Whamcloud - gitweb
b=22108 correct number of objects in slow precreating path
[fs/lustre-release.git] / lustre / obdfilter / filter.c
index e3f13b7..a731840 100644 (file)
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -241,6 +241,7 @@ static int lprocfs_init_rw_stats(struct obd_device *obd,
    plus the procfs overhead :( */
 static int filter_export_stats_init(struct obd_device *obd,
                                     struct obd_export *exp,
+                                    int reconnect,
                                     void *client_nid)
 {
         int rc, newnid = 0;
@@ -250,7 +251,7 @@ static int filter_export_stats_init(struct obd_device *obd,
                 /* Self-export gets no proc entry */
                 RETURN(0);
 
-        rc = lprocfs_exp_setup(exp, client_nid, &newnid);
+        rc = lprocfs_exp_setup(exp, client_nid, reconnect, &newnid);
         if (rc) {
                 /* Mask error for already created
                  * /proc entries */
@@ -282,16 +283,7 @@ static int filter_export_stats_init(struct obd_device *obd,
                                             tmp->nid_stats);
                 if (rc)
                         GOTO(clean, rc);
-                /* Always add in ldlm_stats */
-                tmp->nid_ldlm_stats = 
-                        lprocfs_alloc_stats(LDLM_LAST_OPC - LDLM_FIRST_OPC,
-                                            LPROCFS_STATS_FLAG_NOPERCPU);
-                if (tmp->nid_ldlm_stats == NULL)
-                        GOTO(clean, rc = -ENOMEM);
-
-                lprocfs_init_ldlm_stats(tmp->nid_ldlm_stats);
-                rc = lprocfs_register_stats(tmp->nid_proc, "ldlm_stats",
-                                            tmp->nid_ldlm_stats);
+                rc = lprocfs_nid_ldlm_stats_init(tmp);
                 if (rc)
                         GOTO(clean, rc);
         }
@@ -643,14 +635,18 @@ static void filter_fmd_cleanup(struct obd_export *exp)
 
 static int filter_init_export(struct obd_export *exp)
 {
+        int rc;
         cfs_spin_lock_init(&exp->exp_filter_data.fed_lock);
         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
 
         cfs_spin_lock(&exp->exp_lock);
         exp->exp_connecting = 1;
         cfs_spin_unlock(&exp->exp_lock);
+        rc = lut_client_alloc(exp);
+        if (rc == 0)
+                rc = ldlm_init_export(exp);
 
-        return ldlm_init_export(exp);
+        return rc;
 }
 
 static int filter_free_server_data(struct obd_device_target *obt)
@@ -698,7 +694,7 @@ int filter_update_last_objid(struct obd_device *obd, obd_seq group,
         }
 
         CDEBUG(D_INODE, "%s: server last_objid for "POSTID"\n",
-               obd->obd_name, group, filter->fo_last_objids[group]);
+               obd->obd_name, filter->fo_last_objids[group], group);
 
         tmp = cpu_to_le64(filter->fo_last_objids[group]);
         rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
@@ -821,18 +817,16 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp)
                 GOTO(out, rc = 0);
         }
 
+        OBD_ALLOC_PTR(lcd);
+        if (!lcd)
+                GOTO(err_client, rc = -ENOMEM);
+
         for (cl_idx = 0, off = le32_to_cpu(lsd->lsd_client_start);
              off < last_rcvd_size; cl_idx++) {
                 __u64 last_rcvd;
                 struct obd_export *exp;
                 struct filter_export_data *fed;
 
-                if (!lcd) {
-                        OBD_ALLOC_PTR(lcd);
-                        if (!lcd)
-                                GOTO(err_client, rc = -ENOMEM);
-                }
-
                 /* Don't assume off is incremented properly by
                  * fsfilt_read_record(), in case sizeof(*lcd)
                  * isn't the same as lsd->lsd_client_size.  */
@@ -855,60 +849,52 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp)
 
                 last_rcvd = le64_to_cpu(lcd->lcd_last_transno);
 
