1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 # define EXPORT_SYMTAB
40 #define DEBUG_SUBSYSTEM S_LDLM
43 # include <libcfs/libcfs.h>
45 # include <liblustre.h>
48 #include <lustre_mds.h>
49 #include <lustre_dlm.h>
50 #include <lustre_net.h>
51 #include "ldlm_internal.h"
53 /* @priority: if non-zero, move the selected to the list head
54 * @create: if zero, only search in existed connections
56 static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid,
57 int priority, int create)
59 struct ptlrpc_connection *ptlrpc_conn;
60 struct obd_import_conn *imp_conn = NULL, *item;
64 if (!create && !priority) {
65 CDEBUG(D_HA, "Nothing to do\n");
69 ptlrpc_conn = ptlrpc_uuid_to_connection(uuid);
71 CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid);
76 OBD_ALLOC(imp_conn, sizeof(*imp_conn));
78 GOTO(out_put, rc = -ENOMEM);
82 spin_lock(&imp->imp_lock);
83 list_for_each_entry(item, &imp->imp_conn_list, oic_item) {
84 if (obd_uuid_equals(uuid, &item->oic_uuid)) {
86 list_del(&item->oic_item);
87 list_add(&item->oic_item, &imp->imp_conn_list);
88 item->oic_last_attempt = 0;
90 CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n",
91 imp, imp->imp_obd->obd_name, uuid->uuid,
92 (priority ? ", moved to head" : ""));
93 spin_unlock(&imp->imp_lock);
94 GOTO(out_free, rc = 0);
99 imp_conn->oic_conn = ptlrpc_conn;
100 imp_conn->oic_uuid = *uuid;
101 imp_conn->oic_last_attempt = 0;
103 list_add(&imp_conn->oic_item, &imp->imp_conn_list);
105 list_add_tail(&imp_conn->oic_item, &imp->imp_conn_list);
106 CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n",
107 imp, imp->imp_obd->obd_name, uuid->uuid,
108 (priority ? "head" : "tail"));
110 spin_unlock(&imp->imp_lock);
111 GOTO(out_free, rc = -ENOENT);
115 spin_unlock(&imp->imp_lock);
119 OBD_FREE(imp_conn, sizeof(*imp_conn));
121 ptlrpc_connection_put(ptlrpc_conn);
125 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid)
127 return import_set_conn(imp, uuid, 1, 0);
130 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
133 return import_set_conn(imp, uuid, priority, 1);
136 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
138 struct obd_import_conn *imp_conn;
139 struct obd_export *dlmexp;
143 spin_lock(&imp->imp_lock);
144 if (list_empty(&imp->imp_conn_list)) {
145 LASSERT(!imp->imp_connection);
149 list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) {
150 if (!obd_uuid_equals(uuid, &imp_conn->oic_uuid))
152 LASSERT(imp_conn->oic_conn);
154 /* is current conn? */
155 if (imp_conn == imp->imp_conn_current) {
156 LASSERT(imp_conn->oic_conn == imp->imp_connection);
158 if (imp->imp_state != LUSTRE_IMP_CLOSED &&
159 imp->imp_state != LUSTRE_IMP_DISCON) {
160 CERROR("can't remove current connection\n");
161 GOTO(out, rc = -EBUSY);
164 ptlrpc_connection_put(imp->imp_connection);
165 imp->imp_connection = NULL;
167 dlmexp = class_conn2export(&imp->imp_dlm_handle);
168 if (dlmexp && dlmexp->exp_connection) {
169 LASSERT(dlmexp->exp_connection ==
171 ptlrpc_connection_put(dlmexp->exp_connection);
172 dlmexp->exp_connection = NULL;
176 list_del(&imp_conn->oic_item);
177 ptlrpc_connection_put(imp_conn->oic_conn);
178 OBD_FREE(imp_conn, sizeof(*imp_conn));
179 CDEBUG(D_HA, "imp %p@%s: remove connection %s\n",
180 imp, imp->imp_obd->obd_name, uuid->uuid);
185 spin_unlock(&imp->imp_lock);
187 CERROR("connection %s not found\n", uuid->uuid);
191 /* configure an RPC client OBD device
196 * 3 - inactive-on-startup
198 int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf)
200 struct lustre_cfg* lcfg = buf;
201 struct client_obd *cli = &obddev->u.cli;
202 struct obd_import *imp;
203 struct obd_uuid server_uuid;
204 int rq_portal, rp_portal, connect_op;
205 char *name = obddev->obd_type->typ_name;
209 /* In a more perfect world, we would hang a ptlrpc_client off of
210 * obd_type and just use the values from there. */
211 if (!strcmp(name, LUSTRE_OSC_NAME)) {
212 rq_portal = OST_REQUEST_PORTAL;
213 rp_portal = OSC_REPLY_PORTAL;
214 connect_op = OST_CONNECT;
215 } else if (!strcmp(name, LUSTRE_MDC_NAME)) {
216 rq_portal = MDS_REQUEST_PORTAL;
217 rp_portal = MDC_REPLY_PORTAL;
218 connect_op = MDS_CONNECT;
219 } else if (!strcmp(name, LUSTRE_MGC_NAME)) {
220 rq_portal = MGS_REQUEST_PORTAL;
221 rp_portal = MGC_REPLY_PORTAL;
222 connect_op = MGS_CONNECT;
224 CERROR("unknown client OBD type \"%s\", can't setup\n",
229 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
230 CERROR("requires a TARGET UUID\n");
234 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) {
235 CERROR("client UUID must be less than 38 characters\n");
239 if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
240 CERROR("setup requires a SERVER UUID\n");
244 if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) {
245 CERROR("target UUID must be less than 38 characters\n");
249 init_rwsem(&cli->cl_sem);
250 sema_init(&cli->cl_mgc_sem, 1);
251 cli->cl_conn_count = 0;
252 memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2),
253 min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2),
254 sizeof(server_uuid)));
257 cli->cl_avail_grant = 0;
258 /* FIXME: should limit this for the sum of all cl_dirty_max */
259 cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024;
260 if (cli->cl_dirty_max >> CFS_PAGE_SHIFT > num_physpages / 8)
261 cli->cl_dirty_max = num_physpages << (CFS_PAGE_SHIFT - 3);
262 CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters);
263 CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list);
264 CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list);
265 CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list);
266 CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list);
267 client_obd_list_lock_init(&cli->cl_loi_list_lock);
268 cli->cl_r_in_flight = 0;
269 cli->cl_w_in_flight = 0;
270 spin_lock_init(&cli->cl_read_rpc_hist.oh_lock);
271 spin_lock_init(&cli->cl_write_rpc_hist.oh_lock);
272 spin_lock_init(&cli->cl_read_page_hist.oh_lock);
273 spin_lock_init(&cli->cl_write_page_hist.oh_lock);
274 spin_lock_init(&cli->cl_read_offset_hist.oh_lock);
275 spin_lock_init(&cli->cl_write_offset_hist.oh_lock);
276 cfs_waitq_init(&cli->cl_destroy_waitq);
277 atomic_set(&cli->cl_destroy_in_flight, 0);
278 #ifdef ENABLE_CHECKSUM
279 /* Turn on checksumming by default. */
280 cli->cl_checksum = 1;
282 * The supported checksum types will be worked out at connect time
283 * Set cl_chksum* to CRC32 for now to avoid returning screwed info
286 cli->cl_cksum_type = cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
288 atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
289 atomic_set(&cli->cl_quota_resends, CLIENT_QUOTA_DEFAULT_RESENDS);
291 /* This value may be changed at connect time in
292 ptlrpc_connect_interpret. */
293 cli->cl_max_pages_per_rpc = min((int)PTLRPC_MAX_BRW_PAGES,
294 (int)(1024 * 1024 >> CFS_PAGE_SHIFT));
296 if (!strcmp(name, LUSTRE_MDC_NAME)) {
297 cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
298 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 128 /* MB */) {
299 cli->cl_max_rpcs_in_flight = 2;
300 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 256 /* MB */) {
301 cli->cl_max_rpcs_in_flight = 3;
302 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 512 /* MB */) {
303 cli->cl_max_rpcs_in_flight = 4;
305 cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
309 CERROR("ldlm_get_ref failed: %d\n", rc);
313 ptlrpc_init_client(rq_portal, rp_portal, name,
314 &obddev->obd_ldlm_client);
316 imp = class_new_import(obddev);
318 GOTO(err_ldlm, rc = -ENOENT);
319 imp->imp_client = &obddev->obd_ldlm_client;
320 imp->imp_connect_op = connect_op;
321 imp->imp_initial_recov = 1;
322 imp->imp_initial_recov_bk = 0;
323 CFS_INIT_LIST_HEAD(&imp->imp_pinger_chain);
324 memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
325 LUSTRE_CFG_BUFLEN(lcfg, 1));
326 class_import_put(imp);
328 rc = client_import_add_conn(imp, &server_uuid, 1);
330 CERROR("can't add initial connection\n");
331 GOTO(err_import, rc);
334 cli->cl_import = imp;
335 /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
336 cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3);
337 cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
339 if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
340 if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) {
