Whamcloud - gitweb
LU-3963 obdclass: convert to linux list api
[fs/lustre-release.git] / lustre / obdclass / cl_object.c
index 261181b..357e73f 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) 2008, 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/
@@ -36,6 +36,7 @@
  * Client Lustre Object.
  *
  *   Author: Nikita Danilov <nikita.danilov@sun.com>
+ *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  */
 
 /*
  *
  *  i_mutex
  *      PG_locked
- *          ->coh_page_guard
  *          ->coh_lock_guard
  *          ->coh_attr_guard
  *          ->ls_guard
  */
 
 #define DEBUG_SUBSYSTEM S_CLASS
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 
 #include <libcfs/libcfs.h>
 /* class_put_type() */
 #include <lustre_fid.h>
 #include <libcfs/list.h>
 #include <libcfs/libcfs_hash.h> /* for cfs_hash stuff */
-/* lu_time_global_{init,fini}() */
-#include <lu_time.h>
-
 #include <cl_object.h>
 #include "cl_internal.h"
 
-static cfs_mem_cache_t *cl_env_kmem;
+static struct kmem_cache *cl_env_kmem;
 
-/** Lock class of cl_object_header::coh_page_guard */
-static cfs_lock_class_key_t cl_page_guard_class;
 /** Lock class of cl_object_header::coh_lock_guard */
-static cfs_lock_class_key_t cl_lock_guard_class;
+static struct lock_class_key cl_lock_guard_class;
 /** Lock class of cl_object_header::coh_attr_guard */
-static cfs_lock_class_key_t cl_attr_guard_class;
+static struct lock_class_key cl_attr_guard_class;
 
+extern __u32 lu_context_tags_default;
+extern __u32 lu_session_tags_default;
 /**
  * Initialize cl_object_header.
  */
 int cl_object_header_init(struct cl_object_header *h)
 {
-        int result;
-
-        ENTRY;
-        result = lu_object_header_init(&h->coh_lu);
-        if (result == 0) {
-                cfs_spin_lock_init(&h->coh_page_guard);
-                cfs_spin_lock_init(&h->coh_lock_guard);
-                cfs_spin_lock_init(&h->coh_attr_guard);
-                cfs_lockdep_set_class(&h->coh_attr_guard, &cl_page_guard_class);
-                cfs_lockdep_set_class(&h->coh_attr_guard, &cl_lock_guard_class);
-                cfs_lockdep_set_class(&h->coh_attr_guard, &cl_attr_guard_class);
-                h->coh_pages = 0;
-                /* XXX hard coded GFP_* mask. */
-                INIT_RADIX_TREE(&h->coh_tree, GFP_ATOMIC);
-                CFS_INIT_LIST_HEAD(&h->coh_locks);
-        }
-        RETURN(result);
+       int result;
+
+       ENTRY;
+       result = lu_object_header_init(&h->coh_lu);
+       if (result == 0) {
+               spin_lock_init(&h->coh_lock_guard);
+               spin_lock_init(&h->coh_attr_guard);
+               lockdep_set_class(&h->coh_lock_guard, &cl_lock_guard_class);
+               lockdep_set_class(&h->coh_attr_guard, &cl_attr_guard_class);
+               INIT_LIST_HEAD(&h->coh_locks);
+               h->coh_page_bufsize = 0;
+       }
+       RETURN(result);
 }
 EXPORT_SYMBOL(cl_object_header_init);
 
@@ -106,7 +96,7 @@ EXPORT_SYMBOL(cl_object_header_init);
  */
 void cl_object_header_fini(struct cl_object_header *h)
 {
-        LASSERT(cfs_list_empty(&h->coh_locks));
+       LASSERT(list_empty(&h->coh_locks));
         lu_object_header_fini(&h->coh_lu);
 }
 EXPORT_SYMBOL(cl_object_header_fini);
@@ -123,7 +113,7 @@ struct cl_object *cl_object_find(const struct lu_env *env,
                                  struct cl_device *cd, const struct lu_fid *fid,
                                  const struct cl_object_conf *c)
 {
-        cfs_might_sleep();
+       might_sleep();
         return lu2cl(lu_object_find_slice(env, cl2lu_dev(cd), fid, &c->coc_lu));
 }
 EXPORT_SYMBOL(cl_object_find);
@@ -159,7 +149,7 @@ EXPORT_SYMBOL(cl_object_get);
 /**
  * Returns the top-object for a given \a o.
  *
- * \see cl_page_top(), cl_io_top()
+ * \see cl_io_top()
  */
 struct cl_object *cl_object_top(struct cl_object *o)
 {
@@ -184,9 +174,9 @@ EXPORT_SYMBOL(cl_object_top);
  *
  * \see cl_attr, cl_object_attr_lock(), cl_object_operations::coo_attr_get().
  */
-static cfs_spinlock_t *cl_object_attr_guard(struct cl_object *o)
+static spinlock_t *cl_object_attr_guard(struct cl_object *o)
 {
-        return &cl_object_header(cl_object_top(o))->coh_attr_guard;
+       return &cl_object_header(cl_object_top(o))->coh_attr_guard;
 }
 
 /**
@@ -198,7 +188,7 @@ static cfs_spinlock_t *cl_object_attr_guard(struct cl_object *o)
  */
 void cl_object_attr_lock(struct cl_object *o)
 {
-        cfs_spin_lock(cl_object_attr_guard(o));
+       spin_lock(cl_object_attr_guard(o));
 }
 EXPORT_SYMBOL(cl_object_attr_lock);
 