+                CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
+                       " srv lr: "LPU64"\n", lcd->lcd_uuid, cl_idx,
+                       last_rcvd, le64_to_cpu(lsd->lsd_last_transno));
+
                 /* These exports are cleaned up by filter_disconnect(), so they
                  * need to be set up like real exports as filter_connect() does.
                  */
                 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
-
-                CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
-                       " srv lr: "LPU64"\n", lcd->lcd_uuid, cl_idx,
-                       last_rcvd, le64_to_cpu(lsd->lsd_last_transno));
                 if (IS_ERR(exp)) {
                         if (PTR_ERR(exp) == -EALREADY) {
                                 /* export already exists, zero out this one */
-                                CERROR("Zeroing out duplicate export due to "
-                                       "bug 10479.\n");
-                                lcd->lcd_uuid[0] = '\0';
-                        } else {
-                                GOTO(err_client, rc = PTR_ERR(exp));
+                                CERROR("Duplicate export %s!\n", lcd->lcd_uuid);
+                                continue;
                         }
-                } else {
-                        fed = &exp->exp_filter_data;
-                        fed->fed_ted.ted_lcd = lcd;
-                        fed->fed_group = 0; /* will be assigned at connect */
-                        filter_export_stats_init(obd, exp, NULL);
-                        rc = filter_client_add(obd, exp, cl_idx);
-                        /* can't fail for existing client */
-                        LASSERTF(rc == 0, "rc = %d\n", rc);
-
-                        /* VBR: set export last committed */
-                        exp->exp_last_committed = last_rcvd;
-                        cfs_spin_lock(&exp->exp_lock);
-                        exp->exp_connecting = 0;
-                        exp->exp_in_recovery = 0;
-                        cfs_spin_unlock(&exp->exp_lock);
-                        cfs_spin_lock_bh(&obd->obd_processing_task_lock);
-                        obd->obd_max_recoverable_clients++;
-                        cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
-                        lcd = NULL;
-                        class_export_put(exp);
+                        OBD_FREE_PTR(lcd);
+                        GOTO(err_client, rc = PTR_ERR(exp));
                 }
 
-                /* Need to check last_rcvd even for duplicated exports. */
-                CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
-                       cl_idx, last_rcvd);
+                fed = &exp->exp_filter_data;
+                *fed->fed_ted.ted_lcd = *lcd;
+                fed->fed_group = 0; /* will be assigned at connect */
+                filter_export_stats_init(obd, exp, 0, NULL);
+                rc = filter_client_add(obd, exp, cl_idx);
+                /* can't fail for existing client */
+                LASSERTF(rc == 0, "rc = %d\n", rc);
+
+                /* VBR: set export last committed */
+                exp->exp_last_committed = last_rcvd;
+                cfs_spin_lock(&exp->exp_lock);
+                exp->exp_connecting = 0;
+                exp->exp_in_recovery = 0;
+                cfs_spin_unlock(&exp->exp_lock);
+                cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+                obd->obd_max_recoverable_clients++;
+                cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+                class_export_put(exp);
 
                 if (last_rcvd > le64_to_cpu(lsd->lsd_last_transno))
                         lsd->lsd_last_transno = cpu_to_le64(last_rcvd);
         }
-
-        if (lcd)
-                OBD_FREE_PTR(lcd);
+        OBD_FREE_PTR(lcd);
 
         obd->obd_last_committed = le64_to_cpu(lsd->lsd_last_transno);
 out:
-        lut->lut_mount_count = mount_count + 1;
-        lsd->lsd_mount_count = cpu_to_le64(lut->lut_mount_count);
+        obd->u.obt.obt_mount_count = mount_count + 1;
+        lsd->lsd_mount_count = cpu_to_le64(obd->u.obt.obt_mount_count);
 
         /* save it, so mount count and last_transno is current */
         rc = filter_update_server_data(obd);
@@ -2034,6 +2020,7 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg,
 
         obd->u.obt.obt_vfsmnt = mnt;
         obd->u.obt.obt_sb = mnt->mnt_sb;
+        obd->u.obt.obt_magic = OBT_MAGIC;
         filter->fo_fstype = mnt->mnt_sb->s_type->name;
         CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
 
@@ -2060,6 +2047,8 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg,
         filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
         filter->fo_fmd_max_num = FILTER_FMD_MAX_NUM_DEFAULT;
         filter->fo_fmd_max_age = FILTER_FMD_MAX_AGE_DEFAULT;
+        filter->fo_syncjournal = 0; /* Don't sync journals on i/o by default */
+        filter_slc_set(filter); /* initialize sync on lock cancel */
 
         rc = filter_prep(obd);
         if (rc)
@@ -2761,7 +2750,7 @@ static int filter_reconnect(const struct lu_env *env,
 
         rc = filter_connect_internal(exp, data, 1);
         if (rc == 0)
-                filter_export_stats_init(obd, exp, localdata);
+                filter_export_stats_init(obd, exp, 1, localdata);
 