341 CDEBUG(D_HA, "marking %s %s->%s as inactive\n",
342 name, obddev->obd_name,
343 cli->cl_target_uuid.uuid);
344 spin_lock(&imp->imp_lock);
345 imp->imp_deactive = 1;
346 spin_unlock(&imp->imp_lock);
350 obddev->obd_namespace = ldlm_namespace_new(obddev, obddev->obd_name,
351 LDLM_NAMESPACE_CLIENT,
352 LDLM_NAMESPACE_GREEDY);
353 if (obddev->obd_namespace == NULL) {
354 CERROR("Unable to create client namespace - %s\n",
356 GOTO(err_import, rc = -ENOMEM);
359 cli->cl_qchk_stat = CL_NOT_QUOTACHECKED;
364 class_destroy_import(imp);
372 int client_obd_cleanup(struct obd_device *obddev)
376 ldlm_namespace_free_post(obddev->obd_namespace);
377 obddev->obd_namespace = NULL;
383 /* ->o_connect() method for client side (OSC and MDC and MGC) */
384 int client_connect_import(struct lustre_handle *dlm_handle,
385 struct obd_device *obd, struct obd_uuid *cluuid,
386 struct obd_connect_data *data, void *localdata)
388 struct client_obd *cli = &obd->u.cli;
389 struct obd_import *imp = cli->cl_import;
390 struct obd_export **exp = localdata;
391 struct obd_connect_data *ocd;
395 down_write(&cli->cl_sem);
396 CDEBUG(D_INFO, "connect %s - %d\n", obd->obd_name,
399 if (cli->cl_conn_count > 0)
400 GOTO(out_sem, rc = -EALREADY);
402 rc = class_connect(dlm_handle, obd, cluuid);
406 cli->cl_conn_count++;
407 *exp = class_conn2export(dlm_handle);
409 LASSERT(obd->obd_namespace);
411 imp->imp_dlm_handle = *dlm_handle;
412 rc = ptlrpc_init_import(imp);
416 ocd = &imp->imp_connect_data;
419 imp->imp_connect_flags_orig = data->ocd_connect_flags;
422 rc = ptlrpc_connect_import(imp, NULL);
424 LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
427 LASSERT((*exp)->exp_connection);
430 LASSERT((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
431 ocd->ocd_connect_flags);
432 data->ocd_connect_flags = ocd->ocd_connect_flags;
435 ptlrpc_pinger_add_import(imp);
440 cli->cl_conn_count--;
441 class_disconnect(*exp);
445 up_write(&cli->cl_sem);
449 int client_disconnect_export(struct obd_export *exp)
451 struct obd_device *obd = class_exp2obd(exp);
452 struct client_obd *cli;
453 struct obd_import *imp;
458 CERROR("invalid export for disconnect: exp %p cookie "LPX64"\n",
459 exp, exp ? exp->exp_handle.h_cookie : -1);
464 imp = cli->cl_import;
466 down_write(&cli->cl_sem);
467 CDEBUG(D_INFO, "disconnect %s - %d\n", obd->obd_name,
470 if (!cli->cl_conn_count) {
471 CERROR("disconnecting disconnected device (%s)\n",
473 GOTO(out_disconnect, rc = -EINVAL);
476 cli->cl_conn_count--;
477 if (cli->cl_conn_count)
478 GOTO(out_disconnect, rc = 0);
480 /* Mark import deactivated now, so we don't try to reconnect if any
481 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
482 * fully deactivate the import, or that would drop all requests. */
483 spin_lock(&imp->imp_lock);
484 imp->imp_deactive = 1;
485 spin_unlock(&imp->imp_lock);
487 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
488 * delete it regardless. (It's safe to delete an import that was
490 (void)ptlrpc_pinger_del_import(imp);
492 if (obd->obd_namespace != NULL) {
493 /* obd_force == local only */
494 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
495 obd->obd_force ? LDLM_FL_LOCAL_ONLY:0,
497 ldlm_namespace_free_prior(obd->obd_namespace, imp,
501 rc = ptlrpc_disconnect_import(imp, 0);
503 ptlrpc_invalidate_import(imp);
505 if (imp->imp_rq_pool) {
506 ptlrpc_free_rq_pool(imp->imp_rq_pool);
507 imp->imp_rq_pool = NULL;
509 class_destroy_import(imp);
510 cli->cl_import = NULL;
515 /* use server style - class_disconnect should be always called for
517 err = class_disconnect(exp);
520 up_write(&cli->cl_sem);
525 /* --------------------------------------------------------------------------
526 * from old lib/target.c
527 * -------------------------------------------------------------------------- */
529 static int target_handle_reconnect(struct lustre_handle *conn,
530 struct obd_export *exp,
531 struct obd_uuid *cluuid)
534 if (exp->exp_connection && exp->exp_imp_reverse) {
535 struct lustre_handle *hdl;
536 hdl = &exp->exp_imp_reverse->imp_remote_handle;
537 /* Might be a re-connect after a partition. */
538 if (!memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) {
539 CWARN("%s: %s reconnecting\n", exp->exp_obd->obd_name,
541 conn->cookie = exp->exp_handle.h_cookie;
542 /* target_handle_connect() treats EALREADY and
543 * -EALREADY differently. EALREADY means we are
544 * doing a valid reconnect from the same client. */
547 CERROR("%s reconnecting from %s, "
548 "handle mismatch (ours "LPX64", theirs "
549 LPX64")\n", cluuid->uuid,
550 exp->exp_connection->c_remote_uuid.uuid,
551 hdl->cookie, conn->cookie);
552 memset(conn, 0, sizeof *conn);
553 /* target_handle_connect() treats EALREADY and
554 * -EALREADY differently. -EALREADY is an error
555 * (same UUID, different handle). */
560 conn->cookie = exp->exp_handle.h_cookie;
561 CDEBUG(D_HA, "connect export for UUID '%s' at %p, cookie "LPX64"\n",
562 cluuid->uuid, exp, conn->cookie);
566 void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data,
569 struct obd_export *exp = cb_data;
571 CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n",
572 obd->obd_name, exp->exp_client_uuid.uuid);
574 spin_lock(&exp->exp_lock);
575 exp->exp_need_sync = 0;
576 spin_unlock(&exp->exp_lock);
578 EXPORT_SYMBOL(target_client_add_cb);
581 target_start_and_reset_recovery_timer(struct obd_device *obd,
582 svc_handler_t handler,
583 struct ptlrpc_request *req,
585 void target_stop_recovery(void *, int);
586 static void reset_recovery_timer(struct obd_device *obd, int duration,
588 int target_recovery_check_and_stop(struct obd_device *obd)
590 int abort_recovery = 0;
592 spin_lock_bh(&obd->obd_processing_task_lock);
593 abort_recovery = obd->obd_abort_recovery;
594 obd->obd_abort_recovery = 0;
595 spin_unlock_bh(&obd->obd_processing_task_lock);
598 /** check is fs version-capable */
599 if (target_fs_version_capable(obd)) {
600 class_handle_stale_exports(obd);
602 CWARN("Versions are not supported by ldiskfs, VBR is OFF\n");
603 class_disconnect_stale_exports(obd, exp_flags_from_obd(obd));
605 /* VBR: no clients are remained to replay, stop recovery */
606 spin_lock_bh(&obd->obd_processing_task_lock);
607 if (obd->obd_recovering && obd->obd_recoverable_clients == 0) {
608 spin_unlock_bh(&obd->obd_processing_task_lock);
609 target_stop_recovery(obd, 0);
612 /* always check versions now */
613 obd->obd_version_recov = 1;
614 spin_unlock_bh(&obd->obd_processing_task_lock);
615 /* reset timer, recovery will proceed with versions now */
616 reset_recovery_timer(obd, OBD_RECOVERY_TIME_SOFT, 1);
619 EXPORT_SYMBOL(target_recovery_check_and_stop);
621 int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
623 struct obd_device *target, *targref = NULL;
624 struct obd_export *export = NULL;
625 struct obd_import *revimp;
626 struct lustre_handle conn;
627 struct obd_uuid tgtuuid;
628 struct obd_uuid cluuid;
629 struct obd_uuid remote_uuid;
632 struct obd_connect_data *data;
633 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*data) };
634 lnet_nid_t *client_nid = NULL;
638 OBD_RACE(OBD_FAIL_TGT_CONN_RACE);
640 lustre_set_req_swabbed(req, REQ_REC_OFF);
641 str = lustre_msg_string(req->rq_reqmsg, REQ_REC_OFF, sizeof(tgtuuid)-1);
643 DEBUG_REQ(D_ERROR, req, "bad target UUID for connect");
644 GOTO(out, rc = -EINVAL);
647 obd_str2uuid (&tgtuuid, str);
648 target = class_uuid2obd(&tgtuuid);
650 /* old (pre 1.6) lustre_process_log tries to connect to mdsname
651 (eg. mdsA) instead of uuid. */
653 snprintf((char *)tgtuuid.uuid, sizeof(tgtuuid), "%s_UUID", str);
654 target = class_uuid2obd(&tgtuuid);
657 target = class_name2obd(str);
660 if (!target || target->obd_stopping || !target->obd_set_up) {
661 LCONSOLE_ERROR_MSG(0x137, "UUID '%s' is not available "
662 " for connect (%s)\n", str,
663 !target ? "no target" :
664 (target->obd_stopping ? "stopping" :
666 GOTO(out, rc = -ENODEV);
669 if (target->obd_no_conn) {
670 LCONSOLE_WARN("%s: temporarily refusing client connection "
671 "from %s\n", target->obd_name,
672 libcfs_nid2str(req->rq_peer.nid));
673 GOTO(out, rc = -EAGAIN);
676 /* Make sure the target isn't cleaned up while we're here. Yes,
677 there's still a race between the above check and our incref here.
