drivers/staging/lustre/lustre/ldlm/ldlm_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 /**
  38  * This file deals with various client/target related logic including recovery.
  39  *
  40  * TODO: This code more logically belongs in the ptlrpc module than in ldlm and
  41  * should be moved.
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LDLM
  45
  46 # include <linux/libcfs/libcfs.h>
  47 #include <obd.h>
  48 #include <obd_class.h>
  49 #include <lustre_dlm.h>
  50 #include <lustre_net.h>
  51 #include <lustre_sec.h>
  52 #include "ldlm_internal.h"
  53
  54 /* @priority: If non-zero, move the selected connection to the list head.
  55  * @create: If zero, only search in existing connections.
  56  */
  57 static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid,
  58                            int priority, int create)
  59 {
  60         struct ptlrpc_connection *ptlrpc_conn;
  61         struct obd_import_conn *imp_conn = NULL, *item;
  62         int rc = 0;
  63
  64         if (!create && !priority) {
  65                 CDEBUG(D_HA, "Nothing to do\n");
  66                 return -EINVAL;
  67         }
  68
  69         ptlrpc_conn = ptlrpc_uuid_to_connection(uuid);
  70         if (!ptlrpc_conn) {
  71                 CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid);
  72                 return -ENOENT;
  73         }
  74
  75         if (create) {
  76                 OBD_ALLOC(imp_conn, sizeof(*imp_conn));
  77                 if (!imp_conn) {
  78                         GOTO(out_put, rc = -ENOMEM);
  79                 }
  80         }
  81
  82         spin_lock(&imp->imp_lock);
  83         list_for_each_entry(item, &imp->imp_conn_list, oic_item) {
  84                 if (obd_uuid_equals(uuid, &item->oic_uuid)) {
  85                         if (priority) {
  86                                 list_del(&item->oic_item);
  87                                 list_add(&item->oic_item,
  88                                              &imp->imp_conn_list);
  89                                 item->oic_last_attempt = 0;
  90                         }
  91                         CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n",
  92                                imp, imp->imp_obd->obd_name, uuid->uuid,
  93                                (priority ? ", moved to head" : ""));
  94                         spin_unlock(&imp->imp_lock);
  95                         GOTO(out_free, rc = 0);
  96                 }
  97         }
  98         /* No existing import connection found for \a uuid. */
  99         if (create) {
 100                 imp_conn->oic_conn = ptlrpc_conn;
 101                 imp_conn->oic_uuid = *uuid;
 102                 imp_conn->oic_last_attempt = 0;
 103                 if (priority)
 104                         list_add(&imp_conn->oic_item, &imp->imp_conn_list);
 105                 else
 106                         list_add_tail(&imp_conn->oic_item,
 107                                           &imp->imp_conn_list);
 108                 CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n",
 109                        imp, imp->imp_obd->obd_name, uuid->uuid,
 110                        (priority ? "head" : "tail"));
 111         } else {
 112                 spin_unlock(&imp->imp_lock);
 113                 GOTO(out_free, rc = -ENOENT);
 114         }
 115
 116         spin_unlock(&imp->imp_lock);
 117         return 0;
 118 out_free:
 119         if (imp_conn)
 120                 OBD_FREE(imp_conn, sizeof(*imp_conn));
 121 out_put:
 122         ptlrpc_connection_put(ptlrpc_conn);
 123         return rc;
 124 }
 125
 126 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid)
 127 {
 128         return import_set_conn(imp, uuid, 1, 0);
 129 }
 130
 131 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
 132                            int priority)
 133 {
 134         return import_set_conn(imp, uuid, priority, 1);
 135 }
 136 EXPORT_SYMBOL(client_import_add_conn);
 137
 138 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
 139 {
 140         struct obd_import_conn *imp_conn;
 141         struct obd_export *dlmexp;
