Whamcloud - gitweb
LU-1842 ldlm: support for sending GL ASTs to multiple locks
[fs/lustre-release.git] / lustre / obdfilter / filter.c
index 63a0664..ef25f36 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.
@@ -53,9 +51,6 @@
 
 #define DEBUG_SUBSYSTEM S_FILTER
 
-#ifndef AUTOCONF_INCLUDED
-#include <linux/config.h>
-#endif
 #include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/dcache.h>
@@ -1423,8 +1418,8 @@ obd_id filter_last_id(struct filter_obd *filter, obd_seq group)
 
 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
 {
-        LOCK_INODE_MUTEX_PARENT(dparent->d_inode);
-        return 0;
+       mutex_lock_nested(&dparent->d_inode->i_mutex, I_MUTEX_PARENT);
+       return 0;
 }
 
 /* We never dget the object parent, so DON'T dput it either */
@@ -1464,7 +1459,7 @@ struct dentry *filter_parent_lock(struct obd_device *obd, obd_seq group,
 /* We never dget the object parent, so DON'T dput it either */
 static void filter_parent_unlock(struct dentry *dparent)
 {
-        UNLOCK_INODE_MUTEX(dparent->d_inode);
+       mutex_unlock(&dparent->d_inode->i_mutex);
 }
 
 /* How to get files, dentries, inodes from object id's.
@@ -1573,9 +1568,9 @@ int filter_vfs_unlink(struct inode *dir, struct dentry *dentry,
 
         /* don't need dir->i_zombie for 2.4, it is for rename/unlink of dir
          * itself we already hold dir->i_mutex for child create/unlink ops */
-        LASSERT(dentry->d_inode != NULL);
-        LASSERT(TRYLOCK_INODE_MUTEX(dir) == 0);
-        LASSERT(TRYLOCK_INODE_MUTEX(dentry->d_inode) == 0);
+       LASSERT(dentry->d_inode != NULL);
+       LASSERT(mutex_trylock(&dir->i_mutex) == 0);
+       LASSERT(mutex_trylock(&dentry->d_inode->i_mutex) == 0);
 
 
         /* may_delete() */
@@ -1604,12 +1599,12 @@ int filter_vfs_unlink(struct inode *dir, struct dentry *dentry,
 
         rc = dir->i_op->unlink(dir, dentry);
 out:
-        /* need to drop i_mutex before we lose inode reference */
-        UNLOCK_INODE_MUTEX(dentry->d_inode);
-        if (rc == 0)
-                d_delete(dentry);
+       /* need to drop i_mutex before we lose inode reference */
+       mutex_unlock(&dentry->d_inode->i_mutex);
+       if (rc == 0)
+               d_delete(dentry);
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 /* Caller must hold LCK_PW on parent and push us into kernel context.
@@ -1703,6 +1698,8 @@ static int filter_intent_policy(struct ldlm_namespace *ns,
         __u32 repsize[3] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
                            [DLM_LOCKREPLY_OFF]   = sizeof(*rep),
                            [DLM_REPLY_REC_OFF]   = sizeof(*reply_lvb) };
+       struct ldlm_glimpse_work        gl_work;
+       CFS_LIST_HEAD(gl_list);
         ENTRY;
 
         policy = ldlm_get_processing_policy(res);
@@ -1825,7 +1822,21 @@ static int filter_intent_policy(struct ldlm_namespace *ns,
         }
 
         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
-        rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
+
+       /* Populate the gl_work structure.
+        * Grab additional reference on the lock which will be released in
+        * ldlm_work_gl_ast_lock() */
+       gl_work.gl_lock = LDLM_LOCK_GET(l);
+       /* The glimpse callback is sent to one single extent lock. As a result,
+        * the gl_work list is just composed of one element */
+       cfs_list_add_tail(&gl_work.gl_list, &gl_list);
+       /* the ldlm_glimpse_work structure is allocated on the stack */
+       gl_work.gl_flags = LDLM_GL_WORK_NOFREE;
+
+       rc = ldlm_glimpse_locks(res, &gl_list); /* this will update the LVB */
+
+       if (!cfs_list_empty(&gl_list))
+               LDLM_LOCK_RELEASE(l);
 
         lock_res(res);
         *reply_lvb = *res_lvb;
@@ -1850,15 +1861,16 @@ static int filter_intent_policy(struct ldlm_namespace *ns,
  * at the OST layer there are only (potentially) multiple obd_device of type
  * unknown at the time of OST thread creation.
  *
- * Instead array of iobuf's is attached to struct filter_obd (->fo_iobuf_pool
- * field). This array has size OST_MAX_THREADS, so that each OST thread uses
- * it's very own iobuf.
+ * We create a cfs_hash for struct filter_obd (->fo_iobuf_hash field) on
+ * initializing, each OST thread will create it's own iobuf on the first
+ * access and insert it into ->fo_iobuf_hash with thread ID as key,
+ * so the iobuf can be found again by thread ID.
  *
  * Functions below
  *
- *     filter_kiobuf_pool_init()
+ *     filter_iobuf_pool_init()
  *
- *     filter_kiobuf_pool_done()
+ *     filter_iobuf_pool_done()
  *
  *     filter_iobuf_get()
  *
@@ -1871,21 +1883,13 @@ static int filter_intent_policy(struct ldlm_namespace *ns,
  */
 static void filter_iobuf_pool_done(struct filter_obd *filter)
 {
-        struct filter_iobuf **pool;
-        int i;
+       ENTRY;
 