678 Really, class_uuid2obd should take the ref. */
679 targref = class_incref(target);
681 lustre_set_req_swabbed(req, REQ_REC_OFF + 1);
682 str = lustre_msg_string(req->rq_reqmsg, REQ_REC_OFF + 1,
685 DEBUG_REQ(D_ERROR, req, "bad client UUID for connect");
686 GOTO(out, rc = -EINVAL);
689 obd_str2uuid (&cluuid, str);
691 /* XXX extract a nettype and format accordingly */
692 switch (sizeof(lnet_nid_t)) {
693 /* NB the casts only avoid compiler warnings */
695 snprintf(remote_uuid.uuid, sizeof remote_uuid,
696 "NET_"LPX64"_UUID", (__u64)req->rq_peer.nid);
699 snprintf(remote_uuid.uuid, sizeof remote_uuid,
700 "NET_%x_UUID", (__u32)req->rq_peer.nid);
706 target_recovery_check_and_stop(target);
708 tmp = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, sizeof conn);
710 GOTO(out, rc = -EPROTO);
712 memcpy(&conn, tmp, sizeof conn);
714 data = lustre_swab_reqbuf(req, REQ_REC_OFF + 3, sizeof(*data),
715 lustre_swab_connect);
716 rc = lustre_pack_reply(req, 2, size, NULL);
720 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
722 DEBUG_REQ(D_WARNING, req, "Refusing old (unversioned) "
723 "libclient connection attempt");
724 GOTO(out, rc = -EPROTO);
725 } else if (data->ocd_version < LUSTRE_VERSION_CODE -
726 LUSTRE_VERSION_ALLOWED_OFFSET ||
727 data->ocd_version > LUSTRE_VERSION_CODE +
728 LUSTRE_VERSION_ALLOWED_OFFSET) {
729 DEBUG_REQ(D_WARNING, req, "Refusing %s (%d.%d.%d.%d) "
730 "libclient connection attempt",
731 data->ocd_version < LUSTRE_VERSION_CODE ?
733 OBD_OCD_VERSION_MAJOR(data->ocd_version),
734 OBD_OCD_VERSION_MINOR(data->ocd_version),
735 OBD_OCD_VERSION_PATCH(data->ocd_version),
736 OBD_OCD_VERSION_FIX(data->ocd_version));
737 data = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
738 offsetof(typeof(*data),
740 sizeof(data->ocd_version));
742 data->ocd_connect_flags = OBD_CONNECT_VERSION;
743 data->ocd_version = LUSTRE_VERSION_CODE;
745 GOTO(out, rc = -EPROTO);
749 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) &&
750 (data->ocd_connect_flags & OBD_CONNECT_MDS))
753 /* lctl gets a backstage, all-access pass. */
754 if (obd_uuid_equals(&cluuid, &target->obd_uuid))
755 goto dont_check_exports;
757 export = lustre_hash_lookup(target->obd_uuid_hash, &cluuid);
761 /* we've found an export in the hash */
762 if (export->exp_connecting) {
763 /* bug 9635, et. al. */
764 CWARN("%s: exp %p already connecting\n",
765 export->exp_obd->obd_name, export);
766 class_export_put(export);
769 } else if (mds_conn && export->exp_connection) {
770 if (req->rq_peer.nid != export->exp_connection->c_peer.nid)
771 /* mds reconnected after failover */
772 CWARN("%s: received MDS connection from NID %s,"
773 " removing former export from NID %s\n",
774 target->obd_name, libcfs_nid2str(req->rq_peer.nid),
775 libcfs_nid2str(export->exp_connection->c_peer.nid));
777 /* new mds connection from the same nid */
778 CWARN("%s: received new MDS connection from NID %s,"
779 " removing former export from same NID\n",
780 target->obd_name, libcfs_nid2str(req->rq_peer.nid));
781 class_fail_export(export);
782 class_export_put(export);
785 } else if (export->exp_connection &&
786 req->rq_peer.nid != export->exp_connection->c_peer.nid &&
787 (lustre_msg_get_op_flags(req->rq_reqmsg) &
788 MSG_CONNECT_INITIAL)) {
789 CWARN("%s: cookie %s seen on new NID %s when "
790 "existing NID %s is already connected\n",
791 target->obd_name, cluuid.uuid,
792 libcfs_nid2str(req->rq_peer.nid),
793 libcfs_nid2str(export->exp_connection->c_peer.nid));
795 class_export_put(export);
797 } else if (export->exp_failed) { /* bug 11327 */
798 CDEBUG(D_HA, "%s: exp %p evict in progress - new cookie needed "
799 "for connect\n", export->exp_obd->obd_name, export);
800 class_export_put(export);
803 } else if (export->exp_delayed &&
804 !(data && data->ocd_connect_flags & OBD_CONNECT_VBR)) {
805 class_fail_export(export);
806 class_export_put(export);
808 GOTO(out, rc = -ENODEV);
810 spin_lock(&export->exp_lock);
811 export->exp_connecting = 1;
812 spin_unlock(&export->exp_lock);
813 class_export_put(export);
814 LASSERT(export->exp_obd == target);
816 rc = target_handle_reconnect(&conn, export, &cluuid);
819 /* If we found an export, we already unlocked. */
822 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout);
823 } else if (req->rq_export == NULL &&
824 atomic_read(&export->exp_rpc_count) > 0) {
825 CWARN("%s: refuse connection from %s/%s to 0x%p; still busy "
826 "with %d references\n", target->obd_name, cluuid.uuid,
827 libcfs_nid2str(req->rq_peer.nid),
828 export, atomic_read(&export->exp_refcount));
829 GOTO(out, rc = -EBUSY);
830 } else if (req->rq_export != NULL &&
831 atomic_read(&export->exp_rpc_count) > 1) {
832 /* the current connect rpc has increased exp_rpc_count */
833 CWARN("%s: refuse reconnection from %s@%s to 0x%p; still busy "
834 "with %d active RPCs\n", target->obd_name, cluuid.uuid,
835 libcfs_nid2str(req->rq_peer.nid),
836 export, atomic_read(&export->exp_rpc_count) - 1);
837 spin_lock(&export->exp_lock);
838 if (req->rq_export->exp_conn_cnt <
839 lustre_msg_get_conn_cnt(req->rq_reqmsg))
840 /* try to abort active requests */
841 req->rq_export->exp_abort_active_req = 1;
842 spin_unlock(&export->exp_lock);
843 GOTO(out, rc = -EBUSY);
844 } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) {
845 CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; "
846 "cookies not random?\n", target->obd_name,
847 libcfs_nid2str(req->rq_peer.nid), cluuid.uuid);
848 GOTO(out, rc = -EALREADY);
849 } else if (export->exp_delayed && target->obd_recovering) {
850 /* VBR: don't allow delayed connection during recovery */
851 CWARN("%s: NID %s (%s) export was already marked as delayed "
852 "and will wait for end of recovery\n", target->obd_name,
853 libcfs_nid2str(req->rq_peer.nid), cluuid.uuid);
854 GOTO(out, rc = -EBUSY);
856 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
862 /* Tell the client if we're in recovery. */
863 if (target->obd_recovering) {
864 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
865 /* If this is the first time a client connects,
866 reset the recovery timer */
868 target_start_and_reset_recovery_timer(target, handler,
872 /* We want to handle EALREADY but *not* -EALREADY from
873 * target_handle_reconnect(), return reconnection state in a flag */
874 if (rc == EALREADY) {
875 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
881 /* Tell the client if we support replayable requests */
882 if (target->obd_replayable)
883 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_REPLAYABLE);
884 client_nid = &req->rq_peer.nid;
886 /* VBR: for delayed connections we start recovery */
887 if (export && export->exp_delayed && !export->exp_in_recovery) {
888 LASSERT(!target->obd_recovering);
889 LASSERT(data && data->ocd_connect_flags & OBD_CONNECT_VBR);
890 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_DELAYED |
891 MSG_CONNECT_RECOVERING);
892 spin_lock_bh(&target->obd_processing_task_lock);
893 target->obd_version_recov = 1;
894 spin_unlock_bh(&target->obd_processing_task_lock);