@@ -207,7 +197,7 @@ EXPORT_SYMBOL(cl_object_attr_lock);
  */
 void cl_object_attr_unlock(struct cl_object *o)
 {
-        cfs_spin_unlock(cl_object_attr_guard(o));
+       spin_unlock(cl_object_attr_guard(o));
 }
 EXPORT_SYMBOL(cl_object_attr_unlock);
 
@@ -221,15 +211,15 @@ EXPORT_SYMBOL(cl_object_attr_unlock);
 int cl_object_attr_get(const struct lu_env *env, struct cl_object *obj,
                        struct cl_attr *attr)
 {
-        struct lu_object_header *top;
-        int result;
+       struct lu_object_header *top;
+       int result;
 
-        LASSERT_SPIN_LOCKED(cl_object_attr_guard(obj));
-        ENTRY;
+       assert_spin_locked(cl_object_attr_guard(obj));
+       ENTRY;
 
         top = obj->co_lu.lo_header;
         result = 0;
-        cfs_list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
+       list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
                 if (obj->co_ops->coo_attr_get != NULL) {
                         result = obj->co_ops->coo_attr_get(env, obj, attr);
                         if (result != 0) {
@@ -253,26 +243,25 @@ EXPORT_SYMBOL(cl_object_attr_get);
 int cl_object_attr_set(const struct lu_env *env, struct cl_object *obj,
                        const struct cl_attr *attr, unsigned v)
 {
-        struct lu_object_header *top;
-        int result;
-
-        LASSERT_SPIN_LOCKED(cl_object_attr_guard(obj));
-        ENTRY;
-
-        top = obj->co_lu.lo_header;
-        result = 0;
-        cfs_list_for_each_entry_reverse(obj, &top->loh_layers,
-                                        co_lu.lo_linkage) {
-                if (obj->co_ops->coo_attr_set != NULL) {
-                        result = obj->co_ops->coo_attr_set(env, obj, attr, v);
-                        if (result != 0) {
-                                if (result > 0)
-                                        result = 0;
-                                break;
-                        }
-                }
-        }
-        RETURN(result);
+       struct lu_object_header *top;
+       int result;
+
+       assert_spin_locked(cl_object_attr_guard(obj));
+       ENTRY;
+
+       top = obj->co_lu.lo_header;
+       result = 0;
+       list_for_each_entry_reverse(obj, &top->loh_layers, co_lu.lo_linkage) {
+               if (obj->co_ops->coo_attr_set != NULL) {
+                       result = obj->co_ops->coo_attr_set(env, obj, attr, v);
+                       if (result != 0) {
+                               if (result > 0)
+                                       result = 0;
+                               break;
+                       }
+               }
+       }
+       RETURN(result);
 }
 EXPORT_SYMBOL(cl_object_attr_set);
 
@@ -293,8 +282,7 @@ int cl_object_glimpse(const struct lu_env *env, struct cl_object *obj,
         ENTRY;
         top = obj->co_lu.lo_header;
         result = 0;
-        cfs_list_for_each_entry_reverse(obj, &top->loh_layers,
-                                        co_lu.lo_linkage) {
+       list_for_each_entry_reverse(obj, &top->loh_layers, co_lu.lo_linkage) {
                 if (obj->co_ops->coo_glimpse != NULL) {
                         result = obj->co_ops->coo_glimpse(env, obj, lvb);
                         if (result != 0)
@@ -322,7 +310,7 @@ int cl_conf_set(const struct lu_env *env, struct cl_object *obj,
         ENTRY;
         top = obj->co_lu.lo_header;
         result = 0;
-        cfs_list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
+       list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
                 if (obj->co_ops->coo_conf_set != NULL) {
                         result = obj->co_ops->coo_conf_set(env, obj, conf);
                         if (result != 0)
@@ -334,6 +322,33 @@ int cl_conf_set(const struct lu_env *env, struct cl_object *obj,
 EXPORT_SYMBOL(cl_conf_set);
 
 /**
+ * Prunes caches of pages and locks for this object.
+ */
+void cl_object_prune(const struct lu_env *env, struct cl_object *obj)
+{
+       struct lu_object_header *top;
+       struct cl_object *o;
+       int result;
+       ENTRY;
+
+       top = obj->co_lu.lo_header;
+       result = 0;
+       list_for_each_entry(o, &top->loh_layers, co_lu.lo_linkage) {
+               if (o->co_ops->coo_prune != NULL) {
+                       result = o->co_ops->coo_prune(env, o);
+                       if (result != 0)
+                               break;
+               }
+       }
+
+       /* TODO: pruning locks will be moved into layers after cl_lock
+        * simplification is done */
+       cl_locks_prune(env, obj, 1);
+       EXIT;
+}
+EXPORT_SYMBOL(cl_object_prune);
+
+/**
  * Helper function removing all object locks, and marking object for
  * deletion. All object pages must have been deleted at this point.
  *
@@ -345,10 +360,8 @@ void cl_object_kill(const struct lu_env *env, struct cl_object *obj)
         struct cl_object_header *hdr;
 
         hdr = cl_object_header(obj);
-        LASSERT(hdr->coh_tree.rnode == NULL);
-        LASSERT(hdr->coh_pages == 0);
 