-        ENTRY;
-
-        pool = filter->fo_iobuf_pool;
-        if (pool != NULL) {
-                for (i = 0; i < filter->fo_iobuf_count; ++ i) {
-                        if (pool[i] != NULL)
-                                filter_free_iobuf(pool[i]);
-                }
-                OBD_FREE(pool, filter->fo_iobuf_count * sizeof pool[0]);
-                filter->fo_iobuf_pool = NULL;
-        }
-        EXIT;
+       if (filter->fo_iobuf_hash != NULL) {
+               cfs_hash_putref(filter->fo_iobuf_hash);
+               filter->fo_iobuf_hash = NULL;
+       }
+       EXIT;
 }
 
 static int filter_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
@@ -1912,50 +1916,126 @@ static int filter_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
         return 0;
 }
 
-/*
- * pre-allocate pool of iobuf's to be used by filter_{prep,commit}rw_write().
- */
-static int filter_iobuf_pool_init(struct filter_obd *filter)
+static unsigned
+filter_iobuf_hop_hash(cfs_hash_t *hs, const void *key, unsigned mask)
 {
-        void **pool;
+       __u64   val = *((__u64 *)key);
 
-        ENTRY;
+       return cfs_hash_long(val, hs->hs_cur_bits);
+}
 
+static void *
+filter_iobuf_hop_key(cfs_hlist_node_t *hnode)
+{
+       struct filter_iobuf     *pool;
 
-        OBD_ALLOC_GFP(filter->fo_iobuf_pool, OSS_THREADS_MAX * sizeof(*pool),
-                      GFP_KERNEL);
-        if (filter->fo_iobuf_pool == NULL)
-                RETURN(-ENOMEM);
+       pool = cfs_hlist_entry(hnode, struct filter_iobuf, dr_hlist);
+       return &pool->dr_hkey;
+}
 
-        filter->fo_iobuf_count = OSS_THREADS_MAX;
+static int
+filter_iobuf_hop_keycmp(const void *key, cfs_hlist_node_t *hnode)
+{
+       struct filter_iobuf     *pool;
 
-        RETURN(0);
+       pool = cfs_hlist_entry(hnode, struct filter_iobuf, dr_hlist);
+       return pool->dr_hkey == *((__u64 *)key);
 }
 
-/* Return iobuf allocated for @thread_id.  We don't know in advance how
- * many threads there will be so we allocate a large empty array and only
- * fill in those slots that are actually in use.
- * If we haven't allocated a pool entry for this thread before, do so now. */
-void *filter_iobuf_get(struct filter_obd *filter, struct obd_trans_info *oti)
+static void *
+filter_iobuf_hop_object(cfs_hlist_node_t *hnode)
 {
-        int thread_id                    = (oti && oti->oti_thread) ?
-                                           oti->oti_thread->t_id : -1;
-        struct filter_iobuf  *pool       = NULL;
-        struct filter_iobuf **pool_place = NULL;
+       return cfs_hlist_entry(hnode, struct filter_iobuf, dr_hlist);
+}
 
-        if (thread_id >= 0) {
-                LASSERT(thread_id < filter->fo_iobuf_count);
-                pool = *(pool_place = &filter->fo_iobuf_pool[thread_id]);
-        }
+static void
+filter_iobuf_hop_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+       /* dummy, required by cfs_hash */
+}
 
-        if (unlikely(pool == NULL)) {
-                pool = filter_alloc_iobuf(filter, OBD_BRW_WRITE,
-                                          PTLRPC_MAX_BRW_PAGES);
-                if (pool_place != NULL)
-                        *pool_place = pool;
-        }
+static void
+filter_iobuf_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+       /* dummy, required by cfs_hash */
+}
+
+static void
+filter_iobuf_hop_exit(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+       struct filter_iobuf     *pool;
+
+       pool = cfs_hlist_entry(hnode, struct filter_iobuf, dr_hlist);
+       filter_free_iobuf(pool);
+}
+
+static struct cfs_hash_ops filter_iobuf_hops = {
+       .hs_hash        = filter_iobuf_hop_hash,
+       .hs_key         = filter_iobuf_hop_key,
+       .hs_keycmp      = filter_iobuf_hop_keycmp,
+       .hs_object      = filter_iobuf_hop_object,
+       .hs_get         = filter_iobuf_hop_get,
+       .hs_put_locked  = filter_iobuf_hop_put_locked,
+       .hs_exit        = filter_iobuf_hop_exit
+};
+
+#define FILTER_IOBUF_HASH_BITS 9
+#define FILTER_IOBUF_HBKT_BITS 4
 