         RETURN(rc);
 }
@@ -2775,8 +2764,6 @@ static int filter_connect(const struct lu_env *env,
         struct lvfs_run_ctxt saved;
         struct lustre_handle conn = { 0 };
         struct obd_export *lexp;
-        struct tg_export_data *ted;
-        struct lsd_client_data *lcd = NULL;
         __u32 group;
         int rc;
         ENTRY;
@@ -2790,22 +2777,15 @@ static int filter_connect(const struct lu_env *env,
         lexp = class_conn2export(&conn);
         LASSERT(lexp != NULL);
 
-        ted = &lexp->exp_target_data;
-
         rc = filter_connect_internal(lexp, data, 0);
         if (rc)
                 GOTO(cleanup, rc);
 
-        filter_export_stats_init(obd, lexp, localdata);
+        filter_export_stats_init(obd, lexp, 0, localdata);
         if (obd->obd_replayable) {
-                OBD_ALLOC(lcd, sizeof(*lcd));
-                if (!lcd) {
-                        CERROR("filter: out of memory for client data\n");
-                        GOTO(cleanup, rc = -ENOMEM);
-                }
-
+                struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
+                LASSERT(lcd);
                 memcpy(lcd->lcd_uuid, cluuid, sizeof(lcd->lcd_uuid));
-                ted->ted_lcd = lcd;
                 rc = filter_client_add(obd, lexp, -1);
                 if (rc)
                         GOTO(cleanup, rc);
@@ -2960,12 +2940,11 @@ static int filter_destroy_export(struct obd_export *exp)
 
         target_destroy_export(exp);
         ldlm_destroy_export(exp);
+        lut_client_free(exp);
 
         if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
                 RETURN(0);
 
-        lut_client_free(exp);
-
         if (!exp->exp_obd->obd_replayable)
                 fsfilt_sync(exp->exp_obd, exp->exp_obd->u.obt.obt_sb);
 
@@ -3105,7 +3084,7 @@ struct dentry *__filter_oa2dentry(struct obd_device *obd, struct ost_id *ostid,
         if (dchild->d_inode == NULL) {
                 if (!quiet)
                         CERROR("%s: %s on non-existent object: "POSTID" \n",
-                               obd->obd_name, what, ostid->oi_seq,ostid->oi_id);
+                               obd->obd_name, what, ostid->oi_id,ostid->oi_seq);
                 f_dput(dchild);
                 RETURN(ERR_PTR(-ENOENT));
         }
@@ -3120,10 +3099,6 @@ static int filter_getattr(struct obd_export *exp, struct obd_info *oinfo)
         int rc = 0;
         ENTRY;
 
-        rc = filter_validate_obdo(oinfo->oi_oa, exp);
-        if (rc)
-                RETURN(rc);
-
         rc = filter_auth_capa(exp, NULL, oinfo->oi_oa->o_seq,
                               oinfo_capa(oinfo), CAPA_OPC_META_READ);
         if (rc)
@@ -3302,9 +3277,9 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
         if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_CREDITS))
                 fsfilt_extend(exp->exp_obd, inode, 0, handle);
 
-        /* The truncate might have used up our transaction credits.  Make
-         * sure we have one left for the last_rcvd update. */
-        err = fsfilt_extend(exp->exp_obd, inode, 1, handle);
+       /* The truncate might have used up our transaction credits.  Make sure
+        * we have two left for the last_rcvd and VBR inode version updates. */
+        err = fsfilt_extend(exp->exp_obd, inode, 2, handle);
 
         rc = filter_finish_transno(exp, inode, oti, rc, sync);
         if (sync) {
@@ -3372,10 +3347,6 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo,
         int rc;
         ENTRY;
 
-        rc = filter_validate_obdo(oinfo->oi_oa, exp);
-        if (rc)
-                RETURN(rc);
-
         if (oa->o_valid & OBD_FL_TRUNC)
                 opc |= CAPA_OPC_OSS_TRUNC;
 
@@ -3405,7 +3376,7 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo,
 
                 CERROR("%s: setattr from %s trying to truncate objid "POSTID
                        "%s\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
-                       oa->o_seq, oa->o_id, mdsinum);
+                       oa->o_id, oa->o_seq, mdsinum);
                 RETURN(-EPERM);
         }
 
@@ -3527,10 +3498,6 @@ static int filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
         int skip_orphan;
         ENTRY;
 
