lustre/ldlm/ldlm_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LDLM
  41
  42 #ifdef __KERNEL__
  43 # include <libcfs/libcfs.h>
  44 #else
  45 # include <liblustre.h>
  46 #endif
  47 #include <obd.h>
  48 #include <lustre_mds.h>
  49 #include <lustre_dlm.h>
  50 #include <lustre_net.h>
  51 #include <lustre_sec.h>
  52 #include "ldlm_internal.h"
  53
  54
  55 /* @priority: if non-zero, move the selected to the list head
  56  * @create: if zero, only search in existed connections
  57  */
  58 static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid,
  59                            int priority, int create)
  60 {
  61         struct ptlrpc_connection *ptlrpc_conn;
  62         struct obd_import_conn *imp_conn = NULL, *item;
  63         int rc = 0;
  64         ENTRY;
  65
  66         if (!create && !priority) {
  67                 CDEBUG(D_HA, "Nothing to do\n");
  68                 RETURN(-EINVAL);
  69         }
  70
  71         ptlrpc_conn = ptlrpc_uuid_to_connection(uuid);
  72         if (!ptlrpc_conn) {
  73                 CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid);
  74                 RETURN (-ENOENT);
  75         }
  76
  77         if (create) {
  78                 OBD_ALLOC(imp_conn, sizeof(*imp_conn));
  79                 if (!imp_conn) {
  80                         GOTO(out_put, rc = -ENOMEM);
  81                 }
  82         }
  83
  84         spin_lock(&imp->imp_lock);
  85         list_for_each_entry(item, &imp->imp_conn_list, oic_item) {
  86                 if (obd_uuid_equals(uuid, &item->oic_uuid)) {
  87                         if (priority) {
  88                                 list_del(&item->oic_item);
  89                                 list_add(&item->oic_item, &imp->imp_conn_list);
  90                                 item->oic_last_attempt = 0;
  91                         }
  92                         CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n",
  93                                imp, imp->imp_obd->obd_name, uuid->uuid,
  94                                (priority ? ", moved to head" : ""));
  95                         spin_unlock(&imp->imp_lock);
  96                         GOTO(out_free, rc = 0);
  97                 }
  98         }
  99         /* not found */
 100         if (create) {
 101                 imp_conn->oic_conn = ptlrpc_conn;
 102                 imp_conn->oic_uuid = *uuid;
 103                 item->oic_last_attempt = 0;
 104                 if (priority)
 105                         list_add(&imp_conn->oic_item, &imp->imp_conn_list);
 106                 else
 107                         list_add_tail(&imp_conn->oic_item, &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
 137 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
 138 {
 139         struct obd_import_conn *imp_conn;
 140         struct obd_export *dlmexp;
 141         int rc = -ENOENT;
 142         ENTRY;
 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                 /* is current conn? */
 156                 if (imp_conn == imp->imp_conn_current) {
 157                         LASSERT(imp_conn->oic_conn == imp->imp_connection);
 158
 159                         if (imp->imp_state != LUSTRE_IMP_CLOSED &&
 160                             imp->imp_state != LUSTRE_IMP_DISCON) {
 161                                 CERROR("can't remove current connection\n");
 162                                 GOTO(out, rc = -EBUSY);
 163                         }
 164
 165                         ptlrpc_connection_put(imp->imp_connection);
 166                         imp->imp_connection = NULL;
 167
 168                         dlmexp = class_conn2export(&imp->imp_dlm_handle);
 169                         if (dlmexp && dlmexp->exp_connection) {
 170                                 LASSERT(dlmexp->exp_connection ==
 171                                         imp_conn->oic_conn);
 172                                 ptlrpc_connection_put(dlmexp->exp_connection);
 173                                 dlmexp->exp_connection = NULL;
 174                         }
 175                 }
 176
 177                 list_del(&imp_conn->oic_item);
 178                 ptlrpc_connection_put(imp_conn->oic_conn);
 179                 OBD_FREE(imp_conn, sizeof(*imp_conn));
 180                 CDEBUG(D_HA, "imp %p@%s: remove connection %s\n",
 181                        imp, imp->imp_obd->obd_name, uuid->uuid);
 182                 rc = 0;
 183                 break;
 184         }
 185 out:
 186         spin_unlock(&imp->imp_lock);
 187         if (rc == -ENOENT)
 188                 CERROR("connection %s not found\n", uuid->uuid);
 189         RETURN(rc);
 190 }
 191
 192 static void destroy_import(struct obd_import *imp)
 193 {
 194         /* drop security policy instance after all rpc finished/aborted
 195          * to let all busy contexts be released. */
 196         class_import_get(imp);
 197         class_destroy_import(imp);
 198         sptlrpc_import_sec_put(imp);
 199         class_import_put(imp);
 200 }
 201
 202 /* configure an RPC client OBD device
 203  *
 204  * lcfg parameters:
 205  * 1 - client UUID
 206  * 2 - server UUID
 207  * 3 - inactive-on-startup
 208  */
 209 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
 210 {
 211         struct client_obd *cli = &obddev->u.cli;
 212         struct obd_import *imp;
 213         struct obd_uuid server_uuid;
 214         int rq_portal, rp_portal, connect_op;
 215         char *name = obddev->obd_type->typ_name;
 216         int rc;
 217         ENTRY;
 218
 219         /* In a more perfect world, we would hang a ptlrpc_client off of
 220          * obd_type and just use the values from there. */
 221         if (!strcmp(name, LUSTRE_OSC_NAME)) {
 222                 rq_portal = OST_REQUEST_PORTAL;
 223                 rp_portal = OSC_REPLY_PORTAL;
 224                 connect_op = OST_CONNECT;
 225         } else if (!strcmp(name, LUSTRE_MDC_NAME)) {
 226                 rq_portal = MDS_REQUEST_PORTAL;
 227                 rp_portal = MDC_REPLY_PORTAL;
 228                 connect_op = MDS_CONNECT;
 229         } else if (!strcmp(name, LUSTRE_MGC_NAME)) {
 230                 rq_portal = MGS_REQUEST_PORTAL;
 231                 rp_portal = MGC_REPLY_PORTAL;
 232                 connect_op = MGS_CONNECT;
 233         } else {
 234                 CERROR("unknown client OBD type \"%s\", can't setup\n",
 235                        name);
 236                 RETURN(-EINVAL);
 237         }
 238
 239         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 240                 CERROR("requires a TARGET UUID\n");
 241                 RETURN(-EINVAL);
 242         }
 243
 244         if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) {
 245                 CERROR("client UUID must be less than 38 characters\n");
 246                 RETURN(-EINVAL);
 247         }
 248
 249         if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
 250                 CERROR("setup requires a SERVER UUID\n");
 251                 RETURN(-EINVAL);
 252         }
 253
 254         if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) {
 255                 CERROR("target UUID must be less than 38 characters\n");
 256                 RETURN(-EINVAL);
 257         }
 258
 259         init_rwsem(&cli->cl_sem);
 260         sema_init(&cli->cl_mgc_sem, 1);
 261         sptlrpc_rule_set_init(&cli->cl_sptlrpc_rset);
 262         cli->cl_sec_part = LUSTRE_SP_ANY;
 263         cli->cl_conn_count = 0;
 264         memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2),
 265                min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2),
 266                      sizeof(server_uuid)));
 267
 268         cli->cl_dirty = 0;
 269         cli->cl_avail_grant = 0;
 270         /* FIXME: should limit this for the sum of all cl_dirty_max */
 271         cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024;
 272         if (cli->cl_dirty_max >> CFS_PAGE_SHIFT > num_physpages / 8)
 273                 cli->cl_dirty_max = num_physpages << (CFS_PAGE_SHIFT - 3);
 274         CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters);
 275         CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list);
 276         CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list);
 277         CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list);
 278         client_obd_list_lock_init(&cli->cl_loi_list_lock);
 279         cli->cl_r_in_flight = 0;
 280         cli->cl_w_in_flight = 0;
 281
 282         spin_lock_init(&cli->cl_read_rpc_hist.oh_lock);
 283         spin_lock_init(&cli->cl_write_rpc_hist.oh_lock);
 284         spin_lock_init(&cli->cl_read_page_hist.oh_lock);
 285         spin_lock_init(&cli->cl_write_page_hist.oh_lock);
 286         spin_lock_init(&cli->cl_read_offset_hist.oh_lock);
 287         spin_lock_init(&cli->cl_write_offset_hist.oh_lock);
 288         cfs_waitq_init(&cli->cl_destroy_waitq);
 289         atomic_set(&cli->cl_destroy_in_flight, 0);
 290 #ifdef ENABLE_CHECKSUM
 291         /* Turn on checksumming by default. */
 292         cli->cl_checksum = 1;
 293         /*
 294          * The supported checksum types will be worked out at connect time
 295          * Set cl_chksum* to CRC32 for now to avoid returning screwed info
 296          * through procfs.
