Whamcloud - gitweb
LU-2059 llog: MGC to use OSD API for backup logs
[fs/lustre-release.git] / lustre / obdclass / lu_object.c
index e430eb1..ba45a0d 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -28,6 +26,8 @@
 /*
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -43,9 +43,6 @@
  */
 
 #define DEBUG_SUBSYSTEM S_CLASS
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 
 #include <libcfs/libcfs.h>
 
@@ -60,9 +57,8 @@
 #include <lustre_disk.h>
 #include <lustre_fid.h>
 #include <lu_object.h>
+#include <lu_ref.h>
 #include <libcfs/list.h>
-/* lu_time_global_{init,fini}() */
-#include <lu_time.h>
 
 static void lu_object_free(const struct lu_env *env, struct lu_object *o);
 
@@ -78,11 +74,32 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o)
         struct lu_site          *site;
         struct lu_object        *orig;
         cfs_hash_bd_t            bd;
+       const struct lu_fid     *fid;
 
         top  = o->lo_header;
         site = o->lo_dev->ld_site;
         orig = o;
 
+       /*
+        * till we have full fids-on-OST implemented anonymous objects
+        * are possible in OSP. such an object isn't listed in the site
+        * so we should not remove it from the site.
+        */
+       fid = lu_object_fid(o);
+       if (fid_is_zero(fid)) {
+               LASSERT(top->loh_hash.next == NULL
+                       && top->loh_hash.pprev == NULL);
+               LASSERT(cfs_list_empty(&top->loh_lru));
+               if (!cfs_atomic_dec_and_test(&top->loh_ref))
+                       return;
+               cfs_list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
+                       if (o->lo_ops->loo_object_release != NULL)
+                               o->lo_ops->loo_object_release(env, o);
+               }
+               lu_object_free(env, orig);
+               return;
+       }
+
         cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
         bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
 
@@ -110,6 +127,8 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o)
         }
 
         if (!lu_object_is_dying(top)) {
+                LASSERT(cfs_list_empty(&top->loh_lru));
+                cfs_list_add_tail(&top->loh_lru, &bkt->lsb_lru);
                 cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
                 return;
         }
@@ -125,8 +144,8 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o)
          * and LRU lock, no race with concurrent object lookup is possible
          * and we can safely destroy object below.
          */
-        cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
-        cfs_list_del_init(&top->loh_lru);
+       if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags))
+               cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
         cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
         /*
          * Object was already removed from hash and lru above, can
@@ -137,6 +156,39 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o)
 EXPORT_SYMBOL(lu_object_put);
 
 /**
+ * Put object and don't keep in cache. This is temporary solution for
+ * multi-site objects when its layering is not constant.
+ */
+void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o)
+{
+       set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
+       return lu_object_put(env, o);
+}
+EXPORT_SYMBOL(lu_object_put_nocache);
+
+/**
+ * Kill the object and take it out of LRU cache.
+ * Currently used by client code for layout change.
+ */
+void lu_object_unhash(const struct lu_env *env, struct lu_object *o)
+{
+       struct lu_object_header *top;
+
+       top = o->lo_header;
+       set_bit(LU_OBJECT_HEARD_BANSHEE, &top->loh_flags);
+       if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags)) {
+               cfs_hash_t *obj_hash = o->lo_dev->ld_site->ls_obj_hash;
+               cfs_hash_bd_t bd;
+
+               cfs_hash_bd_get_and_lock(obj_hash, &top->loh_fid, &bd, 1);
+               cfs_list_del_init(&top->loh_lru);
+               cfs_hash_bd_del_locked(obj_hash, &bd, &top->loh_hash);
+               cfs_hash_bd_unlock(obj_hash, &bd, 1);
+       }
+}
+EXPORT_SYMBOL(lu_object_unhash);
+
+/**
  * Allocate new object.
  *
  * This follows object creation protocol, described in the comment within
@@ -154,18 +206,19 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env,
         int result;
         ENTRY;
 
-        /*
-         * Create top-level object slice. This will also create
-         * lu_object_header.
-         */
-        top = dev->ld_ops->ldo_object_alloc(env, NULL, dev);
-        if (top == NULL)
-                RETURN(ERR_PTR(-ENOMEM));
+       /*
+        * Create top-level object slice. This will also create
+        * lu_object_header.
+        */
+       top = dev->ld_ops->ldo_object_alloc(env, NULL, dev);
+       if (top == NULL)
+               RETURN(ERR_PTR(-ENOMEM));
+       if (IS_ERR(top))
+               RETURN(top);
         /*
          * This is the only place where object fid is assigned. It's constant
          * after this point.
          */
-        LASSERT(fid_is_igif(f) || fid_ver(f) == 0);
         top->lo_header->loh_fid = *f;
         layers = &top->lo_header->loh_layers;
         do {
@@ -265,6 +318,9 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
         int                      bnr;
         int                      i;
 
+       if (OBD_FAIL_CHECK(OBD_FAIL_OBD_NO_LRU))
+               RETURN(0);
+
         CFS_INIT_LIST_HEAD(&dispose);
         /*
          * Under LRU list lock, scan LRU list and move unreferenced objects to
@@ -282,20 +338,7 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 
                 cfs_list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
-                        /*
-                         * Objects are sorted in lru order, and "busy"
-                         * objects (ones with h->loh_ref > 0) naturally tend to
-                         * live near hot end that we scan last. Unfortunately,
-                         * sites usually have small (less then ten) number of
-                         * busy yet rarely accessed objects (some global
-                         * objects, accessed directly through pointers,
-                         * bypassing hash table).
-                         * Currently algorithm scans them over and over again.
-                         * Probably we should move busy objects out of LRU,
-                         * or we can live with that.
-                         */
-                        if (cfs_atomic_read(&h->loh_ref) > 0)
-                                continue;
+                        LASSERT(cfs_atomic_read(&h->loh_ref) == 0);
 
                         cfs_hash_bd_get(s->ls_obj_hash, &h->loh_fid, &bd2);
                         LASSERT(bd.bd_bucket == bd2.bd_bucket);
@@ -364,7 +407,7 @@ enum {
          *
          * XXX overflow is not handled correctly.
          */
-        LU_CDEBUG_LINE = 256
+        LU_CDEBUG_LINE = 512
 };
 
 struct lu_cdebug_data {
@@ -382,9 +425,10 @@ LU_KEY_INIT_FINI(lu_global, struct lu_cdebug_data);
  * lu_global_init().
  */
 struct lu_context_key lu_global_key = {
-        .lct_tags = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD,
-        .lct_init = lu_global_key_init,
-        .lct_fini = lu_global_key_fini
+       .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD |
+                   LCT_MG_THREAD | LCT_CL_THREAD | LCT_LOCAL,
+       .lct_init = lu_global_key_init,
+       .lct_fini = lu_global_key_fini
 };
 
