Whamcloud - gitweb
LU-8773 llite: refactor lov_object_fiemap()
[fs/lustre-release.git] / lustre / lov / lov_object.c
index ae80756..7ee0ebc 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #define DEBUG_SUBSYSTEM S_LOV
 
 #include "lov_cl_internal.h"
-#include <lustre_debug.h>
+
+static inline struct lov_device *lov_object_dev(struct lov_object *obj)
+{
+       return lu2lov_dev(obj->lo_cl.co_lu.lo_dev);
+}
 
 /** \addtogroup lov
  *  @{
  */
 
 struct lov_layout_operations {
-        int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
-                        struct lov_object *lov,
-                        const struct cl_object_conf *conf,
-                        union lov_layout_state *state);
+       int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
+                       struct lov_object *lov, struct lov_stripe_md *lsm,
+                       const struct cl_object_conf *conf,
+                       union lov_layout_state *state);
        int (*llo_delete)(const struct lu_env *env, struct lov_object *lov,
                            union lov_layout_state *state);
         void (*llo_fini)(const struct lu_env *env, struct lov_object *lov,
@@ -68,7 +68,7 @@ struct lov_layout_operations {
         int  (*llo_print)(const struct lu_env *env, void *cookie,
                           lu_printer_t p, const struct lu_object *o);
         int  (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
-                               struct cl_page *page, struct page *vmpage);
+                             struct cl_page *page, pgoff_t index);
         int  (*llo_lock_init)(const struct lu_env *env,
                               struct cl_object *obj, struct cl_lock *lock,
                               const struct cl_io *io);
@@ -80,6 +80,12 @@ struct lov_layout_operations {
 
 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov);
 
+static void lov_lsm_put(struct lov_stripe_md *lsm)
+{
+       if (lsm != NULL)
+               lov_free_memmd(&lsm);
+}
+
 /*****************************************************************************
  *
  * Lov object layout operations.
@@ -95,12 +101,12 @@ static void lov_install_empty(const struct lu_env *env,
          */
 }
 
-static int lov_init_empty(const struct lu_env *env,
-                          struct lov_device *dev, struct lov_object *lov,
-                          const struct cl_object_conf *conf,
-                          union  lov_layout_state *state)
+static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
+                         struct lov_object *lov, struct lov_stripe_md *lsm,
+                         const struct cl_object_conf *conf,
+                         union lov_layout_state *state)
 {
-        return 0;
+       return 0;
 }
 
 static void lov_install_raid0(const struct lu_env *env,
@@ -185,18 +191,34 @@ static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
                }
 
                LU_OBJECT_DEBUG(mask, env, &stripe->co_lu,
-                               "stripe %d is already owned.\n", idx);
-               LU_OBJECT_DEBUG(mask, env, old_obj, "owned.\n");
+                               "stripe %d is already owned.", idx);
+               LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
                LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
                cl_object_put(env, stripe);
        }
        return result;
 }
 
-static int lov_init_raid0(const struct lu_env *env,
-                          struct lov_device *dev, struct lov_object *lov,
-                          const struct cl_object_conf *conf,
-                          union  lov_layout_state *state)
+static int lov_page_slice_fixup(struct lov_object *lov,
+                               struct cl_object *stripe)
+{
+       struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
+       struct cl_object *o;
+
+       if (stripe == NULL)
+               return hdr->coh_page_bufsize - lov->lo_cl.co_slice_off -
+                      cfs_size_round(sizeof(struct lov_page));
+
+       cl_object_for_each(o, stripe)
+               o->co_slice_off += hdr->coh_page_bufsize;
+
+       return cl_object_header(stripe)->coh_page_bufsize;
+}
+
+static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
+                         struct lov_object *lov, struct lov_stripe_md *lsm,
+                         const struct cl_object_conf *conf,
+                         union lov_layout_state *state)
 {
         int result;
         int i;
@@ -204,7 +226,6 @@ static int lov_init_raid0(const struct lu_env *env,
         struct cl_object        *stripe;
         struct lov_thread_info  *lti     = lov_env_info(env);
         struct cl_object_conf   *subconf = &lti->lti_stripe_conf;
-        struct lov_stripe_md    *lsm     = conf->u.coc_md->lsm;
         struct lu_fid           *ofid    = &lti->lti_fid;
         struct lov_layout_raid0 *r0      = &state->raid0;
 
@@ -218,13 +239,17 @@ static int lov_init_raid0(const struct lu_env *env,
 
        LASSERT(lov->lo_lsm == NULL);
        lov->lo_lsm = lsm_addref(lsm);
-       r0->lo_nr  = lsm->lsm_stripe_count;
-        LASSERT(r0->lo_nr <= lov_targets_nr(dev));
+       r0->lo_nr = lsm->lsm_stripe_count;
+       LASSERT(r0->lo_nr <= lov_targets_nr(dev));
 
-        OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
-        if (r0->lo_sub != NULL) {
-                result = 0;
-                subconf->coc_inode = conf->coc_inode;
+       lov->lo_layout_invalid = true;
+
+       OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
+       if (r0->lo_sub != NULL) {
+               int psz = 0;
+
+               result = 0;
+               subconf->coc_inode = conf->coc_inode;
                spin_lock_init(&r0->lo_sub_lock);
                 /*
                  * Create stripe cl_objects.
@@ -234,41 +259,58 @@ static int lov_init_raid0(const struct lu_env *env,
                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
                         int ost_idx = oinfo->loi_ost_idx;
 
-                        result = ostid_to_fid(ofid, &oinfo->loi_oi,
+                       if (lov_oinfo_is_dummy(oinfo))
+                               continue;
+
+                       result = ostid_to_fid(ofid, &oinfo->loi_oi,
                                              oinfo->loi_ost_idx);
                        if (result != 0)
                                GOTO(out, result);
 