 142         int rc = -ENOENT;
 143
 144         spin_lock(&imp->imp_lock);
 145         if (list_empty(&imp->imp_conn_list)) {
 146                 LASSERT(!imp->imp_connection);
 147                 GOTO(out, rc);
 148         }
 149
 150         list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) {
 151                 if (!obd_uuid_equals(uuid, &imp_conn->oic_uuid))
 152                         continue;
 153                 LASSERT(imp_conn->oic_conn);
 154
 155                 if (imp_conn == imp->imp_conn_current) {
 156                         LASSERT(imp_conn->oic_conn == imp->imp_connection);
 157
 158                         if (imp->imp_state != LUSTRE_IMP_CLOSED &&
 159                             imp->imp_state != LUSTRE_IMP_DISCON) {
 160                                 CERROR("can't remove current connection\n");
 161                                 GOTO(out, rc = -EBUSY);
 162                         }
 163
 164                         ptlrpc_connection_put(imp->imp_connection);
 165                         imp->imp_connection = NULL;
 166
 167                         dlmexp = class_conn2export(&imp->imp_dlm_handle);
 168                         if (dlmexp && dlmexp->exp_connection) {
 169                                 LASSERT(dlmexp->exp_connection ==
 170                                         imp_conn->oic_conn);
 171                                 ptlrpc_connection_put(dlmexp->exp_connection);
 172                                 dlmexp->exp_connection = NULL;
 173                         }
 174                 }
 175
 176                 list_del(&imp_conn->oic_item);
 177                 ptlrpc_connection_put(imp_conn->oic_conn);
 178                 OBD_FREE(imp_conn, sizeof(*imp_conn));
 179                 CDEBUG(D_HA, "imp %p@%s: remove connection %s\n",
 180                        imp, imp->imp_obd->obd_name, uuid->uuid);
 181                 rc = 0;
 182                 break;
 183         }
 184 out:
 185         spin_unlock(&imp->imp_lock);
 186         if (rc == -ENOENT)
 187                 CERROR("connection %s not found\n", uuid->uuid);
 188         return rc;
 189 }
 190 EXPORT_SYMBOL(client_import_del_conn);
 191
 192 /**
 193  * Find conn UUID by peer NID. \a peer is a server NID. This function is used
 194  * to find a conn uuid of \a imp which can reach \a peer.
 195  */
 196 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
 197                             struct obd_uuid *uuid)
 198 {
 199         struct obd_import_conn *conn;
 200         int rc = -ENOENT;
 201
 202         spin_lock(&imp->imp_lock);
 203         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 204                 /* Check if conn UUID does have this peer NID. */
 205                 if (class_check_uuid(&conn->oic_uuid, peer)) {
 206                         *uuid = conn->oic_uuid;
 207                         rc = 0;
 208                         break;
 209                 }
 210         }
 211         spin_unlock(&imp->imp_lock);
 212         return rc;
 213 }
 214 EXPORT_SYMBOL(client_import_find_conn);
 215
 216 void client_destroy_import(struct obd_import *imp)
 217 {
 218         /* Drop security policy instance after all RPCs have finished/aborted
 219          * to let all busy contexts be released. */
 220         class_import_get(imp);
 221         class_destroy_import(imp);
 222         sptlrpc_import_sec_put(imp);
 223         class_import_put(imp);
 224 }
 225 EXPORT_SYMBOL(client_destroy_import);
 226
 227 /**
 228  * Check whether or not the OSC is on MDT.
 229  * In the config log,
 230  * osc on MDT
 231  *      setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
 232  * osc on client
 233  *      setup 0:{fsname}-OSTxxxx-osc 1:lustre-OST0000_UUID 2:NID
 234  *
 235  **/
 236 static int osc_on_mdt(char *obdname)
 237 {
 238         char *ptr;
 239
 240         ptr = strrchr(obdname, '-');
 241         if (ptr == NULL)
 242                 return 0;
 243
 244         if (strncmp(ptr + 1, "MDT", 3) == 0)
 245                 return 1;
 246
 247         return 0;
 248 }
 249
 250 /* Configure an RPC client OBD device.