-        return pool;
+/*
+ * pre-allocate pool of iobuf's to be used by filter_{prep,commit}rw_write().
+ */
+static int filter_iobuf_pool_init(struct filter_obd *filter)
+{
+       filter->fo_iobuf_hash = cfs_hash_create("filter_iobuf",
+                                               FILTER_IOBUF_HASH_BITS,
+                                               FILTER_IOBUF_HASH_BITS,
+                                               FILTER_IOBUF_HBKT_BITS, 0,
+                                               CFS_HASH_MIN_THETA,
+                                               CFS_HASH_MAX_THETA,
+                                               &filter_iobuf_hops,
+                                               CFS_HASH_RW_BKTLOCK |
+                                               CFS_HASH_NO_ITEMREF);
+
+       return filter->fo_iobuf_hash != NULL ? 0 : -ENOMEM;
+}
+
+/* Return iobuf allocated for @thread_id.
+ * If we haven't allocated a pool entry for this thread before, do so now and
+ * insert it into fo_iobuf_hash, otherwise we can find it from fo_iobuf_hash */
+void *filter_iobuf_get(struct filter_obd *filter, struct obd_trans_info *oti)
+{
+       struct filter_iobuf     *pool = NULL;
+       __u64                   key = 0;
+       int                     thread_id;
+       int                     rc;
+
+       thread_id = (oti && oti->oti_thread) ? oti->oti_thread->t_id : -1;
+       if (thread_id >= 0) {
+               struct ptlrpc_service_part *svcpt;
+
+               svcpt = oti->oti_thread->t_svcpt;
+               LASSERT(svcpt != NULL);
+
+               key = (__u64)(svcpt->scp_cpt) << 32 | thread_id;
+               pool = cfs_hash_lookup(filter->fo_iobuf_hash, &key);
+               if (pool != NULL)
+                       return pool;
+       }
+
+       pool = filter_alloc_iobuf(filter, OBD_BRW_WRITE, PTLRPC_MAX_BRW_PAGES);
+       if (pool == NULL)
+               return NULL;
+
+       if (thread_id >= 0) {
+               pool->dr_hkey = key;
+               rc = cfs_hash_add_unique(filter->fo_iobuf_hash,
+                                        &key, &pool->dr_hlist);
+               /* ptlrpc service thould guarantee thread ID is unique */
+               LASSERT(rc != -EALREADY);
+       }
+
+       return pool;
 }
 
 /* mount the file system (secretly).  lustre_cfg parameters are:
@@ -1969,6 +2049,7 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg,
 {
         struct filter_obd *filter = &obd->u.filter;
         struct vfsmount *mnt;
+        struct file_system_type *type;
         struct lustre_mount_info *lmi;
         struct obd_uuid uuid;
         __u8 *uuid_ptr;
@@ -1989,13 +2070,18 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg,
                    lcfg bufs 1, 2, 4 (device, fstype, mount opts) are ignored.*/
                 struct lustre_sb_info *lsi = s2lsi(lmi->lmi_sb);
                 mnt = lmi->lmi_mnt;
-                obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
+               obd->obd_fsops = fsfilt_get_ops(lsi->lsi_fstype);
         } else {
                 /* old path - used by lctl */
                 CERROR("Using old MDS mount method\n");
-                mnt = ll_kern_mount(lustre_cfg_string(lcfg, 2),
-                                    MS_NOATIME|MS_NODIRATIME,
-                                    lustre_cfg_string(lcfg, 1), option);
+                type = get_fs_type(lustre_cfg_string(lcfg, 2));
+                if (!type) {
+                        CERROR("get_fs_type failed\n");
+                        RETURN(-ENODEV);
+                }
+                mnt = vfs_kern_mount(type, MS_NOATIME|MS_NODIRATIME,
+                                     lustre_cfg_string(lcfg, 1), option);
+                cfs_module_put(type->owner);
                 if (IS_ERR(mnt)) {
                         rc = PTR_ERR(mnt);
                         LCONSOLE_ERROR_MSG(0x135, "Can't mount disk %s (%d)\n",
@@ -2038,7 +2124,9 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg,
         filter->fo_fstype = mnt->mnt_sb->s_type->name;
         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
 
-        fsfilt_setup(obd, obd->u.obt.obt_sb);
+        rc = fsfilt_setup(obd, obd->u.obt.obt_sb);
+        if (rc)
+                GOTO(err_ops, rc);
 