 /**
@@ -393,8 +437,8 @@ struct lu_context_key lu_global_key = {
 int lu_cdebug_printer(const struct lu_env *env,
                       void *cookie, const char *format, ...)
 {
-        struct lu_cdebug_print_info *info = cookie;
-        struct lu_cdebug_data       *key;
+        struct libcfs_debug_msg_data *msgdata = cookie;
+        struct lu_cdebug_data        *key;
         int used;
         int complete;
         va_list args;
@@ -412,10 +456,8 @@ int lu_cdebug_printer(const struct lu_env *env,
         vsnprintf(key->lck_area + used,
                   ARRAY_SIZE(key->lck_area) - used, format, args);
         if (complete) {
-                if (cfs_cdebug_show(info->lpi_mask, info->lpi_subsys))
-                        libcfs_debug_msg(NULL, info->lpi_subsys, info->lpi_mask,
-                                         (char *)info->lpi_file, info->lpi_fn,
-                                         info->lpi_line, "%s", key->lck_area);
+                if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
+                        libcfs_debug_msg(msgdata, "%s", key->lck_area);
                 key->lck_area[0] = 0;
         }
         va_end(args);
@@ -498,21 +540,23 @@ static struct lu_object *htable_lookup(struct lu_site *s,
         __u64  ver = cfs_hash_bd_version_get(bd);
 
         if (*version == ver)
-                return NULL;
+               return ERR_PTR(-ENOENT);
 
         *version = ver;
         bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
-        /* cfs_hash_bd_lookup_intent is a somehow "internal" function
-         * of cfs_hash, but we don't want refcount on object right now */
-        hnode = cfs_hash_bd_lookup_locked(s->ls_obj_hash, bd, (void *)f);
+       /* cfs_hash_bd_peek_locked is a somehow "internal" function
+        * of cfs_hash, it doesn't add refcount on object. */
+       hnode = cfs_hash_bd_peek_locked(s->ls_obj_hash, bd, (void *)f);
         if (hnode == NULL) {
                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
-                return NULL;
+               return ERR_PTR(-ENOENT);
         }
 
         h = container_of0(hnode, struct lu_object_header, loh_hash);
         if (likely(!lu_object_is_dying(h))) {
+               cfs_hash_get(s->ls_obj_hash, hnode);
                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
+                cfs_list_del_init(&h->loh_lru);
                 return lu_object_top(h);
         }
 
@@ -521,7 +565,6 @@ static struct lu_object *htable_lookup(struct lu_site *s,
          * returned (to assure that references to dying objects are eventually
          * drained), and moreover, lookup has to wait until object is freed.
          */
-        cfs_atomic_dec(&h->loh_ref);
 
         cfs_waitlink_init(waiter);
         cfs_waitq_add(&bkt->lsb_marche_funebre, waiter);
@@ -530,6 +573,31 @@ static struct lu_object *htable_lookup(struct lu_site *s,
         return ERR_PTR(-EAGAIN);
 }
 
+static struct lu_object *htable_lookup_nowait(struct lu_site *s,
+                                             cfs_hash_bd_t *bd,
+                                             const struct lu_fid *f)
+{
+       cfs_hlist_node_t        *hnode;
+       struct lu_object_header *h;
+
+       /* cfs_hash_bd_peek_locked is a somehow "internal" function
+        * of cfs_hash, it doesn't add refcount on object. */
+       hnode = cfs_hash_bd_peek_locked(s->ls_obj_hash, bd, (void *)f);
+       if (hnode == NULL) {
+               lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
+               return ERR_PTR(-ENOENT);
+       }
+
+       h = container_of0(hnode, struct lu_object_header, loh_hash);
+       if (unlikely(lu_object_is_dying(h)))
+               return ERR_PTR(-ENOENT);
+
+       cfs_hash_get(s->ls_obj_hash, hnode);
+       lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
+       cfs_list_del_init(&h->loh_lru);
+       return lu_object_top(h);
+}
+
 /**
  * Search cache for an object with the fid \a f. If such object is found,
  * return it. Otherwise, create new object, insert it into cache and return
@@ -543,6 +611,29 @@ struct lu_object *lu_object_find(const struct lu_env *env,
 }
 EXPORT_SYMBOL(lu_object_find);
 
+static struct lu_object *lu_object_new(const struct lu_env *env,
+                                       struct lu_device *dev,
+                                       const struct lu_fid *f,
+                                       const struct lu_object_conf *conf)
+{
+        struct lu_object        *o;
+        cfs_hash_t              *hs;
+        cfs_hash_bd_t            bd;
+        struct lu_site_bkt_data *bkt;
+
+        o = lu_object_alloc(env, dev, f, conf);
+        if (unlikely(IS_ERR(o)))
+                return o;
+
+        hs = dev->ld_site->ls_obj_hash;
+        cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
+        bkt = cfs_hash_bd_extra_get(hs, &bd);
+        cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
+        bkt->lsb_busy++;
+        cfs_hash_bd_unlock(hs, &bd, 1);
+        return o;
+}
+
 /**
  * Core logic of lu_object_find*() functions.
  */
@@ -572,15 +663,22 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
          *     - unlock index;
          *     - return object.
          *
+         * For "LOC_F_NEW" case, we are sure the object is new established.
+         * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
+         * just alloc and insert directly.
+         *
          * If dying object is found during index search, add @waiter to the
          * site wait-queue and return ERR_PTR(-EAGAIN).
          */
+        if (conf != NULL && conf->loc_flags & LOC_F_NEW)
+                return lu_object_new(env, dev, f, conf);
+
         s  = dev->ld_site;
         hs = s->ls_obj_hash;
         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
         o = htable_lookup(s, &bd, f, waiter, &version);
         cfs_hash_bd_unlock(hs, &bd, 1);
-        if (o != NULL)
+       if (!IS_ERR(o) || PTR_ERR(o) != -ENOENT)
                 return o;
 