 251  *
 252  * lcfg parameters:
 253  * 1 - client UUID
 254  * 2 - server UUID
 255  * 3 - inactive-on-startup
 256  */
 257 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
 258 {
 259         struct client_obd *cli = &obddev->u.cli;
 260         struct obd_import *imp;
 261         struct obd_uuid server_uuid;
 262         int rq_portal, rp_portal, connect_op;
 263         char *name = obddev->obd_type->typ_name;
 264         ldlm_ns_type_t ns_type = LDLM_NS_TYPE_UNKNOWN;
 265         int rc;
 266         char    *cli_name = lustre_cfg_buf(lcfg, 0);
 267
 268         /* In a more perfect world, we would hang a ptlrpc_client off of
 269          * obd_type and just use the values from there. */
 270         if (!strcmp(name, LUSTRE_OSC_NAME) ||
 271             (!(strcmp(name, LUSTRE_OSP_NAME)) &&
 272              (is_osp_on_mdt(cli_name) &&
 273                strstr(lustre_cfg_buf(lcfg, 1), "OST") != NULL))) {
 274                 /* OSC or OSP_on_MDT for OSTs */
 275                 rq_portal = OST_REQUEST_PORTAL;
 276                 rp_portal = OSC_REPLY_PORTAL;
 277                 connect_op = OST_CONNECT;
 278                 cli->cl_sp_me = LUSTRE_SP_CLI;
 279                 cli->cl_sp_to = LUSTRE_SP_OST;
 280                 ns_type = LDLM_NS_TYPE_OSC;
 281         } else if (!strcmp(name, LUSTRE_MDC_NAME) ||
 282                    !strcmp(name, LUSTRE_LWP_NAME) ||
 283                    (!strcmp(name, LUSTRE_OSP_NAME) &&
 284                     (is_osp_on_mdt(cli_name) &&
 285                      strstr(lustre_cfg_buf(lcfg, 1), "OST") == NULL))) {
 286                 /* MDC or OSP_on_MDT for other MDTs */
 287                 rq_portal = MDS_REQUEST_PORTAL;
 288                 rp_portal = MDC_REPLY_PORTAL;
 289                 connect_op = MDS_CONNECT;
 290                 cli->cl_sp_me = LUSTRE_SP_CLI;
 291                 cli->cl_sp_to = LUSTRE_SP_MDT;
 292                 ns_type = LDLM_NS_TYPE_MDC;
 293         } else if (!strcmp(name, LUSTRE_MGC_NAME)) {
 294                 rq_portal = MGS_REQUEST_PORTAL;
 295                 rp_portal = MGC_REPLY_PORTAL;
 296                 connect_op = MGS_CONNECT;
 297                 cli->cl_sp_me = LUSTRE_SP_MGC;
 298                 cli->cl_sp_to = LUSTRE_SP_MGS;
 299                 cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID;
 300                 ns_type = LDLM_NS_TYPE_MGC;
 301         } else {
 302                 CERROR("unknown client OBD type \"%s\", can't setup\n",
 303                        name);
 304                 return -EINVAL;
 305         }
 306
 307         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 308                 CERROR("requires a TARGET UUID\n");
 309                 return -EINVAL;
 310         }
 311
 312         if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) {
 313                 CERROR("client UUID must be less than 38 characters\n");
 314                 return -EINVAL;
 315         }
 316
 317         if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
 318                 CERROR("setup requires a SERVER UUID\n");
 319                 return -EINVAL;
 320         }
 321
 322         if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) {
 323                 CERROR("target UUID must be less than 38 characters\n");
 324                 return -EINVAL;
 325         }
 326
 327         init_rwsem(&cli->cl_sem);
 328         sema_init(&cli->cl_mgc_sem, 1);
 329         cli->cl_conn_count = 0;
 330         memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2),
 331                min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2),
 332                      sizeof(server_uuid)));
 333
 334         cli->cl_dirty = 0;
 335         cli->cl_avail_grant = 0;
 336         /* FIXME: Should limit this for the sum of all cl_dirty_max. */
 337         cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024;
 338         if (cli->cl_dirty_max >> PAGE_CACHE_SHIFT > totalram_pages / 8)
 339                 cli->cl_dirty_max = totalram_pages << (PAGE_CACHE_SHIFT - 3);
 340         INIT_LIST_HEAD(&cli->cl_cache_waiters);
 341         INIT_LIST_HEAD(&cli->cl_loi_ready_list);