-                        subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
-                        subconf->u.coc_oinfo = oinfo;
-                        LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
+                       if (dev->ld_target[ost_idx] == NULL) {
+                               CERROR("%s: OST %04x is not initialized\n",
+                                      lov2obd(dev->ld_lov)->obd_name, ost_idx);
+                               GOTO(out, result = -EIO);
+                       }
+
+                       subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
+                       subconf->u.coc_oinfo = oinfo;
+                       LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
                        /* In the function below, .hs_keycmp resolves to
                         * lu_obj_hop_keycmp() */
                        /* coverity[overrun-buffer-val] */
-                        stripe = lov_sub_find(env, subdev, ofid, subconf);
-                        if (!IS_ERR(stripe)) {
-                                result = lov_init_sub(env, lov, stripe, r0, i);
+                       stripe = lov_sub_find(env, subdev, ofid, subconf);
+                       if (!IS_ERR(stripe)) {
+                               result = lov_init_sub(env, lov, stripe, r0, i);
                                if (result == -EAGAIN) { /* try again */
                                        --i;
                                        result = 0;
+                                       continue;
                                }
-                        } else {
-                                result = PTR_ERR(stripe);
+                       } else {
+                               result = PTR_ERR(stripe);
+                       }
+
+                       if (result == 0) {
+                               int sz = lov_page_slice_fixup(lov, stripe);
+                               LASSERT(ergo(psz > 0, psz == sz));
+                               psz = sz;
                        }
                 }
-        } else
-                result = -ENOMEM;
+               if (result == 0)
+                       cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
+       } else
+               result = -ENOMEM;
 out:
-        RETURN(result);
+       RETURN(result);
 }
 
 static int lov_init_released(const struct lu_env *env,
-                       struct lov_device *dev, struct lov_object *lov,
-                       const struct cl_object_conf *conf,
-                       union  lov_layout_state *state)
+                            struct lov_device *dev, struct lov_object *lov,
+                            struct lov_stripe_md *lsm,
+                            const struct cl_object_conf *conf,
+                            union lov_layout_state *state)
 {
-       struct lov_stripe_md *lsm = conf->u.coc_md->lsm;
-
        LASSERT(lsm != NULL);
        LASSERT(lsm_is_released(lsm));
        LASSERT(lov->lo_lsm == NULL);
@@ -277,14 +319,42 @@ static int lov_init_released(const struct lu_env *env,
        return 0;
 }
 
+static struct cl_object *lov_find_subobj(const struct lu_env *env,
+                                        struct lov_object *lov,
+                                        struct lov_stripe_md *lsm,
+                                        int stripe_idx)
+{
+       struct lov_device       *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
+       struct lov_oinfo        *oinfo = lsm->lsm_oinfo[stripe_idx];
+       struct lov_thread_info  *lti = lov_env_info(env);
+       struct lu_fid           *ofid = &lti->lti_fid;
+       struct cl_device        *subdev;
+       int                     ost_idx;
+       int                     rc;
+       struct cl_object        *result;
+
+       if (lov->lo_type != LLT_RAID0)
+               GOTO(out, result = NULL);
+
+       ost_idx = oinfo->loi_ost_idx;
+       rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
+       if (rc != 0)
+               GOTO(out, result = NULL);
+
+       subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
+       result = lov_sub_find(env, subdev, ofid, NULL);
+out:
+       if (result == NULL)
+               result = ERR_PTR(-EINVAL);
+       return result;
+}
+
 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
                            union lov_layout_state *state)
 {
        LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
 
        lov_layout_wait(env, lov);
-
-       cl_object_prune(env, &lov->lo_cl);
        return 0;
 }
 
@@ -313,7 +383,7 @@ static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
          * ->lo_sub[] slot in lovsub_object_fini() */
        if (r0->lo_sub[idx] == los) {
                waiter = &lov_env_info(env)->lti_waiter;
-               init_waitqueue_entry_current(waiter);
+               init_waitqueue_entry(waiter, current);
                add_wait_queue(&bkt->lsb_marche_funebre, waiter);
                set_current_state(TASK_UNINTERRUPTIBLE);
                while (1) {
@@ -323,7 +393,7 @@ static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
                        spin_lock(&r0->lo_sub_lock);
                        if (r0->lo_sub[idx] == los) {
                                spin_unlock(&r0->lo_sub_lock);
-                               waitq_wait(waiter, TASK_UNINTERRUPTIBLE);
+                               schedule();
                        } else {
                                spin_unlock(&r0->lo_sub_lock);
                                set_current_state(TASK_RUNNING);
@@ -352,16 +422,15 @@ static int lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
                         struct lovsub_object *los = r0->lo_sub[i];
 
                         if (los != NULL) {
-                               cl_locks_prune(env, &los->lso_cl, 1);
+                               cl_object_prune(env, &los->lso_cl);
                                 /*
                                  * If top-level object is to be evicted from
                                  * the cache, so are its sub-objects.
                                  */
                                 lov_subobject_kill(env, lov, los, i);
                        }
-                }
-        }
-       cl_object_prune(env, &lov->lo_cl);
+               }
+       }
        RETURN(0);
 }
 
@@ -414,7 +483,7 @@ static int lov_print_raid0(const struct lu_env *env, void *cookie,
 
        (*p)(env, cookie, "stripes: %d, %s, lsm{%p 0x%08X %d %u %u}:\n",
                r0->lo_nr, lov->lo_layout_invalid ? "invalid" : "valid", lsm,
-               lsm->lsm_magic, cfs_atomic_read(&lsm->lsm_refc),
+               lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
                lsm->lsm_stripe_count, lsm->lsm_layout_gen);
        for (i = 0; i < r0->lo_nr; ++i) {
                struct lu_object *sub;
@@ -438,7 +507,7 @@ static int lov_print_released(const struct lu_env *env, void *cookie,
        (*p)(env, cookie,
                "released: %s, lsm{%p 0x%08X %d %u %u}:\n",
                lov->lo_layout_invalid ? "invalid" : "valid", lsm,
-               lsm->lsm_magic, cfs_atomic_read(&lsm->lsm_refc),
+               lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
                lsm->lsm_stripe_count, lsm->lsm_layout_gen);
        return 0;
 }
@@ -475,7 +544,7 @@ static int lov_attr_get_raid0(const struct lu_env *env, struct cl_object *obj,
         * hit this assertion.
         * Anyway, it's still okay to call attr_get w/o type guard as layout
         * can't go if locks exist. */
-       /* LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 1); */
+       /* LASSERT(atomic_read(&lsm->lsm_refc) > 1); */
 