 297          */
 298         cli->cl_cksum_type = cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
 299 #endif
 300         atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
 301
 302         /* This value may be changed at connect time in
 303            ptlrpc_connect_interpret. */
 304         cli->cl_max_pages_per_rpc = min((int)PTLRPC_MAX_BRW_PAGES,
 305                                         (int)(1024 * 1024 >> CFS_PAGE_SHIFT));
 306
 307         if (!strcmp(name, LUSTRE_MDC_NAME)) {
 308                 cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
 309         } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 128 /* MB */) {
 310                 cli->cl_max_rpcs_in_flight = 2;
 311         } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 256 /* MB */) {
 312                 cli->cl_max_rpcs_in_flight = 3;
 313         } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 512 /* MB */) {
 314                 cli->cl_max_rpcs_in_flight = 4;
 315         } else {
 316                 cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
 317         }
 318
 319         rc = ldlm_get_ref();
 320         if (rc) {
 321                 CERROR("ldlm_get_ref failed: %d\n", rc);
 322                 GOTO(err, rc);
 323         }
 324
 325         ptlrpc_init_client(rq_portal, rp_portal, name,
 326                            &obddev->obd_ldlm_client);
 327
 328         imp = class_new_import(obddev);
 329         if (imp == NULL)
 330                 GOTO(err_ldlm, rc = -ENOENT);
 331         imp->imp_client = &obddev->obd_ldlm_client;
 332         imp->imp_connect_op = connect_op;
 333         imp->imp_initial_recov = 1;
 334         imp->imp_initial_recov_bk = 0;
 335         CFS_INIT_LIST_HEAD(&imp->imp_pinger_chain);
 336         memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
 337                LUSTRE_CFG_BUFLEN(lcfg, 1));
 338         class_import_put(imp);
 339
 340         rc = client_import_add_conn(imp, &server_uuid, 1);
 341         if (rc) {
 342                 CERROR("can't add initial connection\n");
 343                 GOTO(err_import, rc);
 344         }
 345
 346         cli->cl_import = imp;
 347         /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
 348         cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3);
 349         cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
 350
 351         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
 352                 if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) {
 353                         CDEBUG(D_HA, "marking %s %s->%s as inactive\n",
 354                                name, obddev->obd_name,
 355                                cli->cl_target_uuid.uuid);
 356                         spin_lock(&imp->imp_lock);
 357                         imp->imp_invalid = 1;
 358                         spin_unlock(&imp->imp_lock);
 359                 }
 360         }
 361
 362         cli->cl_qchk_stat = CL_NOT_QUOTACHECKED;
 363
 364         RETURN(rc);
 365
 366 err_import:
 367         class_destroy_import(imp);
 368 err_ldlm:
 369         ldlm_put_ref();
 370 err:
 371         RETURN(rc);
 372
 373 }
 374
 375 int client_obd_cleanup(struct obd_device *obddev)
 376 {
 377         ENTRY;
 378         sptlrpc_rule_set_free(&obddev->u.cli.cl_sptlrpc_rset);
 379         ldlm_put_ref();
 380         RETURN(0);
 381 }
 382
 383 /* ->o_connect() method for client side (OSC and MDC and MGC) */
 384 int client_connect_import(const struct lu_env *env,
 385                           struct lustre_handle *dlm_handle,
 386                           struct obd_device *obd, struct obd_uuid *cluuid,
 387                           struct obd_connect_data *data, void *localdata)
 388 {
 389         struct client_obd *cli = &obd->u.cli;
 390         struct obd_import *imp = cli->cl_import;
 391         struct obd_export *exp;
 392         struct obd_connect_data *ocd;
 393         struct ldlm_namespace *to_be_freed = NULL;
 394         int rc;
 395         ENTRY;
 396
 397         down_write(&cli->cl_sem);
 398         rc = class_connect(dlm_handle, obd, cluuid);
 399         if (rc)
 400                 GOTO(out_sem, rc);
 401
 402         cli->cl_conn_count++;
 403         if (cli->cl_conn_count > 1)
 404                 GOTO(out_sem, rc);
 405         exp = class_conn2export(dlm_handle);
 406
 407         if (obd->obd_namespace != NULL)
 408                 CERROR("already have namespace!\n");
 409         obd->obd_namespace = ldlm_namespace_new(obd, obd->obd_name,
 410                                                 LDLM_NAMESPACE_CLIENT,
 411                                                 LDLM_NAMESPACE_GREEDY);
 412         if (obd->obd_namespace == NULL)
 413                 GOTO(out_disco, rc = -ENOMEM);
 414
 415         imp->imp_dlm_handle = *dlm_handle;
 416         rc = ptlrpc_init_import(imp);
 417         if (rc != 0)
 418                 GOTO(out_ldlm, rc);
 419
 420         ocd = &imp->imp_connect_data;
 421         if (data) {
 422                 *ocd = *data;
 423                 imp->imp_connect_flags_orig = data->ocd_connect_flags;
 424         }
 425
 426         rc = ptlrpc_connect_import(imp, NULL);
 427         if (rc != 0) {
 428                 LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
 429                 GOTO(out_ldlm, rc);
 430         }
 431         LASSERT(exp->exp_connection);
 432
 433         if (data) {
 434                 LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
 435                          ocd->ocd_connect_flags, "old "LPX64", new "LPX64"\n",
 436                          data->ocd_connect_flags, ocd->ocd_connect_flags);
 437                 data->ocd_connect_flags = ocd->ocd_connect_flags;
 438         }
 439
 440         ptlrpc_pinger_add_import(imp);
 441
 442         EXIT;
 443
 444         if (rc) {
 445 out_ldlm:
 446                 ldlm_namespace_free_prior(obd->obd_namespace, imp, 0);
 447                 to_be_freed = obd->obd_namespace;
 448                 obd->obd_namespace = NULL;
 449 out_disco:
 450                 cli->cl_conn_count--;
 451                 class_disconnect(exp);
 452         } else {
 453                 class_export_put(exp);
 454         }
 455 out_sem:
 456         up_write(&cli->cl_sem);
 457         if (to_be_freed)
 458                 ldlm_namespace_free_post(to_be_freed);
 459
 460         return rc;
 461 }
 462
 463 int client_disconnect_export(struct obd_export *exp)
 464 {
 465         struct obd_device *obd = class_exp2obd(exp);
 466         struct client_obd *cli;
 467         struct obd_import *imp;
 468         int rc = 0, err;
 469         struct ldlm_namespace *to_be_freed = NULL;
 470         ENTRY;
 471
 472         if (!obd) {
 473                 CERROR("invalid export for disconnect: exp %p cookie "LPX64"\n",
 474                        exp, exp ? exp->exp_handle.h_cookie : -1);
 475                 RETURN(-EINVAL);
 476         }
 477
 478         cli = &obd->u.cli;
 479         imp = cli->cl_import;
 480
 481         down_write(&cli->cl_sem);
 482         if (!cli->cl_conn_count) {
 483                 CERROR("disconnecting disconnected device (%s)\n",
 484                        obd->obd_name);
 485                 GOTO(out_sem, rc = -EINVAL);
 486         }
 487
 488         cli->cl_conn_count--;
 489         if (cli->cl_conn_count)
 490                 GOTO(out_no_disconnect, rc = 0);
 491
 492         /* Mark import deactivated now, so we don't try to reconnect if any
 493          * of the cleanup RPCs fails (e.g. ldlm cancel, etc).  We don't
 494          * fully deactivate the import, or that would drop all requests. */
 495         spin_lock(&imp->imp_lock);
 496         imp->imp_deactive = 1;
 497         spin_unlock(&imp->imp_lock);
 498
 499         /* Some non-replayable imports (MDS's OSCs) are pinged, so just
 500          * delete it regardless.  (It's safe to delete an import that was
 501          * never added.) */
 502         (void)ptlrpc_pinger_del_import(imp);
 503
 504         if (obd->obd_namespace != NULL) {
 505                 /* obd_force == local only */
 506                 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
 507                                        obd->obd_force ? LDLM_FL_LOCAL_ONLY:0,
 508                                        NULL);
 509                 ldlm_namespace_free_prior(obd->obd_namespace, imp, obd->obd_force);
 510                 to_be_freed = obd->obd_namespace;
 511         }
 512
 513         rc = ptlrpc_disconnect_import(imp, 0);
 514
 515         ptlrpc_invalidate_import(imp);
 516         /* set obd_namespace to NULL only after invalidate, because we can have
 517          * some connect requests in flight, and his need store a connect flags
 518          * in obd_namespace. bug 14260 */
 519         obd->obd_namespace = NULL;
 520
 521         if (imp->imp_rq_pool) {
 522                 ptlrpc_free_rq_pool(imp->imp_rq_pool);
 523                 imp->imp_rq_pool = NULL;
 524         }
 525         destroy_import(imp);
 526         cli->cl_import = NULL;
 527
 528         EXIT;
 529  out_no_disconnect:
 530         err = class_disconnect(exp);
 531         if (!rc && err)
 532                 rc = err;
 533  out_sem:
 534         up_write(&cli->cl_sem);
 535         if (to_be_freed)
 536                 ldlm_namespace_free_post(to_be_freed);
 537
 538         RETURN(rc);
 539 }
 540
 541 /* --------------------------------------------------------------------------
 542  * from old lib/target.c
 543  * -------------------------------------------------------------------------- */
 544
 545 int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
 546                             struct obd_uuid *cluuid, int initial_conn)
 547 {
 548         ENTRY;
 549         if (exp->exp_connection && exp->exp_imp_reverse && !initial_conn) {
 550                 struct lustre_handle *hdl;
 551                 hdl = &exp->exp_imp_reverse->imp_remote_handle;
 552                 /* Might be a re-connect after a partition. */
 553                 if (!memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) {
 554                         CWARN("%s: %s reconnecting\n", exp->exp_obd->obd_name,
 555                               cluuid->uuid);
 556                         conn->cookie = exp->exp_handle.h_cookie;
 557                         /* target_handle_connect() treats EALREADY and
 558                          * -EALREADY differently.  EALREADY means we are
 559                          * doing a valid reconnect from the same client. */
 560                         RETURN(EALREADY);
 561                 } else {
 562                         CERROR("%s reconnecting from %s, "
 563                                "handle mismatch (ours "LPX64", theirs "
 564                                LPX64")\n", cluuid->uuid,
 565                                exp->exp_connection->c_remote_uuid.uuid,
 566                                hdl->cookie, conn->cookie);
 567                         memset(conn, 0, sizeof *conn);
 568                         /* target_handle_connect() treats EALREADY and
 569                          * -EALREADY differently.  -EALREADY is an error
 570                          * (same UUID, different handle). */
 571                         RETURN(-EALREADY);
 572                 }
 573         }
 574
 575         conn->cookie = exp->exp_handle.h_cookie;
 576         CDEBUG(D_HA, "connect export for UUID '%s' at %p, cookie "LPX64"\n",
 577                cluuid->uuid, exp, conn->cookie);
 578         RETURN(0);
 579 }
 580
 581 void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data,
 582                           int error)
 583 {
 584         struct obd_export *exp = cb_data;
 585
 586         CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n",
 587                obd->obd_name, exp->exp_client_uuid.uuid);
 588
 589         spin_lock(&exp->exp_lock);
 590         exp->exp_need_sync = 0;
 591         spin_unlock(&exp->exp_lock);
 592 }
 593 EXPORT_SYMBOL(target_client_add_cb);
 594 static void
 595 target_start_and_reset_recovery_timer(struct obd_device *obd,
 596                                       struct ptlrpc_request *req,
 597                                       int new_client);
 598
 599 int target_handle_connect(struct ptlrpc_request *req)
 600 {
 601         struct obd_device *target, *targref = NULL;
 602         struct obd_export *export = NULL;
 603         struct obd_import *revimp;
 604         struct lustre_handle conn;
 605         struct lustre_handle *tmp;
 606         struct obd_uuid tgtuuid;
 607         struct obd_uuid cluuid;
 608         struct obd_uuid remote_uuid;
 609         char *str;
 610         int rc = 0;
 611         int initial_conn = 0;
 612         struct obd_connect_data *data, *tmpdata;
 613         lnet_nid_t *client_nid = NULL;
 614         ENTRY;
 615
 616         OBD_RACE(OBD_FAIL_TGT_CONN_RACE);
 617
 618         str = req_capsule_client_get(&req->rq_pill, &RMF_TGTUUID);
 619         if (str == NULL) {
 620                 DEBUG_REQ(D_ERROR, req, "bad target UUID for connect");
 621                 GOTO(out, rc = -EINVAL);
 622         }
 623
 624         obd_str2uuid(&tgtuuid, str);
 625         target = class_uuid2obd(&tgtuuid);
 626         if (!target)
 627                 target = class_name2obd(str);
 628
 629         if (!target || target->obd_stopping || !target->obd_set_up) {
 630                 LCONSOLE_ERROR_MSG(0x137, "UUID '%s' is not available "
 631                                    " for connect (%s)\n", str,
 632                                    !target ? "no target" :
 633                                    (target->obd_stopping ? "stopping" :
 634                                    "not set up"));
 635                 GOTO(out, rc = -ENODEV);
 636         }
 637
 638         if (target->obd_no_conn) {
 639                 LCONSOLE_WARN("%s: temporarily refusing client connection "
 640                               "from %s\n", target->obd_name,
 641                               libcfs_nid2str(req->rq_peer.nid));
 642                 GOTO(out, rc = -EAGAIN);
 643         }
 644
 645         /* Make sure the target isn't cleaned up while we're here. Yes,
 646            there's still a race between the above check and our incref here.