 342         INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list);
 343         INIT_LIST_HEAD(&cli->cl_loi_write_list);
 344         INIT_LIST_HEAD(&cli->cl_loi_read_list);
 345         client_obd_list_lock_init(&cli->cl_loi_list_lock);
 346         atomic_set(&cli->cl_pending_w_pages, 0);
 347         atomic_set(&cli->cl_pending_r_pages, 0);
 348         cli->cl_r_in_flight = 0;
 349         cli->cl_w_in_flight = 0;
 350
 351         spin_lock_init(&cli->cl_read_rpc_hist.oh_lock);
 352         spin_lock_init(&cli->cl_write_rpc_hist.oh_lock);
 353         spin_lock_init(&cli->cl_read_page_hist.oh_lock);
 354         spin_lock_init(&cli->cl_write_page_hist.oh_lock);
 355         spin_lock_init(&cli->cl_read_offset_hist.oh_lock);
 356         spin_lock_init(&cli->cl_write_offset_hist.oh_lock);
 357
 358         /* lru for osc. */
 359         INIT_LIST_HEAD(&cli->cl_lru_osc);
 360         atomic_set(&cli->cl_lru_shrinkers, 0);
 361         atomic_set(&cli->cl_lru_busy, 0);
 362         atomic_set(&cli->cl_lru_in_list, 0);
 363         INIT_LIST_HEAD(&cli->cl_lru_list);
 364         client_obd_list_lock_init(&cli->cl_lru_list_lock);
 365
 366         init_waitqueue_head(&cli->cl_destroy_waitq);
 367         atomic_set(&cli->cl_destroy_in_flight, 0);
 368         /* Turn on checksumming by default. */
 369         cli->cl_checksum = 1;
 370         /*
 371          * The supported checksum types will be worked out at connect time
 372          * Set cl_chksum* to CRC32 for now to avoid returning screwed info
 373          * through procfs.
 374          */
 375         cli->cl_cksum_type = cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
 376         atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
 377
 378         /* This value may be reduced at connect time in
 379          * ptlrpc_connect_interpret() . We initialize it to only
 380          * 1MB until we know what the performance looks like.
 381          * In the future this should likely be increased. LU-1431 */
 382         cli->cl_max_pages_per_rpc = min_t(int, PTLRPC_MAX_BRW_PAGES,
 383                                           LNET_MTU >> PAGE_CACHE_SHIFT);
 384
 385         if (!strcmp(name, LUSTRE_MDC_NAME)) {
 386                 cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
 387         } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 128 /* MB */) {
 388                 cli->cl_max_rpcs_in_flight = 2;
 389         } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 256 /* MB */) {
 390                 cli->cl_max_rpcs_in_flight = 3;
 391         } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 512 /* MB */) {
 392                 cli->cl_max_rpcs_in_flight = 4;
 393         } else {
 394                 if (osc_on_mdt(obddev->obd_name))
 395                         cli->cl_max_rpcs_in_flight = MDS_OSC_MAX_RIF_DEFAULT;
 396                 else
 397                         cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
 398         }
 399         rc = ldlm_get_ref();
 400         if (rc) {
 401                 CERROR("ldlm_get_ref failed: %d\n", rc);
 402                 GOTO(err, rc);
 403         }
 404
 405         ptlrpc_init_client(rq_portal, rp_portal, name,
 406                            &obddev->obd_ldlm_client);
 407
 408         imp = class_new_import(obddev);
 409         if (imp == NULL)
 410                 GOTO(err_ldlm, rc = -ENOENT);
 411         imp->imp_client = &obddev->obd_ldlm_client;
 412         imp->imp_connect_op = connect_op;
 413         memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
 414                LUSTRE_CFG_BUFLEN(lcfg, 1));
 415         class_import_put(imp);
 416
 417         rc = client_import_add_conn(imp, &server_uuid, 1);
 418         if (rc) {
 419                 CERROR("can't add initial connection\n");
 420                 GOTO(err_import, rc);
 421         }
 422
 423         cli->cl_import = imp;
 424         /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
 425         cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3);
 426         cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
 427
 428         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