         /*
@@ -596,12 +694,11 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
         cfs_hash_bd_lock(hs, &bd, 1);
 
         shadow = htable_lookup(s, &bd, f, waiter, &version);
-        if (likely(shadow == NULL)) {
+       if (likely(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT)) {
                 struct lu_site_bkt_data *bkt;
 
                 bkt = cfs_hash_bd_extra_get(hs, &bd);
                 cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
-                cfs_list_add_tail(&o->lo_header->loh_lru, &bkt->lsb_lru);
                 bkt->lsb_busy++;
                 cfs_hash_bd_unlock(hs, &bd, 1);
                 return o;
@@ -643,6 +740,30 @@ struct lu_object *lu_object_find_at(const struct lu_env *env,
 EXPORT_SYMBOL(lu_object_find_at);
 
 /**
+ * Try to find the object in cache without waiting for the dead object
+ * to be released nor allocating object if no cached one was found.
+ *
+ * The found object will be set as LU_OBJECT_HEARD_BANSHEE for purging.
+ */
+void lu_object_purge(const struct lu_env *env, struct lu_device *dev,
+                    const struct lu_fid *f)
+{
+       struct lu_site          *s  = dev->ld_site;
+       cfs_hash_t              *hs = s->ls_obj_hash;
+       cfs_hash_bd_t            bd;
+       struct lu_object        *o;
+
+       cfs_hash_bd_get_and_lock(hs, f, &bd, 1);
+       o = htable_lookup_nowait(s, &bd, f);
+       cfs_hash_bd_unlock(hs, &bd, 1);
+       if (!IS_ERR(o)) {
+               set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
+               lu_object_put(env, o);
+       }
+}
+EXPORT_SYMBOL(lu_object_purge);
+
+/**
  * Find object with given fid, and return its slice belonging to given device.
  */
 struct lu_object *lu_object_find_slice(const struct lu_env *env,
@@ -671,20 +792,22 @@ static CFS_LIST_HEAD(lu_device_types);
 
 int lu_device_type_init(struct lu_device_type *ldt)
 {
-        int result;
+       int result = 0;
 
-        CFS_INIT_LIST_HEAD(&ldt->ldt_linkage);
-        result = ldt->ldt_ops->ldto_init(ldt);
-        if (result == 0)
-                cfs_list_add(&ldt->ldt_linkage, &lu_device_types);
-        return result;
+       CFS_INIT_LIST_HEAD(&ldt->ldt_linkage);
+       if (ldt->ldt_ops->ldto_init)
+               result = ldt->ldt_ops->ldto_init(ldt);
+       if (result == 0)
+               cfs_list_add(&ldt->ldt_linkage, &lu_device_types);
+       return result;
 }
 EXPORT_SYMBOL(lu_device_type_init);
 
 void lu_device_type_fini(struct lu_device_type *ldt)
 {
-        cfs_list_del_init(&ldt->ldt_linkage);
-        ldt->ldt_ops->ldto_fini(ldt);
+       cfs_list_del_init(&ldt->ldt_linkage);
+       if (ldt->ldt_ops->ldto_fini)
+               ldt->ldt_ops->ldto_fini(ldt);
 }
 EXPORT_SYMBOL(lu_device_type_fini);
 
@@ -692,10 +815,10 @@ void lu_types_stop(void)
 {
         struct lu_device_type *ldt;
 
-        cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
-                if (ldt->ldt_device_nr == 0)
-                        ldt->ldt_ops->ldto_stop(ldt);
-        }
+       cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
+               if (ldt->ldt_device_nr == 0 && ldt->ldt_ops->ldto_stop)
+                       ldt->ldt_ops->ldto_stop(ldt);
+       }
 }
 EXPORT_SYMBOL(lu_types_stop);
 
@@ -703,7 +826,7 @@ EXPORT_SYMBOL(lu_types_stop);
  * Global list of all sites on this node
  */
 static CFS_LIST_HEAD(lu_sites);
-static CFS_DECLARE_MUTEX(lu_sites_guard);
+static DEFINE_MUTEX(lu_sites_guard);
 
 /**
  * Global environment used by site shrinker.
@@ -754,9 +877,14 @@ void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
 EXPORT_SYMBOL(lu_site_print);
 
 enum {
-        LU_CACHE_PERCENT   = 20,
+        LU_CACHE_PERCENT_MAX     = 50,
+        LU_CACHE_PERCENT_DEFAULT = 20
 };
 
+static unsigned int lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
+CFS_MODULE_PARM(lu_cache_percent, "i", int, 0644,
+                "Percentage of memory to be used as lu_object cache");
+
 /**
  * Return desired hash table order.
  */
@@ -772,16 +900,25 @@ static int lu_htable_order(void)
          *
          * Size of lu_object is (arbitrary) taken as 1K (together with inode).
          */
-        cache_size = cfs_num_physpages;
+       cache_size = num_physpages;
 
 #if BITS_PER_LONG == 32
         /* limit hashtable size for lowmem systems to low RAM */
-        if (cache_size > 1 << (30 - CFS_PAGE_SHIFT))
-                cache_size = 1 << (30 - CFS_PAGE_SHIFT) * 3 / 4;
+       if (cache_size > 1 << (30 - PAGE_CACHE_SHIFT))
+               cache_size = 1 << (30 - PAGE_CACHE_SHIFT) * 3 / 4;
 #endif
 
-        cache_size = cache_size / 100 * LU_CACHE_PERCENT *
-                (CFS_PAGE_SIZE / 1024);
+        /* clear off unreasonable cache setting. */
+        if (lu_cache_percent == 0 || lu_cache_percent > LU_CACHE_PERCENT_MAX) {
+                CWARN("obdclass: invalid lu_cache_percent: %u, it must be in"
+                      " the range of (0, %u]. Will use default value: %u.\n",
+                      lu_cache_percent, LU_CACHE_PERCENT_MAX,
+                      LU_CACHE_PERCENT_DEFAULT);
+
+                lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
+        }
+        cache_size = cache_size / 100 * lu_cache_percent *
+               (PAGE_CACHE_SIZE / 1024);
 
         for (bits = 1; (1 << bits) < cache_size; ++bits) {
                 ;
@@ -793,10 +930,18 @@ static unsigned lu_obj_hop_hash(cfs_hash_t *hs,
                                 const void *key, unsigned mask)
 {
         struct lu_fid  *fid = (struct lu_fid *)key;
-        unsigned        hash;
+        __u32           hash;
+
+        hash = fid_flatten32(fid);
+        hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
+        hash = cfs_hash_long(hash, hs->hs_bkt_bits);
+
+        /* give me another random factor */
+        hash -= cfs_hash_long((unsigned long)hs, fid_oid(fid) % 11 + 3);
+
+        hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
+        hash |= (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
 
-        hash = (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
-        hash += fid_hash(fid, hs->hs_bkt_bits) << hs->hs_bkt_bits;
         return hash & mask;
 }
 
