_zsize = xv_get_total_size_bytes(cli->xvpool);
-                       zv_mean_zsize = div_u64(total_zsize,
-                                               curr_pers_pampd_count);
-                       if (zv_mean_zsize > zv_max_mean_zsize) {
-                               zcache_mean_compress_poor++;
-                               goto out;
-                       }
-               }
-               pampd = (void *)zv_create(cli->xvpool, pool->pool_id,
-                                               oid, index, cdata, clen);
-               if (pampd == NULL)
-                       goto out;
-               count = atomic_inc_return(&zcache_curr_pers_pampd_count);
-               if (count > zcache_curr_pers_pampd_count_max)
-                       zcache_curr_pers_pampd_count_max = count;
        }
+ok_to_create:
+       *pampd = (void *)zv_create(cli, pool->pool_id, oid, index, cdata, clen);
+       if (*pampd == NULL) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       ret = 0;
+       count = atomic_inc_return(&zcache_curr_pers_pampd_count);
+       if (count > zcache_curr_pers_pampd_count_max)
+               zcache_curr_pers_pampd_count_max = count;
+       if (is_local_client(cli))
+               goto out;
+       zv = *(struct zv_hdr **)pampd;
+       count = atomic_inc_return(&ramster_foreign_pers_pampd_count);
+       if (count > ramster_foreign_pers_pampd_count_max)
+               ramster_foreign_pers_pampd_count_max = count;
 out:
+       return ret;
+}
+
+static void *zcache_pampd_create(char *data, size_t size, bool raw, int eph,
+                               struct tmem_pool *pool, struct tmem_oid *oid,
+                               uint32_t index)
+{
+       void *pampd = NULL;
+       int ret;
+       bool ephemeral;
+
+       BUG_ON(preemptible());
+       ephemeral = (eph == 1) || ((eph == 0) && is_ephemeral(pool));
+       if (ephemeral)
+               ret = zcache_pampd_eph_create(data, size, raw, pool,
+                                               oid, index, &pampd);
+       else
+               ret = zcache_pampd_pers_create(data, size, raw, pool,
+                                               oid, index, &pampd);
+       /* FIXME add some counters here for failed creates? */
        return pampd;
 }
 
@@ -1211,75 +1845,343 @@ static int zcache_pampd_get_data(char *data, size_t *bufsize, bool raw,
 {
        int ret = 0;
 
-       BUG_ON(is_ephemeral(pool));
-       zv_decompress((struct page *)(data), pampd);
-       return ret;
-}
-
-/*
- * fill the pageframe corresponding to the struct page with the data
- * from the passed pampd
- */
+       BUG_ON(preemptible());
+       BUG_ON(is_ephemeral(pool)); /* Fix later for shared pools? */
+       BUG_ON(pampd_is_remote(pampd));
+       if (raw)
+               zv_copy_from_pampd(data, bufsize, pampd);
+       else
+               zv_decompress(virt_to_page(data), pampd);
+       return ret;
+}
+
 static int zcache_pampd_get_data_and_free(char *data, size_t *bufsize, bool raw,
                                        void *pampd, struct tmem_pool *pool,
                                        struct tmem_oid *oid, uint32_t index)
 {
        int ret = 0;
+       unsigned long flags;
+       struct zcache_client *cli = pool->client;
 
-       BUG_ON(!is_ephemeral(pool));
-       zbud_decompress((struct page *)(data), pampd);
-       zbud_free_and_delist((struct zbud_hdr *)pampd);
-       atomic_dec(&zcache_curr_eph_pampd_count);
+       BUG_ON(preemptible());
+       BUG_ON(pampd_is_remote(pampd));
+       if (is_ephemeral(pool)) {
+               local_irq_save(flags);
+               if (raw)
+                       zbud_copy_from_pampd(data, bufsize, pampd);
+               else
+                       ret = zbud_decompress(virt_to_page(data), pampd);
+               zbud_free_and_delist((struct zbud_hdr *)pampd);
+               local_irq_restore(flags);
+               if (!is_local_client(cli))
+                       dec_and_check(&ramster_foreign_eph_pampd_count);
+               dec_and_check(&zcache_curr_eph_pampd_count);
+       } else {
+               if (is_local_client(cli))
+                       BUG();
+               if (raw)
+                       zv_copy_from_pampd(data, bufsize, pampd);
+               else
+                       zv_decompress(virt_to_page(data), pampd);
+               zv_free(cli->xvpool, pampd);
+               if (!is_local_client(cli))
+                       dec_and_check(&ramster_foreign_pers_pampd_count);
+               dec_and_check(&zcache_curr_pers_pampd_count);
+               ret = 0;
+       }
        return ret;
 }
 
+static bool zcache_pampd_is_remote(void *pampd)
+{
+       return pampd_is_remote(pampd);
+}
+
 /*
  * free the pampd and remove it from any zcache lists
  * pampd must no longer be pointed to from any tmem data structures!
  */
 static void zcache_pampd_free(void *pampd, struct tmem_pool *pool,
-                               struct tmem_oid *oid, uint32_t index)
+                             struct tmem_oid *oid, uint32_t index, bool acct)
 {
        struct zcache_client *cli = pool->client;
-
-       if (is_ephemeral(pool)) {
+       bool eph = is_ephemeral(pool);
+       struct zv_hdr *zv;
+
+       BUG_ON(preemptible());
+       if (pampd_is_remote(pampd)) {
+               WARN_ON(acct == false);
+               if (oid == NULL) {
+                       /*
+                        * a NULL oid means to ignore this pampd free
+                        * as the remote freeing will be handled elsewhere
+                        */
+               } else if (eph) {
+                       /* FIXME remote flush optional but probably good idea */
+                       /* FIXME get these working properly again */
+                       dec_and_check(&zcache_curr_eph_pampd_count);
+               } else if (pampd_is_intransit(pampd)) {
+                       /* did a pers remote get_and_free, so just free local */
+                       pampd = pampd_mask_intransit_and_remote(pampd);
+                       goto local_pers;
+               } else {
+                       struct flushlist_node *flnode =
+                               ramster_flnode_alloc(pool);
+
+                       flnode->xh.client_id = pampd_remote_node(pampd);
+                       flnode->xh.pool_id = pool->pool_id;
+                       flnode->xh.oid = *oid;
+                       flnode->xh.index = index;
+                       flnode->rem_op.op = RAMSTER_REMOTIFY_FLUSH_PAGE;
+                       spin_lock(&zcache_rem_op_list_lock);
+                       list_add(&flnode->rem_op.list, &zcache_rem_op_list);
+                       spin_unlock(&zcache_rem_op_list_lock);
+                       dec_and_check(&zcache_curr_pers_pampd_count);
+                       dec_and_check(&ramster_remote_pers_pages);
+               }
+       } else if (eph) {
                zbud_free_and_delist((struct zbud_hdr *)pampd);
-               atomic_dec(&zcache_curr_eph_pampd_count);
-               BUG_ON(atomic_read(&zcache_curr_eph_pampd_count) < 0);
+               if (!is_local_client(pool->client))
+                       dec_and_check(&ramster_foreign_eph_pampd_count);
+               if (acct)
+                       /* FIXME get these working properly again */
+                       dec_and_check(&zcache_curr_eph_pampd_count);
        } else {
-               zv_free(cli->xvpool, (struct zv_hdr *)pampd);
-               atomic_dec(&zcache_curr_pers_pampd_count);
-               BUG_ON(atomic_read(&zcache_curr_pers_pampd_count) < 0);
+local_pers:
+               zv = (struct zv_hdr *)pampd;
+               if (!is_local_client(pool->client))
+                       dec_and_check(&ramster_foreign_pers_pampd_count);
+               zv_free(cli->xvpool, zv);
+               if (acct)
+                       /* FIXME get these working properly again */
+                       dec_and_check(&zcache_curr_pers_pampd_count);
        }
 }
 