        if (!r0->lo_attr_valid) {
                struct lov_stripe_md    *lsm = lov->lo_lsm;
@@ -534,7 +603,7 @@ const static struct lov_layout_operations lov_dispatch[] = {
                 .llo_page_init = lov_page_init_empty,
                 .llo_lock_init = lov_lock_init_empty,
                 .llo_io_init   = lov_io_init_empty,
-                .llo_getattr   = lov_attr_get_empty
+               .llo_getattr   = lov_attr_get_empty,
         },
         [LLT_RAID0] = {
                 .llo_init      = lov_init_raid0,
@@ -545,7 +614,7 @@ const static struct lov_layout_operations lov_dispatch[] = {
                 .llo_page_init = lov_page_init_raid0,
                 .llo_lock_init = lov_lock_init_raid0,
                 .llo_io_init   = lov_io_init_raid0,
-                .llo_getattr   = lov_attr_get_raid0
+               .llo_getattr   = lov_attr_get_raid0,
        },
         [LLT_RELEASED] = {
                 .llo_init      = lov_init_released,
@@ -556,27 +625,27 @@ const static struct lov_layout_operations lov_dispatch[] = {
                 .llo_page_init = lov_page_init_empty,
                 .llo_lock_init = lov_lock_init_empty,
                 .llo_io_init   = lov_io_init_released,
-                .llo_getattr   = lov_attr_get_empty
+               .llo_getattr   = lov_attr_get_empty,
         }
 };
 
 /**
  * Performs a double-dispatch based on the layout type of an object.
  */
-#define LOV_2DISPATCH_NOLOCK(obj, op, ...)                              \
-({                                                                      \
-        struct lov_object                      *__obj = (obj);          \
-        enum lov_layout_type                    __llt;                  \
-                                                                        \
-        __llt = __obj->lo_type;                                         \
-        LASSERT(0 <= __llt && __llt < ARRAY_SIZE(lov_dispatch));        \
-        lov_dispatch[__llt].op(__VA_ARGS__);                            \
+#define LOV_2DISPATCH_NOLOCK(obj, op, ...)             \
+({                                                     \
+       struct lov_object *__obj = (obj);               \
+       enum lov_layout_type __llt;                     \
+                                                       \
+       __llt = __obj->lo_type;                         \
+       LASSERT(__llt < ARRAY_SIZE(lov_dispatch));      \
+       lov_dispatch[__llt].op(__VA_ARGS__);            \
 })
 
 /**
  * Return lov_layout_type associated with a given lsm
  */
-enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
+static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
 {
        if (lsm == NULL)
                return LLT_EMPTY;
@@ -587,12 +656,16 @@ enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
 
 static inline void lov_conf_freeze(struct lov_object *lov)
 {
+       CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
+               lov, lov->lo_owner, current);
        if (lov->lo_owner != current)
                down_read(&lov->lo_type_guard);
 }
 
 static inline void lov_conf_thaw(struct lov_object *lov)
 {
+       CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
+               lov, lov->lo_owner, current);
        if (lov->lo_owner != current)
                up_read(&lov->lo_type_guard);
 }
@@ -624,7 +697,7 @@ do {                                                                    \
                                                                         \
        lov_conf_freeze(__obj);                                         \
         __llt = __obj->lo_type;                                         \
-        LASSERT(0 <= __llt && __llt < ARRAY_SIZE(lov_dispatch));        \
+       LASSERT(__llt < ARRAY_SIZE(lov_dispatch));                      \
         lov_dispatch[__llt].op(__VA_ARGS__);                            \
        lov_conf_thaw(__obj);                                           \
 } while (0)
@@ -635,10 +708,14 @@ static void lov_conf_lock(struct lov_object *lov)
        down_write(&lov->lo_type_guard);
        LASSERT(lov->lo_owner == NULL);
        lov->lo_owner = current;
+       CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
+               lov, lov->lo_owner);
 }
 
 static void lov_conf_unlock(struct lov_object *lov)
 {
+       CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
+               lov, lov->lo_owner);
        lov->lo_owner = NULL;
        up_write(&lov->lo_type_guard);
 }
@@ -648,45 +725,38 @@ static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
        struct l_wait_info lwi = { 0 };
        ENTRY;
 
-       while (cfs_atomic_read(&lov->lo_active_ios) > 0) {
+       while (atomic_read(&lov->lo_active_ios) > 0) {
                CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
                        PFID(lu_object_fid(lov2lu(lov))),
-                       cfs_atomic_read(&lov->lo_active_ios));
+                       atomic_read(&lov->lo_active_ios));
 
                l_wait_event(lov->lo_waitq,
-                            cfs_atomic_read(&lov->lo_active_ios) == 0, &lwi);
+                            atomic_read(&lov->lo_active_ios) == 0, &lwi);
        }
        RETURN(0);
 }
 
 static int lov_layout_change(const struct lu_env *unused,
-                            struct lov_object *lov,
+                            struct lov_object *lov, struct lov_stripe_md *lsm,
                             const struct cl_object_conf *conf)
 {
-       int result;
-       enum lov_layout_type llt = LLT_EMPTY;
+       enum lov_layout_type llt = lov_type(lsm);
        union lov_layout_state *state = &lov->u;
        const struct lov_layout_operations *old_ops;
        const struct lov_layout_operations *new_ops;
-
-       struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
-       void *cookie;
+       struct lov_device *lov_dev = lov_object_dev(lov);
        struct lu_env *env;
-       int refcheck;
+       __u16 refcheck;
+       int rc;
        ENTRY;
 
-       LASSERT(0 <= lov->lo_type && lov->lo_type < ARRAY_SIZE(lov_dispatch));
+       LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
 
-       if (conf->u.coc_md != NULL)
-               llt = lov_type(conf->u.coc_md->lsm);
-       LASSERT(0 <= llt && llt < ARRAY_SIZE(lov_dispatch));
-
-       cookie = cl_env_reenter();
        env = cl_env_get(&refcheck);
-       if (IS_ERR(env)) {
-               cl_env_reexit(cookie);
+       if (IS_ERR(env))
                RETURN(PTR_ERR(env));
-       }
+
+       LASSERT(llt < ARRAY_SIZE(lov_dispatch));
 
        CDEBUG(D_INODE, DFID" from %s to %s\n",
               PFID(lu_object_fid(lov2lu(lov))),
@@ -695,31 +765,45 @@ static int lov_layout_change(const struct lu_env *unused,
        old_ops = &lov_dispatch[lov->lo_type];
        new_ops = &lov_dispatch[llt];
 