@@ -850,31 +995,50 @@ cfs_hash_ops_t lu_site_hash_ops = {
         .hs_put_locked  = lu_obj_hop_put_locked,
 };
 
+void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d)
+{
+       spin_lock(&s->ls_ld_lock);
+       if (cfs_list_empty(&d->ld_linkage))
+               cfs_list_add(&d->ld_linkage, &s->ls_ld_linkage);
+       spin_unlock(&s->ls_ld_lock);
+}
+EXPORT_SYMBOL(lu_dev_add_linkage);
+
+void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d)
+{
+       spin_lock(&s->ls_ld_lock);
+       cfs_list_del_init(&d->ld_linkage);
+       spin_unlock(&s->ls_ld_lock);
+}
+EXPORT_SYMBOL(lu_dev_del_linkage);
+
 /**
  * Initialize site \a s, with \a d as the top level device.
  */
 #define LU_SITE_BITS_MIN    12
-#define LU_SITE_BITS_MAX    23
+#define LU_SITE_BITS_MAX    24
 /**
- * total 128 buckets, we don't want too many buckets because:
+ * total 256 buckets, we don't want too many buckets because:
  * - consume too much memory
  * - avoid unbalanced LRU list
  */
-#define LU_SITE_BKT_BITS    7
+#define LU_SITE_BKT_BITS    8
 
 int lu_site_init(struct lu_site *s, struct lu_device *top)
 {
         struct lu_site_bkt_data *bkt;
         cfs_hash_bd_t bd;
+        char name[16];
         int bits;
         int i;
         ENTRY;
 
         memset(s, 0, sizeof *s);
         bits = lu_htable_order();
+        snprintf(name, 16, "lu_site_%s", top->ld_type->ldt_name);
         for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX);
              bits >= LU_SITE_BITS_MIN; bits--) {
-                s->ls_obj_hash = cfs_hash_create("lu_site", bits, bits,
+                s->ls_obj_hash = cfs_hash_create(name, bits, bits,
                                                  bits - LU_SITE_BKT_BITS,
                                                  sizeof(*bkt), 0, 0,
                                                  &lu_site_hash_ops,
@@ -923,7 +1087,12 @@ int lu_site_init(struct lu_site *s, struct lu_device *top)
         lu_device_get(top);
         lu_ref_add(&top->ld_reference, "site-top", s);
 
-        RETURN(0);
+        CFS_INIT_LIST_HEAD(&s->ls_ld_linkage);
+       spin_lock_init(&s->ls_ld_lock);
+
+       lu_dev_add_linkage(s, top);
+
+       RETURN(0);
 }
 EXPORT_SYMBOL(lu_site_init);
 
@@ -932,9 +1101,9 @@ EXPORT_SYMBOL(lu_site_init);
  */
 void lu_site_fini(struct lu_site *s)
 {
-        cfs_down(&lu_sites_guard);
+       mutex_lock(&lu_sites_guard);
         cfs_list_del_init(&s->ls_linkage);
-        cfs_up(&lu_sites_guard);
+       mutex_unlock(&lu_sites_guard);
 
         if (s->ls_obj_hash != NULL) {
                 cfs_hash_putref(s->ls_obj_hash);
@@ -959,11 +1128,11 @@ EXPORT_SYMBOL(lu_site_fini);
 int lu_site_init_finish(struct lu_site *s)
 {
         int result;
-        cfs_down(&lu_sites_guard);
+       mutex_lock(&lu_sites_guard);
         result = lu_context_refill(&lu_shrink_env.le_ctx);
         if (result == 0)
                 cfs_list_add(&s->ls_linkage, &lu_sites);
-        cfs_up(&lu_sites_guard);
+       mutex_unlock(&lu_sites_guard);
         return result;
 }
 EXPORT_SYMBOL(lu_site_init_finish);
@@ -998,6 +1167,7 @@ int lu_device_init(struct lu_device *d, struct lu_device_type *t)
         cfs_atomic_set(&d->ld_ref, 0);
         d->ld_type = t;
         lu_ref_init(&d->ld_reference);
+        CFS_INIT_LIST_HEAD(&d->ld_linkage);
         return 0;
 }
 EXPORT_SYMBOL(lu_device_init);
@@ -1028,16 +1198,17 @@ EXPORT_SYMBOL(lu_device_fini);
  * Initialize object \a o that is part of compound object \a h and was created
  * by device \a d.
  */
-int lu_object_init(struct lu_object *o,
-                   struct lu_object_header *h, struct lu_device *d)
+int lu_object_init(struct lu_object *o, struct lu_object_header *h,
+                  struct lu_device *d)
 {
-        memset(o, 0, sizeof *o);
-        o->lo_header = h;
-        o->lo_dev    = d;
-        lu_device_get(d);
-        o->lo_dev_ref = lu_ref_add(&d->ld_reference, "lu_object", o);
-        CFS_INIT_LIST_HEAD(&o->lo_linkage);
-        return 0;
+       memset(o, 0, sizeof(*o));
+       o->lo_header = h;
+       o->lo_dev = d;
+       lu_device_get(d);
+       lu_ref_add_at(&d->ld_reference, &o->lo_dev_ref, "lu_object", o);
+       CFS_INIT_LIST_HEAD(&o->lo_linkage);
+
+       return 0;
 }
 EXPORT_SYMBOL(lu_object_init);
 
@@ -1046,16 +1217,16 @@ EXPORT_SYMBOL(lu_object_init);
  */
 void lu_object_fini(struct lu_object *o)
 {
-        struct lu_device *dev = o->lo_dev;
+       struct lu_device *dev = o->lo_dev;
 
-        LASSERT(cfs_list_empty(&o->lo_linkage));
+       LASSERT(cfs_list_empty(&o->lo_linkage));
 
-        if (dev != NULL) {
-                lu_ref_del_at(&dev->ld_reference,
-                              o->lo_dev_ref , "lu_object", o);
-                lu_device_put(dev);
-                o->lo_dev = NULL;
-        }
+       if (dev != NULL) {
+               lu_ref_del_at(&dev->ld_reference, &o->lo_dev_ref,
+                             "lu_object", o);
+               lu_device_put(dev);
+               o->lo_dev = NULL;
+       }
 }
 EXPORT_SYMBOL(lu_object_fini);
 
@@ -1152,15 +1323,6 @@ void lu_stack_fini(const struct lu_env *env, struct lu_device *top)
         /* purge again. */
         lu_site_purge(env, site, ~0);
 