-static void zcache_pampd_free_obj(struct tmem_pool *pool, struct tmem_obj *obj)
+static void zcache_pampd_free_obj(struct tmem_pool *pool,
+                                       struct tmem_obj *obj)
 {
+       struct flushlist_node *flnode;
+
+       BUG_ON(preemptible());
+       if (obj->extra == NULL)
+               return;
+       BUG_ON(!pampd_is_remote(obj->extra));
+       flnode = ramster_flnode_alloc(pool);
+       flnode->xh.client_id = pampd_remote_node(obj->extra);
+       flnode->xh.pool_id = pool->pool_id;
+       flnode->xh.oid = obj->oid;
+       flnode->xh.index = FLUSH_ENTIRE_OBJECT;
+       flnode->rem_op.op = RAMSTER_REMOTIFY_FLUSH_OBJ;
+       spin_lock(&zcache_rem_op_list_lock);
+       list_add(&flnode->rem_op.list, &zcache_rem_op_list);
+       spin_unlock(&zcache_rem_op_list_lock);
 }
 
-static void zcache_pampd_new_obj(struct tmem_obj *obj)
+void zcache_pampd_new_obj(struct tmem_obj *obj)
 {
+       obj->extra = NULL;
 }
 
-static int zcache_pampd_replace_in_obj(void *pampd, struct tmem_obj *obj)
+int zcache_pampd_replace_in_obj(void *new_pampd, struct tmem_obj *obj)
 {
-       return -1;
+       int ret = -1;
+
+       if (new_pampd != NULL) {
+               if (obj->extra == NULL)
+                       obj->extra = new_pampd;
+               /* enforce that all remote pages in an object reside
+                * in the same node! */
+               else if (pampd_remote_node(new_pampd) !=
+                               pampd_remote_node((void *)(obj->extra)))
+                       BUG();
+               ret = 0;
+       }
+       return ret;
 }
 
-static bool zcache_pampd_is_remote(void *pampd)
+/*
+ * Called by the message handler after a (still compressed) page has been
+ * fetched from the remote machine in response to an "is_remote" tmem_get
+ * or persistent tmem_localify.  For a tmem_get, "extra" is the address of
+ * the page that is to be filled to succesfully resolve the tmem_get; for
+ * a (persistent) tmem_localify, "extra" is NULL (as the data is placed only
+ * in the local zcache).  "data" points to "size" bytes of (compressed) data
+ * passed in the message.  In the case of a persistent remote get, if
+ * pre-allocation was successful (see zcache_repatriate_preload), the page
+ * is placed into both local zcache and at "extra".
+ */
+int zcache_localify(int pool_id, struct tmem_oid *oidp,
+                       uint32_t index, char *data, size_t size,
+                       void *extra)
 {
-       return 0;
+       int ret = -ENOENT;
+       unsigned long flags;
+       struct tmem_pool *pool;
+       bool ephemeral, delete = false;
+       size_t clen = PAGE_SIZE;
+       void *pampd, *saved_hb;
+       struct tmem_obj *obj;
+
+       pool = zcache_get_pool_by_id(LOCAL_CLIENT, pool_id);
+       if (unlikely(pool == NULL))
+               /* pool doesn't exist anymore */
+               goto out;
+       ephemeral = is_ephemeral(pool);
+       local_irq_save(flags);  /* FIXME: maybe only disable softirqs? */
+       pampd = tmem_localify_get_pampd(pool, oidp, index, &obj, &saved_hb);
+       if (pampd == NULL) {
+               /* hmmm... must have been a flush while waiting */
+#ifdef RAMSTER_TESTING
+               pr_err("UNTESTED pampd==NULL in zcache_localify\n");
+#endif
+               if (ephemeral)
+                       ramster_remote_eph_pages_unsucc_get++;
+               else
+                       ramster_remote_pers_pages_unsucc_get++;
+               obj = NULL;
+               goto finish;
+       } else if (unlikely(!pampd_is_remote(pampd))) {
+               /* hmmm... must have been a dup put while waiting */
+#ifdef RAMSTER_TESTING
+               pr_err("UNTESTED dup while waiting in zcache_localify\n");
+#endif
+               if (ephemeral)
+                       ramster_remote_eph_pages_unsucc_get++;
+               else
+                       ramster_remote_pers_pages_unsucc_get++;
+               obj = NULL;
+               pampd = NULL;
+               ret = -EEXIST;
+               goto finish;
+       } else if (size == 0) {
+               /* no remote data, delete the local is_remote pampd */
+               pampd = NULL;
+               if (ephemeral)
+                       ramster_remote_eph_pages_unsucc_get++;
+               else
+                       BUG();
+               delete = true;
+               goto finish;
+       }
+       if (!ephemeral && pampd_is_intransit(pampd)) {
+               /* localify to zcache */
+               pampd = pampd_mask_intransit_and_remote(pampd);
+               zv_copy_to_pampd(pampd, data, size);
+       } else {
+               pampd = NULL;
+               obj = NULL;
+       }
+       if (extra != NULL) {
+               /* decompress direct-to-memory to complete remotify */
+               ret = lzo1x_decompress_safe((char *)data, size,
+                                               (char *)extra, &clen);
+               BUG_ON(ret != LZO_E_OK);
+               BUG_ON(clen != PAGE_SIZE);
+       }
+       if (ephemeral)
+               ramster_remote_eph_pages_succ_get++;
+       else
+               ramster_remote_pers_pages_succ_get++;
+       ret = 0;
+finish:
+       tmem_localify_finish(obj, index, pampd, saved_hb, delete);
+       zcache_put_pool(pool);
+       local_irq_restore(flags);
+out:
+       return ret;
+}
+
+/*
+ * Called on a remote persistent tmem_get to attempt to preallocate
+ * local storage for the data contained in the remote persistent page.
+ * If succesfully preallocated, returns the pampd, marked as remote and
+ * in_transit.  Else returns NULL.  Note that the appropriate tmem data
+ * structure must be locked.
+ */
+static void *zcache_pampd_repatriate_preload(void *pampd,
+                                               struct tmem_pool *pool,
+                                               struct tmem_oid *oid,
+                                               uint32_t index,
+                                               bool *intransit)
+{
+       int clen = pampd_remote_size(pampd);
+       void *ret_pampd = NULL;
+       unsigned long flags;
+
+       if (!pampd_is_remote(pampd))
+               BUG();
+       if (is_ephemeral(pool))
+               BUG();
+       if (pampd_is_intransit(pampd)) {
+               /*
+                * to avoid multiple allocations (and maybe a memory leak)
+                * don't preallocate if already in the process of being
+                * repatriated
+                */
+               *intransit = true;
+               goto out;
+       }
+       *intransit = false;
+       local_irq_save(flags);
+       ret_pampd = (void *)zv_alloc(pool, oid, index, clen);
+       if (ret_pampd != NULL) {
+               /*
+                *  a pampd is marked intransit if it is remote and space has
+                *  been allocated for it locally (note, only happens for
+                *  persistent pages, in which case the remote copy is freed)
+                */
+               ret_pampd = pampd_mark_intransit(ret_pampd);
+               dec_and_check(&ramster_remote_pers_pages);
+       } else
+               ramster_pers_pages_remote_nomem++;
+       local_irq_restore(flags);
+out:
+       return ret_pampd;
+}
+
+/*
+ * Called on a remote tmem_get to invoke a message to fetch the page.
+ * Might sleep so no tmem locks can be held.  "extra" is passed
+ * all the way through the round-trip messaging to zcache_localify.
+ */
+static int zcache_pampd_repatriate(void *fake_pampd, void *real_pampd,
+                                  struct tmem_pool *pool,
+                                  struct tmem_oid *oid, uint32_t index,
+                                  bool free, void *extra)
+{
+       struct tmem_xhandle xh;
+       int ret;
+
+       if (pampd_is_intransit(real_pampd))
+               /* have local space pre-reserved, so free remote copy */
+               free = true;
+       xh = tmem_xhandle_fill(LOCAL_CLIENT, pool, oid, index);
+       /* unreliable request/response for now */
+       ret = ramster_remote_async_get(&xh, free,
+                                       pampd_remote_node(fake_pampd),
+                                       pampd_remote_size(fake_pampd),
+                                       pampd_remote_cksum(fake_pampd),
+                                       extra);
+#ifdef RAMSTER_TESTING
+       if (ret != 0 && ret != -ENOENT)
+               pr_err("TESTING zcache_pampd_repatriate returns, ret=%d\n",
+                       ret);
+#endif
+       return ret;
 }
 