 647            Really, class_uuid2obd should take the ref. */
 648         targref = class_incref(target, __FUNCTION__, cfs_current());
 649
 650
 651         str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID);
 652         if (str == NULL) {
 653                 DEBUG_REQ(D_ERROR, req, "bad client UUID for connect");
 654                 GOTO(out, rc = -EINVAL);
 655         }
 656
 657         obd_str2uuid(&cluuid, str);
 658
 659         /* XXX extract a nettype and format accordingly */
 660         switch (sizeof(lnet_nid_t)) {
 661                 /* NB the casts only avoid compiler warnings */
 662         case 8:
 663                 snprintf(remote_uuid.uuid, sizeof remote_uuid,
 664                          "NET_"LPX64"_UUID", (__u64)req->rq_peer.nid);
 665                 break;
 666         case 4:
 667                 snprintf(remote_uuid.uuid, sizeof remote_uuid,
 668                          "NET_%x_UUID", (__u32)req->rq_peer.nid);
 669                 break;
 670         default:
 671                 LBUG();
 672         }
 673
 674         tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
 675         if (tmp == NULL)
 676                 GOTO(out, rc = -EPROTO);
 677
 678         conn = *tmp;
 679
 680         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 681         if (!data)
 682                 GOTO(out, rc = -EPROTO);
 683
 684         rc = req_capsule_server_pack(&req->rq_pill);
 685         if (rc)
 686                 GOTO(out, rc);
 687
 688         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
 689                 if (!data) {
 690                         DEBUG_REQ(D_WARNING, req, "Refusing old (unversioned) "
 691                                   "libclient connection attempt");
 692                         GOTO(out, rc = -EPROTO);
 693                 } else if (data->ocd_version < LUSTRE_VERSION_CODE -
 694                                                LUSTRE_VERSION_ALLOWED_OFFSET ||
 695                            data->ocd_version > LUSTRE_VERSION_CODE +
 696                                                LUSTRE_VERSION_ALLOWED_OFFSET) {
 697                         DEBUG_REQ(D_WARNING, req, "Refusing %s (%d.%d.%d.%d) "
 698                                   "libclient connection attempt",
 699                                   data->ocd_version < LUSTRE_VERSION_CODE ?
 700                                   "old" : "new",
 701                                   OBD_OCD_VERSION_MAJOR(data->ocd_version),
 702                                   OBD_OCD_VERSION_MINOR(data->ocd_version),
 703                                   OBD_OCD_VERSION_PATCH(data->ocd_version),
 704                                   OBD_OCD_VERSION_FIX(data->ocd_version));
 705                         data = req_capsule_server_sized_get(&req->rq_pill,
 706                                                         &RMF_CONNECT_DATA,
 707                                     offsetof(typeof(*data), ocd_version) +
 708                                              sizeof(data->ocd_version));
 709                         if (data) {
 710                                 data->ocd_connect_flags = OBD_CONNECT_VERSION;
 711                                 data->ocd_version = LUSTRE_VERSION_CODE;
 712                         }
 713                         GOTO(out, rc = -EPROTO);
 714                 }
 715         }
 716
 717         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL)
 718                 initial_conn = 1;
 719
 720         /* lctl gets a backstage, all-access pass. */
 721         if (obd_uuid_equals(&cluuid, &target->obd_uuid))
 722                 goto dont_check_exports;
 723
 724         spin_lock(&target->obd_dev_lock);
 725         export = lustre_hash_lookup(target->obd_uuid_hash, &cluuid);
 726
 727         if (export != NULL && export->exp_connecting) { /* bug 9635, et. al. */
 728                 CWARN("%s: exp %p already connecting\n",
 729                       export->exp_obd->obd_name, export);
 730                 class_export_put(export);
 731                 export = NULL;
 732                 rc = -EALREADY;
 733         } else if (export != NULL && export->exp_connection != NULL &&
 734                    req->rq_peer.nid != export->exp_connection->c_peer.nid) {
 735                 /* make darn sure this is coming from the same peer
 736                  * if the UUIDs matched */
 737                   CWARN("%s: cookie %s seen on new NID %s when "
 738                           "existing NID %s is already connected\n",
 739                         target->obd_name, cluuid.uuid,
 740                   libcfs_nid2str(req->rq_peer.nid),
 741                   libcfs_nid2str(export->exp_connection->c_peer.nid));
 742                   class_export_put(export);
 743                   export = NULL;
 744                   rc = -EALREADY;
 745         } else if (export != NULL) {
 746                 spin_lock(&export->exp_lock);
 747                 export->exp_connecting = 1;
 748                 spin_unlock(&export->exp_lock);
 749                 class_export_put(export);
 750                 spin_unlock(&target->obd_dev_lock);
 751                 LASSERT(export->exp_obd == target);
 752
 753                 rc = target_handle_reconnect(&conn, export, &cluuid, initial_conn);
 754         }
 755
 756         /* If we found an export, we already unlocked. */
 757         if (!export) {
 758                 spin_unlock(&target->obd_dev_lock);
 759                 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout);
 760         } else if (req->rq_export == NULL &&
 761                    atomic_read(&export->exp_rpc_count) > 0) {
 762                 CWARN("%s: refuse connection from %s/%s to 0x%p/%d\n",
 763                       target->obd_name, cluuid.uuid,
 764                       libcfs_nid2str(req->rq_peer.nid),
 765                       export, atomic_read(&export->exp_refcount));
 766                 GOTO(out, rc = -EBUSY);
 767         } else if (req->rq_export != NULL &&
 768                    (atomic_read(&export->exp_rpc_count) > 1)) {
 769                 CWARN("%s: refuse reconnection from %s@%s to 0x%p/%d\n",
 770                       target->obd_name, cluuid.uuid,
 771                       libcfs_nid2str(req->rq_peer.nid),
 772                       export, atomic_read(&export->exp_rpc_count));
 773                 GOTO(out, rc = -EBUSY);
 774         } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 &&
 775                    !initial_conn) {
 776                 CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; "
 777                        "cookies not random?\n", target->obd_name,
 778                        libcfs_nid2str(req->rq_peer.nid), cluuid.uuid);
 779                 GOTO(out, rc = -EALREADY);
 780         } else {
 781                 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
 782                 if (req->rq_export == NULL && initial_conn)
 783                        export->exp_last_request_time =
 784                                max(export->exp_last_request_time,
 785                                    (time_t)cfs_time_current_sec());
 786         }
 787
 788         if (rc < 0) {
 789                 GOTO(out, rc);
 790         }
 791
 792         CWARN("%s: connection from %s@%s %st"LPU64" exp %p cur %ld last %ld\n",
 793                target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
 794               target->obd_recovering ? "recovering/" : "", data->ocd_transno,
 795               export, (long)cfs_time_current_sec(),
 796               export ? (long)export->exp_last_request_time : 0);
 797
 798         /* Tell the client if we're in recovery. */
 799         if (target->obd_recovering) {
 800                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
 801                 /* If this is the first time a client connects,
 802                    reset the recovery timer */
 803                 if (rc == 0)
 804                         target_start_and_reset_recovery_timer(target, req,
 805                                                               !export);
 806         }
 807
 808         /* We want to handle EALREADY but *not* -EALREADY from
 809          * target_handle_reconnect(), return reconnection state in a flag */
 810         if (rc == EALREADY) {
 811                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
 812                 rc = 0;
 813         } else {
 814                 LASSERT(rc == 0);
 815         }
 816
 817         /* Tell the client if we support replayable requests */
 818         if (target->obd_replayable)
 819                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_REPLAYABLE);
 820         client_nid = &req->rq_peer.nid;
 821
 822         if (export == NULL) {
 823                 if (target->obd_recovering) {
 824                         cfs_time_t t;
 825
 826                         t = cfs_timer_deadline(&target->obd_recovery_timer);
 827                         t = cfs_time_sub(t, cfs_time_current());
 828                         CERROR("%s: denying connection for new client %s (%s): "
 829                                "%d clients in recovery for "CFS_TIME_T"s\n",
 830                                target->obd_name,
 831                                libcfs_nid2str(req->rq_peer.nid), cluuid.uuid,
 832                                target->obd_recoverable_clients,
 833                                cfs_duration_sec(t));
 834                         rc = -EBUSY;
 835                 } else {
 836 dont_check_exports:
 837                         rc = obd_connect(req->rq_svc_thread->t_env,
 838                                          &conn, target, &cluuid, data,
 839                                          client_nid);
 840                 }
 841         } else {
 842                 rc = obd_reconnect(req->rq_svc_thread->t_env,
 843                                    export, target, &cluuid, data, client_nid);
 844         }
 845         if (rc)
 846                 GOTO(out, rc);
 847         /* Return only the parts of obd_connect_data that we understand, so the
 848          * client knows that we don't understand the rest. */
 849         if (data) {
 850                  tmpdata = req_capsule_server_get(&req->rq_pill,
 851                                                   &RMF_CONNECT_DATA);
 852                   //data->ocd_connect_flags &= OBD_CONNECT_SUPPORTED;
 853                  *tmpdata = *data;
 854         }
 855
 856         /* If all else goes well, this is our RPC return code. */
 857         req->rq_status = 0;
 858
 859         lustre_msg_set_handle(req->rq_repmsg, &conn);
 860
 861         /* ownership of this export ref transfers to the request AFTER we
 862          * drop any previous reference the request had, but we don't want
 863          * that to go to zero before we get our new export reference. */
 864         export = class_conn2export(&conn);
 865         if (!export) {
 866                 DEBUG_REQ(D_ERROR, req, "Missing export!");
 867                 GOTO(out, rc = -ENODEV);
 868         }
 869
 870         /* If the client and the server are the same node, we will already
 871          * have an export that really points to the client's DLM export,
 872          * because we have a shared handles table.
 873          *
 874          * XXX this will go away when shaver stops sending the "connect" handle
 875          * in the real "remote handle" field of the request --phik 24 Apr 2003
 876          */
 877         if (req->rq_export != NULL)
 878                 class_export_put(req->rq_export);
 879
 880         req->rq_export = export;
 881
 882         spin_lock(&export->exp_lock);
 883         if (initial_conn) {
 884                 lustre_msg_set_conn_cnt(req->rq_repmsg, export->exp_conn_cnt + 1);
 885         } else if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
 886                 spin_unlock(&export->exp_lock);
 887                 CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n",
 888                        cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
 889                        export->exp_conn_cnt,
 890                        lustre_msg_get_conn_cnt(req->rq_reqmsg));
 891
 892                 GOTO(out, rc = -EALREADY);
 893         }
 894         export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
 895
 896         /* request from liblustre?  Don't evict it for not pinging. */
 897         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
 898                 export->exp_libclient = 1;
 899                 spin_unlock(&export->exp_lock);
 900
 901                 spin_lock(&target->obd_dev_lock);
 902                 list_del_init(&export->exp_obd_chain_timed);
 903                 spin_unlock(&target->obd_dev_lock);
 904         } else {
 905                 spin_unlock(&export->exp_lock);
 906         }
 907
 908         if (export->exp_connection != NULL)
 909                 ptlrpc_connection_put(export->exp_connection);
 910         export->exp_connection = ptlrpc_connection_get(req->rq_peer,
 911                                                        req->rq_self,
 912                                                        &remote_uuid);
 913         if (hlist_unhashed(&export->exp_nid_hash)) {
 914                 lustre_hash_add_unique(export->exp_obd->obd_nid_hash,
 915                                        &export->exp_connection->c_peer.nid,
 916                                        &export->exp_nid_hash);
 917         }
 918
 919         spin_lock_bh(&target->obd_processing_task_lock);
 920         if (target->obd_recovering && !export->exp_in_recovery) {
 921                 spin_lock(&export->exp_lock);
 922                 export->exp_in_recovery = 1;
 923                 export->exp_req_replay_needed = 1;
 924                 export->exp_lock_replay_needed = 1;
 925                 spin_unlock(&export->exp_lock);
 926                 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
 927                      && (data->ocd_transno == 0))
 928                         CWARN("Connect with zero transno!\n");
 929
 930                 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
 931                      && data->ocd_transno < target->obd_next_recovery_transno)
 932                         target->obd_next_recovery_transno = data->ocd_transno;
 933                 target->obd_connected_clients++;
 934                 /* each connected client is counted as recoverable */
 935                 target->obd_recoverable_clients++;
 936                 atomic_inc(&target->obd_req_replay_clients);
 937                 atomic_inc(&target->obd_lock_replay_clients);
 938                 if (target->obd_connected_clients ==
 939                     target->obd_max_recoverable_clients)
 940                         wake_up(&target->obd_next_transno_waitq);
 941         }
 942         spin_unlock_bh(&target->obd_processing_task_lock);
 943         tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
 944         conn = *tmp;
 945
 946         if (export->exp_imp_reverse != NULL) {
 947                 /* destroyed import can be still referenced in ctxt */
 948                 obd_set_info_async(export, sizeof(KEY_REVIMP_UPD),
 949                                    KEY_REVIMP_UPD, 0, NULL, NULL);
 950
 951                 /* in some recovery senarios, previous ctx init rpc handled
 952                  * in sptlrpc_target_export_check() might be used to install
 953                  * a reverse ctx in this reverse import, and later OBD_CONNECT
 954                  * using the same gss ctx could reach here and following new
 955                  * reverse import. note all reverse ctx in new/old import are
 956                  * actually based on the same gss ctx. so we invalidate ctx
 957                  * here before destroy import, otherwise flush old import will
 958                  * lead to remote reverse ctx be destroied, thus the reverse
 959                  * ctx of new import will lost its peer.