-        if (!cfs_hash_is_empty(site->ls_obj_hash)) {
-                /*
-                 * Uh-oh, objects still exist.
-                 */
-                static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR);
-
-                lu_site_print(env, site, &cookie, lu_cdebug_printer);
-        }
-
         for (scan = top; scan != NULL; scan = next) {
                 const struct lu_device_type *ldt = scan->ld_type;
                 struct obd_type             *type;
@@ -1179,12 +1341,12 @@ enum {
         /**
          * Maximal number of tld slots.
          */
-        LU_CONTEXT_KEY_NR = 32
+        LU_CONTEXT_KEY_NR = 40
 };
 
 static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
 
-static cfs_spinlock_t lu_keys_guard = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(lu_keys_guard);
 
 /**
  * Global counter incremented whenever key is registered, unregistered,
@@ -1208,7 +1370,7 @@ int lu_context_key_register(struct lu_context_key *key)
         LASSERT(key->lct_owner != NULL);
 
         result = -ENFILE;
-        cfs_spin_lock(&lu_keys_guard);
+       spin_lock(&lu_keys_guard);
         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
                 if (lu_keys[i] == NULL) {
                         key->lct_index = i;
@@ -1220,8 +1382,8 @@ int lu_context_key_register(struct lu_context_key *key)
                         break;
                 }
         }
-        cfs_spin_unlock(&lu_keys_guard);
-        return result;
+       spin_unlock(&lu_keys_guard);
+       return result;
 }
 EXPORT_SYMBOL(lu_context_key_register);
 
@@ -1238,13 +1400,14 @@ static void key_fini(struct lu_context *ctx, int index)
                 key->lct_fini(ctx, key, ctx->lc_value[index]);
                 lu_ref_del(&key->lct_reference, "ctx", ctx);
                 cfs_atomic_dec(&key->lct_used);
-                LASSERT(key->lct_owner != NULL);
-                if (!(ctx->lc_tags & LCT_NOREF)) {
-                        LASSERT(cfs_module_refcount(key->lct_owner) > 0);
-                        cfs_module_put(key->lct_owner);
-                }
-                ctx->lc_value[index] = NULL;
-        }
+
+               LASSERT(key->lct_owner != NULL);
+               if ((ctx->lc_tags & LCT_NOREF) == 0) {
+                       LINVRNT(cfs_module_refcount(key->lct_owner) > 0);
+                       cfs_module_put(key->lct_owner);
+               }
+               ctx->lc_value[index] = NULL;
+       }
 }
 
 /**
@@ -1252,23 +1415,23 @@ static void key_fini(struct lu_context *ctx, int index)
  */
 void lu_context_key_degister(struct lu_context_key *key)
 {
-        LASSERT(cfs_atomic_read(&key->lct_used) >= 1);
-        LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
+       LASSERT(cfs_atomic_read(&key->lct_used) >= 1);
+       LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
 
-        lu_context_key_quiesce(key);
+       lu_context_key_quiesce(key);
 
-        ++key_set_version;
-        cfs_spin_lock(&lu_keys_guard);
-        key_fini(&lu_shrink_env.le_ctx, key->lct_index);
-        if (lu_keys[key->lct_index]) {
-                lu_keys[key->lct_index] = NULL;
-                lu_ref_fini(&key->lct_reference);
-        }
-        cfs_spin_unlock(&lu_keys_guard);
+       ++key_set_version;
+       spin_lock(&lu_keys_guard);
+       key_fini(&lu_shrink_env.le_ctx, key->lct_index);
+       if (lu_keys[key->lct_index]) {
+               lu_keys[key->lct_index] = NULL;
+               lu_ref_fini(&key->lct_reference);
+       }
+       spin_unlock(&lu_keys_guard);
 
-        LASSERTF(cfs_atomic_read(&key->lct_used) == 1,
-                 "key has instances: %d\n",
-                 cfs_atomic_read(&key->lct_used));
+       LASSERTF(cfs_atomic_read(&key->lct_used) == 1,
+                "key has instances: %d\n",
+                cfs_atomic_read(&key->lct_used));
 }
 EXPORT_SYMBOL(lu_context_key_degister);
 
@@ -1379,24 +1542,22 @@ static CFS_LIST_HEAD(lu_context_remembered);
 void lu_context_key_quiesce(struct lu_context_key *key)
 {
         struct lu_context *ctx;
-        extern unsigned cl_env_cache_purge(unsigned nr);
 
         if (!(key->lct_tags & LCT_QUIESCENT)) {
                 /*
                  * XXX layering violation.
                  */
-                cl_env_cache_purge(~0);
                 key->lct_tags |= LCT_QUIESCENT;
                 /*
                  * XXX memory barrier has to go here.
                  */
-                cfs_spin_lock(&lu_keys_guard);
-                cfs_list_for_each_entry(ctx, &lu_context_remembered,
-                                        lc_remember)
-                        key_fini(ctx, key->lct_index);
-                cfs_spin_unlock(&lu_keys_guard);
-                ++key_set_version;
-        }
+               spin_lock(&lu_keys_guard);
+               cfs_list_for_each_entry(ctx, &lu_context_remembered,
+                                       lc_remember)
+                       key_fini(ctx, key->lct_index);
+               spin_unlock(&lu_keys_guard);
+               ++key_set_version;
+       }
 }
 EXPORT_SYMBOL(lu_context_key_quiesce);
 
@@ -1409,23 +1570,23 @@ EXPORT_SYMBOL(lu_context_key_revive);
 
 static void keys_fini(struct lu_context *ctx)
 {
-        int i;
+       int     i;
 
-        cfs_spin_lock(&lu_keys_guard);
-        if (ctx->lc_value != NULL) {
-                for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
-                        key_fini(ctx, i);
-                OBD_FREE(ctx->lc_value,
-                         ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
-                ctx->lc_value = NULL;
-        }
-        cfs_spin_unlock(&lu_keys_guard);
+       if (ctx->lc_value == NULL)
+               return;
+
+       for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
+               key_fini(ctx, i);
+
+       OBD_FREE(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
+       ctx->lc_value = NULL;
 }
 
 static int keys_fill(struct lu_context *ctx)
 {
         int i;
 
+        LINVRNT(ctx->lc_value != NULL);
         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
                 struct lu_context_key *key;
 
@@ -1467,17 +1628,11 @@ static int keys_fill(struct lu_context *ctx)
 
 static int keys_init(struct lu_context *ctx)
 {
-        int result;
-
-        OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
-        if (likely(ctx->lc_value != NULL))
-                result = keys_fill(ctx);
-        else
-                result = -ENOMEM;
+       OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
+       if (likely(ctx->lc_value != NULL))
+               return keys_fill(ctx);
 