         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
         obd->obd_lvfs_ctxt.pwdmnt = mnt;
@@ -2218,10 +2306,15 @@ static int filter_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
                 GOTO(free_obd_stats, rc);
         }
 
+       rc = lprocfs_job_stats_init(obd, LPROC_FILTER_STATS_LAST,
+                                   filter_stats_counter_init);
+       if (rc)
+               GOTO(remove_entry_clear, rc);
+
         /* 2.6.9 selinux wants a full option page for do_kern_mount (bug6471) */
         OBD_PAGE_ALLOC(page, CFS_ALLOC_STD);
         if (!page)
-                GOTO(remove_entry_clear, rc = -ENOMEM);
+               GOTO(job_stats_fini, rc = -ENOMEM);
         addr = (unsigned long)cfs_page_address(page);
         clear_page((void *)addr);
         memcpy((void *)addr, lustre_cfg_buf(lcfg, 4),
@@ -2231,11 +2324,13 @@ static int filter_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
         if (rc) {
                 CERROR("%s: filter_common_setup failed: %d.\n",
                        obd->obd_name, rc);
-                GOTO(remove_entry_clear, rc);
+               GOTO(job_stats_fini, rc);
         }
 
         RETURN(0);
 
+job_stats_fini:
+       lprocfs_job_stats_fini(obd);
 remove_entry_clear:
         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 free_obd_stats:
@@ -2395,7 +2490,7 @@ static int filter_llog_finish(struct obd_device *obd, int count)
                  * We actually do sync in disconnect time, but disconnect
                  * may not come being marked rq_no_resend = 1.
                  */
-                llog_sync(ctxt, NULL);
+               llog_sync(ctxt, NULL, OBD_LLOG_FL_EXIT);
 
                 /*
                  * Balance class_import_get() in llog_receptor_accept().
@@ -2407,16 +2502,16 @@ static int filter_llog_finish(struct obd_device *obd, int count)
                         class_import_put(ctxt->loc_imp);
                         ctxt->loc_imp = NULL;
                 }
+
+               if (filter->fo_lcm) {
+                       llog_recov_thread_fini(filter->fo_lcm, obd->obd_force);
+                       filter->fo_lcm = NULL;
+               }
+
                 cfs_mutex_unlock(&ctxt->loc_mutex);
                 llog_ctxt_put(ctxt);
         }
 
-        if (filter->fo_lcm) {
-                cfs_mutex_lock(&ctxt->loc_mutex);
-                llog_recov_thread_fini(filter->fo_lcm, obd->obd_force);
-                filter->fo_lcm = NULL;
-                cfs_mutex_unlock(&ctxt->loc_mutex);
-        }
         RETURN(filter_olg_fini(&obd->obd_olg));
 }
 
@@ -2542,9 +2637,9 @@ static int filter_llog_connect(struct obd_export *exp,
         LASSERTF(ctxt != NULL, "ctxt is not null, ctxt idx %d \n",
                  body->lgdc_ctxt_idx);
 
-        CWARN("%s: Recovery from log "LPX64"/"LPX64":%x\n",
-              obd->obd_name, body->lgdc_logid.lgl_oid,
-              body->lgdc_logid.lgl_oseq, body->lgdc_logid.lgl_ogen);
+        CDEBUG(D_HA, "%s: Recovery from log "LPX64"/"LPX64":%x\n",
+               obd->obd_name, body->lgdc_logid.lgl_oid,
+               body->lgdc_logid.lgl_oseq, body->lgdc_logid.lgl_ogen);
 
         cfs_spin_lock(&obd->u.filter.fo_flags_lock);
         obd->u.filter.fo_mds_ost_sync = 1;
@@ -2612,6 +2707,7 @@ static int filter_precleanup(struct obd_device *obd,
                 obd_zombie_barrier();
 
                 rc = filter_llog_preclean(obd);
+               lprocfs_job_stats_fini(obd);
                 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
                 lprocfs_free_per_client_stats(obd);
                 lprocfs_obd_cleanup(obd);
@@ -2687,8 +2783,8 @@ static int filter_connect_internal(struct obd_export *exp,
 
         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
-                CWARN("%s: Received MDS connection for group %u\n",
-                      exp->exp_obd->obd_name, data->ocd_group);
+                CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
+                       exp->exp_obd->obd_name, data->ocd_group);
         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
                 RETURN(-EPROTO);
 
@@ -2763,11 +2859,9 @@ static int filter_connect_internal(struct obd_export *exp,
                 /* The client set in ocd_cksum_types the checksum types it
                  * supports. We have to mask off the algorithms that we don't
                  * support */
-                data->ocd_cksum_types &= cksum_types_supported();
+               data->ocd_cksum_types &= cksum_types_supported_server();
 