 960                  * there might be a better way to deal with this???
 961                  */
 962                 sptlrpc_import_inval_all_ctx(export->exp_imp_reverse);
 963
 964                 destroy_import(export->exp_imp_reverse);
 965         }
 966
 967         /* for the rest part, we return -ENOTCONN in case of errors
 968          * in order to let client initialize connection again.
 969          */
 970         revimp = export->exp_imp_reverse = class_new_import(target);
 971         if (!revimp) {
 972                 CERROR("fail to alloc new reverse import.\n");
 973                 GOTO(out, rc = -ENOTCONN);
 974         }
 975
 976         revimp->imp_connection = ptlrpc_connection_addref(export->exp_connection);
 977         revimp->imp_client = &export->exp_obd->obd_ldlm_client;
 978         revimp->imp_remote_handle = conn;
 979         revimp->imp_dlm_fake = 1;
 980         revimp->imp_state = LUSTRE_IMP_FULL;
 981
 982         /* unknown versions will be caught in
 983          * ptlrpc_handle_server_req_in->lustre_unpack_msg() */
 984         revimp->imp_msg_magic = req->rq_reqmsg->lm_magic;
 985
 986         if ((export->exp_connect_flags & OBD_CONNECT_AT) &&
 987             (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1))
 988                 revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT;
 989         else
 990                 revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
 991
 992         rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx,
 993                                       req->rq_flvr.sf_rpc);
 994         if (rc) {
 995                 CERROR("Failed to get sec for reverse import: %d\n", rc);
 996                 export->exp_imp_reverse = NULL;
 997                 class_destroy_import(revimp);
 998         }
 999
1000         class_import_put(revimp);
1001 out:
1002         if (export) {
1003                 spin_lock(&export->exp_lock);
1004                 export->exp_connecting = 0;
1005                 spin_unlock(&export->exp_lock);
1006         }
1007         if (targref)
1008                 class_decref(targref, __FUNCTION__, cfs_current());
1009         if (rc)
1010                 req->rq_status = rc;
1011         RETURN(rc);
1012 }
1013
1014 int target_handle_disconnect(struct ptlrpc_request *req)
1015 {
1016         int rc;
1017         ENTRY;
1018
1019         rc = req_capsule_server_pack(&req->rq_pill);
1020         if (rc)
1021                 RETURN(rc);
1022
1023         /* keep the rq_export around so we can send the reply */
1024         req->rq_status = obd_disconnect(class_export_get(req->rq_export));
1025
1026         RETURN(0);
1027 }
1028
1029 void target_destroy_export(struct obd_export *exp)
1030 {
1031         /* exports created from last_rcvd data, and "fake"
1032            exports created by lctl don't have an import */
1033         if (exp->exp_imp_reverse != NULL)
1034                 destroy_import(exp->exp_imp_reverse);
1035
1036         /* We cancel locks at disconnect time, but this will catch any locks
1037          * granted in a race with recovery-induced disconnect. */
1038         if (exp->exp_obd->obd_namespace != NULL)
1039                 ldlm_cancel_locks_for_export(exp);
1040 }
1041
1042 /*
1043  * Recovery functions
1044  */
1045
1046 struct ptlrpc_request *ptlrpc_clone_req( struct ptlrpc_request *orig_req)
1047 {
1048         struct ptlrpc_request *copy_req;
1049         struct lustre_msg *copy_reqmsg;
1050         struct ptlrpc_user_desc *udesc = NULL;
1051
1052         OBD_ALLOC_PTR(copy_req);
1053         if (!copy_req)
1054                 return NULL;
1055         OBD_ALLOC(copy_reqmsg, orig_req->rq_reqlen);
1056         if (!copy_reqmsg){
1057                 OBD_FREE_PTR(copy_req);
1058                 return NULL;
1059         }
1060
1061         if (orig_req->rq_user_desc) {
1062                 int ngroups = orig_req->rq_user_desc->pud_ngroups;
1063
1064                 OBD_ALLOC(udesc, sptlrpc_user_desc_size(ngroups));
1065                 if (!udesc) {
1066                         OBD_FREE(copy_reqmsg, orig_req->rq_reqlen);
1067                         OBD_FREE_PTR(copy_req);
1068                         return NULL;
1069                 }
1070                 memcpy(udesc, orig_req->rq_user_desc,
1071                        sptlrpc_user_desc_size(ngroups));
1072         }
1073
1074         *copy_req = *orig_req;
1075         memcpy(copy_reqmsg, orig_req->rq_reqmsg, orig_req->rq_reqlen);
1076         copy_req->rq_reqmsg = copy_reqmsg;
1077         copy_req->rq_user_desc = udesc;
1078
1079         class_export_rpc_get(copy_req->rq_export);
1080         CFS_INIT_LIST_HEAD(&copy_req->rq_list);
1081         CFS_INIT_LIST_HEAD(&copy_req->rq_replay_list);
1082         sptlrpc_svc_ctx_addref(copy_req);
1083
1084         if (copy_req->rq_reply_state) {
1085                 /* the copied req takes over the reply state */
1086                 orig_req->rq_reply_state = NULL;
1087                 /* to catch further access */
1088                 orig_req->rq_repmsg = NULL;
1089                 orig_req->rq_replen = 0;
1090         }
1091
1092         return copy_req;
1093 }
1094
1095 void ptlrpc_free_clone(struct ptlrpc_request *req)
1096 {
1097         LASSERT(list_empty(&req->rq_replay_list));
1098
1099         ptlrpc_req_drop_rs(req);
1100         sptlrpc_svc_ctx_decref(req);
1101         class_export_rpc_put(req->rq_export);
1102         list_del(&req->rq_list);
1103
1104         if (req->rq_user_desc) {
1105                 int ngroups = req->rq_user_desc->pud_ngroups;
1106                 OBD_FREE(req->rq_user_desc, sptlrpc_user_desc_size(ngroups));
1107         }
1108         OBD_FREE(req->rq_reqmsg, req->rq_reqlen);
1109         OBD_FREE_PTR(req);
1110 }
1111
1112 static int target_exp_enqueue_req_replay(struct ptlrpc_request *req)
1113 {
1114         __u64                  transno = lustre_msg_get_transno(req->rq_reqmsg);
1115         struct obd_export     *exp = req->rq_export;
1116         struct ptlrpc_request *reqiter;
1117         int                    dup = 0;
1118
1119         LASSERT(exp);
1120
1121         spin_lock(&exp->exp_lock);
1122         list_for_each_entry(reqiter, &exp->exp_req_replay_queue,
1123                             rq_replay_list) {
1124                 if (lustre_msg_get_transno(reqiter->rq_reqmsg) == transno) {
1125                         dup = 1;
1126                         break;
1127                 }
1128         }
1129
1130         if (dup) {
1131                 /* we expect it with RESENT and REPLAY flags */
1132                 if ((lustre_msg_get_flags(req->rq_reqmsg) &
1133                      (MSG_RESENT | MSG_REPLAY)) != (MSG_RESENT | MSG_REPLAY))
1134                         CERROR("invalid flags %x of resent replay\n",
1135                                lustre_msg_get_flags(req->rq_reqmsg));
1136         } else {
1137                 list_add_tail(&req->rq_replay_list, &exp->exp_req_replay_queue);
1138         }
1139
1140         spin_unlock(&exp->exp_lock);
1141         return dup;
1142 }
1143
1144 static void target_exp_dequeue_req_replay(struct ptlrpc_request *req)
1145 {
1146         LASSERT(!list_empty(&req->rq_replay_list));
1147         LASSERT(req->rq_export);
1148
1149         spin_lock(&req->rq_export->exp_lock);
1150         list_del_init(&req->rq_replay_list);
1151         spin_unlock(&req->rq_export->exp_lock);
1152 }
1153
1154 #ifdef __KERNEL__
1155 static void target_finish_recovery(struct obd_device *obd)
1156 {
1157         ENTRY;
1158         LCONSOLE_INFO("%s: sending delayed replies to recovered clients\n",
1159                       obd->obd_name);
1160
1161         ldlm_reprocess_all_ns(obd->obd_namespace);
1162
1163         /* when recovery finished, cleanup orphans on mds and ost */
1164         if (OBT(obd) && OBP(obd, postrecov)) {
1165                 int rc = OBP(obd, postrecov)(obd);
1166                 LCONSOLE_WARN("%s: recovery %s: rc %d\n", obd->obd_name,
1167                               rc < 0 ? "failed" : "complete", rc);
1168         }
1169
1170         obd->obd_recovery_end = cfs_time_current_sec();
1171         EXIT;
1172 }
1173
1174 static void abort_req_replay_queue(struct obd_device *obd)
1175 {
1176         struct ptlrpc_request *req, *n;
1177
1178         list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) {
1179                 DEBUG_REQ(D_WARNING, req, "aborted:");
1180                 req->rq_status = -ENOTCONN;
1181                 if (ptlrpc_error(req)) {
1182                         DEBUG_REQ(D_ERROR, req,
1183                                   "failed abort_req_reply; skipping");
1184                 }
1185                 target_exp_dequeue_req_replay(req);
1186                 ptlrpc_free_clone(req);
1187         }
1188 }
1189
1190 static void abort_lock_replay_queue(struct obd_device *obd)
1191 {
1192         struct ptlrpc_request *req, *n;
1193
1194         list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){
1195                 DEBUG_REQ(D_ERROR, req, "aborted:");
1196                 req->rq_status = -ENOTCONN;
1197                 if (ptlrpc_error(req)) {
1198                         DEBUG_REQ(D_ERROR, req,
1199                                   "failed abort_lock_reply; skipping");
1200                 }
1201                 ptlrpc_free_clone(req);