895 target_start_and_reset_recovery_timer(target, handler, req, 1);
898 if (export == NULL) {
899 if (target->obd_recovering) {
900 CERROR("%s: denying connection for new client %s (%s): "
901 "%d clients in recovery for %lds\n",
903 libcfs_nid2str(req->rq_peer.nid), cluuid.uuid,
904 target->obd_recoverable_clients,
905 cfs_duration_sec(cfs_time_sub(cfs_timer_deadline(&target->obd_recovery_timer),
906 cfs_time_current())));
910 rc = obd_connect(&conn, target, &cluuid, data,
914 rc = obd_reconnect(export, target, &cluuid, data, client_nid);
920 /* Return only the parts of obd_connect_data that we understand, so the
921 * client knows that we don't understand the rest. */
923 memcpy(lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
925 data, sizeof(*data));
927 /* If all else goes well, this is our RPC return code. */
930 lustre_msg_set_handle(req->rq_repmsg, &conn);
932 /* ownership of this export ref transfers to the request AFTER we
933 * drop any previous reference the request had, but we don't want
934 * that to go to zero before we get our new export reference. */
935 export = class_conn2export(&conn);
937 DEBUG_REQ(D_ERROR, req, "Missing export!");
938 GOTO(out, rc = -ENODEV);
941 /* If the client and the server are the same node, we will already
942 * have an export that really points to the client's DLM export,
943 * because we have a shared handles table.
945 * XXX this will go away when shaver stops sending the "connect" handle
946 * in the real "remote handle" field of the request --phik 24 Apr 2003
948 if (req->rq_export != NULL)
949 class_export_put(req->rq_export);
951 req->rq_export = export;
953 spin_lock(&export->exp_lock);
954 if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
955 CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n",
956 cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
957 export->exp_conn_cnt,
958 lustre_msg_get_conn_cnt(req->rq_reqmsg));
960 spin_unlock(&export->exp_lock);
961 GOTO(out, rc = -EALREADY);
963 export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
964 export->exp_abort_active_req = 0;
966 /* request from liblustre? Don't evict it for not pinging. */
967 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
968 export->exp_libclient = 1;
969 spin_unlock(&export->exp_lock);
971 spin_lock(&target->obd_dev_lock);
972 list_del_init(&export->exp_obd_chain_timed);
973 spin_unlock(&target->obd_dev_lock);
975 spin_unlock(&export->exp_lock);
978 if (export->exp_connection != NULL) {
979 /* Check to see if connection came from another NID */
980 if ((export->exp_connection->c_peer.nid != req->rq_peer.nid) &&
981 !hlist_unhashed(&export->exp_nid_hash))
982 lustre_hash_del(export->exp_obd->obd_nid_hash,
983 &export->exp_connection->c_peer.nid,
984 &export->exp_nid_hash);
986 ptlrpc_connection_put(export->exp_connection);
989 export->exp_connection = ptlrpc_connection_get(req->rq_peer,
993 if (hlist_unhashed(&export->exp_nid_hash)) {
994 lustre_hash_add_unique(export->exp_obd->obd_nid_hash,
995 &export->exp_connection->c_peer.nid,
996 &export->exp_nid_hash);
999 if (lustre_msg_get_op_flags(req->rq_repmsg) & MSG_CONNECT_RECONNECT) {
1000 revimp = class_import_get(export->exp_imp_reverse);
1001 ptlrpc_connection_put(revimp->imp_connection);
1002 revimp->imp_connection = NULL;
1003 GOTO(set_flags, rc = 0);
1006 if (target->obd_recovering && !export->exp_in_recovery) {
1007 spin_lock(&export->exp_lock);
1008 export->exp_in_recovery = 1;
1009 spin_unlock(&export->exp_lock);
1010 target->obd_connected_clients++;
1013 lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, sizeof conn),
1016 if (export->exp_imp_reverse != NULL)
1017 class_destroy_import(export->exp_imp_reverse);
1018 revimp = export->exp_imp_reverse = class_new_import(target);
1019 revimp->imp_client = &export->exp_obd->obd_ldlm_client;
1020 revimp->imp_remote_handle = conn;
1021 revimp->imp_dlm_fake = 1;
1022 revimp->imp_state = LUSTRE_IMP_FULL;
1025 revimp->imp_connection = ptlrpc_connection_addref(export->exp_connection);
1026 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V1 &&
1027 lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_NEXT_VER) {
1028 revimp->imp_msg_magic = LUSTRE_MSG_MAGIC_V2;
1029 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_NEXT_VER);
1031 /* unknown versions will be caught in
1032 * ptlrpc_handle_server_req_in->lustre_unpack_msg() */
1033 revimp->imp_msg_magic = req->rq_reqmsg->lm_magic;
1036 if (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1) {
1037 if (export->exp_connect_flags & OBD_CONNECT_AT)
1038 revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT;
1040 revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
1043 class_import_put(revimp);
1046 spin_lock(&export->exp_lock);
1047 export->exp_connecting = 0;
1048 spin_unlock(&export->exp_lock);
1051 class_decref(targref);
1053 req->rq_status = rc;
1057 int target_handle_disconnect(struct ptlrpc_request *req)
1062 rc = lustre_pack_reply(req, 1, NULL, NULL);
1066 /* keep the rq_export around so we can send the reply */
1067 req->rq_status = obd_disconnect(class_export_get(req->rq_export));
1071 void target_destroy_export(struct obd_export *exp)
1073 /* exports created from last_rcvd data, and "fake"
1074 exports created by lctl don't have an import */
1075 if (exp->exp_imp_reverse != NULL)
1076 class_destroy_import(exp->exp_imp_reverse);
1078 /* We cancel locks at disconnect time, but this will catch any locks
1079 * granted in a race with recovery-induced disconnect. */
1080 if (exp->exp_obd->obd_namespace != NULL)
1081 ldlm_cancel_locks_for_export(exp);
1085 * Recovery functions
1088 static int target_exp_enqueue_req_replay(struct ptlrpc_request *req)
1090 __u64 transno = lustre_msg_get_transno(req->rq_reqmsg);
1091 struct obd_export *exp = req->rq_export;
1092 struct ptlrpc_request *reqiter;
1097 spin_lock(&exp->exp_lock);
1098 list_for_each_entry(reqiter, &exp->exp_req_replay_queue,
1100 if (lustre_msg_get_transno(reqiter->rq_reqmsg) == transno) {
1107 /* we expect it with RESENT and REPLAY flags */
1108 if ((lustre_msg_get_flags(req->rq_reqmsg) &
1109 (MSG_RESENT | MSG_REPLAY)) != (MSG_RESENT | MSG_REPLAY))
1110 CERROR("invalid flags %x of resent replay\n",
1111 lustre_msg_get_flags(req->rq_reqmsg));
1113 list_add_tail(&req->rq_replay_list, &exp->exp_req_replay_queue);
1116 spin_unlock(&exp->exp_lock);
1120 static void target_exp_dequeue_req_replay(struct ptlrpc_request *req)
1122 LASSERT(!list_empty(&req->rq_replay_list));
1123 LASSERT(req->rq_export);
1125 spin_lock(&req->rq_export->exp_lock);
1126 list_del_init(&req->rq_replay_list);
1127 spin_unlock(&req->rq_export->exp_lock);
1130 static void target_release_saved_req(struct ptlrpc_request *req)
1132 ptlrpc_req_drop_rs(req);
1133 class_export_put(req->rq_export);
1134 OBD_FREE(req->rq_reqmsg, req->rq_reqlen);
1135 OBD_FREE(req, sizeof *req);
1138 static void target_send_delayed_replies(struct obd_device *obd)
1140 int max_clients = obd->obd_max_recoverable_clients;
1141 struct ptlrpc_request *req, *tmp;
1142 time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() -