-        cfs_set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags);
+       set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags);
         /*
          * Destroy all locks. Object destruction (including cl_inode_fini())
          * cannot cancel the locks, because in the case of a local client,
@@ -361,65 +374,55 @@ void cl_object_kill(const struct lu_env *env, struct cl_object *obj)
 EXPORT_SYMBOL(cl_object_kill);
 
 /**
- * Prunes caches of pages and locks for this object.
- */
-void cl_object_prune(const struct lu_env *env, struct cl_object *obj)
-{
-        ENTRY;
-        cl_pages_prune(env, obj);
-        cl_locks_prune(env, obj, 1);
-        EXIT;
-}
-EXPORT_SYMBOL(cl_object_prune);
-
-/**
  * Check if the object has locks.
  */
 int cl_object_has_locks(struct cl_object *obj)
 {
-        struct cl_object_header *head = cl_object_header(obj);
-        int has;
+       struct cl_object_header *head = cl_object_header(obj);
+       int has;
 
-        cfs_spin_lock(&head->coh_lock_guard);
-        has = cfs_list_empty(&head->coh_locks);
-        cfs_spin_unlock(&head->coh_lock_guard);
+       spin_lock(&head->coh_lock_guard);
+       has = list_empty(&head->coh_locks);
+       spin_unlock(&head->coh_lock_guard);
 
-        return (has == 0);
+       return (has == 0);
 }
 EXPORT_SYMBOL(cl_object_has_locks);
 
 void cache_stats_init(struct cache_stats *cs, const char *name)
 {
+       int i;
+
         cs->cs_name = name;
-        cfs_atomic_set(&cs->cs_lookup, 0);
-        cfs_atomic_set(&cs->cs_hit,    0);
-        cfs_atomic_set(&cs->cs_total,  0);
-        cfs_atomic_set(&cs->cs_busy,   0);
+       for (i = 0; i < CS_NR; i++)
+               atomic_set(&cs->cs_stats[i], 0);
 }
 
-int cache_stats_print(const struct cache_stats *cs,
-                      char *page, int count, int h)
+int cache_stats_print(const struct cache_stats *cs, struct seq_file *m, int h)
 {
-        int nob = 0;
-/*
-       lookup    hit  total cached create
-  env: ...... ...... ...... ...... ......
-*/
-        if (h)
-                nob += snprintf(page, count,
-                                "       lookup    hit  total   busy create\n");
-
-        nob += snprintf(page + nob, count - nob,
-                        "%5.5s: %6u %6u %6u %6u %6u",
-                        cs->cs_name,
-                        cfs_atomic_read(&cs->cs_lookup),
-                        cfs_atomic_read(&cs->cs_hit),
-                        cfs_atomic_read(&cs->cs_total),
-                        cfs_atomic_read(&cs->cs_busy),
-                        cfs_atomic_read(&cs->cs_created));
-        return nob;
+       int i;
+
+       /*
+        *   lookup    hit    total  cached create
+        * env: ...... ...... ...... ...... ......
+        */
+       if (h) {
+               const char *names[CS_NR] = CS_NAMES;
+
+               seq_printf(m, "%6s", " ");
+               for (i = 0; i < CS_NR; i++)
+                       seq_printf(m, "%8s", names[i]);
+               seq_printf(m, "\n");
+       }
+
+       seq_printf(m, "%5.5s:", cs->cs_name);
+       for (i = 0; i < CS_NR; i++)
+               seq_printf(m, "%8u", atomic_read(&cs->cs_stats[i]));
+       return 0;
 }
 
+static void cl_env_percpu_refill(void);
+
 /**
  * Initialize client site.
  *
@@ -436,11 +439,12 @@ int cl_site_init(struct cl_site *s, struct cl_device *d)
                 cache_stats_init(&s->cs_pages, "pages");
                 cache_stats_init(&s->cs_locks, "locks");
                 for (i = 0; i < ARRAY_SIZE(s->cs_pages_state); ++i)
-                        cfs_atomic_set(&s->cs_pages_state[0], 0);
+                       atomic_set(&s->cs_pages_state[0], 0);
                 for (i = 0; i < ARRAY_SIZE(s->cs_locks_state); ++i)
-                        cfs_atomic_set(&s->cs_locks_state[i], 0);
-        }
-        return result;
+                       atomic_set(&s->cs_locks_state[i], 0);
+               cl_env_percpu_refill();
+       }
+       return result;
 }
 EXPORT_SYMBOL(cl_site_init);
 
@@ -455,61 +459,55 @@ EXPORT_SYMBOL(cl_site_fini);
 
 static struct cache_stats cl_env_stats = {
         .cs_name    = "envs",
-        .cs_created = CFS_ATOMIC_INIT(0),
-        .cs_lookup  = CFS_ATOMIC_INIT(0),
-        .cs_hit     = CFS_ATOMIC_INIT(0),
-        .cs_total   = CFS_ATOMIC_INIT(0),
-        .cs_busy    = CFS_ATOMIC_INIT(0)
+       .cs_stats = { ATOMIC_INIT(0), }
 };
 