-       result = old_ops->llo_delete(env, lov, &lov->u);
-       if (result == 0) {
-               old_ops->llo_fini(env, lov, &lov->u);
+       rc = cl_object_prune(env, &lov->lo_cl);
+       if (rc != 0)
+               GOTO(out, rc);
 
-               LASSERT(cfs_atomic_read(&lov->lo_active_ios) == 0);
-               LASSERT(hdr->coh_tree.rnode == NULL);
-               LASSERT(hdr->coh_pages == 0);
+       rc = old_ops->llo_delete(env, lov, &lov->u);
+       if (rc != 0)
+               GOTO(out, rc);
 
-               lov->lo_type = LLT_EMPTY;
-               result = new_ops->llo_init(env,
-                                       lu2lov_dev(lov->lo_cl.co_lu.lo_dev),
-                                       lov, conf, state);
-               if (result == 0) {
-                       new_ops->llo_install(env, lov, state);
-                       lov->lo_type = llt;
-               } else {
-                       new_ops->llo_delete(env, lov, state);
-                       new_ops->llo_fini(env, lov, state);
-                       /* this file becomes an EMPTY file. */
-               }
+       old_ops->llo_fini(env, lov, &lov->u);
+
+       LASSERT(atomic_read(&lov->lo_active_ios) == 0);
+
+       CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
+              PFID(lu_object_fid(lov2lu(lov))), lov, llt);
+
+       lov->lo_type = LLT_EMPTY;
+
+       /* page bufsize fixup */
+       cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
+               lov_page_slice_fixup(lov, NULL);
+
+       rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
+       if (rc != 0) {
+               struct obd_device *obd = lov2obd(lov_dev->ld_lov);
+
+               CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
+                      obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
+               new_ops->llo_delete(env, lov, state);
+               new_ops->llo_fini(env, lov, state);
+               /* this file becomes an EMPTY file. */
+               GOTO(out, rc);
        }
 
+       new_ops->llo_install(env, lov, state);
+       lov->lo_type = llt;
+
+out:
        cl_env_put(env, &refcheck);
-       cl_env_reexit(cookie);
-       RETURN(result);
+       RETURN(rc);
 }
 
 /*****************************************************************************
@@ -728,29 +812,44 @@ static int lov_layout_change(const struct lu_env *unused,
  *
  */
 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
-                    const struct lu_object_conf *conf)
+                   const struct lu_object_conf *conf)
 {
-        struct lov_device            *dev   = lu2lov_dev(obj->lo_dev);
-        struct lov_object            *lov   = lu2lov(obj);
-        const struct cl_object_conf  *cconf = lu2cl_conf(conf);
-        union  lov_layout_state      *set   = &lov->u;
-        const struct lov_layout_operations *ops;
-        int result;
+       struct lov_object            *lov   = lu2lov(obj);
+       struct lov_device            *dev   = lov_object_dev(lov);
+       const struct cl_object_conf  *cconf = lu2cl_conf(conf);
+       union lov_layout_state       *set   = &lov->u;
+       const struct lov_layout_operations *ops;
+       struct lov_stripe_md *lsm = NULL;
+       int rc;
+       ENTRY;
 
-        ENTRY;
        init_rwsem(&lov->lo_type_guard);
-       cfs_atomic_set(&lov->lo_active_ios, 0);
+       atomic_set(&lov->lo_active_ios, 0);
        init_waitqueue_head(&lov->lo_waitq);
-
        cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
 
-        /* no locking is necessary, as object is being created */
-       lov->lo_type = lov_type(cconf->u.coc_md->lsm);
-        ops = &lov_dispatch[lov->lo_type];
-        result = ops->llo_init(env, dev, lov, cconf, set);
-        if (result == 0)
-                ops->llo_install(env, lov, set);
-        RETURN(result);
+       lov->lo_type = LLT_EMPTY;
+       if (cconf->u.coc_layout.lb_buf != NULL) {
+               lsm = lov_unpackmd(dev->ld_lov,
+                                  cconf->u.coc_layout.lb_buf,
+                                  cconf->u.coc_layout.lb_len);
+               if (IS_ERR(lsm))
+                       RETURN(PTR_ERR(lsm));
+       }
+
+       /* no locking is necessary, as object is being created */
+       lov->lo_type = lov_type(lsm);
+       ops = &lov_dispatch[lov->lo_type];
+       rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
+       if (rc != 0)
+               GOTO(out_lsm, rc);
+
+       ops->llo_install(env, lov, set);
+
+out_lsm:
+       lov_lsm_put(lsm);
+
+       RETURN(rc);
 }
 
 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
@@ -761,6 +860,15 @@ static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
        int                      result = 0;
        ENTRY;
 
+       if (conf->coc_opc == OBJECT_CONF_SET &&
+           conf->u.coc_layout.lb_buf != NULL) {
+               lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
+                                  conf->u.coc_layout.lb_buf,
+                                  conf->u.coc_layout.lb_len);
+               if (IS_ERR(lsm))
+                       RETURN(PTR_ERR(lsm));
+       }
+
        lov_conf_lock(lov);
        if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
                lov->lo_layout_invalid = true;
@@ -769,7 +877,7 @@ static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
 
        if (conf->coc_opc == OBJECT_CONF_WAIT) {
                if (lov->lo_layout_invalid &&
-                   cfs_atomic_read(&lov->lo_active_ios) > 0) {
+                   atomic_read(&lov->lo_active_ios) > 0) {
                        lov_conf_unlock(lov);
                        result = lov_layout_wait(env, lov);
                        lov_conf_lock(lov);
@@ -779,8 +887,6 @@ static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
 
        LASSERT(conf->coc_opc == OBJECT_CONF_SET);
 
-       if (conf->u.coc_md != NULL)
-               lsm = conf->u.coc_md->lsm;
        if ((lsm == NULL && lov->lo_lsm == NULL) ||
            ((lsm != NULL && lov->lo_lsm != NULL) &&
             (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
@@ -791,16 +897,18 @@ static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
        }
 
        /* will change layout - check if there still exists active IO. */
-       if (cfs_atomic_read(&lov->lo_active_ios) > 0) {
+       if (atomic_read(&lov->lo_active_ios) > 0) {
                lov->lo_layout_invalid = true;
                GOTO(out, result = -EBUSY);
        }
 