1202         }
1203 }
1204 #endif
1205
1206 /* Called from a cleanup function if the device is being cleaned up
1207    forcefully.  The exports should all have been disconnected already,
1208    the only thing left to do is
1209      - clear the recovery flags
1210      - cancel the timer
1211      - free queued requests and replies, but don't send replies
1212    Because the obd_stopping flag is set, no new requests should be received.
1213
1214 */
1215 void target_cleanup_recovery(struct obd_device *obd)
1216 {
1217         struct ptlrpc_request *req, *n;
1218         ENTRY;
1219
1220         LASSERT(obd->obd_stopping);
1221
1222         spin_lock_bh(&obd->obd_processing_task_lock);
1223         if (!obd->obd_recovering) {
1224                 spin_unlock_bh(&obd->obd_processing_task_lock);
1225                 EXIT;
1226                 return;
1227         }
1228         obd->obd_recovering = obd->obd_abort_recovery = 0;
1229         target_cancel_recovery_timer(obd);
1230         spin_unlock_bh(&obd->obd_processing_task_lock);
1231
1232         list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) {
1233                 LASSERT (req->rq_reply_state == 0);
1234                 target_exp_dequeue_req_replay(req);
1235                 ptlrpc_free_clone(req);
1236         }
1237         list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){
1238                 LASSERT (req->rq_reply_state == 0);
1239                 ptlrpc_free_clone(req);
1240         }
1241         list_for_each_entry_safe(req, n, &obd->obd_final_req_queue, rq_list) {
1242                 LASSERT (req->rq_reply_state == 0);
1243                 ptlrpc_free_clone(req);
1244         }
1245
1246         EXIT;
1247 }
1248
1249 /* obd_processing_task_lock should be held */
1250 void target_cancel_recovery_timer(struct obd_device *obd)
1251 {
1252         CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
1253         cfs_timer_disarm(&obd->obd_recovery_timer);
1254 }
1255
1256 /* extend = 1 means require at least "duration" seconds left in the timer,
1257    extend = 0 means set the total duration (start_recovery_timer) */
1258 static void reset_recovery_timer(struct obd_device *obd, int duration,
1259                                  int extend)
1260 {
1261         cfs_time_t now = cfs_time_current_sec();
1262         cfs_duration_t left;
1263
1264         spin_lock_bh(&obd->obd_processing_task_lock);
1265         if (!obd->obd_recovering || obd->obd_abort_recovery) {
1266                 spin_unlock_bh(&obd->obd_processing_task_lock);
1267                 return;
1268         }
1269
1270         left = cfs_time_sub(obd->obd_recovery_end, now);
1271
1272         if (extend && (duration > left))
1273                 obd->obd_recovery_timeout += duration - left;
1274         else if (!extend && (duration > obd->obd_recovery_timeout))
1275                 /* Track the client's largest expected replay time */
1276                 obd->obd_recovery_timeout = duration;
1277 #ifdef CRAY_XT3
1278         /*
1279          * If total recovery time already exceed the
1280          * obd_recovery_max_time, then CRAY XT3 will
1281          * abort the recovery
1282          */
1283         if(obd->obd_recovery_timeout > obd->obd_recovery_max_time)
1284                 obd->obd_recovery_timeout = obd->obd_recovery_max_time;
1285 #endif
1286         obd->obd_recovery_end = obd->obd_recovery_start +
1287                                 obd->obd_recovery_timeout;
1288         if (!cfs_timer_is_armed(&obd->obd_recovery_timer) ||
1289             cfs_time_before(now, obd->obd_recovery_end)) {
1290                 left = cfs_time_sub(obd->obd_recovery_end, now);
1291                 cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(left));
1292         }
1293         spin_unlock_bh(&obd->obd_processing_task_lock);
1294         CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n",
1295                obd->obd_name, (unsigned)left);
1296 }
1297
1298 static void check_and_start_recovery_timer(struct obd_device *obd)
1299 {
1300         spin_lock_bh(&obd->obd_processing_task_lock);
1301         if (cfs_timer_is_armed(&obd->obd_recovery_timer)) {
1302                 spin_unlock_bh(&obd->obd_processing_task_lock);
1303                 return;
1304         }
1305         CWARN("%s: starting recovery timer\n", obd->obd_name);
1306         obd->obd_recovery_start = cfs_time_current_sec();
1307         /* minimum */
1308         obd->obd_recovery_timeout = OBD_RECOVERY_FACTOR * obd_timeout;
1309         spin_unlock_bh(&obd->obd_processing_task_lock);
1310
1311         reset_recovery_timer(obd, obd->obd_recovery_timeout, 0);
1312 }
1313
1314 /* Reset the timer with each new client connection */
1315 /*
1316  * This timer is actually reconnect_timer, which is for making sure
1317  * the total recovery window is at least as big as my reconnect
1318  * attempt timing. So the initial recovery time_out will be set to
1319  * OBD_RECOVERY_FACTOR * obd_timeout. If the timeout coming
1320  * from client is bigger than this, then the recovery time_out will
1321  * be extend to make sure the client could be reconnected, in the
1322  * process, the timeout from the new client should be ignored.
1323  */
1324
1325 static void
1326 target_start_and_reset_recovery_timer(struct obd_device *obd,
1327                                       struct ptlrpc_request *req,
1328                                       int new_client)
1329 {
1330         int req_timeout = lustre_msg_get_timeout(req->rq_reqmsg);
1331
1332         /* teach server about old server's estimates */
1333         if (!new_client)
1334                 at_add(&req->rq_rqbd->rqbd_service->srv_at_estimate,
1335                        at_timeout2est(req_timeout));
1336
1337         check_and_start_recovery_timer(obd);
1338
1339         req_timeout *= OBD_RECOVERY_FACTOR;
1340         if (req_timeout > obd->obd_recovery_timeout && !new_client)
1341                 reset_recovery_timer(obd, req_timeout, 0);
1342 }
1343
1344 #ifdef __KERNEL__
1345 static int check_for_next_transno(struct obd_device *obd)
1346 {
1347         struct ptlrpc_request *req = NULL;
1348         int wake_up = 0, connected, completed, queue_len, max;
1349         __u64 next_transno, req_transno;
1350         ENTRY;
1351         spin_lock_bh(&obd->obd_processing_task_lock);
1352
1353         if (!list_empty(&obd->obd_req_replay_queue)) {
1354                 req = list_entry(obd->obd_req_replay_queue.next,
1355                                  struct ptlrpc_request, rq_list);
1356                 req_transno = lustre_msg_get_transno(req->rq_reqmsg);
1357         } else {
1358                 req_transno = 0;
1359         }
1360
1361         max = obd->obd_max_recoverable_clients;
1362         connected = obd->obd_connected_clients;
1363         completed = connected - obd->obd_recoverable_clients;
1364         queue_len = obd->obd_requests_queued_for_recovery;
1365         next_transno = obd->obd_next_recovery_transno;
1366
1367         CDEBUG(D_HA, "max: %d, connected: %d, completed: %d, queue_len: %d, "
1368                "req_transno: "LPU64", next_transno: "LPU64"\n",
1369                max, connected, completed, queue_len, req_transno, next_transno);
1370
1371         if (obd->obd_abort_recovery) {
1372                 CDEBUG(D_HA, "waking for aborted recovery\n");
1373                 wake_up = 1;
1374         } else if (atomic_read(&obd->obd_req_replay_clients) == 0) {
1375                 CDEBUG(D_HA, "waking for completed recovery\n");
1376                 wake_up = 1;
1377         } else if (req_transno == next_transno) {
1378                 CDEBUG(D_HA, "waking for next ("LPD64")\n", next_transno);
1379                 wake_up = 1;
1380         } else if (queue_len + completed == max) {
1381                 /* handle gaps occured due to lost reply. It is allowed gaps
1382                  * because all clients are connected and there will be resend
1383                  * for missed transaction */
1384                 LASSERTF(req_transno >= next_transno,
1385                          "req_transno: "LPU64", next_transno: "LPU64"\n",
1386                          req_transno, next_transno);
1387
1388                 CDEBUG(req_transno > obd->obd_last_committed ? D_ERROR : D_HA,
1389                        "waking for skipped transno (skip: "LPD64
1390                        ", ql: %d, comp: %d, conn: %d, next: "LPD64")\n",
1391                        next_transno, queue_len, completed, connected, req_transno);
1392                 obd->obd_next_recovery_transno = req_transno;
1393                 wake_up = 1;
1394         } else if (OBD_FAIL_CHECK(OBD_FAIL_MDS_RECOVERY_ACCEPTS_GAPS)) {
1395                 CDEBUG(D_HA, "accepting transno gaps is explicitly allowed"
1396                        " by fail_lock, waking up ("LPD64")\n", next_transno);
1397                 obd->obd_next_recovery_transno = req_transno;
1398                 wake_up = 1;
1399         } else if (queue_len == atomic_read(&obd->obd_req_replay_clients)) {
1400                 /* some clients haven't connected in time, but we can try
1401                  * to replay requests that demand on already committed ones
1402                  * also, we can replay first non-committed transation */
1403                 LASSERT(req_transno != 0);
1404                 if (req_transno == obd->obd_last_committed + 1) {
1405                         obd->obd_next_recovery_transno = req_transno;
1406                 } else if (req_transno > obd->obd_last_committed) {
1407                         /* can't continue recovery: have no needed transno */
1408                         obd->obd_abort_recovery = 1;
1409                         CDEBUG(D_ERROR, "abort due to missed clients. max: %d, "
1410                                "connected: %d, completed: %d, queue_len: %d, "
1411                                "req_transno: "LPU64", next_transno: "LPU64"\n",
1412                                max, connected, completed, queue_len,
1413                                req_transno, next_transno);
1414                 }
1415                 wake_up = 1;
1416         }
1417
1418         spin_unlock_bh(&obd->obd_processing_task_lock);
1419         return wake_up;
1420 }
1421
1422 static struct ptlrpc_request *target_next_replay_req(struct obd_device *obd)
1423 {
1424         struct l_wait_info lwi = { 0 };
1425         struct ptlrpc_request *req;
1426
1427         CDEBUG(D_HA, "Waiting for transno "LPD64"\n",
1428                obd->obd_next_recovery_transno);
1429         l_wait_event(obd->obd_next_transno_waitq,
1430                      check_for_next_transno(obd), &lwi);
1431
1432         spin_lock_bh(&obd->obd_processing_task_lock);
1433         if (obd->obd_abort_recovery) {
1434                 req = NULL;
1435         } else if (!list_empty(&obd->obd_req_replay_queue)) {
1436                 req = list_entry(obd->obd_req_replay_queue.next,
1437                                  struct ptlrpc_request, rq_list);
1438                 target_exp_dequeue_req_replay(req);
1439                 list_del_init(&req->rq_list);
1440                 obd->obd_requests_queued_for_recovery--;
1441         } else {
1442                 req = NULL;
1443         }
1444         spin_unlock_bh(&obd->obd_processing_task_lock);
1445         RETURN(req);
1446 }
1447
1448 static int check_for_next_lock(struct obd_device *obd)
1449 {
1450         struct ptlrpc_request *req = NULL;
1451         int wake_up = 0;
1452
1453         spin_lock_bh(&obd->obd_processing_task_lock);
1454         if (!list_empty(&obd->obd_lock_replay_queue)) {
1455                 req = list_entry(obd->obd_lock_replay_queue.next,
1456                                  struct ptlrpc_request, rq_list);
1457                 CDEBUG(D_HA, "waking for next lock\n");
1458                 wake_up = 1;
1459         } else if (atomic_read(&obd->obd_lock_replay_clients) == 0) {
1460                 CDEBUG(D_HA, "waking for completed lock replay\n");
1461                 wake_up = 1;
1462         } else if (obd->obd_abort_recovery) {