-        if (result != 0)
-                keys_fini(ctx);
-        return result;
+       return -ENOMEM;
 }
 
 /**
@@ -1485,16 +1640,24 @@ static int keys_init(struct lu_context *ctx)
  */
 int lu_context_init(struct lu_context *ctx, __u32 tags)
 {
-        memset(ctx, 0, sizeof *ctx);
-        ctx->lc_state = LCS_INITIALIZED;
-        ctx->lc_tags = tags;
-        if (tags & LCT_REMEMBER) {
-                cfs_spin_lock(&lu_keys_guard);
-                cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
-                cfs_spin_unlock(&lu_keys_guard);
-        } else
-                CFS_INIT_LIST_HEAD(&ctx->lc_remember);
-        return keys_init(ctx);
+       int     rc;
+
+       memset(ctx, 0, sizeof *ctx);
+       ctx->lc_state = LCS_INITIALIZED;
+       ctx->lc_tags = tags;
+       if (tags & LCT_REMEMBER) {
+               spin_lock(&lu_keys_guard);
+               cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
+               spin_unlock(&lu_keys_guard);
+       } else {
+               CFS_INIT_LIST_HEAD(&ctx->lc_remember);
+       }
+
+       rc = keys_init(ctx);
+       if (rc != 0)
+               lu_context_fini(ctx);
+
+       return rc;
 }
 EXPORT_SYMBOL(lu_context_init);
 
@@ -1503,12 +1666,19 @@ EXPORT_SYMBOL(lu_context_init);
  */
 void lu_context_fini(struct lu_context *ctx)
 {
-        LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
-        ctx->lc_state = LCS_FINALIZED;
-        keys_fini(ctx);
-        cfs_spin_lock(&lu_keys_guard);
-        cfs_list_del_init(&ctx->lc_remember);
-        cfs_spin_unlock(&lu_keys_guard);
+       LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
+       ctx->lc_state = LCS_FINALIZED;
+
+       if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
+               LASSERT(cfs_list_empty(&ctx->lc_remember));
+               keys_fini(ctx);
+
+       } else { /* could race with key degister */
+               spin_lock(&lu_keys_guard);
+               keys_fini(ctx);
+               cfs_list_del_init(&ctx->lc_remember);
+               spin_unlock(&lu_keys_guard);
+       }
 }
 EXPORT_SYMBOL(lu_context_fini);
 
@@ -1549,15 +1719,61 @@ EXPORT_SYMBOL(lu_context_exit);
 
 /**
  * Allocate for context all missing keys that were registered after context
- * creation.
+ * creation. key_set_version is only changed in rare cases when modules
+ * are loaded and removed.
  */
 int lu_context_refill(struct lu_context *ctx)
 {
-        LINVRNT(ctx->lc_value != NULL);
-        return ctx->lc_version == key_set_version ? 0 : keys_fill(ctx);
+        return likely(ctx->lc_version == key_set_version) ? 0 : keys_fill(ctx);
 }
 EXPORT_SYMBOL(lu_context_refill);
 
+/**
+ * lu_ctx_tags/lu_ses_tags will be updated if there are new types of
+ * obd being added. Currently, this is only used on client side, specifically
+ * for echo device client, for other stack (like ptlrpc threads), context are
+ * predefined when the lu_device type are registered, during the module probe
+ * phase.
+ */
+__u32 lu_context_tags_default = 0;
+__u32 lu_session_tags_default = 0;
+
+void lu_context_tags_update(__u32 tags)
+{
+       spin_lock(&lu_keys_guard);
+       lu_context_tags_default |= tags;
+       key_set_version++;
+       spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_context_tags_update);
+
+void lu_context_tags_clear(__u32 tags)
+{
+       spin_lock(&lu_keys_guard);
+       lu_context_tags_default &= ~tags;
+       key_set_version++;
+       spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_context_tags_clear);
+
+void lu_session_tags_update(__u32 tags)
+{
+       spin_lock(&lu_keys_guard);
+       lu_session_tags_default |= tags;
+       key_set_version++;
+       spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_session_tags_update);
+
+void lu_session_tags_clear(__u32 tags)
+{
+       spin_lock(&lu_keys_guard);
+       lu_session_tags_default &= ~tags;
+       key_set_version++;
+       spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_session_tags_clear);
+
 int lu_env_init(struct lu_env *env, __u32 tags)
 {
         int result;
@@ -1589,7 +1805,35 @@ int lu_env_refill(struct lu_env *env)
 }
 EXPORT_SYMBOL(lu_env_refill);
 
-static struct cfs_shrinker *lu_site_shrinker = NULL;
+/**
+ * Currently, this API will only be used by echo client.
+ * Because echo client and normal lustre client will share
+ * same cl_env cache. So echo client needs to refresh
+ * the env context after it get one from the cache, especially
+ * when normal client and echo client co-exist in the same client.
+ */
+int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags,
+                          __u32 stags)
+{
+        int    result;
+
+        if ((env->le_ctx.lc_tags & ctags) != ctags) {
+                env->le_ctx.lc_version = 0;
+                env->le_ctx.lc_tags |= ctags;
+        }
+
+        if (env->le_ses && (env->le_ses->lc_tags & stags) != stags) {
+                env->le_ses->lc_version = 0;
+                env->le_ses->lc_tags |= stags;
+        }
+
+        result = lu_env_refill(env);
+
+        return result;
+}
+EXPORT_SYMBOL(lu_env_refill_by_tags);
+
+static struct shrinker *lu_site_shrinker;
 
 typedef struct lu_site_stats{
         unsigned        lss_populated;
@@ -1627,25 +1871,57 @@ static void lu_site_stats_get(cfs_hash_t *hs,
 }
 
 #ifdef __KERNEL__
-static int lu_cache_shrink(SHRINKER_FIRST_ARG int nr_to_scan,
-                           unsigned int gfp_mask)
+
+/*
+ * There exists a potential lock inversion deadlock scenario when using
+ * Lustre on top of ZFS. This occurs between one of ZFS's
+ * buf_hash_table.ht_lock's, and Lustre's lu_sites_guard lock. Essentially,
+ * thread A will take the lu_sites_guard lock and sleep on the ht_lock,
+ * while thread B will take the ht_lock and sleep on the lu_sites_guard
+ * lock. Obviously neither thread will wake and drop their respective hold
+ * on their lock.
+ *
+ * To prevent this from happening we must ensure the lu_sites_guard lock is
+ * not taken while down this code path. ZFS reliably does not set the
+ * __GFP_FS bit in its code paths, so this can be used to determine if it
+ * is safe to take the lu_sites_guard lock.
+ *
+ * Ideally we should accurately return the remaining number of cached
+ * objects without taking the  lu_sites_guard lock, but this is not
+ * possible in the current implementation.
+ */
+static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
 {
         lu_site_stats_t stats;
         struct lu_site *s;
         struct lu_site *tmp;
         int cached = 0;
-        int remain = nr_to_scan;
+        int remain = shrink_param(sc, nr_to_scan);
         CFS_LIST_HEAD(splice);
 