 static struct tmem_pamops zcache_pamops = {
        .create = zcache_pampd_create,
        .get_data = zcache_pampd_get_data,
-       .get_data_and_free = zcache_pampd_get_data_and_free,
        .free = zcache_pampd_free,
+       .get_data_and_free = zcache_pampd_get_data_and_free,
        .free_obj = zcache_pampd_free_obj,
+       .is_remote = zcache_pampd_is_remote,
+       .repatriate_preload = zcache_pampd_repatriate_preload,
+       .repatriate = zcache_pampd_repatriate,
        .new_obj = zcache_pampd_new_obj,
        .replace_in_obj = zcache_pampd_replace_in_obj,
-       .is_remote = zcache_pampd_is_remote,
 };
 
 /*
@@ -1327,9 +2229,13 @@ static int zcache_cpu_notifier(struct notifier_block *nb,
                per_cpu(zcache_workmem, cpu) =
                        kzalloc(LZO1X_MEM_COMPRESS,
                                GFP_KERNEL | __GFP_REPEAT);
+               per_cpu(zcache_remoteputmem, cpu) =
+                       kzalloc(PAGE_SIZE, GFP_KERNEL | __GFP_REPEAT);
                break;
        case CPU_DEAD:
        case CPU_UP_CANCELED:
+               kfree(per_cpu(zcache_remoteputmem, cpu));
+               per_cpu(zcache_remoteputmem, cpu) = NULL;
                free_pages((unsigned long)per_cpu(zcache_dstmem, cpu),
                                LZO_DSTMEM_PAGE_ORDER);
                per_cpu(zcache_dstmem, cpu) = NULL;
@@ -1346,6 +2252,10 @@ static int zcache_cpu_notifier(struct notifier_block *nb,
                        kmem_cache_free(zcache_obj_cache, kp->obj);
                        kp->obj = NULL;
                }
+               if (kp->flnode) {
+                       kmem_cache_free(ramster_flnode_cache, kp->flnode);
+                       kp->flnode = NULL;
+               }
                if (kp->page) {
                        free_page((unsigned long)kp->page);
                        kp->page = NULL;
@@ -1402,12 +2312,12 @@ ZCACHE_SYSFS_RO(flush_found);
 ZCACHE_SYSFS_RO(flobj_total);
 ZCACHE_SYSFS_RO(flobj_found);
 ZCACHE_SYSFS_RO(failed_eph_puts);
+ZCACHE_SYSFS_RO(nonactive_puts);
 ZCACHE_SYSFS_RO(failed_pers_puts);
 ZCACHE_SYSFS_RO(zbud_curr_zbytes);
 ZCACHE_SYSFS_RO(zbud_cumul_zpages);
 ZCACHE_SYSFS_RO(zbud_cumul_zbytes);
 ZCACHE_SYSFS_RO(zbud_buddied_count);
-ZCACHE_SYSFS_RO(zbpg_unused_list_count);
 ZCACHE_SYSFS_RO(evicted_raw_pages);
 ZCACHE_SYSFS_RO(evicted_unbuddied_pages);
 ZCACHE_SYSFS_RO(evicted_buddied_pages);
@@ -1416,6 +2326,7 @@ ZCACHE_SYSFS_RO(failed_alloc);
 ZCACHE_SYSFS_RO(put_to_flush);
 ZCACHE_SYSFS_RO(compress_poor);
 ZCACHE_SYSFS_RO(mean_compress_poor);
+ZCACHE_SYSFS_RO(policy_percent_exceeded);
 ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_raw_pages);
 ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_zpages);
 ZCACHE_SYSFS_RO_ATOMIC(curr_obj_count);
@@ -1439,7 +2350,9 @@ static struct attribute *zcache_attrs[] = {
        &zcache_flush_found_attr.attr,
        &zcache_flobj_found_attr.attr,
        &zcache_failed_eph_puts_attr.attr,
+       &zcache_nonactive_puts_attr.attr,
        &zcache_failed_pers_puts_attr.attr,
+       &zcache_policy_percent_exceeded_attr.attr,
        &zcache_compress_poor_attr.attr,
        &zcache_mean_compress_poor_attr.attr,
        &zcache_zbud_curr_raw_pages_attr.attr,
@@ -1448,7 +2361,6 @@ static struct attribute *zcache_attrs[] = {
        &zcache_zbud_cumul_zpages_attr.attr,
        &zcache_zbud_cumul_zbytes_attr.attr,
        &zcache_zbud_buddied_count_attr.attr,
-       &zcache_zbpg_unused_list_count_attr.attr,
        &zcache_evicted_raw_pages_attr.attr,
        &zcache_evicted_unbuddied_pages_attr.attr,
        &zcache_evicted_buddied_pages_attr.attr,
@@ -1470,6 +2382,202 @@ static struct attribute_group zcache_attr_group = {
        .name = "zcache",
 };
 