1463                 CDEBUG(D_HA, "waking for aborted recovery\n");
1464                 wake_up = 1;
1465         }
1466         spin_unlock_bh(&obd->obd_processing_task_lock);
1467
1468         return wake_up;
1469 }
1470
1471 static struct ptlrpc_request *target_next_replay_lock(struct obd_device *obd)
1472 {
1473         struct l_wait_info lwi = { 0 };
1474         struct ptlrpc_request *req;
1475
1476         CDEBUG(D_HA, "Waiting for lock\n");
1477         l_wait_event(obd->obd_next_transno_waitq,
1478                      check_for_next_lock(obd), &lwi);
1479
1480         spin_lock_bh(&obd->obd_processing_task_lock);
1481         if (obd->obd_abort_recovery) {
1482                 req = NULL;
1483         } else if (!list_empty(&obd->obd_lock_replay_queue)) {
1484                 req = list_entry(obd->obd_lock_replay_queue.next,
1485                                  struct ptlrpc_request, rq_list);
1486                 list_del_init(&req->rq_list);
1487         } else {
1488                 req = NULL;
1489         }
1490         spin_unlock_bh(&obd->obd_processing_task_lock);
1491         return req;
1492 }
1493
1494 static struct ptlrpc_request *target_next_final_ping(struct obd_device *obd)
1495 {
1496         struct ptlrpc_request *req;
1497
1498         spin_lock_bh(&obd->obd_processing_task_lock);
1499         if (!list_empty(&obd->obd_final_req_queue)) {
1500                 req = list_entry(obd->obd_final_req_queue.next,
1501                                  struct ptlrpc_request, rq_list);
1502                 list_del_init(&req->rq_list);
1503         } else {
1504                 req = NULL;
1505         }
1506         spin_unlock_bh(&obd->obd_processing_task_lock);
1507         return req;
1508 }
1509
1510 static inline int req_replay_done(struct obd_export *exp)
1511 {
1512         return (exp->exp_req_replay_needed == 0);
1513 }
1514
1515 static inline int lock_replay_done(struct obd_export *exp)
1516 {
1517         return (exp->exp_lock_replay_needed == 0);
1518 }
1519
1520 static inline int connect_done(struct obd_export *exp)
1521 {
1522         return (exp->exp_in_recovery != 0);
1523 }
1524
1525 static int check_for_clients(struct obd_device *obd)
1526 {
1527         if (obd->obd_abort_recovery)
1528                 return 1;
1529         LASSERT(obd->obd_connected_clients <= obd->obd_max_recoverable_clients);
1530         if (obd->obd_no_conn == 0 &&
1531             obd->obd_connected_clients == obd->obd_max_recoverable_clients)
1532                 return 1;
1533         return 0;
1534 }
1535
1536 static int handle_recovery_req(struct ptlrpc_thread *thread,
1537                                struct ptlrpc_request *req,
1538                                svc_handler_t handler)
1539 {
1540         int rc;
1541         ENTRY;
1542
1543         rc = lu_context_init(&req->rq_session, LCT_SESSION);
1544         if (rc) {
1545                 CERROR("Failure to initialize session: %d\n", rc);
1546                 return rc;
1547         }
1548         req->rq_session.lc_thread = thread;
1549         lu_context_enter(&req->rq_session);
1550         req->rq_svc_thread = thread;
1551         req->rq_svc_thread->t_env->le_ses = &req->rq_session;
1552
1553         /* thread context */
1554         lu_context_enter(&thread->t_env->le_ctx);
1555         (void)handler(req);
1556         lu_context_exit(&thread->t_env->le_ctx);
1557
1558         lu_context_exit(&req->rq_session);
1559         lu_context_fini(&req->rq_session);
1560         /* don't reset timer for final stage */
1561         if (!req_replay_done(req->rq_export) ||
1562             !lock_replay_done(req->rq_export))
1563                 reset_recovery_timer(class_exp2obd(req->rq_export),
1564                        OBD_RECOVERY_FACTOR * AT_OFF ? obd_timeout :
1565                        at_get(&req->rq_rqbd->rqbd_service->srv_at_estimate), 1);
1566         ptlrpc_free_clone(req);
1567         RETURN(0);
1568 }
1569
1570 static void resume_recovery_timer(struct obd_device *obd)
1571 {
1572         /* to be safe, make it at least OBD_RECOVERY_FACTOR * obd_timeout */
1573         reset_recovery_timer(obd, OBD_RECOVERY_FACTOR * obd_timeout, 1);
1574 }
1575
1576 static int target_recovery_thread(void *arg)
1577 {
1578         struct obd_device *obd = arg;
1579         struct ptlrpc_request *req;
1580         struct target_recovery_data *trd = &obd->obd_recovery_data;
1581         struct l_wait_info lwi = { 0 };
1582         unsigned long delta;
1583         unsigned long flags;
1584         struct lu_env env;
1585         struct ptlrpc_thread fake_svc_thread, *thread = &fake_svc_thread;
1586         int rc = 0;
1587         ENTRY;
1588
1589         cfs_daemonize("tgt_recov");
1590
1591         SIGNAL_MASK_LOCK(current, flags);
1592         sigfillset(&current->blocked);
1593         RECALC_SIGPENDING;
1594         SIGNAL_MASK_UNLOCK(current, flags);
1595
1596         rc = lu_context_init(&env.le_ctx, LCT_MD_THREAD);
1597         if (rc)
1598                 RETURN(rc);
1599
1600         thread->t_env = &env;
1601         env.le_ctx.lc_thread = thread;
1602
1603         CERROR("%s: started recovery thread pid %d\n", obd->obd_name,
1604                current->pid);
1605         trd->trd_processing_task = current->pid;
1606
1607         obd->obd_recovering = 1;
1608         complete(&trd->trd_starting);
1609
1610         /* first of all, we have to know the first transno to replay */
1611         obd->obd_abort_recovery = 0;
1612         l_wait_event(obd->obd_next_transno_waitq,
1613                      check_for_clients(obd), &lwi);
1614
1615         spin_lock_bh(&obd->obd_processing_task_lock);
1616         target_cancel_recovery_timer(obd);
1617         spin_unlock_bh(&obd->obd_processing_task_lock);
1618
1619         /* If some clients haven't connected in time, evict them */
1620         if (obd->obd_abort_recovery) {
1621                 CWARN("Some clients haven't connect in time (%d/%d),"
1622                        "evict them\n", obd->obd_connected_clients,
1623                        obd->obd_max_recoverable_clients);
1624                 obd->obd_abort_recovery = obd->obd_stopping;
1625                 class_disconnect_stale_exports(obd, connect_done);
1626         }
1627         /* next stage: replay requests */
1628         delta = jiffies;
1629         obd->obd_req_replaying = 1;
1630         CDEBUG(D_INFO, "1: request replay stage - %d clients from t"LPU64"\n",
1631               atomic_read(&obd->obd_req_replay_clients),
1632               obd->obd_next_recovery_transno);
1633         resume_recovery_timer(obd);
1634         while ((req = target_next_replay_req(obd))) {
1635                 LASSERT(trd->trd_processing_task == current->pid);
1636                 DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s",
1637                           lustre_msg_get_transno(req->rq_reqmsg),
1638                           libcfs_nid2str(req->rq_peer.nid));
1639                 handle_recovery_req(thread, req,
1640                                     trd->trd_recovery_handler);
1641                 obd->obd_replayed_requests++;
1642                 spin_lock_bh(&obd->obd_processing_task_lock);
1643                 obd->obd_next_recovery_transno++;
1644                 spin_unlock_bh(&obd->obd_processing_task_lock);
1645         }
1646
1647         spin_lock_bh(&obd->obd_processing_task_lock);
1648         target_cancel_recovery_timer(obd);
1649         spin_unlock_bh(&obd->obd_processing_task_lock);
1650
1651         /* If some clients haven't replayed requests in time, evict them */
1652         if (obd->obd_abort_recovery) {
1653                 CDEBUG(D_ERROR, "req replay timed out, aborting ...\n");
1654                 obd->obd_abort_recovery = obd->obd_stopping;
1655                 class_disconnect_stale_exports(obd, req_replay_done);
1656                 abort_req_replay_queue(obd);
1657         }
1658
1659         /* The second stage: replay locks */
1660         CDEBUG(D_INFO, "2: lock replay stage - %d clients\n",
1661                atomic_read(&obd->obd_lock_replay_clients));
1662         resume_recovery_timer(obd);
1663         while ((req = target_next_replay_lock(obd))) {
1664                 LASSERT(trd->trd_processing_task == current->pid);
1665                 DEBUG_REQ(D_HA|D_WARNING, req, "processing lock from %s: ",
1666                           libcfs_nid2str(req->rq_peer.nid));
1667                 handle_recovery_req(thread, req,
1668                                     trd->trd_recovery_handler);
1669                 obd->obd_replayed_locks++;
1670         }
1671
1672         spin_lock_bh(&obd->obd_processing_task_lock);
1673         target_cancel_recovery_timer(obd);
1674         spin_unlock_bh(&obd->obd_processing_task_lock);
1675         /* If some clients haven't replayed requests in time, evict them */
1676         if (obd->obd_abort_recovery) {
1677                 int stale;
1678                 CERROR("lock replay timed out, aborting ...\n");
1679                 obd->obd_abort_recovery = obd->obd_stopping;
1680                 stale = class_disconnect_stale_exports(obd, lock_replay_done);
1681                 abort_lock_replay_queue(obd);
1682         }
1683
1684         /* We drop recoverying flag to forward all new requests
1685          * to regular mds_handle() since now */
1686         spin_lock_bh(&obd->obd_processing_task_lock);
1687         obd->obd_recovering = obd->obd_abort_recovery = 0;
1688         spin_unlock_bh(&obd->obd_processing_task_lock);
1689         /* The third stage: reply on final pings */
1690         CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n");
1691         while ((req = target_next_final_ping(obd))) {
1692                 LASSERT(trd->trd_processing_task == current->pid);
1693                 DEBUG_REQ(D_HA, req, "processing final ping from %s: ",
1694                           libcfs_nid2str(req->rq_peer.nid));
1695                 handle_recovery_req(thread, req,
1696                                     trd->trd_recovery_handler);
1697         }
1698
1699         delta = (jiffies - delta) / HZ;
1700         CDEBUG(D_INFO,"4: recovery completed in %lus - %d/%d reqs/locks\n",
1701               delta, obd->obd_replayed_requests, obd->obd_replayed_locks);
1702         LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0);
1703         LASSERT(atomic_read(&obd->obd_lock_replay_clients) == 0);
1704         if (delta > obd_timeout * 2) {
1705                 CWARN("too long recovery - read logs\n");
1706                 libcfs_debug_dumplog();
1707         }
1708
1709         target_finish_recovery(obd);
1710
1711         lu_context_fini(&env.le_ctx);
1712         trd->trd_processing_task = 0;
1713         complete(&trd->trd_finishing);
1714         RETURN(rc);
1715 }
1716
1717 int target_start_recovery_thread(struct obd_device *obd, svc_handler_t handler)
1718 {
1719         int rc = 0;
1720         struct target_recovery_data *trd = &obd->obd_recovery_data;
1721
1722         memset(trd, 0, sizeof(*trd));
1723         init_completion(&trd->trd_starting);
1724         init_completion(&trd->trd_finishing);
1725         trd->trd_recovery_handler = handler;
1726
1727         if (kernel_thread(target_recovery_thread, obd, 0) > 0) {
1728                 wait_for_completion(&trd->trd_starting);
1729                 LASSERT(obd->obd_recovering != 0);
1730         } else
1731                 rc = -ECHILD;
1732
1733         return rc;
1734 }
1735
1736 void target_stop_recovery_thread(struct obd_device *obd)
1737 {
1738         spin_lock_bh(&obd->obd_processing_task_lock);
1739         if (obd->obd_recovery_data.trd_processing_task > 0) {
1740                 struct target_recovery_data *trd = &obd->obd_recovery_data;
1741                 CERROR("%s: Aborting recovery\n", obd->obd_name);
1742                 obd->obd_abort_recovery = 1;
1743                 wake_up(&obd->obd_next_transno_waitq);
1744                 spin_unlock_bh(&obd->obd_processing_task_lock);
1745                 wait_for_completion(&trd->trd_finishing);
1746         } else {