-       lov->lo_layout_invalid = lov_layout_change(env, lov, conf);
+       result = lov_layout_change(env, lov, lsm, conf);
+       lov->lo_layout_invalid = result != 0;
        EXIT;
 
 out:
        lov_conf_unlock(lov);
+       lov_lsm_put(lsm);
        CDEBUG(D_INODE, DFID" lo_layout_invalid=%d\n",
               PFID(lu_object_fid(lov2lu(lov))), lov->lo_layout_invalid);
        RETURN(result);
@@ -833,10 +941,10 @@ static int lov_object_print(const struct lu_env *env, void *cookie,
 }
 
 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
-                 struct cl_page *page, struct page *vmpage)
+                 struct cl_page *page, pgoff_t index)
 {
-        return LOV_2DISPATCH_NOLOCK(cl2lov(obj),
-                                   llo_page_init, env, obj, page, vmpage);
+       return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
+                                   index);
 }
 
 /**
@@ -847,6 +955,11 @@ int lov_io_init(const struct lu_env *env, struct cl_object *obj,
                struct cl_io *io)
 {
        CL_IO_SLICE_CLEAN(lov_env_io(env), lis_cl);
+
+       CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
+              PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
+              io->ci_ignore_layout, io->ci_verify_layout);
+
        return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
                                     !io->ci_ignore_layout, env, obj, io);
 }
@@ -864,13 +977,13 @@ static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
 }
 
-static int lov_attr_set(const struct lu_env *env, struct cl_object *obj,
-                        const struct cl_attr *attr, unsigned valid)
+static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
+                          const struct cl_attr *attr, unsigned valid)
 {
-        /*
-         * No dispatch is required here, as no layout implements this.
-         */
-        return 0;
+       /*
+        * No dispatch is required here, as no layout implements this.
+        */
+       return 0;
 }
 
 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
@@ -881,13 +994,543 @@ int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
                                    io);
 }
 