+#define RAMSTER_SYSFS_RO(_name) \
+       static ssize_t ramster_##_name##_show(struct kobject *kobj, \
+                               struct kobj_attribute *attr, char *buf) \
+       { \
+               return sprintf(buf, "%lu\n", ramster_##_name); \
+       } \
+       static struct kobj_attribute ramster_##_name##_attr = { \
+               .attr = { .name = __stringify(_name), .mode = 0444 }, \
+               .show = ramster_##_name##_show, \
+       }
+
+#define RAMSTER_SYSFS_RW(_name) \
+       static ssize_t ramster_##_name##_show(struct kobject *kobj, \
+                               struct kobj_attribute *attr, char *buf) \
+       { \
+               return sprintf(buf, "%lu\n", ramster_##_name); \
+       } \
+       static ssize_t ramster_##_name##_store(struct kobject *kobj, \
+               struct kobj_attribute *attr, const char *buf, size_t count) \
+       { \
+               int err; \
+               unsigned long enable; \
+               err = kstrtoul(buf, 10, &enable); \
+               if (err) \
+                       return -EINVAL; \
+               ramster_##_name = enable; \
+               return count; \
+       } \
+       static struct kobj_attribute ramster_##_name##_attr = { \
+               .attr = { .name = __stringify(_name), .mode = 0644 }, \
+               .show = ramster_##_name##_show, \
+               .store = ramster_##_name##_store, \
+       }
+
+#define RAMSTER_SYSFS_RO_ATOMIC(_name) \
+       static ssize_t ramster_##_name##_show(struct kobject *kobj, \
+                               struct kobj_attribute *attr, char *buf) \
+       { \
+           return sprintf(buf, "%d\n", atomic_read(&ramster_##_name)); \
+       } \
+       static struct kobj_attribute ramster_##_name##_attr = { \
+               .attr = { .name = __stringify(_name), .mode = 0444 }, \
+               .show = ramster_##_name##_show, \
+       }
+
+RAMSTER_SYSFS_RO(interface_revision);
+RAMSTER_SYSFS_RO_ATOMIC(remote_pers_pages);
+RAMSTER_SYSFS_RW(pers_remotify_enable);
+RAMSTER_SYSFS_RW(eph_remotify_enable);
+RAMSTER_SYSFS_RO(eph_pages_remoted);
+RAMSTER_SYSFS_RO(eph_pages_remote_failed);
+RAMSTER_SYSFS_RO(pers_pages_remoted);
+RAMSTER_SYSFS_RO(pers_pages_remote_failed);
+RAMSTER_SYSFS_RO(pers_pages_remote_nomem);
+RAMSTER_SYSFS_RO(remote_pages_flushed);
+RAMSTER_SYSFS_RO(remote_page_flushes_failed);
+RAMSTER_SYSFS_RO(remote_objects_flushed);
+RAMSTER_SYSFS_RO(remote_object_flushes_failed);
+RAMSTER_SYSFS_RO(remote_eph_pages_succ_get);
+RAMSTER_SYSFS_RO(remote_eph_pages_unsucc_get);
+RAMSTER_SYSFS_RO(remote_pers_pages_succ_get);
+RAMSTER_SYSFS_RO(remote_pers_pages_unsucc_get);
+RAMSTER_SYSFS_RO_ATOMIC(foreign_eph_pampd_count);
+RAMSTER_SYSFS_RO(foreign_eph_pampd_count_max);
+RAMSTER_SYSFS_RO_ATOMIC(foreign_pers_pampd_count);
+RAMSTER_SYSFS_RO(foreign_pers_pampd_count_max);
+RAMSTER_SYSFS_RO_ATOMIC(curr_flnode_count);
+RAMSTER_SYSFS_RO(curr_flnode_count_max);
+
+#define MANUAL_NODES 8
+static bool ramster_nodes_manual_up[MANUAL_NODES];
+static ssize_t ramster_manual_node_up_show(struct kobject *kobj,
+                               struct kobj_attribute *attr, char *buf)
+{
+       int i;
+       char *p = buf;
+       for (i = 0; i < MANUAL_NODES; i++)
+               if (ramster_nodes_manual_up[i])
+                       p += sprintf(p, "%d ", i);
+       p += sprintf(p, "\n");
+       return p - buf;
+}
+
+static ssize_t ramster_manual_node_up_store(struct kobject *kobj,
+               struct kobj_attribute *attr, const char *buf, size_t count)
+{
+       int err;
+       unsigned long node_num;
+
+       err = kstrtoul(buf, 10, &node_num);
+       if (err) {
+               pr_err("ramster: bad strtoul?\n");
+               return -EINVAL;
+       }
+       if (node_num >= MANUAL_NODES) {
+               pr_err("ramster: bad node_num=%lu?\n", node_num);
+               return -EINVAL;
+       }
+       if (ramster_nodes_manual_up[node_num]) {
+               pr_err("ramster: node %d already up, ignoring\n",
+                                                       (int)node_num);
+       } else {
+               ramster_nodes_manual_up[node_num] = true;
+               r2net_hb_node_up_manual((int)node_num);
+       }
+       return count;
+}
+
+static struct kobj_attribute ramster_manual_node_up_attr = {
+       .attr = { .name = "manual_node_up", .mode = 0644 },
+       .show = ramster_manual_node_up_show,
+       .store = ramster_manual_node_up_store,
+};
+
+static ssize_t ramster_remote_target_nodenum_show(struct kobject *kobj,
+                               struct kobj_attribute *attr, char *buf)
+{
+       if (ramster_remote_target_nodenum == -1UL)
+               return sprintf(buf, "unset\n");
+       else
+               return sprintf(buf, "%d\n", ramster_remote_target_nodenum);
+}
+
+static ssize_t ramster_remote_target_nodenum_store(struct kobject *kobj,
+               struct kobj_attribute *attr, const char *buf, size_t count)
+{
+       int err;
+       unsigned long node_num;
+
+       err = kstrtoul(buf, 10, &node_num);
+       if (err) {
+               pr_err("ramster: bad strtoul?\n");
+               return -EINVAL;
+       } else if (node_num == -1UL) {
+               pr_err("ramster: disabling all remotification, "
+                       "data may still reside on remote nodes however\n");
+               return -EINVAL;
+       } else if (node_num >= MANUAL_NODES) {
+               pr_err("ramster: bad node_num=%lu?\n", node_num);
+               return -EINVAL;
+       } else if (!ramster_nodes_manual_up[node_num]) {
+               pr_err("ramster: node %d not up, ignoring setting "
+                       "of remotification target\n", (int)node_num);
+       } else if (r2net_remote_target_node_set((int)node_num) >= 0) {
+               pr_info("ramster: node %d set as remotification target\n",
+                               (int)node_num);
+               ramster_remote_target_nodenum = (int)node_num;
+       } else {
+               pr_err("ramster: bad num to node node_num=%d?\n",
+                               (int)node_num);
+               return -EINVAL;
+       }
+       return count;
+}
+
+static struct kobj_attribute ramster_remote_target_nodenum_attr = {
+       .attr = { .name = "remote_target_nodenum", .mode = 0644 },
+       .show = ramster_remote_target_nodenum_show,
+       .store = ramster_remote_target_nodenum_store,
+};
+
+
+static struct attribute *ramster_attrs[] = {
+       &ramster_interface_revision_attr.attr,
+       &ramster_pers_remotify_enable_attr.attr,
+       &ramster_eph_remotify_enable_attr.attr,
+       &ramster_remote_pers_pages_attr.attr,
+       &ramster_eph_pages_remoted_attr.attr,
+       &ramster_eph_pages_remote_failed_attr.attr,
+       &ramster_pers_pages_remoted_attr.attr,
+       &ramster_pers_pages_remote_failed_attr.attr,
+       &ramster_pers_pages_remote_nomem_attr.attr,
+       &ramster_remote_pages_flushed_attr.attr,
+       &ramster_remote_page_flushes_failed_attr.attr,
+       &ramster_remote_objects_flushed_attr.attr,
+       &ramster_remote_object_flushes_failed_attr.attr,
+       &ramster_remote_eph_pages_succ_get_attr.attr,
+       &ramster_remote_eph_pages_unsucc_get_attr.attr,
+       &ramster_remote_pers_pages_succ_get_attr.attr,
+       &ramster_remote_pers_pages_unsucc_get_attr.attr,
+       &ramster_foreign_eph_pampd_count_attr.attr,
+       &ramster_foreign_eph_pampd_count_max_attr.attr,
+       &ramster_foreign_pers_pampd_count_attr.attr,
+       &ramster_foreign_pers_pampd_count_max_attr.attr,
+       &ramster_curr_flnode_count_attr.attr,
+       &ramster_curr_flnode_count_max_attr.attr,
+       &ramster_manual_node_up_attr.attr,
+       &ramster_remote_target_nodenum_attr.attr,
+       NULL,
+};
+
+static struct attribute_group ramster_attr_group = {
+       .attrs = ramster_attrs,
+       .name = "ramster",
+};
+
 #endif /* CONFIG_SYSFS */
 /*
  * When zcache is disabled ("frozen"), pools can be created and destroyed,
@@ -1510,8 +2618,9 @@ static struct shrinker zcache_shrinker = {
  * zcache shims between cleancache/frontswap ops and tmem
  */
 