 /**
  * Outputs client site statistical counters into a buffer. Suitable for
  * ll_rd_*()-style functions.
  */
-int cl_site_stats_print(const struct cl_site *site, char *page, int count)
+int cl_site_stats_print(const struct cl_site *site, struct seq_file *m)
 {
-        int nob;
-        int i;
-        static const char *pstate[] = {
-                [CPS_CACHED]  = "c",
-                [CPS_OWNED]   = "o",
-                [CPS_PAGEOUT] = "w",
-                [CPS_PAGEIN]  = "r",
-                [CPS_FREEING] = "f"
-        };
-        static const char *lstate[] = {
-                [CLS_NEW]       = "n",
-                [CLS_QUEUING]   = "q",
-                [CLS_ENQUEUED]  = "e",
-                [CLS_HELD]      = "h",
-                [CLS_INTRANSIT] = "t",
-                [CLS_CACHED]    = "c",
-                [CLS_FREEING]   = "f"
-        };
+       static const char *pstate[] = {
+               [CPS_CACHED]    = "c",
+               [CPS_OWNED]     = "o",
+               [CPS_PAGEOUT]   = "w",
+               [CPS_PAGEIN]    = "r",
+               [CPS_FREEING]   = "f"
+       };
+       static const char *lstate[] = {
+               [CLS_NEW]       = "n",
+               [CLS_QUEUING]   = "q",
+               [CLS_ENQUEUED]  = "e",
+               [CLS_HELD]      = "h",
+               [CLS_INTRANSIT] = "t",
+               [CLS_CACHED]    = "c",
+               [CLS_FREEING]   = "f"
+       };
+       int i;
+
 /*
        lookup    hit  total   busy create
 pages: ...... ...... ...... ...... ...... [...... ...... ...... ......]
 locks: ...... ...... ...... ...... ...... [...... ...... ...... ...... ......]
   env: ...... ...... ...... ...... ......
  */
-        nob = lu_site_stats_print(&site->cs_lu, page, count);
-        nob += cache_stats_print(&site->cs_pages, page + nob, count - nob, 1);
-        nob += snprintf(page + nob, count - nob, " [");
-        for (i = 0; i < ARRAY_SIZE(site->cs_pages_state); ++i)
-                nob += snprintf(page + nob, count - nob, "%s: %u ",
-                                pstate[i],
-                                cfs_atomic_read(&site->cs_pages_state[i]));
-        nob += snprintf(page + nob, count - nob, "]\n");
-        nob += cache_stats_print(&site->cs_locks, page + nob, count - nob, 0);
-        nob += snprintf(page + nob, count - nob, " [");
-        for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i)
-                nob += snprintf(page + nob, count - nob, "%s: %u ",
-                                lstate[i],
-                                cfs_atomic_read(&site->cs_locks_state[i]));
-        nob += snprintf(page + nob, count - nob, "]\n");
-        nob += cache_stats_print(&cl_env_stats, page + nob, count - nob, 0);
-        nob += snprintf(page + nob, count - nob, "\n");
-        return nob;
+       lu_site_stats_seq_print(&site->cs_lu, m);
+       cache_stats_print(&site->cs_pages, m, 1);
+       seq_printf(m, " [");
+       for (i = 0; i < ARRAY_SIZE(site->cs_pages_state); ++i)
+               seq_printf(m, "%s: %u ", pstate[i],
+                          atomic_read(&site->cs_pages_state[i]));
+       seq_printf(m, "]\n");
+       cache_stats_print(&site->cs_locks, m, 0);
+       seq_printf(m, " [");
+       for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i)
+               seq_printf(m, "%s: %u ", lstate[i],
+                          atomic_read(&site->cs_locks_state[i]));
+       seq_printf(m, "]\n");
+       cache_stats_print(&cl_env_stats, m, 0);
+       seq_printf(m, "\n");
+       return 0;
 }
 EXPORT_SYMBOL(cl_site_stats_print);
 
@@ -537,11 +535,11 @@ EXPORT_SYMBOL(cl_site_stats_print);
  * bz20044, bz22683.
  */
 
-static CFS_LIST_HEAD(cl_envs);
+static struct list_head cl_envs;
 static unsigned cl_envs_cached_nr  = 0;
 static unsigned cl_envs_cached_max = 128; /* XXX: prototype: arbitrary limit
                                            * for now. */
-static cfs_spinlock_t cl_envs_guard = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(cl_envs_guard);
 
 struct cl_env {
         void             *ce_magic;
@@ -555,26 +553,26 @@ struct cl_env {
          * This allows cl_env to be entered into cl_env_hash which implements
          * the current thread -> client environment lookup.
          */
-        cfs_hlist_node_t  ce_node;
+       struct hlist_node  ce_node;
 #endif
-        /**
-         * Owner for the current cl_env.
-         *
-         * If LL_TASK_CL_ENV is defined, this point to the owning cfs_current(),
-         * only for debugging purpose ;
-         * Otherwise hash is used, and this is the key for cfs_hash.
-         * Now current thread pid is stored. Note using thread pointer would
-         * lead to unbalanced hash because of its specific allocation locality
-         * and could be varied for different platforms and OSes, even different
-         * OS versions.
-         */
-        void             *ce_owner;
+       /**
+        * Owner for the current cl_env.
+        *
+        * If LL_TASK_CL_ENV is defined, this point to the owning current,
+        * only for debugging purpose ;
+        * Otherwise hash is used, and this is the key for cfs_hash.
+        * Now current thread pid is stored. Note using thread pointer would
+        * lead to unbalanced hash because of its specific allocation locality
+        * and could be varied for different platforms and OSes, even different
+        * OS versions.
+        */
+       void             *ce_owner;
 
         /*
          * Linkage into global list of all client environments. Used for
          * garbage collection.
          */
-        cfs_list_t        ce_linkage;
+       struct list_head  ce_linkage;
         /*
          *
          */
@@ -586,13 +584,17 @@ struct cl_env {
         void             *ce_debug;
 };
 