 429                 if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) {
 430                         CDEBUG(D_HA, "marking %s %s->%s as inactive\n",
 431                                name, obddev->obd_name,
 432                                cli->cl_target_uuid.uuid);
 433                         spin_lock(&imp->imp_lock);
 434                         imp->imp_deactive = 1;
 435                         spin_unlock(&imp->imp_lock);
 436                 }
 437         }
 438
 439         obddev->obd_namespace = ldlm_namespace_new(obddev, obddev->obd_name,
 440                                                    LDLM_NAMESPACE_CLIENT,
 441                                                    LDLM_NAMESPACE_GREEDY,
 442                                                    ns_type);
 443         if (obddev->obd_namespace == NULL) {
 444                 CERROR("Unable to create client namespace - %s\n",
 445                        obddev->obd_name);
 446                 GOTO(err_import, rc = -ENOMEM);
 447         }
 448
 449         cli->cl_qchk_stat = CL_NOT_QUOTACHECKED;
 450
 451         return rc;
 452
 453 err_import:
 454         class_destroy_import(imp);
 455 err_ldlm:
 456         ldlm_put_ref();
 457 err:
 458         return rc;
 459
 460 }
 461 EXPORT_SYMBOL(client_obd_setup);
 462
 463 int client_obd_cleanup(struct obd_device *obddev)
 464 {
 465         ldlm_namespace_free_post(obddev->obd_namespace);
 466         obddev->obd_namespace = NULL;
 467
 468         LASSERT(obddev->u.cli.cl_import == NULL);
 469
 470         ldlm_put_ref();
 471         return 0;
 472 }
 473 EXPORT_SYMBOL(client_obd_cleanup);
 474
 475 /* ->o_connect() method for client side (OSC and MDC and MGC) */
 476 int client_connect_import(const struct lu_env *env,
 477                           struct obd_export **exp,
 478                           struct obd_device *obd, struct obd_uuid *cluuid,
 479                           struct obd_connect_data *data, void *localdata)
 480 {
 481         struct client_obd       *cli    = &obd->u.cli;
 482         struct obd_import       *imp    = cli->cl_import;
 483         struct obd_connect_data *ocd;
 484         struct lustre_handle    conn    = { 0 };
 485         int                  rc;
 486
 487         *exp = NULL;
 488         down_write(&cli->cl_sem);
 489         if (cli->cl_conn_count > 0 )
 490                 GOTO(out_sem, rc = -EALREADY);
 491
 492         rc = class_connect(&conn, obd, cluuid);
 493         if (rc)
 494                 GOTO(out_sem, rc);
 495
 496         cli->cl_conn_count++;
 497         *exp = class_conn2export(&conn);
 498
 499         LASSERT(obd->obd_namespace);
 500
 501         imp->imp_dlm_handle = conn;
 502         rc = ptlrpc_init_import(imp);
 503         if (rc != 0)
 504                 GOTO(out_ldlm, rc);
 505
 506         ocd = &imp->imp_connect_data;
 507         if (data) {
 508                 *ocd = *data;
 509                 imp->imp_connect_flags_orig = data->ocd_connect_flags;
 510         }
 511
 512         rc = ptlrpc_connect_import(imp);
 513         if (rc != 0) {
 514                 LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
 515                 GOTO(out_ldlm, rc);
 516         }
 517         LASSERT((*exp)->exp_connection);
 518
 519         if (data) {
 520                 LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
 521                          ocd->ocd_connect_flags, "old "LPX64", new "LPX64"\n",
 522                          data->ocd_connect_flags, ocd->ocd_connect_flags);
 523                 data->ocd_connect_flags = ocd->ocd_connect_flags;
 524         }
 525
 526         ptlrpc_pinger_add_import(imp);
 527
 528         if (rc) {
 529 out_ldlm:
 530                 cli->cl_conn_count--;
 531                 class_disconnect(*exp);
 532                 *exp = NULL;
 533         }
 534 out_sem:
 535         up_write(&cli->cl_sem);
 536
 537         return rc;
 538 }
 539 EXPORT_SYMBOL(client_connect_import);
 540
 541 int client_disconnect_export(struct obd_export *exp)
 542 {
 543         struct obd_device *obd = class_exp2obd(exp);
 544         struct client_obd *cli;
 545         struct obd_import *imp;
 546         int rc = 0, err;