-static int zcache_put_page(int cli_id, int pool_id, struct tmem_oid *oidp,
-                               uint32_t index, struct page *page)
+int zcache_put(int cli_id, int pool_id, struct tmem_oid *oidp,
+                       uint32_t index, char *data, size_t size,
+                       bool raw, int ephemeral)
 {
        struct tmem_pool *pool;
        int ret = -1;
@@ -1522,8 +2631,7 @@ static int zcache_put_page(int cli_id, int pool_id, struct tmem_oid *oidp,
                goto out;
        if (!zcache_freeze && zcache_do_preload(pool) == 0) {
                /* preload does preempt_disable on success */
-               ret = tmem_put(pool, oidp, index, (char *)(page),
-                               PAGE_SIZE, 0, is_ephemeral(pool));
+               ret = tmem_put(pool, oidp, index, data, size, raw, ephemeral);
                if (ret < 0) {
                        if (is_ephemeral(pool))
                                zcache_failed_eph_puts++;
@@ -1543,27 +2651,38 @@ out:
        return ret;
 }
 
-static int zcache_get_page(int cli_id, int pool_id, struct tmem_oid *oidp,
-                               uint32_t index, struct page *page)
+int zcache_get(int cli_id, int pool_id, struct tmem_oid *oidp,
+                       uint32_t index, char *data, size_t *sizep,
+                       bool raw, int get_and_free)
 {
        struct tmem_pool *pool;
        int ret = -1;
-       unsigned long flags;
-       size_t size = PAGE_SIZE;
+       bool eph;
 
-       local_irq_save(flags);
+       if (!raw) {
+               BUG_ON(irqs_disabled());
+               BUG_ON(in_softirq());
+       }
        pool = zcache_get_pool_by_id(cli_id, pool_id);
+       eph = is_ephemeral(pool);
        if (likely(pool != NULL)) {
                if (atomic_read(&pool->obj_count) > 0)
-                       ret = tmem_get(pool, oidp, index, (char *)(page),
-                                       &size, 0, is_ephemeral(pool));
+                       ret = tmem_get(pool, oidp, index, data, sizep,
+                                       raw, get_and_free);
                zcache_put_pool(pool);
        }
-       local_irq_restore(flags);
+       WARN_ONCE((!eph && (ret != 0)), "zcache_get fails on persistent pool, "
+                         "bad things are very likely to happen soon\n");
+#ifdef RAMSTER_TESTING
+       if (ret != 0 && ret != -1 && !(ret == -EINVAL && is_ephemeral(pool)))
+               pr_err("TESTING zcache_get tmem_get returns ret=%d\n", ret);
+#endif
+       if (ret == -EAGAIN)
+               BUG(); /* FIXME... don't need this anymore??? let's ensure */
        return ret;
 }
 
-static int zcache_flush_page(int cli_id, int pool_id,
+int zcache_flush(int cli_id, int pool_id,
                                struct tmem_oid *oidp, uint32_t index)
 {
        struct tmem_pool *pool;
@@ -1573,6 +2692,7 @@ static int zcache_flush_page(int cli_id, int pool_id,
        local_irq_save(flags);
        zcache_flush_total++;
        pool = zcache_get_pool_by_id(cli_id, pool_id);
+       ramster_do_preload_flnode_only(pool);
        if (likely(pool != NULL)) {
                if (atomic_read(&pool->obj_count) > 0)
                        ret = tmem_flush_page(pool, oidp, index);
@@ -1584,8 +2704,7 @@ static int zcache_flush_page(int cli_id, int pool_id,
        return ret;
 }
 
-static int zcache_flush_object(int cli_id, int pool_id,
-                               struct tmem_oid *oidp)
+int zcache_flush_object(int cli_id, int pool_id, struct tmem_oid *oidp)
 {
        struct tmem_pool *pool;
        int ret = -1;
@@ -1594,6 +2713,7 @@ static int zcache_flush_object(int cli_id, int pool_id,
        local_irq_save(flags);
        zcache_flobj_total++;
        pool = zcache_get_pool_by_id(cli_id, pool_id);
+       ramster_do_preload_flnode_only(pool);
        if (likely(pool != NULL)) {
                if (atomic_read(&pool->obj_count) > 0)
                        ret = tmem_flush_object(pool, oidp);
@@ -1605,7 +2725,7 @@ static int zcache_flush_object(int cli_id, int pool_id,
        return ret;
 }
 