-#define CL_ENV_INC(counter) cfs_atomic_inc(&cl_env_stats.counter)
+#ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
+#define CL_ENV_INC(counter) atomic_inc(&cl_env_stats.cs_stats[CS_##counter])
 
-#define CL_ENV_DEC(counter)                                             \
-        do {                                                            \
-                LASSERT(cfs_atomic_read(&cl_env_stats.counter) > 0);    \
-                cfs_atomic_dec(&cl_env_stats.counter);                  \
-        } while (0)
+#define CL_ENV_DEC(counter) do {                                              \
+       LASSERT(atomic_read(&cl_env_stats.cs_stats[CS_##counter]) > 0);   \
+       atomic_dec(&cl_env_stats.cs_stats[CS_##counter]);                 \
+} while (0)
+#else
+#define CL_ENV_INC(counter)
+#define CL_ENV_DEC(counter)
+#endif
 
 static void cl_env_init0(struct cl_env *cle, void *debug)
 {
@@ -602,7 +604,7 @@ static void cl_env_init0(struct cl_env *cle, void *debug)
 
         cle->ce_ref = 1;
         cle->ce_debug = debug;
-        CL_ENV_INC(cs_busy);
+        CL_ENV_INC(busy);
 }
 
 
@@ -613,7 +615,8 @@ static void cl_env_init0(struct cl_env *cle, void *debug)
 
 static cfs_hash_t *cl_env_hash;
 
-static unsigned cl_env_hops_hash(cfs_hash_t *lh, void *key, unsigned mask)
+static unsigned cl_env_hops_hash(cfs_hash_t *lh,
+                                 const void *key, unsigned mask)
 {
 #if BITS_PER_LONG == 64
         return cfs_hash_u64_hash((__u64)key, mask);
@@ -622,14 +625,15 @@ static unsigned cl_env_hops_hash(cfs_hash_t *lh, void *key, unsigned mask)
 #endif
 }
 
-static void *cl_env_hops_obj(cfs_hlist_node_t *hn)
+static void *cl_env_hops_obj(struct hlist_node *hn)
 {
-        struct cl_env *cle = cfs_hlist_entry(hn, struct cl_env, ce_node);
-        LASSERT(cle->ce_magic == &cl_env_init0);
-        return (void *)cle;
+       struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node);
+
+       LASSERT(cle->ce_magic == &cl_env_init0);
+       return (void *)cle;
 }
 
-static int cl_env_hops_keycmp(void *key, cfs_hlist_node_t *hn)
+static int cl_env_hops_keycmp(const void *key, struct hlist_node *hn)
 {
         struct cl_env *cle = cl_env_hops_obj(hn);
 
@@ -637,46 +641,52 @@ static int cl_env_hops_keycmp(void *key, cfs_hlist_node_t *hn)
         return (key == cle->ce_owner);
 }
 
+static void cl_env_hops_noop(cfs_hash_t *hs, struct hlist_node *hn)
+{
+       struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node);
+        LASSERT(cle->ce_magic == &cl_env_init0);
+}
+
 static cfs_hash_ops_t cl_env_hops = {
         .hs_hash        = cl_env_hops_hash,
         .hs_key         = cl_env_hops_obj,
         .hs_keycmp      = cl_env_hops_keycmp,
         .hs_object      = cl_env_hops_obj,
-        .hs_get         = cl_env_hops_obj,
-        .hs_put_locked  = cl_env_hops_obj,
+        .hs_get         = cl_env_hops_noop,
+        .hs_put_locked  = cl_env_hops_noop,
 };
 
 static inline struct cl_env *cl_env_fetch(void)
 {
-        struct cl_env *cle;
+       struct cl_env *cle;
 
-        cle = cfs_hash_lookup(cl_env_hash, (void *) (long) cfs_current()->pid);
-        LASSERT(ergo(cle, cle->ce_magic == &cl_env_init0));
-        return cle;
+       cle = cfs_hash_lookup(cl_env_hash, (void *) (long) current->pid);
+       LASSERT(ergo(cle, cle->ce_magic == &cl_env_init0));
+       return cle;
 }
 
 static inline void cl_env_attach(struct cl_env *cle)
 {
-        if (cle) {
-                int rc;
-
-                LASSERT(cle->ce_owner == NULL);
-                cle->ce_owner = (void *) (long) cfs_current()->pid;
-                rc = cfs_hash_add_unique(cl_env_hash, cle->ce_owner,
-                                         &cle->ce_node);
-                LASSERT(rc == 0);
-        }
+       if (cle) {
+               int rc;
+
+               LASSERT(cle->ce_owner == NULL);
+               cle->ce_owner = (void *) (long) current->pid;
+               rc = cfs_hash_add_unique(cl_env_hash, cle->ce_owner,
+                                        &cle->ce_node);
+               LASSERT(rc == 0);
+       }
 }
 
 static inline void cl_env_do_detach(struct cl_env *cle)
 {
-        void *cookie;
+       void *cookie;
 
-        LASSERT(cle->ce_owner == (void *) (long) cfs_current()->pid);
-        cookie = cfs_hash_del(cl_env_hash, cle->ce_owner,
-                              &cle->ce_node);
-        LASSERT(cookie == cle);
-        cle->ce_owner = NULL;
+       LASSERT(cle->ce_owner == (void *) (long) current->pid);
+       cookie = cfs_hash_del(cl_env_hash, cle->ce_owner,
+                             &cle->ce_node);
+       LASSERT(cookie == cle);
+       cle->ce_owner = NULL;
 }
 
 static int cl_env_store_init(void) {
@@ -701,30 +711,30 @@ static void cl_env_store_fini(void) {
 
 static inline struct cl_env *cl_env_fetch(void)
 {
-        struct cl_env *cle;
+       struct cl_env *cle;
 