1143 obd->obd_recovery_start);
1145 LCONSOLE_INFO("%s: Recovery period over after %d:%.02d, of %d clients "
1146 "%d recovered and %d %s evicted.\n", obd->obd_name,
1147 (int)elapsed_time/60, (int)elapsed_time%60, max_clients,
1148 obd->obd_connected_clients,
1149 obd->obd_stale_clients,
1150 obd->obd_stale_clients == 1 ? "was" : "were");
1152 LCONSOLE_INFO("%s: sending delayed replies to recovered clients\n",
1155 list_for_each_entry_safe(req, tmp, &obd->obd_delayed_reply_queue,
1157 list_del_init(&req->rq_list);
1158 DEBUG_REQ(D_HA, req, "delayed:");
1160 target_release_saved_req(req);
1162 obd->obd_recovery_end = cfs_time_current_sec();
1165 static void target_finish_recovery(struct obd_device *obd)
1167 OBD_RACE(OBD_FAIL_TGT_REPLAY_DELAY);
1169 ldlm_reprocess_all_ns(obd->obd_namespace);
1170 spin_lock_bh(&obd->obd_processing_task_lock);
1171 if (list_empty(&obd->obd_recovery_queue)) {
1172 obd->obd_recovery_thread = NULL;
1173 obd->obd_processing_task = 0;
1175 spin_unlock_bh(&obd->obd_processing_task_lock);
1176 CERROR("%s: Recovery queue isn't empty\n", obd->obd_name);
1179 spin_unlock_bh(&obd->obd_processing_task_lock);
1181 /* when recovery finished, cleanup orphans on mds and ost */
1182 if (OBT(obd) && OBP(obd, postrecov)) {
1183 int rc = OBP(obd, postrecov)(obd);
1185 LCONSOLE_WARN("%s: Post recovery failed, rc %d\n",
1188 target_send_delayed_replies(obd);
1191 static void abort_recovery_queue(struct obd_device *obd)
1193 struct ptlrpc_request *req, *n;
1194 struct list_head abort_list;
1197 CFS_INIT_LIST_HEAD(&abort_list);
1198 spin_lock_bh(&obd->obd_processing_task_lock);
1199 list_splice_init(&obd->obd_recovery_queue, &abort_list);
1200 spin_unlock_bh(&obd->obd_processing_task_lock);
1201 /* process abort list unlocked */
1202 list_for_each_entry_safe(req, n, &abort_list, rq_list) {
1203 target_exp_dequeue_req_replay(req);
1204 list_del_init(&req->rq_list);
1205 DEBUG_REQ(D_ERROR, req, "aborted:");
1206 req->rq_status = -ENOTCONN;
1207 req->rq_type = PTL_RPC_MSG_ERR;
1208 rc = lustre_pack_reply(req, 1, NULL, NULL);
1212 DEBUG_REQ(D_ERROR, req,
1213 "packing failed for abort-reply; skipping");
1215 LASSERT(req->rq_copy);
1216 class_export_rpc_put(req->rq_export);
1217 target_release_saved_req(req);
1221 /* Called from a cleanup function if the device is being cleaned up
1222 forcefully. The exports should all have been disconnected already,
1223 the only thing left to do is
1224 - clear the recovery flags
1226 - free queued requests and replies, but don't send replies
1227 Because the obd_stopping flag is set, no new requests should be received.
1230 void target_cleanup_recovery(struct obd_device *obd)
1232 struct list_head *tmp, *n;
1233 struct ptlrpc_request *req;
1234 struct list_head clean_list;
1237 LASSERT(obd->obd_stopping);
1239 spin_lock_bh(&obd->obd_processing_task_lock);
1240 if (!obd->obd_recovering) {
1241 spin_unlock_bh(&obd->obd_processing_task_lock);
1245 obd->obd_recovering = obd->obd_abort_recovery = 0;
1246 target_cancel_recovery_timer(obd);
1247 spin_unlock_bh(&obd->obd_processing_task_lock);
1249 list_for_each_safe(tmp, n, &obd->obd_delayed_reply_queue) {
1250 req = list_entry(tmp, struct ptlrpc_request, rq_list);
1251 list_del(&req->rq_list);
1252 target_release_saved_req(req);
1255 CFS_INIT_LIST_HEAD(&clean_list);
1256 spin_lock_bh(&obd->obd_processing_task_lock);
1257 list_splice_init(&obd->obd_recovery_queue, &clean_list);
1258 spin_unlock_bh(&obd->obd_processing_task_lock);
1259 list_for_each_safe(tmp, n, &clean_list) {
1260 req = list_entry(tmp, struct ptlrpc_request, rq_list);
1261 target_exp_dequeue_req_replay(req);
1262 list_del_init(&req->rq_list);
1264 LASSERT(req->rq_copy);
1265 class_export_rpc_put(req->rq_export);
1266 target_release_saved_req(req);
1271 void target_stop_recovery(void *data, int abort)
1273 struct obd_device *obd = data;
1274 enum obd_option flags;
1277 spin_lock_bh(&obd->obd_processing_task_lock);
1278 if (!obd->obd_recovering) {
1279 spin_unlock_bh(&obd->obd_processing_task_lock);
1283 flags = exp_flags_from_obd(obd) | OBD_OPT_ABORT_RECOV;
1284 obd->obd_recovering = 0;
1285 obd->obd_abort_recovery = 0;
1286 obd->obd_processing_task = 0;
1288 LASSERT(obd->obd_recoverable_clients == 0);
1290 target_cancel_recovery_timer(obd);
1291 spin_unlock_bh(&obd->obd_processing_task_lock);
1294 LCONSOLE_WARN("%s: recovery is aborted by administrative "
1295 "request; %d clients are not recovered "
1296 "(%d clients did)\n", obd->obd_name,
1297 obd->obd_recoverable_clients,
1298 obd->obd_connected_clients);
1299 class_disconnect_stale_exports(obd, flags);
1301 abort_recovery_queue(obd);
1302 target_finish_recovery(obd);
1303 CDEBUG(D_HA, "%s: recovery complete\n", obd_uuid2str(&obd->obd_uuid));
1307 void target_abort_recovery(void *data)
1309 target_stop_recovery(data, 1);
1312 static void reset_recovery_timer(struct obd_device *, int, int);
1313 static void target_recovery_expired(unsigned long castmeharder)
1315 struct obd_device *obd = (struct obd_device *)castmeharder;
1316 CDEBUG(D_HA, "%s: recovery period over; %d clients never reconnected "
1317 "after %lds (%d clients did)\n", obd->obd_name,
1318 obd->obd_recoverable_clients,
1319 cfs_time_current_sec() - obd->obd_recovery_start,
1320 obd->obd_connected_clients);
1322 spin_lock_bh(&obd->obd_processing_task_lock);
1323 obd->obd_abort_recovery = 1;
1324 cfs_waitq_signal(&obd->obd_next_transno_waitq);
1325 spin_unlock_bh(&obd->obd_processing_task_lock);
1328 /* obd_processing_task_lock should be held */
1329 void target_cancel_recovery_timer(struct obd_device *obd)
1331 CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
1332 cfs_timer_disarm(&obd->obd_recovery_timer);
1335 /* extend = 1 means require at least "duration" seconds left in the timer,
1336 extend = 0 means set the total duration (start_recovery_timer) */
1337 static void reset_recovery_timer(struct obd_device *obd, int duration,
1340 cfs_time_t now = cfs_time_current_sec();
1341 cfs_duration_t left;
1343 spin_lock_bh(&obd->obd_processing_task_lock);
1344 if (!obd->obd_recovering) {
1345 spin_unlock_bh(&obd->obd_processing_task_lock);
1349 left = cfs_time_sub(obd->obd_recovery_end, now);
1351 if (extend && (duration > left))
1352 obd->obd_recovery_timeout += duration - left;
1353 else if (!extend && (duration > obd->obd_recovery_timeout))
1354 /* Track the client's largest expected replay time */
1355 obd->obd_recovery_timeout = duration;
1357 /* Hard limit of obd_recovery_time_hard which should not happen */
1358 if(obd->obd_recovery_timeout > obd->obd_recovery_time_hard)
1359 obd->obd_recovery_timeout = obd->obd_recovery_time_hard;
1361 obd->obd_recovery_end = obd->obd_recovery_start +
1362 obd->obd_recovery_timeout;
1363 if (cfs_time_before(now, obd->obd_recovery_end)) {
1364 left = cfs_time_sub(obd->obd_recovery_end, now);
1365 cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(left));
1367 spin_unlock_bh(&obd->obd_processing_task_lock);
1368 CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n",
1369 obd->obd_name, (unsigned)left);