 547
 548         if (!obd) {
 549                 CERROR("invalid export for disconnect: exp %p cookie "LPX64"\n",
 550                        exp, exp ? exp->exp_handle.h_cookie : -1);
 551                 return -EINVAL;
 552         }
 553
 554         cli = &obd->u.cli;
 555         imp = cli->cl_import;
 556
 557         down_write(&cli->cl_sem);
 558         CDEBUG(D_INFO, "disconnect %s - %d\n", obd->obd_name,
 559                cli->cl_conn_count);
 560
 561         if (!cli->cl_conn_count) {
 562                 CERROR("disconnecting disconnected device (%s)\n",
 563                        obd->obd_name);
 564                 GOTO(out_disconnect, rc = -EINVAL);
 565         }
 566
 567         cli->cl_conn_count--;
 568         if (cli->cl_conn_count)
 569                 GOTO(out_disconnect, rc = 0);
 570
 571         /* Mark import deactivated now, so we don't try to reconnect if any
 572          * of the cleanup RPCs fails (e.g. LDLM cancel, etc).  We don't
 573          * fully deactivate the import, or that would drop all requests. */
 574         spin_lock(&imp->imp_lock);
 575         imp->imp_deactive = 1;
 576         spin_unlock(&imp->imp_lock);
 577
 578         /* Some non-replayable imports (MDS's OSCs) are pinged, so just
 579          * delete it regardless.  (It's safe to delete an import that was
 580          * never added.) */
 581         (void)ptlrpc_pinger_del_import(imp);
 582
 583         if (obd->obd_namespace != NULL) {
 584                 /* obd_force == local only */
 585                 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
 586                                        obd->obd_force ? LCF_LOCAL : 0, NULL);
 587                 ldlm_namespace_free_prior(obd->obd_namespace, imp, obd->obd_force);
 588         }
 589
 590         /* There's no need to hold sem while disconnecting an import,
 591          * and it may actually cause deadlock in GSS. */
 592         up_write(&cli->cl_sem);
 593         rc = ptlrpc_disconnect_import(imp, 0);
 594         down_write(&cli->cl_sem);
 595
 596         ptlrpc_invalidate_import(imp);
 597
 598 out_disconnect:
 599         /* Use server style - class_disconnect should be always called for
 600          * o_disconnect. */
 601         err = class_disconnect(exp);
 602         if (!rc && err)
 603                 rc = err;
 604
 605         up_write(&cli->cl_sem);
 606
 607         return rc;
 608 }
 609 EXPORT_SYMBOL(client_disconnect_export);
 610
 611
 612 /**
 613  * Packs current SLV and Limit into \a req.
 614  */
 615 int target_pack_pool_reply(struct ptlrpc_request *req)
 616 {
 617         struct obd_device *obd;
 618
 619         /* Check that we still have all structures alive as this may
 620          * be some late RPC at shutdown time. */
 621         if (unlikely(!req->rq_export || !req->rq_export->exp_obd ||
 622                      !exp_connect_lru_resize(req->rq_export))) {
 623                 lustre_msg_set_slv(req->rq_repmsg, 0);
 624                 lustre_msg_set_limit(req->rq_repmsg, 0);
 625                 return 0;
 626         }
 627
 628         /* OBD is alive here as export is alive, which we checked above. */
 629         obd = req->rq_export->exp_obd;
 630
 631         read_lock(&obd->obd_pool_lock);
 632         lustre_msg_set_slv(req->rq_repmsg, obd->obd_pool_slv);
 633         lustre_msg_set_limit(req->rq_repmsg, obd->obd_pool_limit);
 634         read_unlock(&obd->obd_pool_lock);
 635
 636         return 0;
 637 }
 638 EXPORT_SYMBOL(target_pack_pool_reply);
 639
 640 int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id)
 641 {
 642         if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) {
 643                 DEBUG_REQ(D_ERROR, req, "dropping reply");
 644                 return (-ECOMM);
 645         }
 646
 647         if (unlikely(rc)) {
 648                 DEBUG_REQ(D_NET, req, "processing error (%d)", rc);
 649                 req->rq_status = rc;
 650                 return (ptlrpc_send_error(req, 1));
 651         } else {
 652                 DEBUG_REQ(D_NET, req, "sending reply");
 653         }
 654
 655         return (ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT));