-        cle = cfs_current()->LL_TASK_CL_ENV;
-        if (cle && cle->ce_magic != &cl_env_init0)
-                cle = NULL;
-        return cle;
+       cle = current->LL_TASK_CL_ENV;
+       if (cle && cle->ce_magic != &cl_env_init0)
+               cle = NULL;
+       return cle;
 }
 
 static inline void cl_env_attach(struct cl_env *cle)
 {
-        if (cle) {
-                LASSERT(cle->ce_owner == NULL);
-                cle->ce_owner = cfs_current();
-                cle->ce_prev = cfs_current()->LL_TASK_CL_ENV;
-                cfs_current()->LL_TASK_CL_ENV = cle;
-        }
+       if (cle) {
+               LASSERT(cle->ce_owner == NULL);
+               cle->ce_owner = current;
+               cle->ce_prev = current->LL_TASK_CL_ENV;
+               current->LL_TASK_CL_ENV = cle;
+       }
 }
 
 static inline void cl_env_do_detach(struct cl_env *cle)
 {
-        LASSERT(cle->ce_owner == cfs_current());
-        LASSERT(cfs_current()->LL_TASK_CL_ENV == cle);
-        cfs_current()->LL_TASK_CL_ENV = cle->ce_prev;
-        cle->ce_owner = NULL;
+       LASSERT(cle->ce_owner == current);
+       LASSERT(current->LL_TASK_CL_ENV == cle);
+       current->LL_TASK_CL_ENV = cle->ce_prev;
+       cle->ce_owner = NULL;
 }
 
 static int cl_env_store_init(void) { return 0; }
@@ -743,43 +753,44 @@ static inline struct cl_env *cl_env_detach(struct cl_env *cle)
         return cle;
 }
 
-static struct lu_env *cl_env_new(__u32 tags, void *debug)
+static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug)
 {
-        struct lu_env *env;
-        struct cl_env *cle;
-
-        OBD_SLAB_ALLOC_PTR_GFP(cle, cl_env_kmem, CFS_ALLOC_IO);
-        if (cle != NULL) {
-                int rc;
-
-                CFS_INIT_LIST_HEAD(&cle->ce_linkage);
-                cle->ce_magic = &cl_env_init0;
-                env = &cle->ce_lu;
-                rc = lu_env_init(env, LCT_CL_THREAD|tags);
-                if (rc == 0) {
-                        rc = lu_context_init(&cle->ce_ses, LCT_SESSION|tags);
-                        if (rc == 0) {
-                                lu_context_enter(&cle->ce_ses);
-                                env->le_ses = &cle->ce_ses;
-                                cl_env_init0(cle, debug);
-                        } else
-                                lu_env_fini(env);
-                }
-                if (rc != 0) {
-                        OBD_SLAB_FREE_PTR(cle, cl_env_kmem);
-                        env = ERR_PTR(rc);
-                } else {
-                        CL_ENV_INC(cs_created);
-                        CL_ENV_INC(cs_total);
-                }
-        } else
-                env = ERR_PTR(-ENOMEM);
-        return env;
+       struct lu_env *env;
+       struct cl_env *cle;
+
+       OBD_SLAB_ALLOC_PTR_GFP(cle, cl_env_kmem, GFP_NOFS);
+       if (cle != NULL) {
+               int rc;
+
+               INIT_LIST_HEAD(&cle->ce_linkage);
+               cle->ce_magic = &cl_env_init0;
+               env = &cle->ce_lu;
+               rc = lu_env_init(env, LCT_CL_THREAD|ctx_tags);
+               if (rc == 0) {
+                       rc = lu_context_init(&cle->ce_ses,
+                                            LCT_SESSION | ses_tags);
+                       if (rc == 0) {
+                               lu_context_enter(&cle->ce_ses);
+                               env->le_ses = &cle->ce_ses;
+                               cl_env_init0(cle, debug);
+                       } else
+                               lu_env_fini(env);
+               }
+               if (rc != 0) {
+                       OBD_SLAB_FREE_PTR(cle, cl_env_kmem);
+                       env = ERR_PTR(rc);
+               } else {
+                       CL_ENV_INC(create);
+                       CL_ENV_INC(total);
+               }
+       } else
+               env = ERR_PTR(-ENOMEM);
+       return env;
 }
 
 static void cl_env_fini(struct cl_env *cle)
 {
-        CL_ENV_DEC(cs_total);
+        CL_ENV_DEC(total);
         lu_context_fini(&cle->ce_lu.le_ctx);
         lu_context_fini(&cle->ce_ses);
         OBD_SLAB_FREE_PTR(cle, cl_env_kmem);
@@ -787,19 +798,19 @@ static void cl_env_fini(struct cl_env *cle)
 
 static struct lu_env *cl_env_obtain(void *debug)
 {
-        struct cl_env *cle;
-        struct lu_env *env;
+       struct cl_env *cle;
+       struct lu_env *env;
 