+/**
+ * We calculate on which OST the mapping will end. If the length of mapping
+ * is greater than (stripe_size * stripe_count) then the last_stripe will
+ * will be one just before start_stripe. Else we check if the mapping
+ * intersects each OST and find last_stripe.
+ * This function returns the last_stripe and also sets the stripe_count
+ * over which the mapping is spread
+ *
+ * \param lsm [in]             striping information for the file
+ * \param fm_start [in]                logical start of mapping
+ * \param fm_end [in]          logical end of mapping
+ * \param start_stripe [in]    starting stripe of the mapping
+ * \param stripe_count [out]   the number of stripes across which to map is
+ *                             returned
+ *
+ * \retval last_stripe         return the last stripe of the mapping
+ */
+static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm,
+                                  u64 fm_start, u64 fm_end,
+                                  int start_stripe, int *stripe_count)
+{
+       int last_stripe;
+       u64 obd_start;
+       u64 obd_end;
+       int i, j;
+
+       if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
+               last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
+                                                             start_stripe - 1);
+               *stripe_count = lsm->lsm_stripe_count;
+       } else {
+               for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
+                    i = (i + 1) % lsm->lsm_stripe_count, j++) {
+                       if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
+                                                  &obd_start, &obd_end)) == 0)
+                               break;
+               }
+               *stripe_count = j;
+               last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
+       }
+
+       return last_stripe;
+}
+
+/**
+ * Set fe_device and copy extents from local buffer into main return buffer.
+ *
+ * \param fiemap [out]         fiemap to hold all extents
+ * \param lcl_fm_ext [in]      array of fiemap extents get from OSC layer
+ * \param ost_index [in]       OST index to be written into the fm_device
+ *                             field for each extent
+ * \param ext_count [in]       number of extents to be copied
+ * \param current_extent [in]  where to start copying in the extent array
+ */
+static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
+                                        struct fiemap_extent *lcl_fm_ext,
+                                        int ost_index, unsigned int ext_count,
+                                        int current_extent)
+{
+       char            *to;
+       unsigned int    ext;
+
+       for (ext = 0; ext < ext_count; ext++) {
+               lcl_fm_ext[ext].fe_device = ost_index;
+               lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
+       }
+
+       /* Copy fm_extent's from fm_local to return buffer */
+       to = (char *)fiemap + fiemap_count_to_size(current_extent);
+       memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
+}
+
+#define FIEMAP_BUFFER_SIZE 4096
+
+/**
+ * Non-zero fe_logical indicates that this is a continuation FIEMAP
+ * call. The local end offset and the device are sent in the first
+ * fm_extent. This function calculates the stripe number from the index.
+ * This function returns a stripe_no on which mapping is to be restarted.
+ *
+ * This function returns fm_end_offset which is the in-OST offset at which
+ * mapping should be restarted. If fm_end_offset=0 is returned then caller
+ * will re-calculate proper offset in next stripe.
+ * Note that the first extent is passed to lov_get_info via the value field.
+ *
+ * \param fiemap [in]          fiemap request header
+ * \param lsm [in]             striping information for the file
+ * \param fm_start [in]                logical start of mapping
+ * \param fm_end [in]          logical end of mapping
+ * \param start_stripe [out]   starting stripe will be returned in this
+ */
+static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
+                                    struct lov_stripe_md *lsm,
+                                    u64 fm_start, u64 fm_end,
+                                    int *start_stripe)
+{
+       u64 local_end = fiemap->fm_extents[0].fe_logical;
+       u64 lun_start;
+       u64 lun_end;
+       u64 fm_end_offset;
+       int stripe_no = -1;
+       int i;
+
+       if (fiemap->fm_extent_count == 0 ||
+           fiemap->fm_extents[0].fe_logical == 0)
+               return 0;
+
+       /* Find out stripe_no from ost_index saved in the fe_device */
+       for (i = 0; i < lsm->lsm_stripe_count; i++) {
+               struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
+
+               if (lov_oinfo_is_dummy(oinfo))
+                       continue;
+
+               if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
+                       stripe_no = i;
+                       break;
+               }
+       }
+
+       if (stripe_no == -1)
+               return -EINVAL;
+
+       /* If we have finished mapping on previous device, shift logical
+        * offset to start of next device */
+       if (lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
+                                 &lun_start, &lun_end) != 0 &&
+           local_end < lun_end) {
+               fm_end_offset = local_end;
+               *start_stripe = stripe_no;
+       } else {
+               /* This is a special value to indicate that caller should
+                * calculate offset in next stripe. */
+               fm_end_offset = 0;
+               *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
+       }
+
+       return fm_end_offset;
+}
+
+struct fiemap_state {
+       struct fiemap   *fs_fm;
+       u64             fs_start;
+       u64             fs_length;
+       u64             fs_end;
+       u64             fs_end_offset;
+       int             fs_cur_extent;
+       int             fs_cnt_need;
+       int             fs_start_stripe;
+       int             fs_last_stripe;
+       bool            fs_device_done;
+       bool            fs_finish;
+       bool            fs_enough;
+};
+
+int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
+                     struct lov_stripe_md *lsm,
+                     struct fiemap *fiemap, size_t *buflen,
+                     struct ll_fiemap_info_key *fmkey, int stripeno,
+                     struct fiemap_state *fs)
+{
+       struct cl_object *subobj;
+       struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
+       struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
+       u64 req_fm_len; /* Stores length of required mapping */
+       u64 len_mapped_single_call;
+       u64 lun_start;
+       u64 lun_end;
+       u64 obd_object_end;
+       unsigned int ext_count;
+       /* EOF for object */
+       bool ost_eof = false;
+       /* done with required mapping for this OST? */
+       bool ost_done = false;
+       int ost_index;
+       int rc = 0;
+
+       fs->fs_device_done = false;
+       /* Find out range of mapping on this stripe */
+       if ((lov_stripe_intersects(lsm, stripeno, fs->fs_start, fs->fs_end,
+                                  &lun_start, &obd_object_end)) == 0)
+               return 0;
+
+       if (lov_oinfo_is_dummy(lsm->lsm_oinfo[stripeno]))
+               return -EIO;
+
+       /* If this is a continuation FIEMAP call and we are on
+        * starting stripe then lun_start needs to be set to
+        * end_offset */
+       if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
+               lun_start = fs->fs_end_offset;
+
+       lun_end = fs->fs_length;
+       if (lun_end != ~0ULL) {
+               /* Handle fs->fs_start + fs->fs_length overflow */
+               if (fs->fs_start + fs->fs_length < fs->fs_start)
+                       fs->fs_length = ~0ULL - fs->fs_start;
+               lun_end = lov_size_to_stripe(lsm, fs->fs_start + fs->fs_length,
+                                            stripeno);
+       }
+
+       if (lun_start == lun_end)
+               return 0;
+
+       req_fm_len = obd_object_end - lun_start;
+       fs->fs_fm->fm_length = 0;
+       len_mapped_single_call = 0;
+
+       /* find lobsub object */
+       subobj = lov_find_subobj(env, cl2lov(obj), lsm, stripeno);
+       if (IS_ERR(subobj))
+               return PTR_ERR(subobj);
+       /* If the output buffer is very large and the objects have many
+        * extents we may need to loop on a single OST repeatedly */
+       do {
+               if (fiemap->fm_extent_count > 0) {
+                       /* Don't get too many extents. */
+                       if (fs->fs_cur_extent + fs->fs_cnt_need >
+                           fiemap->fm_extent_count)
+                               fs->fs_cnt_need = fiemap->fm_extent_count -
+                                                 fs->fs_cur_extent;
+               }
+
+               lun_start += len_mapped_single_call;
+               fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
+               req_fm_len = fs->fs_fm->fm_length;
+               fs->fs_fm->fm_extent_count = fs->fs_enough ?
+                                            1 : fs->fs_cnt_need;
+               fs->fs_fm->fm_mapped_extents = 0;
+               fs->fs_fm->fm_flags = fiemap->fm_flags;
+
+               ost_index = lsm->lsm_oinfo[stripeno]->loi_ost_idx;
+
+               if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
+                       GOTO(obj_put, rc = -EINVAL);
+               /* If OST is inactive, return extent with UNKNOWN flag. */
+               if (!lov->lov_tgts[ost_index]->ltd_active) {
+                       fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
+                       fs->fs_fm->fm_mapped_extents = 1;
+
+                       fm_ext[0].fe_logical = lun_start;
+                       fm_ext[0].fe_length = obd_object_end - lun_start;
+                       fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
+
+                       goto inactive_tgt;
+               }
+
+               fs->fs_fm->fm_start = lun_start;
+               fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
+               memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
+               *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
+
+               rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
+               if (rc != 0)
+                       GOTO(obj_put, rc);
+inactive_tgt:
+               ext_count = fs->fs_fm->fm_mapped_extents;
+               if (ext_count == 0) {
+                       ost_done = true;
+                       fs->fs_device_done = true;
+                       /* If last stripe has hold at the end,
+                        * we need to return */
+                       if (stripeno == fs->fs_last_stripe) {