-static int zcache_destroy_pool(int cli_id, int pool_id)
+int zcache_client_destroy_pool(int cli_id, int pool_id)
 {
        struct tmem_pool *pool = NULL;
        struct zcache_client *cli = NULL;
@@ -1632,13 +2752,17 @@ static int zcache_destroy_pool(int cli_id, int pool_id)
        ret = tmem_destroy_pool(pool);
        local_bh_enable();
        kfree(pool);
-       pr_info("zcache: destroyed pool id=%d, cli_id=%d\n",
-                       pool_id, cli_id);
+       pr_info("ramster: destroyed pool id=%d cli_id=%d\n", pool_id, cli_id);
 out:
        return ret;
 }
 
-static int zcache_new_pool(uint16_t cli_id, uint32_t flags)
+static int zcache_destroy_pool(int pool_id)
+{
+       return zcache_client_destroy_pool(LOCAL_CLIENT, pool_id);
+}
+
+int zcache_new_pool(uint16_t cli_id, uint32_t flags)
 {
        int poolid = -1;
        struct tmem_pool *pool;
@@ -1653,7 +2777,7 @@ static int zcache_new_pool(uint16_t cli_id, uint32_t flags)
        atomic_inc(&cli->refcount);
        pool = kmalloc(sizeof(struct tmem_pool), GFP_ATOMIC);
        if (pool == NULL) {
-               pr_info("zcache: pool creation failed: out of memory\n");
+               pr_info("ramster: pool creation failed: out of memory\n");
                goto out;
        }
 
@@ -1661,7 +2785,7 @@ static int zcache_new_pool(uint16_t cli_id, uint32_t flags)
                if (cli->tmem_pools[poolid] == NULL)
                        break;
        if (poolid >= MAX_POOLS_PER_CLIENT) {
-               pr_info("zcache: pool creation failed: max exceeded\n");
+               pr_info("ramster: pool creation failed: max exceeded\n");
                kfree(pool);
                poolid = -1;
                goto out;
@@ -1671,15 +2795,78 @@ static int zcache_new_pool(uint16_t cli_id, uint32_t flags)
        pool->pool_id = poolid;
        tmem_new_pool(pool, flags);
        cli->tmem_pools[poolid] = pool;
-       pr_info("zcache: created %s tmem pool, id=%d, client=%d\n",
-               flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral",
-               poolid, cli_id);
+       if (cli_id == LOCAL_CLIENT)
+               pr_info("ramster: created %s tmem pool, id=%d, local client\n",
+                       flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral",
+                       poolid);
+       else
+               pr_info("ramster: created %s tmem pool, id=%d, client=%d\n",
+                       flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral",
+                       poolid, cli_id);
 out:
        if (cli != NULL)
                atomic_dec(&cli->refcount);
        return poolid;
 }
 
+static int zcache_local_new_pool(uint32_t flags)
+{
+       return zcache_new_pool(LOCAL_CLIENT, flags);
+}
+
+int zcache_autocreate_pool(int cli_id, int pool_id, bool ephemeral)
+{
+       struct tmem_pool *pool;
+       struct zcache_client *cli = NULL;
+       uint32_t flags = ephemeral ? 0 : TMEM_POOL_PERSIST;
+       int ret = -1;
+
+       if (cli_id == LOCAL_CLIENT)
+               goto out;
+       if (pool_id >= MAX_POOLS_PER_CLIENT)
+               goto out;
+       else if ((unsigned int)cli_id < MAX_CLIENTS)
+               cli = &zcache_clients[cli_id];
+       if ((ephemeral && !use_cleancache) || (!ephemeral && !use_frontswap))
+               BUG(); /* FIXME, handle more gracefully later */
+       if (!cli->allocated) {
+               if (zcache_new_client(cli_id))
+                       BUG(); /* FIXME, handle more gracefully later */
+               cli = &zcache_clients[cli_id];
+       }
+       atomic_inc(&cli->refcount);
+       pool = cli->tmem_pools[pool_id];
+       if (pool != NULL) {
+               if (pool->persistent && ephemeral) {
+                       pr_err("zcache_autocreate_pool: type mismatch\n");
+                       goto out;
+               }
+               ret = 0;
+               goto out;
+       }
+       pool = kmalloc(sizeof(struct tmem_pool), GFP_KERNEL);
+       if (pool == NULL) {
+               pr_info("ramster: pool creation failed: out of memory\n");
+               goto out;
+       }
+       atomic_set(&pool->refcount, 0);
+       pool->client = cli;
+       pool->pool_id = pool_id;
+       tmem_new_pool(pool, flags);
+       cli->tmem_pools[pool_id] = pool;
+       pr_info("ramster: AUTOcreated %s tmem poolid=%d, for remote client=%d\n",
+               flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral",
+               pool_id, cli_id);
+       ret = 0;
+out:
+       if (cli == NULL)
+               BUG(); /* FIXME, handle more gracefully later */
+               /* pr_err("zcache_autocreate_pool: failed\n"); */
+       if (cli != NULL)
+               atomic_dec(&cli->refcount);
+       return ret;
+}
+
 /**********
  * Two kernel functionalities currently can be layered on top of tmem.
  * These are "cleancache" which is used as a second-chance cache for clean
@@ -1696,8 +2883,18 @@ static void zcache_cleancache_put_page(int pool_id,
        u32 ind = (u32) index;
        struct tmem_oid oid = *(struct tmem_oid *)&key;
 
-       if (likely(ind == index))
-               (void)zcache_put_page(LOCAL_CLIENT, pool_id, &oid, index, page);
+#ifdef __PG_WAS_ACTIVE
+       if (!PageWasActive(page)) {
+               zcache_nonactive_puts++;
+               return;
+       }
+#endif
+       if (likely(ind == index)) {
+               char *kva = page_address(page);
+
+               (void)zcache_put(LOCAL_CLIENT, pool_id, &oid, index,
+                       kva, PAGE_SIZE, 0, 1);
+       }
 }
 
 static int zcache_cleancache_get_page(int pool_id,
@@ -1708,8 +2905,19 @@ static int zcache_cleancache_get_page(int pool_id,
        struct tmem_oid oid = *(struct tmem_oid *)&key;
        int ret = -1;
 
-       if (likely(ind == index))
-               ret = zcache_get_page(LOCAL_CLIENT, pool_id, &oid, index, page);
+       preempt_disable();
+       if (likely(ind == index)) {
+               char *kva = page_address(page);
+               size_t size = PAGE_SIZE;
+
+               ret = zcache_get(LOCAL_CLIENT, pool_id, &oid, index,
+                       kva, &size, 0, 0);
+#ifdef __PG_WAS_ACTIVE
+               if (ret == 0)
+                       SetPageWasActive(page);
+#endif
+       }
+       preempt_enable();
        return ret;
 }
 