1372 static void check_and_start_recovery_timer(struct obd_device *obd,
1373 svc_handler_t handler)
1375 spin_lock_bh(&obd->obd_processing_task_lock);
1376 if (obd->obd_recovery_handler) {
1377 spin_unlock_bh(&obd->obd_processing_task_lock);
1380 CDEBUG(D_HA, "%s: starting recovery timer\n", obd->obd_name);
1381 obd->obd_recovery_start = cfs_time_current_sec();
1382 obd->obd_recovery_handler = handler;
1383 cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd);
1384 spin_unlock_bh(&obd->obd_processing_task_lock);
1386 reset_recovery_timer(obd, obd->obd_recovery_timeout, 0);
1389 /* Reset the timer with each new client connection */
1391 * This timer is actually reconnect_timer, which is for making sure
1392 * the total recovery window is at least as big as my reconnect
1393 * attempt timing. So the initial recovery time_out will be set to
1394 * OBD_RECOVERY_FACTOR * obd_timeout. If the timeout coming
1395 * from client is bigger than this, then the recovery time_out will
1396 * be extend to make sure the client could be reconnected, in the
1397 * process, the timeout from the new client should be ignored.
1401 target_start_and_reset_recovery_timer(struct obd_device *obd,
1402 svc_handler_t handler,
1403 struct ptlrpc_request *req,
1406 int service_time = lustre_msg_get_service_time(req->rq_reqmsg);
1408 if (!new_client && service_time)
1409 /* Teach server about old server's estimates, as first guess
1410 at how long new requests will take. */
1411 at_add(&req->rq_rqbd->rqbd_service->srv_at_estimate,
1414 check_and_start_recovery_timer(obd, handler);
1416 /* convert the service time to rpc timeout,
1417 * reuse service_time to limit stack usage */
1418 service_time = at_est2timeout(service_time);
1420 /* We expect other clients to timeout within service_time, then try
1421 * to reconnect, then try the failover server. The max delay between
1422 * connect attempts is SWITCH_MAX + SWITCH_INC + INITIAL */
1423 service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC +
1424 INITIAL_CONNECT_TIMEOUT);
1425 if (service_time > obd->obd_recovery_timeout && !new_client)
1426 reset_recovery_timer(obd, service_time, 0);
1429 static int check_for_next_transno(struct obd_device *obd)
1431 struct ptlrpc_request *req;
1432 int wake_up = 0, connected, completed, queue_len, max;
1433 __u64 next_transno, req_transno;
1435 spin_lock_bh(&obd->obd_processing_task_lock);
1436 req = list_entry(obd->obd_recovery_queue.next,
1437 struct ptlrpc_request, rq_list);
1438 max = obd->obd_max_recoverable_clients;
1439 req_transno = lustre_msg_get_transno(req->rq_reqmsg);
1440 connected = obd->obd_connected_clients;
1441 completed = max - obd->obd_recoverable_clients -
1442 obd->obd_delayed_clients;
1443 queue_len = obd->obd_requests_queued_for_recovery;
1444 next_transno = obd->obd_next_recovery_transno;
1446 CDEBUG(D_HA,"max: %d, connected: %d, delayed %d, completed: %d, "
1447 "queue_len: %d, req_transno: "LPU64", next_transno: "LPU64"\n",
1448 max, connected, obd->obd_delayed_clients, completed, queue_len,
1449 req_transno, next_transno);
1450 if (obd->obd_abort_recovery) {
1451 CDEBUG(D_HA, "waking for aborted recovery\n");
1453 } else if (!obd->obd_recovering) {
1454 CDEBUG(D_HA, "waking for completed recovery (?)\n");
1456 } else if (req_transno == next_transno) {
1457 CDEBUG(D_HA, "waking for next ("LPD64")\n", next_transno);
1459 } else if (queue_len == obd->obd_recoverable_clients) {
1461 "waking for skipped transno (skip: "LPD64
1462 ", ql: %d, comp: %d, conn: %d, next: "LPD64")\n",
1463 next_transno, queue_len, completed, max, req_transno);
1464 obd->obd_next_recovery_transno = req_transno;
1467 spin_unlock_bh(&obd->obd_processing_task_lock);
1468 LASSERT(lustre_msg_get_transno(req->rq_reqmsg) >= next_transno);
1472 static void process_recovery_queue(struct obd_device *obd)
1474 struct ptlrpc_request *req;
1475 struct l_wait_info lwi = { 0 };
1479 spin_lock_bh(&obd->obd_processing_task_lock);
1480 LASSERT(obd->obd_processing_task == cfs_curproc_pid());
1481 req = list_entry(obd->obd_recovery_queue.next,
1482 struct ptlrpc_request, rq_list);
1484 if (lustre_msg_get_transno(req->rq_reqmsg) !=
1485 obd->obd_next_recovery_transno) {
1486 spin_unlock_bh(&obd->obd_processing_task_lock);
1487 CDEBUG(D_HA, "Waiting for transno "LPD64" (1st is "
1488 LPD64", x"LPU64")\n",
1489 obd->obd_next_recovery_transno,
1490 lustre_msg_get_transno(req->rq_reqmsg),
1492 l_wait_event(obd->obd_next_transno_waitq,
1493 check_for_next_transno(obd), &lwi);
1494 if (target_recovery_check_and_stop(obd))
1498 list_del_init(&req->rq_list);
1499 LASSERT(obd->obd_recovery_thread);
1500 /* replace request initial thread with current one, bug #18221 */
1501 req->rq_svc_thread = obd->obd_recovery_thread;
1502 obd->obd_requests_queued_for_recovery--;
1503 spin_unlock_bh(&obd->obd_processing_task_lock);
1505 DEBUG_REQ(D_HA, req, "processing: ");
1506 (void)obd->obd_recovery_handler(req);
1507 obd->obd_replayed_requests++;
1508 /* Extend the recovery timer enough to complete the next
1510 reset_recovery_timer(obd, AT_OFF ? obd_timeout :
1511 at_get(&req->rq_rqbd->rqbd_service->srv_at_estimate), 1);
1512 /* bug 1580: decide how to properly sync() in recovery */
1513 //mds_fsync_super(obd->u.obt.obt_sb);
1514 spin_lock_bh(&obd->obd_processing_task_lock);
1515 obd->obd_next_recovery_transno++;
1516 spin_unlock_bh(&obd->obd_processing_task_lock);
1517 target_exp_dequeue_req_replay(req);
1519 LASSERT(req->rq_copy);
1520 class_export_rpc_put(req->rq_export);
1522 class_export_put(req->rq_export);
1523 ptlrpc_req_drop_rs(req);
1524 OBD_FREE(req->rq_reqmsg, req->rq_reqlen);
1525 OBD_FREE(req, sizeof *req);
1526 OBD_RACE(OBD_FAIL_TGT_REPLAY_DELAY);
1527 spin_lock_bh(&obd->obd_processing_task_lock);
1528 if (list_empty(&obd->obd_recovery_queue)) {
1529 obd->obd_processing_task = 0;
1530 obd->obd_recovery_thread = NULL;
1531 spin_unlock_bh(&obd->obd_processing_task_lock);
1534 spin_unlock_bh(&obd->obd_processing_task_lock);
1539 int target_queue_recovery_request(struct ptlrpc_request *req,
1540 struct obd_device *obd)
1542 struct list_head *tmp;
1544 __u64 transno = lustre_msg_get_transno(req->rq_reqmsg);
1545 struct ptlrpc_request *saved_req, *orig_req;
1546 struct lustre_msg *reqmsg;
1549 /* CAVEAT EMPTOR: The incoming request message has been swabbed
1550 * (i.e. buflens etc are in my own byte order), but type-dependent
1551 * buffers (eg mds_body, ost_body etc) have NOT been swabbed. */
1554 CFS_INIT_LIST_HEAD(&req->rq_list);
1555 DEBUG_REQ(D_HA, req, "not queueing");
1559 /* XXX If I were a real man, these LBUGs would be sane cleanups. */
1560 /* XXX just like the request-dup code in queue_final_reply */
1561 OBD_ALLOC(saved_req, sizeof *saved_req);
1564 OBD_ALLOC(reqmsg, req->rq_reqlen);
1568 spin_lock_bh(&obd->obd_processing_task_lock);
1570 /* If we're processing the queue, we want don't want to queue this
1573 * Also, if this request has a transno less than the one we're waiting
1574 * for, we should process it now. It could (and currently always will)
1575 * be an open request for a descriptor that was opened some time ago.