+                               fiemap->fm_mapped_extents = 0;
+                               fs->fs_finish = true;
+                               GOTO(obj_put, rc);
+                       }
+                       break;
+               } else if (fs->fs_enough) {
+                       /*
+                        * We've collected enough extents and there are
+                        * more extents after it.
+                        */
+                       fs->fs_finish = true;
+                       GOTO(obj_put, rc);
+               }
+
+               /* If we just need num of extents, got to next device */
+               if (fiemap->fm_extent_count == 0) {
+                       fs->fs_cur_extent += ext_count;
+                       break;
+               }
+
+               /* prepare to copy retrived map extents */
+               len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
+                                        fm_ext[ext_count - 1].fe_length -
+                                        lun_start;
+
+               /* Have we finished mapping on this device? */
+               if (req_fm_len <= len_mapped_single_call) {
+                       ost_done = true;
+                       fs->fs_device_done = true;
+               }
+
+               /* Clear the EXTENT_LAST flag which can be present on
+                * the last extent */
+               if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
+                       fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
+               if (lov_stripe_size(lsm, fm_ext[ext_count - 1].fe_logical +
+                                        fm_ext[ext_count - 1].fe_length,
+                                   stripeno) >= fmkey->lfik_oa.o_size) {
+                       ost_eof = true;
+                       fs->fs_device_done = true;
+               }
+
+               fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
+                                            ext_count, fs->fs_cur_extent);
+               fs->fs_cur_extent += ext_count;
+
+               /* Ran out of available extents? */
+               if (fs->fs_cur_extent >= fiemap->fm_extent_count)
+                       fs->fs_enough = true;
+       } while (!ost_done && !ost_eof);
+
+       if (stripeno == fs->fs_last_stripe)
+               fs->fs_finish = true;
+obj_put:
+       cl_object_put(env, subobj);
+
+       return rc;
+}
+
+/**
+ * Break down the FIEMAP request and send appropriate calls to individual OSTs.
+ * This also handles the restarting of FIEMAP calls in case mapping overflows
+ * the available number of extents in single call.
+ *
+ * \param env [in]             lustre environment
+ * \param obj [in]             file object
+ * \param fmkey [in]           fiemap request header and other info
+ * \param fiemap [out]         fiemap buffer holding retrived map extents
+ * \param buflen [in/out]      max buffer length of @fiemap, when iterate
+ *                             each OST, it is used to limit max map needed
+ * \retval 0   success
+ * \retval < 0 error
+ */
+static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
+                            struct ll_fiemap_info_key *fmkey,
+                            struct fiemap *fiemap, size_t *buflen)
+{
+       struct lov_stripe_md    *lsm;
+       struct fiemap           *fm_local = NULL;
+       int                     cur_stripe;
+       int                     stripe_count;
+       unsigned int            buffer_size = FIEMAP_BUFFER_SIZE;
+       int                     rc = 0;
+       struct fiemap_state fs = { 0 };
+       ENTRY;
+
+       lsm = lov_lsm_addref(cl2lov(obj));
+       if (lsm == NULL)
+               RETURN(-ENODATA);
+
+       /**
+        * If the stripe_count > 1 and the application does not understand
+        * DEVICE_ORDER flag, it cannot interpret the extents correctly.
+        */
+       if (lsm->lsm_stripe_count > 1 && !(fiemap->fm_flags &
+                                          FIEMAP_FLAG_DEVICE_ORDER))
+               GOTO(out_lsm, rc = -ENOTSUPP);
+
+       if (lsm_is_released(lsm)) {
+               if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
+                       /**
+                        * released file, return a minimal FIEMAP if
+                        * request fits in file-size.
+                        */
+                       fiemap->fm_mapped_extents = 1;
+                       fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
+                       if (fiemap->fm_start + fiemap->fm_length <
+                           fmkey->lfik_oa.o_size)
+                               fiemap->fm_extents[0].fe_length =
+                                       fiemap->fm_length;
+                       else
+                               fiemap->fm_extents[0].fe_length =
+                                       fmkey->lfik_oa.o_size -
+                                       fiemap->fm_start;
+                       fiemap->fm_extents[0].fe_flags |=
+                               FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
+               }
+               GOTO(out_lsm, rc = 0);
+       }
+
+       if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
+               buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
+
+       OBD_ALLOC_LARGE(fm_local, buffer_size);
+       if (fm_local == NULL)
+               GOTO(out_lsm, rc = -ENOMEM);
+
+       fs.fs_fm = fm_local;
+       fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
+
+       fs.fs_start = fiemap->fm_start;
+       /* fs.fs_start is beyond the end of the file */
+       if (fs.fs_start > fmkey->lfik_oa.o_size)
+               GOTO(out_fm_local, rc = -EINVAL);
+
+       fs.fs_length = fiemap->fm_length;
+       /* Calculate start stripe, last stripe and length of mapping */
+       fs.fs_start_stripe = lov_stripe_number(lsm, fs.fs_start);
+       fs.fs_end = (fs.fs_length == ~0ULL) ? fmkey->lfik_oa.o_size :
+                                             fs.fs_start + fs.fs_length - 1;
+       /* If fs_length != ~0ULL but fs_start+fs_length-1 exceeds file size */
+       if (fs.fs_end > fmkey->lfik_oa.o_size) {
+               fs.fs_end = fmkey->lfik_oa.o_size;
+               fs.fs_length = fs.fs_end - fs.fs_start;
+       }
+
+       fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, fs.fs_start, fs.fs_end,
+                                                   fs.fs_start_stripe,
+                                                   &stripe_count);
+       fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fs.fs_start,
+                                                    fs.fs_end,
+                                                    &fs.fs_start_stripe);
+       if (fs.fs_end_offset == -EINVAL)
+               GOTO(out_fm_local, rc = -EINVAL);
+
+       /**
+        * Requested extent count exceeds the fiemap buffer size, shrink our
+        * ambition.
+        */
+       if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
+               fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
+       if (fiemap->fm_extent_count == 0)
+               fs.fs_cnt_need = 0;
+
+       fs.fs_finish = false;
+       fs.fs_enough = false;
+       fs.fs_cur_extent = 0;
+
+       /* Check each stripe */
+       for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
+            --stripe_count,
+            cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
+               rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen, fmkey,
+                                      cur_stripe, &fs);
+               if (rc < 0)
+                       GOTO(out_fm_local, rc);
+               if (fs.fs_finish)
+                       break;
+       } /* for each stripe */
+
+       /* Indicate that we are returning device offsets unless file just has
+        * single stripe */
+       if (lsm->lsm_stripe_count > 1)
+               fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
+
+       if (fiemap->fm_extent_count == 0)
+               goto skip_last_device_calc;
+
+       /* Check if we have reached the last stripe and whether mapping for that
+        * stripe is done. */
+       if ((cur_stripe == fs.fs_last_stripe) && fs.fs_device_done)
+               fiemap->fm_extents[fs.fs_cur_extent - 1].fe_flags |=
+                                                            FIEMAP_EXTENT_LAST;
+skip_last_device_calc:
+       fiemap->fm_mapped_extents = fs.fs_cur_extent;
+out_fm_local:
+       OBD_FREE_LARGE(fm_local, buffer_size);
+
+out_lsm:
+       lov_lsm_put(lsm);
+
+       return rc;
+}
+
+static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
+                               struct lov_user_md __user *lum)
+{
+       struct lov_object       *lov = cl2lov(obj);
+       struct lov_stripe_md    *lsm;
+       int                     rc = 0;
+       ENTRY;
+
+       lsm = lov_lsm_addref(lov);
+       if (lsm == NULL)
+               RETURN(-ENODATA);
+
+       rc = lov_getstripe(cl2lov(obj), lsm, lum);
+       lov_lsm_put(lsm);
+       RETURN(rc);
+}
+
+static int lov_object_layout_get(const struct lu_env *env,
+                                struct cl_object *obj,
+                                struct cl_layout *cl)
+{
+       struct lov_object *lov = cl2lov(obj);
+       struct lov_stripe_md *lsm = lov_lsm_addref(lov);
+       struct lu_buf *buf = &cl->cl_buf;
+       ssize_t rc;
+       ENTRY;
+
+       if (lsm == NULL) {
+               cl->cl_size = 0;
+               cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
+
+               RETURN(0);
+       }
+
+       cl->cl_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
+       cl->cl_layout_gen = lsm->lsm_layout_gen;
+
+       rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
+       lov_lsm_put(lsm);
+
+       RETURN(rc < 0 ? rc : 0);
+}
+
+static loff_t lov_object_maxbytes(struct cl_object *obj)
+{
+       struct lov_object *lov = cl2lov(obj);
+       struct lov_stripe_md *lsm = lov_lsm_addref(lov);
+       loff_t maxbytes;
+
+       if (lsm == NULL)
+               return LLONG_MAX;
+
+       maxbytes = lsm->lsm_maxbytes;
+
+       lov_lsm_put(lsm);
+
+       return maxbytes;
+}
+
 static const struct cl_object_operations lov_ops = {
-        .coo_page_init = lov_page_init,
-        .coo_lock_init = lov_lock_init,
-        .coo_io_init   = lov_io_init,
-        .coo_attr_get  = lov_attr_get,
-        .coo_attr_set  = lov_attr_set,
-        .coo_conf_set  = lov_conf_set
+       .coo_page_init    = lov_page_init,
+       .coo_lock_init    = lov_lock_init,
+       .coo_io_init      = lov_io_init,
+       .coo_attr_get     = lov_attr_get,
+       .coo_attr_update  = lov_attr_update,
+       .coo_conf_set     = lov_conf_set,
+       .coo_getstripe    = lov_object_getstripe,
+       .coo_layout_get   = lov_object_layout_get,
+       .coo_maxbytes     = lov_object_maxbytes,
+       .coo_fiemap       = lov_object_fiemap,
 };
 