@@ -1721,7 +2929,7 @@ static void zcache_cleancache_flush_page(int pool_id,
        struct tmem_oid oid = *(struct tmem_oid *)&key;
 
        if (likely(ind == index))
-               (void)zcache_flush_page(LOCAL_CLIENT, pool_id, &oid, ind);
+               (void)zcache_flush(LOCAL_CLIENT, pool_id, &oid, ind);
 }
 
 static void zcache_cleancache_flush_inode(int pool_id,
@@ -1735,7 +2943,7 @@ static void zcache_cleancache_flush_inode(int pool_id,
 static void zcache_cleancache_flush_fs(int pool_id)
 {
        if (pool_id >= 0)
-               (void)zcache_destroy_pool(LOCAL_CLIENT, pool_id);
+               (void)zcache_destroy_pool(pool_id);
 }
 
 static int zcache_cleancache_init_fs(size_t pagesize)
@@ -1743,7 +2951,7 @@ static int zcache_cleancache_init_fs(size_t pagesize)
        BUG_ON(sizeof(struct cleancache_filekey) !=
                                sizeof(struct tmem_oid));
        BUG_ON(pagesize != PAGE_SIZE);
-       return zcache_new_pool(LOCAL_CLIENT, 0);
+       return zcache_local_new_pool(0);
 }
 
 static int zcache_cleancache_init_shared_fs(char *uuid, size_t pagesize)
@@ -1752,7 +2960,7 @@ static int zcache_cleancache_init_shared_fs(char *uuid, size_t pagesize)
        BUG_ON(sizeof(struct cleancache_filekey) !=
                                sizeof(struct tmem_oid));
        BUG_ON(pagesize != PAGE_SIZE);
-       return zcache_new_pool(LOCAL_CLIENT, 0);
+       return zcache_local_new_pool(0);
 }
 
 static struct cleancache_ops zcache_cleancache_ops = {
@@ -1781,10 +2989,8 @@ static int zcache_frontswap_poolid = -1;
 /*
  * Swizzling increases objects per swaptype, increasing tmem concurrency
  * for heavy swaploads.  Later, larger nr_cpus -> larger SWIZ_BITS
- * Setting SWIZ_BITS to 27 basically reconstructs the swap entry from
- * frontswap_get_page()
  */
-#define SWIZ_BITS              27
+#define SWIZ_BITS              8
 #define SWIZ_MASK              ((1 << SWIZ_BITS) - 1)
 #define _oswiz(_type, _ind)    ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
 #define iswiz(_ind)            (_ind >> SWIZ_BITS)
@@ -1804,12 +3010,14 @@ static int zcache_frontswap_put_page(unsigned type, pgoff_t offset,
        struct tmem_oid oid = oswiz(type, ind);
        int ret = -1;
        unsigned long flags;
+       char *kva;
 
        BUG_ON(!PageLocked(page));
        if (likely(ind64 == ind)) {
                local_irq_save(flags);
-               ret = zcache_put_page(LOCAL_CLIENT, zcache_frontswap_poolid,
-                                       &oid, iswiz(ind), page);
+               kva = page_address(page);
+               ret = zcache_put(LOCAL_CLIENT, zcache_frontswap_poolid,
+                               &oid, iswiz(ind), kva, PAGE_SIZE, 0, 0);
                local_irq_restore(flags);
        }
        return ret;
@@ -1825,10 +3033,16 @@ static int zcache_frontswap_get_page(unsigned type, pgoff_t offset,
        struct tmem_oid oid = oswiz(type, ind);
        int ret = -1;
 
+       preempt_disable(); /* FIXME, remove this? */
        BUG_ON(!PageLocked(page));
-       if (likely(ind64 == ind))
-               ret = zcache_get_page(LOCAL_CLIENT, zcache_frontswap_poolid,
-                                       &oid, iswiz(ind), page);
+       if (likely(ind64 == ind)) {
+               char *kva = page_address(page);
+               size_t size = PAGE_SIZE;
+
+               ret = zcache_get(LOCAL_CLIENT, zcache_frontswap_poolid,
+                                       &oid, iswiz(ind), kva, &size, 0, -1);
+       }
+       preempt_enable(); /* FIXME, remove this? */
        return ret;
 }
 
@@ -1840,7 +3054,7 @@ static void zcache_frontswap_flush_page(unsigned type, pgoff_t offset)
        struct tmem_oid oid = oswiz(type, ind);
 
        if (likely(ind64 == ind))
-               (void)zcache_flush_page(LOCAL_CLIENT, zcache_frontswap_poolid,
+               (void)zcache_flush(LOCAL_CLIENT, zcache_frontswap_poolid,
                                        &oid, iswiz(ind));
 }
 
@@ -1862,7 +3076,7 @@ static void zcache_frontswap_init(unsigned ignored)
        /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
        if (zcache_frontswap_poolid < 0)
                zcache_frontswap_poolid =
-                       zcache_new_pool(LOCAL_CLIENT, TMEM_POOL_PERSIST);
+                               zcache_local_new_pool(TMEM_POOL_PERSIST);
 }
 
 static struct frontswap_ops zcache_frontswap_ops = {
@@ -1882,20 +3096,126 @@ struct frontswap_ops zcache_frontswap_register_ops(void)
 }
 #endif
 