-        ENTRY;
-        cfs_spin_lock(&cl_envs_guard);
-        LASSERT(equi(cl_envs_cached_nr == 0, cfs_list_empty(&cl_envs)));
-        if (cl_envs_cached_nr > 0) {
-                int rc;
+       ENTRY;
+       spin_lock(&cl_envs_guard);
+       LASSERT(equi(cl_envs_cached_nr == 0, list_empty(&cl_envs)));
+       if (cl_envs_cached_nr > 0) {
+               int rc;
 
-                cle = container_of(cl_envs.next, struct cl_env, ce_linkage);
-                cfs_list_del_init(&cle->ce_linkage);
-                cl_envs_cached_nr--;
-                cfs_spin_unlock(&cl_envs_guard);
+               cle = container_of(cl_envs.next, struct cl_env, ce_linkage);
+               list_del_init(&cle->ce_linkage);
+               cl_envs_cached_nr--;
+               spin_unlock(&cl_envs_guard);
 
                 env = &cle->ce_lu;
                 rc = lu_env_refill(env);
@@ -812,10 +823,11 @@ static struct lu_env *cl_env_obtain(void *debug)
                         env = ERR_PTR(rc);
                 }
         } else {
-                cfs_spin_unlock(&cl_envs_guard);
-                env = cl_env_new(0, debug);
-        }
-        RETURN(env);
+               spin_unlock(&cl_envs_guard);
+               env = cl_env_new(lu_context_tags_default,
+                                lu_session_tags_default, debug);
+       }
+       RETURN(env);
 }
 
 static inline struct cl_env *cl_env_container(struct lu_env *env)
@@ -828,7 +840,7 @@ struct lu_env *cl_env_peek(int *refcheck)
         struct lu_env *env;
         struct cl_env *cle;
 
-        CL_ENV_INC(cs_lookup);
+        CL_ENV_INC(lookup);
 
         /* check that we don't go far from untrusted pointer */
         CLASSERT(offsetof(struct cl_env, ce_magic) == 0);
@@ -836,7 +848,7 @@ struct lu_env *cl_env_peek(int *refcheck)
         env = NULL;
         cle = cl_env_fetch();
         if (cle != NULL) {
-                CL_ENV_INC(cs_hit);
+                CL_ENV_INC(hit);
                 env = &cle->ce_lu;
                 *refcheck = ++cle->ce_ref;
         }
@@ -888,7 +900,7 @@ struct lu_env *cl_env_alloc(int *refcheck, __u32 tags)
         struct lu_env *env;
 
         LASSERT(cl_env_peek(refcheck) == NULL);
-        env = cl_env_new(tags, __builtin_return_address(0));
+        env = cl_env_new(tags, tags, __builtin_return_address(0));
         if (!IS_ERR(env)) {
                 struct cl_env *cle;
 
@@ -914,23 +926,23 @@ static void cl_env_exit(struct cl_env *cle)
  */
 unsigned cl_env_cache_purge(unsigned nr)
 {
-        struct cl_env *cle;
-
-        ENTRY;
-        cfs_spin_lock(&cl_envs_guard);
-        for (; !cfs_list_empty(&cl_envs) && nr > 0; --nr) {
-                cle = container_of(cl_envs.next, struct cl_env, ce_linkage);
-                cfs_list_del_init(&cle->ce_linkage);
-                LASSERT(cl_envs_cached_nr > 0);
-                cl_envs_cached_nr--;
-                cfs_spin_unlock(&cl_envs_guard);
-
-                cl_env_fini(cle);
-                cfs_spin_lock(&cl_envs_guard);
-        }
-        LASSERT(equi(cl_envs_cached_nr == 0, cfs_list_empty(&cl_envs)));
-        cfs_spin_unlock(&cl_envs_guard);
-        RETURN(nr);
+       struct cl_env *cle;
+
+       ENTRY;
+       spin_lock(&cl_envs_guard);
+       for (; !list_empty(&cl_envs) && nr > 0; --nr) {
+               cle = container_of(cl_envs.next, struct cl_env, ce_linkage);
+               list_del_init(&cle->ce_linkage);
+               LASSERT(cl_envs_cached_nr > 0);
+               cl_envs_cached_nr--;
+               spin_unlock(&cl_envs_guard);
+
+               cl_env_fini(cle);
+               spin_lock(&cl_envs_guard);
+       }
+       LASSERT(equi(cl_envs_cached_nr == 0, list_empty(&cl_envs)));
+       spin_unlock(&cl_envs_guard);
+       RETURN(nr);
 }
 EXPORT_SYMBOL(cl_env_cache_purge);
 
@@ -952,7 +964,7 @@ void cl_env_put(struct lu_env *env, int *refcheck)
 
         CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
         if (--cle->ce_ref == 0) {
-                CL_ENV_DEC(cs_busy);
+                CL_ENV_DEC(busy);
                 cl_env_detach(cle);
                 cle->ce_debug = NULL;
                 cl_env_exit(cle);
@@ -965,13 +977,13 @@ void cl_env_put(struct lu_env *env, int *refcheck)
                 if (cl_envs_cached_nr < cl_envs_cached_max &&
                     (env->le_ctx.lc_tags & ~LCT_HAS_EXIT) == LCT_CL_THREAD &&
                     (env->le_ses->lc_tags & ~LCT_HAS_EXIT) == LCT_SESSION) {
-                        cfs_spin_lock(&cl_envs_guard);
-                        cfs_list_add(&cle->ce_linkage, &cl_envs);
-                        cl_envs_cached_nr++;
-                        cfs_spin_unlock(&cl_envs_guard);
-                } else
-                        cl_env_fini(cle);
-        }
+                       spin_lock(&cl_envs_guard);
+                       list_add(&cle->ce_linkage, &cl_envs);
+                       cl_envs_cached_nr++;
+                       spin_unlock(&cl_envs_guard);
+               } else
+                       cl_env_fini(cle);
+       }
 }
 EXPORT_SYMBOL(cl_env_put);
 