-        if (nr_to_scan != 0) {
-                if (!(gfp_mask & __GFP_FS))
+       if (!(shrink_param(sc, gfp_mask) & __GFP_FS)) {
+               if (remain != 0)
                         return -1;
-                CDEBUG(D_INODE, "Shrink %d objects\n", nr_to_scan);
+               else
+                       /* We must not take the lu_sites_guard lock when
+                        * __GFP_FS is *not* set because of the deadlock
+                        * possibility detailed above. Additionally,
+                        * since we cannot determine the number of
+                        * objects in the cache without taking this
+                        * lock, we're in a particularly tough spot. As
+                        * a result, we'll just lie and say our cache is
+                        * empty. This _should_ be ok, as we can't
+                        * reclaim objects when __GFP_FS is *not* set
+                        * anyways.
+                        */
+                       return 0;
         }
 
-        cfs_down(&lu_sites_guard);
+       CDEBUG(D_INODE, "Shrink %d objects\n", remain);
+
+       mutex_lock(&lu_sites_guard);
         cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
-                if (nr_to_scan != 0) {
+                if (shrink_param(sc, nr_to_scan) != 0) {
                         remain = lu_site_purge(&lu_shrink_env, s, remain);
                         /*
                          * Move just shrunk site to the tail of site list to
@@ -1657,14 +1933,14 @@ static int lu_cache_shrink(SHRINKER_FIRST_ARG int nr_to_scan,
                 memset(&stats, 0, sizeof(stats));
                 lu_site_stats_get(s->ls_obj_hash, &stats, 0);
                 cached += stats.lss_total - stats.lss_busy;
-                if (nr_to_scan && remain <= 0)
+                if (shrink_param(sc, nr_to_scan) && remain <= 0)
                         break;
         }
         cfs_list_splice(&splice, lu_sites.prev);
-        cfs_up(&lu_sites_guard);
+       mutex_unlock(&lu_sites_guard);
 
         cached = (cached / 100) * sysctl_vfs_cache_pressure;
-        if (nr_to_scan == 0)
+        if (shrink_param(sc, nr_to_scan) == 0)
                 CDEBUG(D_INODE, "%d objects cached\n", cached);
         return cached;
 }
@@ -1692,9 +1968,9 @@ int lu_printk_printer(const struct lu_env *env,
         return 0;
 }
 
-void lu_debugging_setup(void)
+int lu_debugging_setup(void)
 {
-        lu_env_init(&lu_debugging_env, ~0);
+       return lu_env_init(&lu_debugging_env, ~0);
 }
 
 void lu_context_keys_dump(void)
@@ -1724,17 +2000,6 @@ static int lu_cache_shrink(int nr, unsigned int gfp_mask)
 }
 #endif /* __KERNEL__ */
 
-int  cl_global_init(void);
-void cl_global_fini(void);
-int  lu_ref_global_init(void);
-void lu_ref_global_fini(void);
-
-int dt_global_init(void);
-void dt_global_fini(void);
-
-int llo_global_init(void);
-void llo_global_fini(void);
-
 /**
  * Initialization of global lu_* data.
  */
@@ -1742,7 +2007,7 @@ int lu_global_init(void)
 {
         int result;
 
-        CDEBUG(D_CONSOLE, "Lustre LU module (%p).\n", &lu_keys);
+        CDEBUG(D_INFO, "Lustre LU module (%p).\n", &lu_keys);
 
         result = lu_ref_global_init();
         if (result != 0)
@@ -1752,14 +2017,15 @@ int lu_global_init(void)
         result = lu_context_key_register(&lu_global_key);
         if (result != 0)
                 return result;
+
         /*
          * At this level, we don't know what tags are needed, so allocate them
          * conservatively. This should not be too bad, because this
          * environment is global.
          */
-        cfs_down(&lu_sites_guard);
+       mutex_lock(&lu_sites_guard);
         result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
-        cfs_up(&lu_sites_guard);
+       mutex_unlock(&lu_sites_guard);
         if (result != 0)
                 return result;
 
@@ -1768,26 +2034,10 @@ int lu_global_init(void)
          * inode, one for ea. Unfortunately setting this high value results in
          * lu_object/inode cache consuming all the memory.
          */
-        lu_site_shrinker = cfs_set_shrinker(CFS_DEFAULT_SEEKS, lu_cache_shrink);
+       lu_site_shrinker = set_shrinker(DEFAULT_SEEKS, lu_cache_shrink);
         if (lu_site_shrinker == NULL)
                 return -ENOMEM;
 
-        result = lu_time_global_init();
-        if (result)
-                GOTO(out, result);
-
-#ifdef __KERNEL__
-        result = dt_global_init();
-        if (result)
-                GOTO(out, result);
-
-        result = llo_global_init();
-        if (result)
-                GOTO(out, result);
-#endif
-        result = cl_global_init();
-out:
-
         return result;
 }
 
@@ -1796,36 +2046,24 @@ out:
  */
 void lu_global_fini(void)
 {
-        cl_global_fini();
-#ifdef __KERNEL__
-        llo_global_fini();
-        dt_global_fini();
-#endif
-        lu_time_global_fini();
         if (lu_site_shrinker != NULL) {
-                cfs_remove_shrinker(lu_site_shrinker);
+               remove_shrinker(lu_site_shrinker);
                 lu_site_shrinker = NULL;
         }
 
-        lu_context_key_degister(&lu_global_key);
+       lu_context_key_degister(&lu_global_key);
 
         /*
          * Tear shrinker environment down _after_ de-registering
          * lu_global_key, because the latter has a value in the former.
          */
-        cfs_down(&lu_sites_guard);
+       mutex_lock(&lu_sites_guard);
         lu_env_fini(&lu_shrink_env);
-        cfs_up(&lu_sites_guard);
+       mutex_unlock(&lu_sites_guard);
 
         lu_ref_global_fini();
 }
 
-struct lu_buf LU_BUF_NULL = {
-        .lb_buf = NULL,
-        .lb_len = 0
-};
-EXPORT_SYMBOL(LU_BUF_NULL);
-
 static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx)
 {
 #ifdef LPROCFS
@@ -1864,26 +2102,22 @@ int lu_site_stats_print(const struct lu_site *s, char *page, int count)
 }
 EXPORT_SYMBOL(lu_site_stats_print);
 