1747                 spin_unlock_bh(&obd->obd_processing_task_lock);
1748         }
1749 }
1750
1751 void target_recovery_fini(struct obd_device *obd)
1752 {
1753         class_disconnect_exports(obd);
1754         target_stop_recovery_thread(obd);
1755         target_cleanup_recovery(obd);
1756 }
1757 EXPORT_SYMBOL(target_recovery_fini);
1758
1759 static void target_recovery_expired(unsigned long castmeharder)
1760 {
1761         struct obd_device *obd = (struct obd_device *)castmeharder;
1762         LCONSOLE_WARN("%s: recovery timed out; %d clients never reconnected "
1763                       "after %lds (%d clients did)\n",
1764                       obd->obd_name, obd->obd_recoverable_clients,
1765                       cfs_time_current_sec()- obd->obd_recovery_start,
1766                       obd->obd_connected_clients);
1767         spin_lock_bh(&obd->obd_processing_task_lock);
1768         if (obd->obd_recovering)
1769                 obd->obd_abort_recovery = 1;
1770         cfs_waitq_signal(&obd->obd_next_transno_waitq);
1771         spin_unlock_bh(&obd->obd_processing_task_lock);
1772 }
1773
1774 void target_recovery_init(struct obd_device *obd, svc_handler_t handler)
1775 {
1776         if (obd->obd_max_recoverable_clients == 0)
1777                 return;
1778
1779         CWARN("RECOVERY: service %s, %d recoverable clients, "
1780               "last_transno "LPU64"\n", obd->obd_name,
1781               obd->obd_max_recoverable_clients, obd->obd_last_committed);
1782         obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
1783         obd->obd_recovery_start = 0;
1784         obd->obd_recovery_end = 0;
1785         obd->obd_recovery_timeout = OBD_RECOVERY_FACTOR * obd_timeout;
1786         /* bz13079: this should be set to desired value for ost but not for mds */
1787         obd->obd_recovery_max_time = OBD_RECOVERY_MAX_TIME;
1788         cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd);
1789         target_start_recovery_thread(obd, handler);
1790 }
1791 EXPORT_SYMBOL(target_recovery_init);
1792
1793 #endif
1794
1795 int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req)
1796 {
1797         struct obd_export *exp = req->rq_export;
1798         LASSERT(exp != NULL);
1799         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
1800                 /* client declares he's ready to replay locks */
1801                 spin_lock_bh(&obd->obd_processing_task_lock);
1802                 if (exp->exp_req_replay_needed) {
1803                         LASSERT(atomic_read(&obd->obd_req_replay_clients) > 0);
1804                         spin_lock(&exp->exp_lock);
1805                         exp->exp_req_replay_needed = 0;
1806                         spin_unlock(&exp->exp_lock);
1807                         atomic_dec(&obd->obd_req_replay_clients);
1808                         LASSERT(obd->obd_recoverable_clients > 0);
1809                         obd->obd_recoverable_clients--;
1810                         if (atomic_read(&obd->obd_req_replay_clients) == 0)
1811                                 CDEBUG(D_HA, "all clients have replayed reqs\n");
1812                         wake_up(&obd->obd_next_transno_waitq);
1813                 }
1814                 spin_unlock_bh(&obd->obd_processing_task_lock);
1815         }
1816         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
1817                 /* client declares he's ready to complete recovery
1818                  * so, we put the request on th final queue */
1819                 spin_lock_bh(&obd->obd_processing_task_lock);
1820                 if (exp->exp_lock_replay_needed) {
1821                         LASSERT(atomic_read(&obd->obd_lock_replay_clients) > 0);
1822                         spin_lock(&exp->exp_lock);
1823                         exp->exp_lock_replay_needed = 0;
1824                         spin_unlock(&exp->exp_lock);
1825                         atomic_dec(&obd->obd_lock_replay_clients);
1826                         if (atomic_read(&obd->obd_lock_replay_clients) == 0)
1827                                 CDEBUG(D_HA, "all clients have replayed locks\n");
1828                         wake_up(&obd->obd_next_transno_waitq);
1829                 }
1830                 spin_unlock_bh(&obd->obd_processing_task_lock);
1831         }
1832
1833         return 0;
1834 }
1835
1836 int target_queue_recovery_request(struct ptlrpc_request *req,
1837                                   struct obd_device *obd)
1838 {
1839         struct list_head *tmp;
1840         int inserted = 0;
1841         __u64 transno = lustre_msg_get_transno(req->rq_reqmsg);
1842
1843         ENTRY;
1844
1845         if (obd->obd_recovery_data.trd_processing_task == cfs_curproc_pid()) {
1846                 /* Processing the queue right now, don't re-add. */
1847                 RETURN(1);
1848         }
1849
1850         target_process_req_flags(obd, req);
1851
1852         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
1853                 /* client declares he's ready to complete recovery
1854                  * so, we put the request on th final queue */
1855                 req = ptlrpc_clone_req(req);
1856                 if (req == NULL)
1857                         RETURN(-ENOMEM);
1858                 DEBUG_REQ(D_HA, req, "queue final req");
1859                 spin_lock_bh(&obd->obd_processing_task_lock);
1860                 if (obd->obd_recovering)
1861                         list_add_tail(&req->rq_list, &obd->obd_final_req_queue);
1862                 else {
1863                         spin_unlock_bh(&obd->obd_processing_task_lock);
1864                         ptlrpc_free_clone(req);
1865                         if (obd->obd_stopping) {
1866                                 RETURN(-ENOTCONN);
1867                         } else {
1868                                 RETURN(1);
1869                         }
1870                 }
1871                 spin_unlock_bh(&obd->obd_processing_task_lock);
1872                 RETURN(0);
1873         }
1874         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
1875                 /* client declares he's ready to replay locks */
1876                 req = ptlrpc_clone_req(req);
1877                 if (req == NULL)
1878                         RETURN(-ENOMEM);
1879                 DEBUG_REQ(D_HA, req, "queue lock replay req");
1880                 spin_lock_bh(&obd->obd_processing_task_lock);
1881                 LASSERT(obd->obd_recovering);
1882                 /* usually due to recovery abort */
1883                 if (!req->rq_export->exp_in_recovery) {
1884                         spin_unlock_bh(&obd->obd_processing_task_lock);
1885                         ptlrpc_free_clone(req);
1886                         RETURN(-ENOTCONN);
1887                 }
1888                 LASSERT(req->rq_export->exp_lock_replay_needed);
1889                 list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue);
1890                 spin_unlock_bh(&obd->obd_processing_task_lock);
1891                 wake_up(&obd->obd_next_transno_waitq);
1892                 RETURN(0);
1893         }
1894
1895         /* CAVEAT EMPTOR: The incoming request message has been swabbed
1896          * (i.e. buflens etc are in my own byte order), but type-dependent
1897          * buffers (eg mds_body, ost_body etc) have NOT been swabbed. */
1898
1899         if (!transno) {
1900                 CFS_INIT_LIST_HEAD(&req->rq_list);
1901                 DEBUG_REQ(D_HA, req, "not queueing");
1902                 RETURN(1);
1903         }
1904
1905         spin_lock_bh(&obd->obd_processing_task_lock);
1906
1907         /* If we're processing the queue, we want don't want to queue this
1908          * message.
1909          *
1910          * Also, if this request has a transno less than the one we're waiting
1911          * for, we should process it now.  It could (and currently always will)
1912          * be an open request for a descriptor that was opened some time ago.
1913          *
1914          * Also, a resent, replayed request that has already been
1915          * handled will pass through here and be processed immediately.
1916          */
1917         CWARN("Next recovery transno: "LPU64", current: "LPU64", replaying: %i\n",
1918               obd->obd_next_recovery_transno, transno, obd->obd_req_replaying);
1919         if (transno < obd->obd_next_recovery_transno && obd->obd_req_replaying) {
1920                 /* Processing the queue right now, don't re-add. */
1921                 LASSERT(list_empty(&req->rq_list));
1922                 spin_unlock_bh(&obd->obd_processing_task_lock);
1923                 RETURN(1);
1924         }
1925         spin_unlock_bh(&obd->obd_processing_task_lock);
1926
1927         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_REPLAY_DROP))
1928                 RETURN(0);
1929
1930         req = ptlrpc_clone_req(req);
1931         if (req == NULL)
1932                 RETURN(-ENOMEM);
1933
1934         spin_lock_bh(&obd->obd_processing_task_lock);
1935         LASSERT(obd->obd_recovering);
1936         if (!req->rq_export->exp_in_recovery) {
1937                 spin_unlock_bh(&obd->obd_processing_task_lock);
1938                 ptlrpc_free_clone(req);
1939                 RETURN(-ENOTCONN);
1940         }
1941         LASSERT(req->rq_export->exp_req_replay_needed);
1942
1943         if (target_exp_enqueue_req_replay(req)) {
1944                 spin_unlock_bh(&obd->obd_processing_task_lock);
1945                 DEBUG_REQ(D_ERROR, req, "dropping resent queued req");
1946                 ptlrpc_free_clone(req);
1947                 RETURN(0);
1948         }
1949
1950         /* XXX O(n^2) */
1951         list_for_each(tmp, &obd->obd_req_replay_queue) {
1952                 struct ptlrpc_request *reqiter =
1953                         list_entry(tmp, struct ptlrpc_request, rq_list);
1954
1955                 if (lustre_msg_get_transno(reqiter->rq_reqmsg) > transno) {
1956                         list_add_tail(&req->rq_list, &reqiter->rq_list);
1957                         inserted = 1;
1958                         break;
1959                 }
1960
1961                 if (unlikely(lustre_msg_get_transno(reqiter->rq_reqmsg) ==
1962                              transno)) {
1963                         DEBUG_REQ(D_ERROR, req, "dropping replay: transno "
1964                                   "has been claimed by another client");
1965                         spin_unlock_bh(&obd->obd_processing_task_lock);
1966                         target_exp_dequeue_req_replay(req);
1967                         ptlrpc_free_clone(req);
1968                         RETURN(0);
1969                 }
1970         }
1971
1972         if (!inserted)
1973                 list_add_tail(&req->rq_list, &obd->obd_req_replay_queue);
1974
1975         obd->obd_requests_queued_for_recovery++;
1976         wake_up(&obd->obd_next_transno_waitq);
1977         spin_unlock_bh(&obd->obd_processing_task_lock);
1978         RETURN(0);
1979 }
1980
1981 struct obd_device * target_req2obd(struct ptlrpc_request *req)
1982 {
1983         return req->rq_export->exp_obd;
1984 }
1985
1986 static inline struct ldlm_pool *ldlm_exp2pl(struct obd_export *exp)
1987 {
1988         LASSERT(exp != NULL);
1989         return &exp->exp_obd->obd_namespace->ns_pool;
1990 }
1991
1992 /**
1993  * Packs current SLV and Limit into \a req.
1994  */
1995 int target_pack_pool_reply(struct ptlrpc_request *req)
1996 {
1997         struct obd_device *obd;
1998         ENTRY;
1999
2000         /*
2001          * Check that we still have all structures alive as this may
2002          * be some late rpc in shutdown time.
2003          */
2004         if (unlikely(!req->rq_export || !req->rq_export->exp_obd ||
2005                      !exp_connect_lru_resize(req->rq_export))) {
2006                 lustre_msg_set_slv(req->rq_repmsg, 0);
2007                 lustre_msg_set_limit(req->rq_repmsg, 0);
2008                 RETURN(0);
2009         }
2010
2011         /*
2012          * OBD is alive here as export is alive, which we checked above.