-                /* 1.6.4- only support CRC32 and didn't set ocd_cksum_types */
-                if (unlikely(data->ocd_cksum_types == 0))
-                        data->ocd_cksum_types = OBD_CKSUM_CRC32;
+               /* 1.6.4 clients are not supported any more */
 
                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
                                    "%x\n", exp->exp_obd->obd_name,
@@ -3052,7 +3146,7 @@ static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
                         ctxt = llog_group_get_ctxt(olg_min,
                                                    LLOG_MDS_OST_REPL_CTXT);
                         if (ctxt) {
-                                err = llog_sync(ctxt, olg_min->olg_exp);
+                               err = llog_sync(ctxt, olg_min->olg_exp, 0);
                                 llog_ctxt_put(ctxt);
                                 if (err) {
                                         CERROR("error flushing logs to MDS: "
@@ -3108,7 +3202,7 @@ static void filter_revimp_update(struct obd_export *exp)
         EXIT;
 }
 
-static int filter_ping(struct obd_export *exp)
+static int filter_ping(const struct lu_env *env, struct obd_export *exp)
 {
         filter_fmd_expire(exp);
         return 0;
@@ -3135,7 +3229,7 @@ struct dentry *__filter_oa2dentry(struct obd_device *obd, struct ost_id *ostid,
                 RETURN(ERR_PTR(-ENOENT));
         }
 
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
          * timestamps to appear to be in the far future, due old timestamp
          * being stored on disk as an unsigned value.  This fixes up any
@@ -3154,7 +3248,8 @@ struct dentry *__filter_oa2dentry(struct obd_device *obd, struct ost_id *ostid,
         return dchild;
 }
 
-static int filter_getattr(struct obd_export *exp, struct obd_info *oinfo)
+static int filter_getattr(const struct lu_env *env, struct obd_export *exp,
+                          struct obd_info *oinfo)
 {
         struct dentry *dentry = NULL;
         struct obd_device *obd;
@@ -3179,7 +3274,7 @@ static int filter_getattr(struct obd_export *exp, struct obd_info *oinfo)
 
         /* Limit the valid bits in the return data to what we actually use */
         oinfo->oi_oa->o_valid = OBD_MD_FLID;
-        obdo_from_inode(oinfo->oi_oa, dentry->d_inode, NULL, FILTER_VALID_FLAGS);
+        obdo_from_inode(oinfo->oi_oa, dentry->d_inode, FILTER_VALID_FLAGS);
 
         /* Store inode version in reply */
         curr_version = fsfilt_get_version(exp->exp_obd, dentry->d_inode);
@@ -3262,17 +3357,17 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
                 if (fcc != NULL)
                         *fcc = oa->o_lcookie;
         }
-        if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) {
-                unsigned long now = jiffies;
-                /* Filter truncates and writes are serialized by
-                 * i_alloc_sem, see the comment in
-                 * filter_preprw_write.*/
-                if (ia_valid & ATTR_SIZE)
-                        down_write(&inode->i_alloc_sem);
-                LOCK_INODE_MUTEX(inode);
-                fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
-                old_size = i_size_read(inode);
-        }
+       if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) {
+               unsigned long now = jiffies;
+               /* Filter truncates and writes are serialized by
+                * i_alloc_sem, see the comment in
+                * filter_preprw_write.*/
+               if (ia_valid & ATTR_SIZE)
+                       down_write(&inode->i_alloc_sem);
+               mutex_lock(&inode->i_mutex);
+               fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
+               old_size = i_size_read(inode);
+       }
 
         /* VBR: version recovery check */
         rc = filter_version_get_check(exp, oti, inode);
@@ -3391,12 +3486,12 @@ out_unlock:
         if (page)
                 page_cache_release(page);
 
-        if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID))
-                UNLOCK_INODE_MUTEX(inode);
-        if (ia_valid & ATTR_SIZE)
-                up_write(&inode->i_alloc_sem);
-        if (fcc)
-                OBD_FREE(fcc, sizeof(*fcc));
+       if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID))
+               mutex_unlock(&inode->i_mutex);
+       if (ia_valid & ATTR_SIZE)
+               up_write(&inode->i_alloc_sem);
+       if (fcc)
+               OBD_FREE(fcc, sizeof(*fcc));
 
         /* trigger quota release */
         if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) {
@@ -3411,8 +3506,8 @@ out_unlock:
 }
 
 /* this is called from filter_truncate() until we have filter_punch() */
-int filter_setattr(struct obd_export *exp, struct obd_info *oinfo,
-                   struct obd_trans_info *oti)
+int filter_setattr(const struct lu_env *env, struct obd_export *exp,
+                   struct obd_info *oinfo, struct obd_trans_info *oti)
 {
         struct obdo *oa = oinfo->oi_oa;
         struct lustre_capa *capa = oinfo_capa(oinfo);
@@ -3503,9 +3598,11 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo,
         oa->o_valid = OBD_MD_FLID;
 