1577 * Also, a resent, replayed request that has already been
1578 * handled will pass through here and be processed immediately.
1580 if (obd->obd_processing_task == cfs_curproc_pid() ||
1581 transno < obd->obd_next_recovery_transno) {
1582 /* Processing the queue right now, don't re-add. */
1583 LASSERT(list_empty(&req->rq_list));
1584 spin_unlock_bh(&obd->obd_processing_task_lock);
1585 GOTO(err_free, rc = 1);
1588 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_REPLAY_DROP))) {
1589 spin_unlock_bh(&obd->obd_processing_task_lock);
1590 GOTO(err_free, rc = 0);
1593 memcpy(saved_req, req, sizeof *req);
1594 memcpy(reqmsg, req->rq_reqmsg, req->rq_reqlen);
1597 req->rq_reqmsg = reqmsg;
1598 class_export_get(req->rq_export);
1599 CFS_INIT_LIST_HEAD(&req->rq_list);
1600 CFS_INIT_LIST_HEAD(&req->rq_replay_list);
1602 if (target_exp_enqueue_req_replay(req)) {
1603 spin_unlock_bh(&obd->obd_processing_task_lock);
1604 DEBUG_REQ(D_ERROR, req, "dropping resent queued req");
1605 GOTO(err_exp, rc = 0);
1609 list_for_each(tmp, &obd->obd_recovery_queue) {
1610 struct ptlrpc_request *reqiter =
1611 list_entry(tmp, struct ptlrpc_request, rq_list);
1613 if (lustre_msg_get_transno(reqiter->rq_reqmsg) > transno) {
1614 list_add_tail(&req->rq_list, &reqiter->rq_list);
1619 if (unlikely(lustre_msg_get_transno(reqiter->rq_reqmsg) ==
1621 spin_unlock_bh(&obd->obd_processing_task_lock);
1622 DEBUG_REQ(D_ERROR, req, "dropping replay: transno "
1623 "has been claimed by another client");
1624 target_exp_dequeue_req_replay(req);
1625 GOTO(err_exp, rc = 0);
1630 list_add_tail(&req->rq_list, &obd->obd_recovery_queue);
1633 obd->obd_requests_queued_for_recovery++;
1634 orig_req->rq_copy_queued = 1;
1637 if (obd->obd_processing_task != 0) {
1638 /* Someone else is processing this queue, we'll leave it to
1641 cfs_waitq_signal(&obd->obd_next_transno_waitq);
1642 spin_unlock_bh(&obd->obd_processing_task_lock);
1646 /* Nobody is processing, and we know there's (at least) one to process
1647 * now, so we'll do the honours.
1649 obd->obd_processing_task = cfs_curproc_pid();
1650 /* save thread that handle recovery queue */
1651 obd->obd_recovery_thread = req->rq_svc_thread;
1652 spin_unlock_bh(&obd->obd_processing_task_lock);
1654 process_recovery_queue(obd);
1658 class_export_put(req->rq_export);
1660 OBD_FREE(reqmsg, req->rq_reqlen);
1661 OBD_FREE(saved_req, sizeof(*saved_req));
1665 struct obd_device * target_req2obd(struct ptlrpc_request *req)
1667 return req->rq_export->exp_obd;
1670 int target_queue_last_replay_reply(struct ptlrpc_request *req, int rc)
1672 struct obd_device *obd = target_req2obd(req);
1673 struct ptlrpc_request *saved_req;
1674 struct lustre_msg *reqmsg;
1675 struct obd_export *exp = req->rq_export;
1676 int recovery_done = 0, delayed_done = 0;
1678 LASSERT ((rc == 0) == req->rq_packed_final);
1680 if (!req->rq_packed_final) {
1681 /* Just like ptlrpc_error, but without the sending. */
1682 rc = lustre_pack_reply(req, 1, NULL, NULL);
1685 req->rq_type = PTL_RPC_MSG_ERR;
1688 LASSERT(!req->rq_reply_state->rs_difficult);
1689 LASSERT(list_empty(&req->rq_list));
1690 /* XXX a bit like the request-dup code in queue_recovery_request */
1691 OBD_ALLOC(saved_req, sizeof *saved_req);
1694 OBD_ALLOC(reqmsg, req->rq_reqlen);
1696 OBD_FREE(saved_req, sizeof *req);
1700 memcpy(reqmsg, req->rq_reqmsg, req->rq_reqlen);
1702 /* Don't race cleanup */
1703 spin_lock_bh(&obd->obd_processing_task_lock);
1704 if (obd->obd_stopping) {
1705 spin_unlock_bh(&obd->obd_processing_task_lock);
1709 if (!exp->exp_vbr_failed) {
1710 ptlrpc_rs_addref(req->rq_reply_state); /* +1 ref for saved reply */
1712 req->rq_reqmsg = reqmsg;
1713 CFS_INIT_LIST_HEAD(&req->rq_list);
1714 CFS_INIT_LIST_HEAD(&req->rq_replay_list);
1715 class_export_get(exp);
1716 list_add(&req->rq_list, &obd->obd_delayed_reply_queue);
1719 /* only count the first "replay over" request from each
1721 if (exp->exp_replay_needed) {
1722 spin_lock(&exp->exp_lock);
1723 exp->exp_replay_needed = 0;
1724 spin_unlock(&exp->exp_lock);
1726 if (!exp->exp_delayed) {
1727 --obd->obd_recoverable_clients;
1729 spin_lock(&exp->exp_lock);
1730 exp->exp_delayed = 0;
1731 spin_unlock(&exp->exp_lock);
1733 if (obd->obd_delayed_clients == 0) {
1734 spin_unlock_bh(&obd->obd_processing_task_lock);
1737 --obd->obd_delayed_clients;
1740 recovery_done = (obd->obd_recoverable_clients == 0);
1741 spin_unlock_bh(&obd->obd_processing_task_lock);
1744 /* start pinging export */
1745 spin_lock(&obd->obd_dev_lock);
1746 list_add_tail(&exp->exp_obd_chain_timed,
1747 &obd->obd_exports_timed);
1748 list_move_tail(&exp->exp_obd_chain, &obd->obd_exports);
1749 spin_unlock(&obd->obd_dev_lock);
1750 target_send_delayed_replies(obd);
1753 OBD_RACE(OBD_FAIL_LDLM_RECOV_CLIENTS);
1754 if (recovery_done) {
1755 spin_lock_bh(&obd->obd_processing_task_lock);
1756 obd->obd_recovering = 0;
1757 obd->obd_version_recov = 0;
1758 obd->obd_abort_recovery = 0;
1759 target_cancel_recovery_timer(obd);
1760 spin_unlock_bh(&obd->obd_processing_task_lock);
1763 target_finish_recovery(obd);
1764 CDEBUG(D_HA, "%s: recovery complete\n",
1765 obd_uuid2str(&obd->obd_uuid));
1767 CWARN("%s: %d recoverable clients remain\n",
1768 obd->obd_name, obd->obd_recoverable_clients);
1769 cfs_waitq_signal(&obd->obd_next_transno_waitq);
1772 /* VBR: disconnect export with failed recovery */
1773 if (exp->exp_vbr_failed) {
1774 CWARN("%s: disconnect export %s\n", obd->obd_name,
1775 exp->exp_client_uuid.uuid);
1776 class_fail_export(exp);
1777 OBD_FREE(reqmsg, req->rq_reqlen);
1778 OBD_FREE(saved_req, sizeof *req);
1780 ptlrpc_send_reply(req, 0);
1786 OBD_FREE(reqmsg, req->rq_reqlen);
1787 OBD_FREE(saved_req, sizeof *req);