-        rc = filter_validate_obdo(oa, exp);
-        if (rc)
-                RETURN(rc);
-
         LASSERT(down_trylock(&filter->fo_create_locks[oa->o_seq]) != 0);
 
         memset(&doa, 0, sizeof(doa));
@@ -3795,7 +3762,8 @@ 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, cfs_time_current_64() - CFS_HZ,
+                rc = filter_statfs(obd, osfs,
+                                   cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                    0);
                 if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
                         CDEBUG(D_RPCTRACE,"%s: not enough space for create "
@@ -3838,8 +3806,8 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa,
                 /* Temporary solution for oid in CMD before fid-on-OST */
                 if ((fid_seq_is_mdt0(oa->o_seq) && next_id >= IDIF_MAX_OID) &&
                     (fid_seq_is_cmd(oa->o_seq) && next_id >= OBIF_MAX_OID)) {
-                        CERROR("%s:"POSTID" hit the max IDIF_MAX_OID(1 << 48) !\n",
-                                obd->obd_name, group, next_id);
+                        CERROR("%s:"POSTID" hit the max IDIF_MAX_OID(1<<48)!\n",
+                                obd->obd_name, next_id, group);
                         GOTO(cleanup, rc = -ENOSPC);
                 }
 
@@ -3941,6 +3909,7 @@ set_last_id:
                 if (rc)
                         break;
                 if (cfs_time_after(jiffies, enough_time)) {
+                        i++;
                         CDEBUG(D_RPCTRACE,
                                "%s: precreate slow - want %d got %d \n",
                                obd->obd_name, *num, i);
@@ -3951,13 +3920,13 @@ set_last_id:
 
         CDEBUG(D_RPCTRACE,
                "%s: created %d objects for group "POSTID" rc %d\n",
-               obd->obd_name, i, group, filter->fo_last_objids[group], rc);
+               obd->obd_name, i, filter->fo_last_objids[group], group, rc);
 
         RETURN(rc);
 }
 
-static int filter_create(struct obd_export *exp, struct obdo *oa,
-                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
+int filter_create(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;
@@ -3967,10 +3936,6 @@ static int filter_create(struct obd_export *exp, struct obdo *oa,
         int rc = 0, diff;
         ENTRY;
 
-        rc = filter_validate_obdo(oa, exp);
-        if (rc)
-                RETURN(rc);
-
         CDEBUG(D_INODE, "%s: filter_create(group="LPU64",id="
                LPU64")\n", obd->obd_name, oa->o_seq, oa->o_id);
 
@@ -4000,7 +3965,8 @@ static int filter_create(struct obd_export *exp, struct obdo *oa,
 
         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
-                if (oa->o_id > filter_last_id(filter, oa->o_seq)) {
+                if (!obd->obd_recovering ||
+                    oa->o_id > filter_last_id(filter, oa->o_seq)) {
                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
                                oa->o_id, filter_last_id(filter, oa->o_seq));
                         rc = -EINVAL;
@@ -4045,9 +4011,6 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa,
         struct iattr iattr;
         ENTRY;
 
-        rc = filter_validate_obdo(oa, exp);
-        if (rc)
-                RETURN(rc);
         rc = filter_auth_capa(exp, NULL, oa->o_seq,
                               (struct lustre_capa *)capa, CAPA_OPC_OSS_DESTROY);
         if (rc)
@@ -4069,7 +4032,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa,
 
         if (dchild->d_inode == NULL) {
                 CDEBUG(D_INODE, "destroying non-existent object "POSTID"\n",
-                       oa->o_seq, oa->o_id);
+                       oa->o_id, oa->o_seq);
                 /* If object already gone, cancel cookie right now */
                 if (oa->o_valid & OBD_MD_FLCOOKIE) {
                         struct llog_ctxt *ctxt;
@@ -4257,10 +4220,6 @@ static int filter_sync(struct obd_export *exp, struct obdo *oa,
         int rc, rc2;
         ENTRY;
 
-        rc = filter_validate_obdo(oa, exp);
-        if (rc)
-                RETURN(rc);
-
         rc = filter_auth_capa(exp, NULL, oa->o_seq,
                               (struct lustre_capa *)capa, CAPA_OPC_OSS_WRITE);
         if (rc)
@@ -4362,9 +4321,6 @@ static int filter_get_info(struct obd_export *exp, __u32 keylen,
                 struct lvfs_run_ctxt saved;
                 int rc;
 
-                rc = filter_validate_obdo(&fm_key->oa, exp);
-                if (rc)
-                        RETURN(rc);
                 if (fiemap == NULL) {
                         *vallen = fiemap_count_to_size(
                                                 fm_key->fiemap.fm_extent_count);
@@ -4386,6 +4342,12 @@ static int filter_get_info(struct obd_export *exp, __u32 keylen,
                 RETURN(rc);
         }
 
+        if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
+                *((__u32 *) val) = obd->u.filter.fo_sync_lock_cancel;
+                *vallen = sizeof(__u32);
+                RETURN(0);
+        }
+
         CDEBUG(D_IOCTL, "invalid key\n");
         RETURN(-EINVAL);
 }