         /* Quota release need uid/gid info */
-        obdo_from_inode(oa, dentry->d_inode, NULL,
+        obdo_from_inode(oa, dentry->d_inode,
                         FILTER_VALID_FLAGS | OBD_MD_FLUID | OBD_MD_FLGID);
 
+       filter_counter_incr(exp, LPROC_FILTER_STATS_SETATTR,
+                           oti ? oti->oti_jobid : NULL, 1);
         EXIT;
 out_unlock:
         f_dput(dentry);
@@ -3603,7 +3700,7 @@ static int filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
 
         for (id = last; id > oa->o_id; id--) {
                 doa.o_id = id;
-                rc = filter_destroy(exp, &doa, NULL, NULL, NULL, NULL);
+                rc = filter_destroy(NULL, exp, &doa, NULL, NULL, NULL, NULL);
                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
                         CEMERG("error destroying precreate objid "LPU64": %d\n",
                                id, rc);
@@ -3672,7 +3769,7 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa,
                 CDEBUG(D_HA, "filter_last_id() = "LPU64" -> diff = %d\n",
                        last, diff);
 
-                if (-diff > OST_MAX_PRECREATE) {
+                if (-diff > (OST_MAX_PRECREATE * 3) / 2) {
                         CERROR("%s: ignoring bogus orphan destroy request: "
                                "obdid "LPU64" last_id "LPU64"\n", obd->obd_name,
                                oa->o_id, last);
@@ -3704,8 +3801,15 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa,
                 CDEBUG(D_RPCTRACE, "filter_last_id() = "LPU64" -> diff = %d\n",
                        filter_last_id(filter, group), diff);
 
-                LASSERTF(diff >= 0,"%s: "LPU64" - "LPU64" = %d\n",obd->obd_name,
-                         oa->o_id, filter_last_id(filter, group), diff);
+               /*
+                * Check obd->obd_recovering to handle the race condition
+                * while recreating missing precreated objects through
+                * filter_preprw_write() and mds_lov_clear_orphans()
+                * at the same time.
+                */
+               LASSERTF(ergo(!obd->obd_recovering, diff >= 0),
+                        "%s: "LPU64" - "LPU64" = %d\n", obd->obd_name,
+                        oa->o_id, filter_last_id(filter, group), diff);
         }
 
         if (diff > 0) {
@@ -3723,9 +3827,10 @@ out:
         return rc;
 }
 
-static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
-                         __u64 max_age, __u32 flags)
+static int filter_statfs(const struct lu_env *env, struct obd_export *exp,
+                         struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 {
+        struct obd_device *obd = class_exp2obd(exp);
         struct filter_obd *filter = &obd->u.filter;
         int blockbits = obd->u.obt.obt_sb->s_blocksize_bits;
         struct lr_server_data *lsd = class_server_data(obd);
@@ -3849,7 +3954,7 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa,
                 OBD_ALLOC(osfs, sizeof(*osfs));
                 if (osfs == NULL)
                         RETURN(-ENOMEM);
-                rc = filter_statfs(obd, osfs,
+                rc = filter_statfs(NULL, obd->obd_self_export, osfs,
                                    cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                    0);
                 if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
@@ -4051,8 +4156,9 @@ set_last_id:
         RETURN(rc);
 }
 
-int filter_create(struct obd_export *exp, struct obdo *oa,
-                  struct lov_stripe_md **ea, struct obd_trans_info *oti)
+int filter_create(const struct lu_env *env, struct obd_export *exp,
+                  struct obdo *oa, struct lov_stripe_md **ea,
+                  struct obd_trans_info *oti)
 {
         struct obd_device *obd = exp->exp_obd;
         struct filter_export_data *fed;
@@ -4131,9 +4237,10 @@ int filter_create(struct obd_export *exp, struct obdo *oa,
         RETURN(rc);
 }
 