 656 }
 657
 658 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
 659 {
 660         struct ptlrpc_service_part *svcpt;
 661         int                     netrc;
 662         struct ptlrpc_reply_state *rs;
 663         struct obd_export        *exp;
 664
 665         if (req->rq_no_reply) {
 666                 return;
 667         }
 668
 669         svcpt = req->rq_rqbd->rqbd_svcpt;
 670         rs = req->rq_reply_state;
 671         if (rs == NULL || !rs->rs_difficult) {
 672                 /* no notifiers */
 673                 target_send_reply_msg (req, rc, fail_id);
 674                 return;
 675         }
 676
 677         /* must be an export if locks saved */
 678         LASSERT (req->rq_export != NULL);
 679         /* req/reply consistent */
 680         LASSERT(rs->rs_svcpt == svcpt);
 681
 682         /* "fresh" reply */
 683         LASSERT (!rs->rs_scheduled);
 684         LASSERT (!rs->rs_scheduled_ever);
 685         LASSERT (!rs->rs_handled);
 686         LASSERT (!rs->rs_on_net);
 687         LASSERT (rs->rs_export == NULL);
 688         LASSERT (list_empty(&rs->rs_obd_list));
 689         LASSERT (list_empty(&rs->rs_exp_list));
 690
 691         exp = class_export_get (req->rq_export);
 692
 693         /* disable reply scheduling while I'm setting up */
 694         rs->rs_scheduled = 1;
 695         rs->rs_on_net    = 1;
 696         rs->rs_xid       = req->rq_xid;
 697         rs->rs_transno   = req->rq_transno;
 698         rs->rs_export    = exp;
 699         rs->rs_opc       = lustre_msg_get_opc(req->rq_reqmsg);
 700
 701         spin_lock(&exp->exp_uncommitted_replies_lock);
 702         CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n",
 703                rs->rs_transno, exp->exp_last_committed);
 704         if (rs->rs_transno > exp->exp_last_committed) {
 705                 /* not committed already */
 706                 list_add_tail(&rs->rs_obd_list,
 707                                   &exp->exp_uncommitted_replies);
 708         }
 709         spin_unlock(&exp->exp_uncommitted_replies_lock);
 710
 711         spin_lock(&exp->exp_lock);
 712         list_add_tail(&rs->rs_exp_list, &exp->exp_outstanding_replies);
 713         spin_unlock(&exp->exp_lock);
 714
 715         netrc = target_send_reply_msg(req, rc, fail_id);
 716
 717         spin_lock(&svcpt->scp_rep_lock);
 718
 719         atomic_inc(&svcpt->scp_nreps_difficult);
 720
 721         if (netrc != 0) {
 722                 /* error sending: reply is off the net.  Also we need +1
 723                  * reply ref until ptlrpc_handle_rs() is done
 724                  * with the reply state (if the send was successful, there
 725                  * would have been +1 ref for the net, which
 726                  * reply_out_callback leaves alone) */
 727                 rs->rs_on_net = 0;
 728                 ptlrpc_rs_addref(rs);
 729         }
 730
 731         spin_lock(&rs->rs_lock);
 732         if (rs->rs_transno <= exp->exp_last_committed ||
 733             (!rs->rs_on_net && !rs->rs_no_ack) ||
 734             list_empty(&rs->rs_exp_list) ||     /* completed already */
 735             list_empty(&rs->rs_obd_list)) {
 736                 CDEBUG(D_HA, "Schedule reply immediately\n");
 737                 ptlrpc_dispatch_difficult_reply(rs);
 738         } else {
 739                 list_add(&rs->rs_list, &svcpt->scp_rep_active);
 740                 rs->rs_scheduled = 0;   /* allow notifier to schedule */
 741         }
 742         spin_unlock(&rs->rs_lock);
 743         spin_unlock(&svcpt->scp_rep_lock);
 744 }
 745 EXPORT_SYMBOL(target_send_reply);
 746
 747 ldlm_mode_t lck_compat_array[] = {
 748         [LCK_EX]        = LCK_COMPAT_EX,
 749         [LCK_PW]        = LCK_COMPAT_PW,
 750         [LCK_PR]        = LCK_COMPAT_PR,
 751         [LCK_CW]        = LCK_COMPAT_CW,
 752         [LCK_CR]        = LCK_COMPAT_CR,
 753         [LCK_NL]        = LCK_COMPAT_NL,
 754         [LCK_GROUP]     = LCK_COMPAT_GROUP,
 755         [LCK_COS]       = LCK_COMPAT_COS,
 756 };
 757
 758 /**
 759  * Rather arbitrary mapping from LDLM error codes to errno values. This should
 760  * not escape to the user level.