@@ -1097,6 +1109,103 @@ void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb)
 }
 EXPORT_SYMBOL(cl_lvb2attr);
 
+static struct cl_env cl_env_percpu[NR_CPUS];
+
+static int cl_env_percpu_init(void)
+{
+       struct cl_env *cle;
+       int tags = LCT_REMEMBER | LCT_NOREF;
+       int i, j;
+       int rc = 0;
+
+       for_each_possible_cpu(i) {
+               struct lu_env *env;
+
+               cle = &cl_env_percpu[i];
+               env = &cle->ce_lu;
+
+               INIT_LIST_HEAD(&cle->ce_linkage);
+               cle->ce_magic = &cl_env_init0;
+               rc = lu_env_init(env, LCT_CL_THREAD | tags);
+               if (rc == 0) {
+                       rc = lu_context_init(&cle->ce_ses, LCT_SESSION | tags);
+                        if (rc == 0) {
+                                lu_context_enter(&cle->ce_ses);
+                                env->le_ses = &cle->ce_ses;
+                       } else {
+                               lu_env_fini(env);
+                       }
+               }
+               if (rc != 0)
+                       break;
+       }
+       if (rc != 0) {
+               /* Indices 0 to i (excluding i) were correctly initialized,
+                * thus we must uninitialize up to i, the rest are undefined. */
+               for (j = 0; j < i; j++) {
+                       cle = &cl_env_percpu[i];
+                       lu_context_exit(&cle->ce_ses);
+                       lu_context_fini(&cle->ce_ses);
+                       lu_env_fini(&cle->ce_lu);
+               }
+       }
+
+       return rc;
+}
+
+static void cl_env_percpu_fini(void)
+{
+       int i;
+
+       for_each_possible_cpu(i) {
+               struct cl_env *cle = &cl_env_percpu[i];
+
+               lu_context_exit(&cle->ce_ses);
+               lu_context_fini(&cle->ce_ses);
+               lu_env_fini(&cle->ce_lu);
+       }
+}
+
+static void cl_env_percpu_refill(void)
+{
+       int i;
+
+       for_each_possible_cpu(i)
+               lu_env_refill(&cl_env_percpu[i].ce_lu);
+}
+
+void cl_env_percpu_put(struct lu_env *env)
+{
+       struct cl_env *cle;
+       int cpu;
+
+       cpu = smp_processor_id();
+       cle = cl_env_container(env);
+       LASSERT(cle == &cl_env_percpu[cpu]);
+
+       cle->ce_ref--;
+       LASSERT(cle->ce_ref == 0);
+
+       CL_ENV_DEC(busy);
+       cl_env_detach(cle);
+       cle->ce_debug = NULL;
+
+       put_cpu();
+}
+EXPORT_SYMBOL(cl_env_percpu_put);
+
+struct lu_env *cl_env_percpu_get()
+{
+       struct cl_env *cle;
+
+       cle = &cl_env_percpu[get_cpu()];
+       cl_env_init0(cle, __builtin_return_address(0));
+
+       cl_env_attach(cle);
+       return &cle->ce_lu;
+}
+EXPORT_SYMBOL(cl_env_percpu_get);
+
 /*****************************************************************************
  *
  * Temporary prototype thing: mirror obd-devices into cl devices.
@@ -1229,24 +1338,45 @@ static struct lu_kmem_descr cl_object_caches[] = {
  */
 int cl_global_init(void)
 {
-        int result;
+       int result;
 
-        result = cl_env_store_init();
-        if (result)
-                return result;
+       INIT_LIST_HEAD(&cl_envs);
+
+       result = cl_env_store_init();
+       if (result)
+               return result;
 
         result = lu_kmem_init(cl_object_caches);
-        if (result == 0) {
-                LU_CONTEXT_KEY_INIT(&cl_key);
-                result = lu_context_key_register(&cl_key);
-                if (result == 0) {
-                        result = cl_lock_init();
-                        if (result == 0)
-                                result = cl_page_init();
-                }
-        }
         if (result)
-                cl_env_store_fini();
+                goto out_store;
+
+        LU_CONTEXT_KEY_INIT(&cl_key);
+        result = lu_context_key_register(&cl_key);
+        if (result)
+                goto out_kmem;
+
+        result = cl_lock_init();
+        if (result)
+                goto out_context;
+
+        result = cl_page_init();
+        if (result)
+                goto out_lock;
+
+       result = cl_env_percpu_init();
+       if (result)
+               /* no cl_env_percpu_fini on error */
+               goto out_lock;
+
+        return 0;
+out_lock:
+        cl_lock_fini();
+out_context:
+        lu_context_key_degister(&cl_key);
+out_kmem:
+        lu_kmem_fini(cl_object_caches);
+out_store:
+        cl_env_store_fini();
         return result;
 }
 
@@ -1255,6 +1385,7 @@ int cl_global_init(void)
  */
 void cl_global_fini(void)
 {
+       cl_env_percpu_fini();
         cl_lock_fini();
         cl_page_fini();
         lu_context_key_degister(&cl_key);