-int filter_destroy(struct obd_export *exp, struct obdo *oa,
-                   struct lov_stripe_md *md, struct obd_trans_info *oti,
-                   struct obd_export *md_exp, void *capa)
+int filter_destroy(const struct lu_env *env, struct obd_export *exp,
+                   struct obdo *oa, struct lov_stripe_md *md,
+                   struct obd_trans_info *oti, struct obd_export *md_exp,
+                   void *capa)
 {
         unsigned int qcids[MAXQUOTAS] = {0, 0};
         struct obd_device *obd;
@@ -4183,7 +4290,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa,
                         }
                         fcc = &oa->o_lcookie;
                         ctxt = llog_group_get_ctxt(olg, fcc->lgc_subsys + 1);
-                        llog_cancel(ctxt, NULL, 1, fcc, 0);
+                       llog_cancel(NULL, ctxt, NULL, 1, fcc, 0);
                         llog_ctxt_put(ctxt);
                         fcc = NULL; /* we didn't allocate fcc, don't free it */
                 }
@@ -4213,40 +4320,40 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa,
          * between page lock, i_mutex & starting new journal handle.
          * (see bug 20321) -johann
          */
-        now = jiffies;
-        down_write(&dchild->d_inode->i_alloc_sem);
-        LOCK_INODE_MUTEX(dchild->d_inode);
-        fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
-
-        /* VBR: version recovery check */
-        rc = filter_version_get_check(exp, oti, dchild->d_inode);
-        if (rc) {
-                UNLOCK_INODE_MUTEX(dchild->d_inode);
-                up_write(&dchild->d_inode->i_alloc_sem);
-                GOTO(cleanup, rc);
-        }
-
-        handle = fsfilt_start_log(obd, dchild->d_inode, FSFILT_OP_SETATTR,
-                                  NULL, 1);
-        if (IS_ERR(handle)) {
-                UNLOCK_INODE_MUTEX(dchild->d_inode);
-                up_write(&dchild->d_inode->i_alloc_sem);
-                GOTO(cleanup, rc = PTR_ERR(handle));
-        }
-
-        /* Locking order: i_mutex -> journal_lock -> dqptr_sem. LU-952 */
-        ll_vfs_dq_init(dchild->d_inode);
-
-        iattr.ia_valid = ATTR_SIZE;
-        iattr.ia_size = 0;
-        rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
-        rc2 = fsfilt_commit(obd, dchild->d_inode, handle, 0);
-        UNLOCK_INODE_MUTEX(dchild->d_inode);
-        up_write(&dchild->d_inode->i_alloc_sem);
-        if (rc)
-                GOTO(cleanup, rc);
-        if (rc2)
-                GOTO(cleanup, rc = rc2);
+       now = jiffies;
+       down_write(&dchild->d_inode->i_alloc_sem);
+       mutex_lock(&dchild->d_inode->i_mutex);
+       fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
+
+       /* VBR: version recovery check */
+       rc = filter_version_get_check(exp, oti, dchild->d_inode);
+       if (rc) {
+               mutex_unlock(&dchild->d_inode->i_mutex);
+               up_write(&dchild->d_inode->i_alloc_sem);
+               GOTO(cleanup, rc);
+       }
+
+       handle = fsfilt_start_log(obd, dchild->d_inode, FSFILT_OP_SETATTR,
+                                 NULL, 1);
+       if (IS_ERR(handle)) {
+               mutex_unlock(&dchild->d_inode->i_mutex);
+               up_write(&dchild->d_inode->i_alloc_sem);
+               GOTO(cleanup, rc = PTR_ERR(handle));
+       }
+
+       /* Locking order: i_mutex -> journal_lock -> dqptr_sem. LU-952 */
+       ll_vfs_dq_init(dchild->d_inode);
+
+       iattr.ia_valid = ATTR_SIZE;
+       iattr.ia_size = 0;
+       rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
+       rc2 = fsfilt_commit(obd, dchild->d_inode, handle, 0);
+       mutex_unlock(&dchild->d_inode->i_mutex);
+       up_write(&dchild->d_inode->i_alloc_sem);
+       if (rc)
+               GOTO(cleanup, rc);
+       if (rc2)
+               GOTO(cleanup, rc = rc2);
 
         /* We don't actually need to lock the parent until we are unlinking
          * here, and not while truncating above.  That avoids holding the
@@ -4257,16 +4364,17 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa,
                 GOTO(cleanup, rc = PTR_ERR(dparent));
         cleanup_phase = 3; /* filter_parent_unlock */
 
-        LOCK_INODE_MUTEX(dchild->d_inode);
-        handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
-        if (IS_ERR(handle)) {
-                UNLOCK_INODE_MUTEX(dchild->d_inode);
-                GOTO(cleanup, rc = PTR_ERR(handle));
-        }
-        cleanup_phase = 4; /* fsfilt_commit */
+       mutex_lock(&dchild->d_inode->i_mutex);
+       handle = fsfilt_start_log(obd, dparent->d_inode,
+                                 FSFILT_OP_UNLINK, oti, 1);
+       if (IS_ERR(handle)) {
+               mutex_unlock(&dchild->d_inode->i_mutex);
+               GOTO(cleanup, rc = PTR_ERR(handle));
+       }
+       cleanup_phase = 4; /* fsfilt_commit */
 
         /* Quota release need uid/gid of inode */
-        obdo_from_inode(oa, dchild->d_inode, NULL, OBD_MD_FLUID|OBD_MD_FLGID);
+        obdo_from_inode(oa, dchild->d_inode, OBD_MD_FLUID | OBD_MD_FLGID);
 
         filter_fmd_drop(exp, oa->o_id, oa->o_seq);
 