2013          */
2014         obd = req->rq_export->exp_obd;
2015
2016         read_lock(&obd->obd_pool_lock);
2017         lustre_msg_set_slv(req->rq_repmsg, obd->obd_pool_slv);
2018         lustre_msg_set_limit(req->rq_repmsg, obd->obd_pool_limit);
2019         read_unlock(&obd->obd_pool_lock);
2020
2021         RETURN(0);
2022 }
2023
2024 int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id)
2025 {
2026         if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) {
2027                 DEBUG_REQ(D_ERROR, req, "dropping reply");
2028                 return (-ECOMM);
2029         }
2030
2031         if (unlikely(rc)) {
2032                 DEBUG_REQ(D_ERROR, req, "processing error (%d)", rc);
2033                 req->rq_status = rc;
2034                 return (ptlrpc_send_error(req, 1));
2035         } else {
2036                 DEBUG_REQ(D_NET, req, "sending reply");
2037         }
2038
2039         return (ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT));
2040 }
2041
2042 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
2043 {
2044         int                        netrc;
2045         struct ptlrpc_reply_state *rs;
2046         struct obd_device         *obd;
2047         struct obd_export         *exp;
2048         struct ptlrpc_service     *svc;
2049         ENTRY;
2050
2051         if (req->rq_no_reply) {
2052                 EXIT;
2053                 return;
2054         }
2055
2056         svc = req->rq_rqbd->rqbd_service;
2057         rs = req->rq_reply_state;
2058         if (rs == NULL || !rs->rs_difficult) {
2059                 /* no notifiers */
2060                 target_send_reply_msg (req, rc, fail_id);
2061                 EXIT;
2062                 return;
2063         }
2064
2065         /* must be an export if locks saved */
2066         LASSERT (req->rq_export != NULL);
2067         /* req/reply consistent */
2068         LASSERT (rs->rs_service == svc);
2069
2070         /* "fresh" reply */
2071         LASSERT (!rs->rs_scheduled);
2072         LASSERT (!rs->rs_scheduled_ever);
2073         LASSERT (!rs->rs_handled);
2074         LASSERT (!rs->rs_on_net);
2075         LASSERT (rs->rs_export == NULL);
2076         LASSERT (list_empty(&rs->rs_obd_list));
2077         LASSERT (list_empty(&rs->rs_exp_list));
2078
2079         exp = class_export_get (req->rq_export);
2080         obd = exp->exp_obd;
2081
2082         /* disable reply scheduling onto srv_reply_queue while I'm setting up */
2083         rs->rs_scheduled = 1;
2084         rs->rs_on_net    = 1;
2085         rs->rs_xid       = req->rq_xid;
2086         rs->rs_transno   = req->rq_transno;
2087         rs->rs_export    = exp;
2088
2089         spin_lock(&obd->obd_uncommitted_replies_lock);
2090
2091         CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n",
2092                rs->rs_transno, obd->obd_last_committed);
2093         if (rs->rs_transno > obd->obd_last_committed) {
2094                 /* not committed already */
2095                 list_add_tail (&rs->rs_obd_list,
2096                                &obd->obd_uncommitted_replies);
2097         }
2098
2099         spin_unlock (&obd->obd_uncommitted_replies_lock);
2100         spin_lock (&exp->exp_lock);
2101
2102         list_add_tail (&rs->rs_exp_list, &exp->exp_outstanding_replies);
2103
2104         spin_unlock(&exp->exp_lock);
2105
2106         netrc = target_send_reply_msg (req, rc, fail_id);
2107
2108         spin_lock(&svc->srv_lock);
2109
2110         svc->srv_n_difficult_replies++;
2111
2112         if (netrc != 0) {
2113                 /* error sending: reply is off the net.  Also we need +1
2114                  * reply ref until ptlrpc_server_handle_reply() is done
2115                  * with the reply state (if the send was successful, there
2116                  * would have been +1 ref for the net, which
2117                  * reply_out_callback leaves alone) */
2118                 rs->rs_on_net = 0;
2119                 ptlrpc_rs_addref(rs);
2120                 atomic_inc (&svc->srv_outstanding_replies);
2121         }
2122
2123         if (rs->rs_transno <= obd->obd_last_committed ||
2124             (!rs->rs_on_net && !rs->rs_no_ack) ||
2125              list_empty(&rs->rs_exp_list) ||     /* completed already */
2126              list_empty(&rs->rs_obd_list)) {
2127                 CDEBUG(D_HA, "Schedule reply immediately\n");
2128                 list_add_tail (&rs->rs_list, &svc->srv_reply_queue);
2129                 cfs_waitq_signal (&svc->srv_waitq);
2130         } else {
2131                 list_add (&rs->rs_list, &svc->srv_active_replies);
2132                 rs->rs_scheduled = 0;           /* allow notifier to schedule */
2133         }
2134
2135         spin_unlock(&svc->srv_lock);
2136         EXIT;
2137 }
2138
2139 int target_handle_ping(struct ptlrpc_request *req)
2140 {
2141         obd_ping(req->rq_export);
2142         return req_capsule_server_pack(&req->rq_pill);
2143 }
2144
2145 void target_committed_to_req(struct ptlrpc_request *req)
2146 {
2147         struct obd_device *obd;
2148
2149         if (req == NULL || req->rq_export == NULL)
2150                 return;
2151
2152         obd = req->rq_export->exp_obd;
2153         if (obd == NULL)
2154                 return;
2155
2156         if (!obd->obd_no_transno && req->rq_repmsg != NULL)
2157                 lustre_msg_set_last_committed(req->rq_repmsg,
2158                                               obd->obd_last_committed);
2159         else
2160                 DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/"
2161                           "%d)", obd->obd_no_transno, req->rq_repmsg == NULL);
2162
2163         CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n",
2164                obd->obd_last_committed, req->rq_transno, req->rq_xid);
2165 }
2166
2167 EXPORT_SYMBOL(target_committed_to_req);
2168
2169 #ifdef HAVE_QUOTA_SUPPORT
2170 int target_handle_qc_callback(struct ptlrpc_request *req)
2171 {
2172         struct obd_quotactl *oqctl;
2173         struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
2174
2175         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2176         if (oqctl == NULL)
2177                 RETURN(-EPROTO);
2178
2179         cli->cl_qchk_stat = oqctl->qc_stat;
2180
2181         return 0;
2182 }
2183
2184 int target_handle_dqacq_callback(struct ptlrpc_request *req)
2185 {
2186 #ifdef __KERNEL__
2187         struct obd_device *obd = req->rq_export->exp_obd;
2188         struct obd_device *master_obd;
2189         struct lustre_quota_ctxt *qctxt;
2190         struct qunit_data *qdata;
2191         void* rep;
2192         struct qunit_data_old *qdata_old;
2193         int rc = 0;
2194         ENTRY;
2195
2196         rc = req_capsule_server_pack(&req->rq_pill);
2197         if (rc) {
2198                 CERROR("packing reply failed!: rc = %d\n", rc);
2199                 RETURN(rc);
2200         }
2201
2202         LASSERT(req->rq_export);
2203
2204         /* fixed for bug10707 */
2205         if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) &&
2206             !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) {
2207                 CDEBUG(D_QUOTA, "qd_count is 64bit!\n");
2208                 rep = req_capsule_server_get(&req->rq_pill,
2209                                              &RMF_QUNIT_DATA);
2210                 LASSERT(rep);
2211                 qdata = req_capsule_client_swab_get(&req->rq_pill,
2212                                                     &RMF_QUNIT_DATA,
2213                                           (void*)lustre_swab_qdata);
2214         } else {
2215                 CDEBUG(D_QUOTA, "qd_count is 32bit!\n");
2216                 rep = req_capsule_server_get(&req->rq_pill, &RMF_QUNIT_DATA);
2217                 LASSERT(rep);
2218                 qdata_old = req_capsule_client_swab_get(&req->rq_pill,
2219                                                         &RMF_QUNIT_DATA,
2220                                            (void*)lustre_swab_qdata_old);
2221                 qdata = lustre_quota_old_to_new(qdata_old);
2222         }
2223
2224         if (qdata == NULL)
2225                 RETURN(-EPROTO);
2226
2227         /* we use the observer */
2228         LASSERT(obd->obd_observer && obd->obd_observer->obd_observer);
2229         master_obd = obd->obd_observer->obd_observer;
2230         qctxt = &master_obd->u.obt.obt_qctxt;
2231
2232         LASSERT(qctxt->lqc_handler);
2233         rc = qctxt->lqc_handler(master_obd, qdata,
2234                                 lustre_msg_get_opc(req->rq_reqmsg));
2235         if (rc && rc != -EDQUOT)
2236                 CDEBUG(rc == -EBUSY  ? D_QUOTA : D_ERROR,
2237                        "dqacq failed! (rc:%d)\n", rc);
2238
2239         /* the qd_count might be changed in lqc_handler */
2240         if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) &&
2241             !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) {
2242                 memcpy(rep, qdata, sizeof(*qdata));
2243         } else {
2244                 qdata_old = lustre_quota_new_to_old(qdata);
2245                 memcpy(rep, qdata_old, sizeof(*qdata_old));
2246         }
2247         req->rq_status = rc;
2248         rc = ptlrpc_reply(req);
2249
2250         RETURN(rc);
2251 #else
2252         return 0;
2253 #endif /* !__KERNEL__ */
2254 }
2255 #endif /* HAVE_QUOTA_SUPPORT */
2256
2257 ldlm_mode_t lck_compat_array[] = {
2258         [LCK_EX] LCK_COMPAT_EX,
2259         [LCK_PW] LCK_COMPAT_PW,
2260         [LCK_PR] LCK_COMPAT_PR,
2261         [LCK_CW] LCK_COMPAT_CW,
2262         [LCK_CR] LCK_COMPAT_CR,
2263         [LCK_NL] LCK_COMPAT_NL,
2264         [LCK_GROUP] LCK_COMPAT_GROUP,
2265         [LCK_COS] LCK_COMPAT_COS,
2266 };
2267
2268 /**
2269  * Rather arbitrary mapping from LDLM error codes to errno values. This should
2270  * not escape to the user level.
2271  */
2272 int ldlm_error2errno(ldlm_error_t error)
2273 {
2274         int result;
2275
2276         switch (error) {
2277         case ELDLM_OK:
2278                 result = 0;
2279                 break;
2280         case ELDLM_LOCK_CHANGED:
2281                 result = -ESTALE;
2282                 break;
2283         case ELDLM_LOCK_ABORTED:
2284                 result = -ENAVAIL;
2285                 break;
2286         case ELDLM_LOCK_REPLACED:
2287                 result = -ESRCH;
2288                 break;
2289         case ELDLM_NO_LOCK_DATA:
2290                 result = -ENOENT;
2291                 break;
2292         case ELDLM_NAMESPACE_EXISTS:
2293                 result = -EEXIST;
2294                 break;
2295         case ELDLM_BAD_NAMESPACE:
2296                 result = -EBADF;
2297                 break;
2298         default:
2299                 if (((int)error) < 0)  /* cast to signed type */
2300                         result = error; /* as ldlm_error_t can be unsigned */
2301                 else {
2302                         CERROR("Invalid DLM result code: %i\n", error);
2303                         result = -EPROTO;
2304                 }
2305         }
2306         return result;
2307 }
2308 EXPORT_SYMBOL(ldlm_error2errno);
2309
2310 /**
2311  * Dual to ldlm_error2errno(): maps errno values back to ldlm_error_t.
2312  */
2313 ldlm_error_t ldlm_errno2error(int err_no)
2314 {
2315         int error;
2316
2317         switch (err_no) {
2318         case 0:
2319                 error = ELDLM_OK;
2320                 break;
2321         case -ESTALE:
2322                 error = ELDLM_LOCK_CHANGED;
2323                 break;
2324         case -ENAVAIL:
2325                 error = ELDLM_LOCK_ABORTED;
2326                 break;
2327         case -ESRCH:
2328                 error = ELDLM_LOCK_REPLACED;
2329                 break;
2330         case -ENOENT:
2331                 error = ELDLM_NO_LOCK_DATA;
2332                 break;
2333         case -EEXIST:
2334                 error = ELDLM_NAMESPACE_EXISTS;
2335                 break;
2336         case -EBADF:
2337                 error = ELDLM_BAD_NAMESPACE;
2338                 break;
2339         default:
2340                 error = err_no;
2341         }
2342         return error;
2343 }
2344 EXPORT_SYMBOL(ldlm_errno2error);
2345