 761  */
 762 int ldlm_error2errno(ldlm_error_t error)
 763 {
 764         int result;
 765
 766         switch (error) {
 767         case ELDLM_OK:
 768                 result = 0;
 769                 break;
 770         case ELDLM_LOCK_CHANGED:
 771                 result = -ESTALE;
 772                 break;
 773         case ELDLM_LOCK_ABORTED:
 774                 result = -ENAVAIL;
 775                 break;
 776         case ELDLM_LOCK_REPLACED:
 777                 result = -ESRCH;
 778                 break;
 779         case ELDLM_NO_LOCK_DATA:
 780                 result = -ENOENT;
 781                 break;
 782         case ELDLM_NAMESPACE_EXISTS:
 783                 result = -EEXIST;
 784                 break;
 785         case ELDLM_BAD_NAMESPACE:
 786                 result = -EBADF;
 787                 break;
 788         default:
 789                 if (((int)error) < 0)  /* cast to signed type */
 790                         result = error; /* as ldlm_error_t can be unsigned */
 791                 else {
 792                         CERROR("Invalid DLM result code: %d\n", error);
 793                         result = -EPROTO;
 794                 }
 795         }
 796         return result;
 797 }
 798 EXPORT_SYMBOL(ldlm_error2errno);
 799
 800 /**
 801  * Dual to ldlm_error2errno(): maps errno values back to ldlm_error_t.
 802  */
 803 ldlm_error_t ldlm_errno2error(int err_no)
 804 {
 805         int error;
 806
 807         switch (err_no) {
 808         case 0:
 809                 error = ELDLM_OK;
 810                 break;
 811         case -ESTALE:
 812                 error = ELDLM_LOCK_CHANGED;
 813                 break;
 814         case -ENAVAIL:
 815                 error = ELDLM_LOCK_ABORTED;
 816                 break;
 817         case -ESRCH:
 818                 error = ELDLM_LOCK_REPLACED;
 819                 break;
 820         case -ENOENT:
 821                 error = ELDLM_NO_LOCK_DATA;
 822                 break;
 823         case -EEXIST:
 824                 error = ELDLM_NAMESPACE_EXISTS;
 825                 break;
 826         case -EBADF:
 827                 error = ELDLM_BAD_NAMESPACE;
 828                 break;
 829         default:
 830                 error = err_no;
 831         }
 832         return error;
 833 }
 834 EXPORT_SYMBOL(ldlm_errno2error);
 835
 836 #if LUSTRE_TRACKS_LOCK_EXP_REFS
 837 void ldlm_dump_export_locks(struct obd_export *exp)
 838 {
 839         spin_lock(&exp->exp_locks_list_guard);
 840         if (!list_empty(&exp->exp_locks_list)) {
 841                 struct ldlm_lock *lock;
 842
 843                 CERROR("dumping locks for export %p,"
 844                        "ignore if the unmount doesn't hang\n", exp);
 845                 list_for_each_entry(lock, &exp->exp_locks_list,
 846                                         l_exp_refs_link)
 847                         LDLM_ERROR(lock, "lock:");
 848         }
 849         spin_unlock(&exp->exp_locks_list_guard);
 850 }
 851 #endif