+/*
+ * frontswap selfshrinking
+ */
+
+#ifdef CONFIG_FRONTSWAP
+/* In HZ, controls frequency of worker invocation. */
+static unsigned int selfshrink_interval __read_mostly = 5;
+
+static void selfshrink_process(struct work_struct *work);
+static DECLARE_DELAYED_WORK(selfshrink_worker, selfshrink_process);
+
+/* Enable/disable with sysfs. */
+static bool frontswap_selfshrinking __read_mostly;
+
+/* Enable/disable with kernel boot option. */
+static bool use_frontswap_selfshrink __initdata = true;
+
+/*
+ * The default values for the following parameters were deemed reasonable
+ * by experimentation, may be workload-dependent, and can all be
+ * adjusted via sysfs.
+ */
+
+/* Control rate for frontswap shrinking. Higher hysteresis is slower. */
+static unsigned int frontswap_hysteresis __read_mostly = 20;
+
+/*
+ * Number of selfshrink worker invocations to wait before observing that
+ * frontswap selfshrinking should commence. Note that selfshrinking does
+ * not use a separate worker thread.
+ */
+static unsigned int frontswap_inertia __read_mostly = 3;
+
+/* Countdown to next invocation of frontswap_shrink() */
+static unsigned long frontswap_inertia_counter;
+
+/*
+ * Invoked by the selfshrink worker thread, uses current number of pages
+ * in frontswap (frontswap_curr_pages()), previous status, and control
+ * values (hysteresis and inertia) to determine if frontswap should be
+ * shrunk and what the new frontswap size should be.  Note that
+ * frontswap_shrink is essentially a partial swapoff that immediately
+ * transfers pages from the "swap device" (frontswap) back into kernel
+ * RAM; despite the name, frontswap "shrinking" is very different from
+ * the "shrinker" interface used by the kernel MM subsystem to reclaim
+ * memory.
+ */
+static void frontswap_selfshrink(void)
+{
+       static unsigned long cur_frontswap_pages;
+       static unsigned long last_frontswap_pages;
+       static unsigned long tgt_frontswap_pages;
+
+       last_frontswap_pages = cur_frontswap_pages;
+       cur_frontswap_pages = frontswap_curr_pages();
+       if (!cur_frontswap_pages ||
+                       (cur_frontswap_pages > last_frontswap_pages)) {
+               frontswap_inertia_counter = frontswap_inertia;
+               return;
+       }
+       if (frontswap_inertia_counter && --frontswap_inertia_counter)
+               return;
+       if (cur_frontswap_pages <= frontswap_hysteresis)
+               tgt_frontswap_pages = 0;
+       else
+               tgt_frontswap_pages = cur_frontswap_pages -
+                       (cur_frontswap_pages / frontswap_hysteresis);
+       frontswap_shrink(tgt_frontswap_pages);
+}
+
+static int __init ramster_nofrontswap_selfshrink_setup(char *s)
+{
+       use_frontswap_selfshrink = false;
+       return 1;
+}
+
+__setup("noselfshrink", ramster_nofrontswap_selfshrink_setup);
+
+static void selfshrink_process(struct work_struct *work)
+{
+       if (frontswap_selfshrinking && frontswap_enabled) {
+               frontswap_selfshrink();
+               schedule_delayed_work(&selfshrink_worker,
+                       selfshrink_interval * HZ);
+       }
+}
+
+static int ramster_enabled;
+
+static int __init ramster_selfshrink_init(void)
+{
+       frontswap_selfshrinking = ramster_enabled && use_frontswap_selfshrink;
+       if (frontswap_selfshrinking)
+               pr_info("ramster: Initializing frontswap "
+                                       "selfshrinking driver.\n");
+       else
+               return -ENODEV;
+
+       schedule_delayed_work(&selfshrink_worker, selfshrink_interval * HZ);
+
+       return 0;
+}
+
+subsys_initcall(ramster_selfshrink_init);
+#endif
+
 /*
  * zcache initialization
- * NOTE FOR NOW zcache MUST BE PROVIDED AS A KERNEL BOOT PARAMETER OR
+ * NOTE FOR NOW ramster MUST BE PROVIDED AS A KERNEL BOOT PARAMETER OR
  * NOTHING HAPPENS!
  */
 
-static int zcache_enabled;
+static int ramster_enabled;
 
-static int __init enable_zcache(char *s)
+static int __init enable_ramster(char *s)
 {
-       zcache_enabled = 1;
+       ramster_enabled = 1;
        return 1;
 }
-__setup("zcache", enable_zcache);
+__setup("ramster", enable_ramster);
 
 /* allow independent dynamic disabling of cleancache and frontswap */
 
@@ -1903,16 +3223,22 @@ static int use_cleancache = 1;
 
 static int __init no_cleancache(char *s)
 {
+       pr_info("INIT no_cleancache called\n");
        use_cleancache = 0;
        return 1;
 }
 
-__setup("nocleancache", no_cleancache);
+/*
+ * FIXME: need to guarantee this gets checked before zcache_init is called
+ * What is the correct way to achieve this?
+ */
+early_param("nocleancache", no_cleancache);
 
 static int use_frontswap = 1;
 
 static int __init no_frontswap(char *s)
 {
+       pr_info("INIT no_frontswap called\n");
        use_frontswap = 0;
        return 1;
 }
@@ -1925,20 +3251,22 @@ static int __init zcache_init(void)
 
 #ifdef CONFIG_SYSFS
        ret = sysfs_create_group(mm_kobj, &zcache_attr_group);
+       ret = sysfs_create_group(mm_kobj, &ramster_attr_group);
        if (ret) {
-               pr_err("zcache: can't create sysfs\n");
+               pr_err("ramster: can't create sysfs\n");
                goto out;
        }
 #endif /* CONFIG_SYSFS */
 #if defined(CONFIG_CLEANCACHE) || defined(CONFIG_FRONTSWAP)
-       if (zcache_enabled) {
+       if (ramster_enabled) {
                unsigned int cpu;
 
+               (void)r2net_register_handlers();
                tmem_register_hostops(&zcache_hostops);
                tmem_register_pamops(&zcache_pamops);
                ret = register_cpu_notifier(&zcache_cpu_notifier_block);
                if (ret) {
-                       pr_err("zcache: can't register cpu notifier\n");
+                       pr_err("ramster: can't register cpu notifier\n");
                        goto out;
                }
                for_each_online_cpu(cpu) {
@@ -1951,35 +3279,39 @@ static int __init zcache_init(void)
                                sizeof(struct tmem_objnode), 0, 0, NULL);
        zcache_obj_cache = kmem_cache_create("zcache_obj",
                                sizeof(struct tmem_obj), 0, 0, NULL);
-       ret = zcache_new_client(LOCAL_CLIENT);
-       if (ret) {
-               pr_err("zcache: can't create client\n");
-               goto out;
-       }
+       ramster_flnode_cache = kmem_cache_create("ramster_flnode",
+                               sizeof(struct flushlist_node), 0, 0, NULL);
 #endif
 #ifdef CONFIG_CLEANCACHE
-       if (zcache_enabled && use_cleancache) {
+       pr_info("INIT ramster_enabled=%d use_cleancache=%d\n",
+                                       ramster_enabled, use_cleancache);
+       if (ramster_enabled && use_cleancache) {
                struct cleancache_ops old_ops;
 
                zbud_init();
                register_shrinker(&zcache_shrinker);
                old_ops = zcache_cleancache_register_ops();
-               pr_info("zcache: cleancache enabled using kernel "
+               pr_info("ramster: cleancache enabled using kernel "
                        "transcendent memory and compression buddies\n");
                if (old_ops.init_fs != NULL)
-                       pr_warning("zcache: cleancache_ops overridden");
+                       pr_warning("ramster: cleancache_ops overridden");
        }
 #endif
 #ifdef CONFIG_FRONTSWAP
-       if (zcache_enabled && use_frontswap) {
+       pr_info("INIT ramster_enabled=%d use_frontswap=%d\n",
+                                       ramster_enabled, use_frontswap);
+       if (ramster_enabled && use_frontswap) {
                struct frontswap_ops old_ops;
 
+               zcache_new_client(LOCAL_CLIENT);
                old_ops = zcache_frontswap_register_ops();
-               pr_info("zcache: frontswap enabled using kernel "
+               pr_info("ramster: frontswap enabled using kernel "
                        "transcendent memory and xvmalloc\n");
                if (old_ops.init != NULL)
-                       pr_warning("zcache: frontswap_ops overridden");
+                       pr_warning("ramster: frontswap_ops overridden");
        }
+       if (ramster_enabled && (use_frontswap || use_cleancache))
+               ramster_remotify_init();
 #endif
 out:
        return ret;