 static const struct lu_object_operations lov_lu_obj_ops = {
@@ -900,28 +1543,28 @@ static const struct lu_object_operations lov_lu_obj_ops = {
 };
 
 struct lu_object *lov_object_alloc(const struct lu_env *env,
-                                   const struct lu_object_header *unused,
-                                   struct lu_device *dev)
+                                  const struct lu_object_header *unused,
+                                  struct lu_device *dev)
 {
-        struct lov_object *lov;
-        struct lu_object  *obj;
+       struct lov_object *lov;
+       struct lu_object  *obj;
 
-        ENTRY;
-       OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, __GFP_IO);
-        if (lov != NULL) {
-                obj = lov2lu(lov);
-                lu_object_init(obj, NULL, dev);
-                lov->lo_cl.co_ops = &lov_ops;
-                lov->lo_type = -1; /* invalid, to catch uninitialized type */
-                /*
-                 * object io operation vector (cl_object::co_iop) is installed
-                 * later in lov_object_init(), as different vectors are used
-                 * for object with different layouts.
-                 */
-                obj->lo_ops = &lov_lu_obj_ops;
-        } else
-                obj = NULL;
-        RETURN(obj);
+       ENTRY;
+       OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
+       if (lov != NULL) {
+               obj = lov2lu(lov);
+               lu_object_init(obj, NULL, dev);
+               lov->lo_cl.co_ops = &lov_ops;
+               lov->lo_type = -1; /* invalid, to catch uninitialized type */
+               /*
+                * object io operation vector (cl_object::co_iop) is installed
+                * later in lov_object_init(), as different vectors are used
+                * for object with different layouts.
+                */
+               obj->lo_ops = &lov_lu_obj_ops;
+       } else
+               obj = NULL;
+       RETURN(obj);
 }
 
 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
@@ -932,47 +1575,13 @@ struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
        if (lov->lo_lsm != NULL) {
                lsm = lsm_addref(lov->lo_lsm);
                CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
-                       lsm, cfs_atomic_read(&lsm->lsm_refc),
+                       lsm, atomic_read(&lsm->lsm_refc),
                        lov->lo_layout_invalid, current);
        }
        lov_conf_thaw(lov);
        return lsm;
 }
 
-void lov_lsm_decref(struct lov_object *lov, struct lov_stripe_md *lsm)
-{
-       if (lsm == NULL)
-               return;
-
-       CDEBUG(D_INODE, "lsm %p decref %d by %p.\n",
-               lsm, cfs_atomic_read(&lsm->lsm_refc), current);
-
-       lov_free_memmd(&lsm);
-}
-
-struct lov_stripe_md *lov_lsm_get(struct cl_object *clobj)
-{
-       struct lu_object *luobj;
-       struct lov_stripe_md *lsm = NULL;
-
-       if (clobj == NULL)
-               return NULL;
-
-       luobj = lu_object_locate(&cl_object_header(clobj)->coh_lu,
-                                &lov_device_type);
-       if (luobj != NULL)
-               lsm = lov_lsm_addref(lu2lov(luobj));
-       return lsm;
-}
-EXPORT_SYMBOL(lov_lsm_get);
-
-void lov_lsm_put(struct cl_object *unused, struct lov_stripe_md *lsm)
-{
-       if (lsm != NULL)
-               lov_free_memmd(&lsm);
-}
-EXPORT_SYMBOL(lov_lsm_put);
-
 int lov_read_and_clear_async_rc(struct cl_object *clob)
 {
        struct lu_object *luobj;
@@ -994,6 +1603,10 @@ int lov_read_and_clear_async_rc(struct cl_object *clob)
                        LASSERT(lsm != NULL);
                        for (i = 0; i < lsm->lsm_stripe_count; i++) {
                                struct lov_oinfo *loi = lsm->lsm_oinfo[i];
+
+                               if (lov_oinfo_is_dummy(loi))
+                                       continue;
+
                                if (loi->loi_ar.ar_rc && !rc)
                                        rc = loi->loi_ar.ar_rc;
                                loi->loi_ar.ar_rc = 0;