@@ -4327,8 +4435,8 @@ cleanup:
 }
 
 /* NB start and end are used for punch, but not truncate */
-static int filter_truncate(struct obd_export *exp, struct obd_info *oinfo,
-                           struct obd_trans_info *oti,
+static int filter_truncate(const struct lu_env *env, struct obd_export *exp,
+                           struct obd_info *oinfo, struct obd_trans_info *oti,
                            struct ptlrpc_request_set *rqset)
 {
         int rc;
@@ -4345,12 +4453,12 @@ static int filter_truncate(struct obd_export *exp, struct obd_info *oinfo,
                 oinfo->oi_policy.l_extent.start);
 
         oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
-        rc = filter_setattr(exp, oinfo, oti);
+        rc = filter_setattr(env, exp, oinfo, oti);
         RETURN(rc);
 }
 
-static int filter_sync(struct obd_export *exp, struct obd_info *oinfo,
-                       obd_off start, obd_off end,
+static int filter_sync(const struct lu_env *env, struct obd_export *exp,
+                       struct obd_info *oinfo, obd_off start, obd_off end,
                        struct ptlrpc_request_set *set)
 {
         struct lvfs_run_ctxt saved;
@@ -4381,34 +4489,34 @@ static int filter_sync(struct obd_export *exp, struct obd_info *oinfo,
 
         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
 
-        LOCK_INODE_MUTEX(dentry->d_inode);
+       mutex_lock(&dentry->d_inode->i_mutex);
 
-        rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
-        if (rc == 0) {
-                /* just any file to grab fsync method - "file" arg unused */
-                struct file *file = obt->obt_rcvd_filp;
+       rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
+       if (rc == 0) {
+               /* just any file to grab fsync method - "file" arg unused */
+               struct file *file = obt->obt_rcvd_filp;
 
-                if (file->f_op && file->f_op->fsync)
-                        rc = file->f_op->fsync(NULL, dentry, 1);
+               if (file->f_op && file->f_op->fsync)
+                       rc = file->f_op->fsync(NULL, dentry, 1);
 
-                rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
-                if (!rc)
-                        rc = rc2;
-        }
-        UNLOCK_INODE_MUTEX(dentry->d_inode);
+               rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
+               if (!rc)
+                       rc = rc2;
+       }
+       mutex_unlock(&dentry->d_inode->i_mutex);
 
         oinfo->oi_oa->o_valid = OBD_MD_FLID;
-        obdo_from_inode(oinfo->oi_oa, dentry->d_inode, NULL,
-                        FILTER_VALID_FLAGS);
+        obdo_from_inode(oinfo->oi_oa, dentry->d_inode, FILTER_VALID_FLAGS);
 
         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
 
+       filter_counter_incr(exp, LPROC_FILTER_STATS_SYNC, oinfo->oi_jobid, 1);
         f_dput(dentry);
         RETURN(rc);
 }
 
-static int filter_get_info(struct obd_export *exp, __u32 keylen,
-                           void *key, __u32 *vallen, void *val,
+static int filter_get_info(const struct lu_env *env, struct obd_export *exp,
+                           __u32 keylen, void *key, __u32 *vallen, void *val,
                            struct lov_stripe_md *lsm)
 {
         struct obd_device *obd;
@@ -4564,7 +4672,8 @@ out:
         RETURN(rc);
 }
 
-static int filter_set_info_async(struct obd_export *exp, __u32 keylen,
+static int filter_set_info_async(const struct lu_env *env,
+                                 struct obd_export *exp, __u32 keylen,
                                  void *key, __u32 vallen, void *val,
                                  struct ptlrpc_request_set *set)
 {
@@ -4640,8 +4749,8 @@ int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
                 CDEBUG(D_HA, "syncing ost %s\n", obd->obd_name);
                 rc = fsfilt_sync(obd, obd->u.obt.obt_sb);
 
-                lvfs_set_rdonly(obd, obd->u.obt.obt_sb);
-                RETURN(0);
+                rc = lvfs_set_rdonly(obd, obd->u.obt.obt_sb);
+                RETURN(rc);
         }
 
         case OBD_IOC_CATLOGLIST: {
@@ -4673,7 +4782,7 @@ int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
         RETURN(0);
 }
 
-static int filter_health_check(struct obd_device *obd)
+static int filter_health_check(const struct lu_env *env, struct obd_device *obd)
 {
 #ifdef USE_HEALTH_CHECK_WRITE
         struct filter_obd *filter = &obd->u.filter;
@@ -4762,7 +4871,6 @@ static struct obd_ops filter_obd_ops = {
         .o_create         = filter_create,
         .o_setattr        = filter_setattr,
         .o_destroy        = filter_destroy,
-        .o_brw            = filter_brw,
         .o_punch          = filter_truncate,
         .o_sync           = filter_sync,
         .o_preprw         = filter_preprw,