-const char *lu_time_names[LU_TIME_NR] = {
-        [LU_TIME_FIND_LOOKUP] = "find_lookup",
-        [LU_TIME_FIND_ALLOC]  = "find_alloc",
-        [LU_TIME_FIND_INSERT] = "find_insert"
-};
-EXPORT_SYMBOL(lu_time_names);
-
 /**
  * Helper function to initialize a number of kmem slab caches at once.
  */
 int lu_kmem_init(struct lu_kmem_descr *caches)
 {
         int result;
+        struct lu_kmem_descr *iter = caches;
 
-        for (result = 0; caches->ckd_cache != NULL; ++caches) {
-                *caches->ckd_cache = cfs_mem_cache_create(caches->ckd_name,
-                                                          caches->ckd_size,
-                                                          0, 0);
-                if (*caches->ckd_cache == NULL) {
+        for (result = 0; iter->ckd_cache != NULL; ++iter) {
+               *iter->ckd_cache = kmem_cache_create(iter->ckd_name,
+                                                    iter->ckd_size,
+                                                    0, 0, NULL);
+                if (*iter->ckd_cache == NULL) {
                         result = -ENOMEM;
+                        /* free all previously allocated caches */
+                        lu_kmem_fini(caches);
                         break;
                 }
         }
@@ -1897,15 +2131,139 @@ EXPORT_SYMBOL(lu_kmem_init);
  */
 void lu_kmem_fini(struct lu_kmem_descr *caches)
 {
-        int rc;
-
         for (; caches->ckd_cache != NULL; ++caches) {
                 if (*caches->ckd_cache != NULL) {
-                        rc = cfs_mem_cache_destroy(*caches->ckd_cache);
-                        LASSERTF(rc == 0, "couldn't destroy %s slab\n",
-                                 caches->ckd_name);
+                       kmem_cache_destroy(*caches->ckd_cache);
                         *caches->ckd_cache = NULL;
                 }
         }
 }
 EXPORT_SYMBOL(lu_kmem_fini);
+
+/**
+ * Temporary solution to be able to assign fid in ->do_create()
+ * till we have fully-functional OST fids
+ */
+void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
+                         const struct lu_fid *fid)
+{
+       struct lu_site          *s = o->lo_dev->ld_site;
+       struct lu_fid           *old = &o->lo_header->loh_fid;
+       struct lu_site_bkt_data *bkt;
+       struct lu_object        *shadow;
+       cfs_waitlink_t           waiter;
+       cfs_hash_t              *hs;
+       cfs_hash_bd_t            bd;
+       __u64                    version = 0;
+
+       LASSERT(fid_is_zero(old));
+
+       hs = s->ls_obj_hash;
+       cfs_hash_bd_get_and_lock(hs, (void *)fid, &bd, 1);
+       shadow = htable_lookup(s, &bd, fid, &waiter, &version);
+       /* supposed to be unique */
+       LASSERT(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT);
+       *old = *fid;
+       bkt = cfs_hash_bd_extra_get(hs, &bd);
+       cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
+       bkt->lsb_busy++;
+       cfs_hash_bd_unlock(hs, &bd, 1);
+}
+EXPORT_SYMBOL(lu_object_assign_fid);
+
+/**
+ * allocates object with 0 (non-assiged) fid
+ * XXX: temporary solution to be able to assign fid in ->do_create()
+ *      till we have fully-functional OST fids
+ */
+struct lu_object *lu_object_anon(const struct lu_env *env,
+                                struct lu_device *dev,
+                                const struct lu_object_conf *conf)
+{
+       struct lu_fid     fid;
+       struct lu_object *o;
+
+       fid_zero(&fid);
+       o = lu_object_alloc(env, dev, &fid, conf);
+
+       return o;
+}
+EXPORT_SYMBOL(lu_object_anon);
+
+struct lu_buf LU_BUF_NULL = {
+       .lb_buf = NULL,
+       .lb_len = 0
+};
+EXPORT_SYMBOL(LU_BUF_NULL);
+
+void lu_buf_free(struct lu_buf *buf)
+{
+       LASSERT(buf);
+       if (buf->lb_buf) {
+               LASSERT(buf->lb_len > 0);
+               OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
+               buf->lb_buf = NULL;
+               buf->lb_len = 0;
+       }
+}
+EXPORT_SYMBOL(lu_buf_free);
+
+void lu_buf_alloc(struct lu_buf *buf, int size)
+{
+       LASSERT(buf);
+       LASSERT(buf->lb_buf == NULL);
+       LASSERT(buf->lb_len == 0);
+       OBD_ALLOC_LARGE(buf->lb_buf, size);
+       if (likely(buf->lb_buf))
+               buf->lb_len = size;
+}
+EXPORT_SYMBOL(lu_buf_alloc);
+
+void lu_buf_realloc(struct lu_buf *buf, int size)
+{
+       lu_buf_free(buf);
+       lu_buf_alloc(buf, size);
+}
+EXPORT_SYMBOL(lu_buf_realloc);
+
+struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len)
+{
+       if (buf->lb_buf == NULL && buf->lb_len == 0)
+               lu_buf_alloc(buf, len);
+
+       if ((len > buf->lb_len) && (buf->lb_buf != NULL))
+               lu_buf_realloc(buf, len);
+
+       return buf;
+}
+EXPORT_SYMBOL(lu_buf_check_and_alloc);
+
+/**
+ * Increase the size of the \a buf.
+ * preserves old data in buffer
+ * old buffer remains unchanged on error
+ * \retval 0 or -ENOMEM
+ */
+int lu_buf_check_and_grow(struct lu_buf *buf, int len)
+{
+       char *ptr;
+
+       if (len <= buf->lb_len)
+               return 0;
+
+       OBD_ALLOC_LARGE(ptr, len);
+       if (ptr == NULL)
+               return -ENOMEM;
+
+       /* Free the old buf */
+       if (buf->lb_buf != NULL) {
+               memcpy(ptr, buf->lb_buf, buf->lb_len);
+               OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
+       }
+
+       buf->lb_buf = ptr;
+       buf->lb_len = len;
+       return 0;
+}
+EXPORT_SYMBOL(lu_buf_check_and_grow);
+