1788 req->rq_status = -ENOTCONN;
1789 /* rv is ignored anyhow */
1793 int target_handle_reply(struct ptlrpc_request *req, int rc, int fail)
1795 struct obd_device *obd = NULL;
1798 obd = target_req2obd(req);
1800 /* handle replay reply for version recovery */
1801 if (obd && obd->obd_version_recov &&
1802 (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)) {
1803 LASSERT(req->rq_repmsg);
1804 lustre_msg_add_flags(req->rq_repmsg, MSG_VERSION_REPLAY);
1807 /* handle last replay */
1808 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1810 lustre_msg_get_flags(req->rq_reqmsg) & MSG_DELAY_REPLAY) {
1811 DEBUG_REQ(D_HA, req,
1812 "delayed LAST_REPLAY, queuing reply");
1813 rc = target_queue_last_replay_reply(req, rc);
1814 LASSERT(req->rq_export->exp_delayed == 0);
1818 if (obd && obd->obd_recovering) { /* normal recovery */
1819 DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1820 rc = target_queue_last_replay_reply(req, rc);
1824 /* Lost a race with recovery; let the error path DTRT. */
1825 rc = req->rq_status = -ENOTCONN;
1827 target_send_reply(req, rc, fail);
1831 static inline struct ldlm_pool *ldlm_exp2pl(struct obd_export *exp)
1833 LASSERT(exp != NULL);
1834 return &exp->exp_obd->obd_namespace->ns_pool;
1837 int target_pack_pool_reply(struct ptlrpc_request *req)
1839 struct obd_device *obd;
1843 * Check that we still have all structures alive as this may
1844 * be some late rpc in shutdown time.
1846 if (unlikely(!req->rq_export || !req->rq_export->exp_obd ||
1847 !exp_connect_lru_resize(req->rq_export))) {
1848 lustre_msg_set_slv(req->rq_repmsg, 0);
1849 lustre_msg_set_limit(req->rq_repmsg, 0);
1854 * OBD is alive here as export is alive, which we checked above.
1856 obd = req->rq_export->exp_obd;
1858 read_lock(&obd->obd_pool_lock);
1859 lustre_msg_set_slv(req->rq_repmsg, obd->obd_pool_slv);
1860 lustre_msg_set_limit(req->rq_repmsg, obd->obd_pool_limit);
1861 read_unlock(&obd->obd_pool_lock);
1867 target_send_reply_msg (struct ptlrpc_request *req, int rc, int fail_id)
1869 if (OBD_FAIL_CHECK(fail_id | OBD_FAIL_ONCE)) {
1870 obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED;
1871 DEBUG_REQ(D_ERROR, req, "dropping reply");
1876 DEBUG_REQ(D_ERROR, req, "processing error (%d)", rc);
1877 req->rq_status = rc;
1878 return (ptlrpc_send_error(req, 1));
1880 DEBUG_REQ(D_NET, req, "sending reply");
1883 return (ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT));
1887 target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
1890 struct ptlrpc_reply_state *rs;
1891 struct obd_device *obd;
1892 struct obd_export *exp;
1893 struct ptlrpc_service *svc;
1895 svc = req->rq_rqbd->rqbd_service;
1896 rs = req->rq_reply_state;
1897 if (rs == NULL || !rs->rs_difficult) {
1899 target_send_reply_msg (req, rc, fail_id);
1903 /* must be an export if locks saved */
1904 LASSERT (req->rq_export != NULL);
1905 /* req/reply consistent */
1906 LASSERT (rs->rs_service == svc);
1909 LASSERT (!rs->rs_scheduled);
1910 LASSERT (!rs->rs_scheduled_ever);
1911 LASSERT (!rs->rs_handled);
1912 LASSERT (!rs->rs_on_net);
1913 LASSERT (rs->rs_export == NULL);
1914 LASSERT (list_empty(&rs->rs_obd_list));
1915 LASSERT (list_empty(&rs->rs_exp_list));
1917 exp = class_export_get(req->rq_export);
1920 /* disable reply scheduling onto srv_reply_queue while I'm setting up */
1921 rs->rs_scheduled = 1;
1923 rs->rs_xid = req->rq_xid;
1924 rs->rs_transno = req->rq_transno;
1925 rs->rs_export = exp;
1927 spin_lock(&exp->exp_uncommitted_replies_lock);
1929 /* VBR: use exp_last_committed */
1930 if (rs->rs_transno > exp->exp_last_committed) {
1931 /* not committed already */
1932 list_add_tail (&rs->rs_obd_list,
1933 &exp->exp_uncommitted_replies);
1936 spin_unlock (&exp->exp_uncommitted_replies_lock);
1937 spin_lock (&exp->exp_lock);
1939 list_add_tail (&rs->rs_exp_list, &exp->exp_outstanding_replies);
1941 spin_unlock(&exp->exp_lock);
1943 netrc = target_send_reply_msg (req, rc, fail_id);
1945 spin_lock(&svc->srv_lock);
1947 svc->srv_n_difficult_replies++;
1950 /* error sending: reply is off the net. Also we need +1
1951 * reply ref until ptlrpc_server_handle_reply() is done
1952 * with the reply state (if the send was successful, there
1953 * would have been +1 ref for the net, which
1954 * reply_out_callback leaves alone) */
1956 ptlrpc_rs_addref(rs);
1957 atomic_inc (&svc->srv_outstanding_replies);
1960 if (!rs->rs_on_net || /* some notifier */
1961 list_empty(&rs->rs_exp_list) || /* completed already */
1962 list_empty(&rs->rs_obd_list)) {
1963 list_add_tail (&rs->rs_list, &svc->srv_reply_queue);
1964 cfs_waitq_signal (&svc->srv_waitq);
1966 list_add (&rs->rs_list, &svc->srv_active_replies);
1967 rs->rs_scheduled = 0; /* allow notifier to schedule */
1970 spin_unlock(&svc->srv_lock);
1973 int target_handle_ping(struct ptlrpc_request *req)
1975 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY &&
1976 req->rq_export->exp_in_recovery) {
1977 spin_lock(&req->rq_export->exp_lock);
1978 req->rq_export->exp_in_recovery = 0;
1979 spin_unlock(&req->rq_export->exp_lock);
1981 obd_ping(req->rq_export);
1982 return lustre_pack_reply(req, 1, NULL, NULL);
1985 void target_committed_to_req(struct ptlrpc_request *req)
1987 struct obd_export *exp = req->rq_export;
1988 if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL) {
1989 lustre_msg_set_last_committed(req->rq_repmsg,
1990 exp->exp_last_committed);
1992 DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/"
1993 "%d)", exp->exp_obd->obd_no_transno,
1994 req->rq_repmsg == NULL);
1996 CDEBUG(D_INFO, "last_committed x"LPU64", this req x"LPU64"\n",
1997 exp->exp_obd->obd_last_committed, req->rq_xid);
2000 EXPORT_SYMBOL(target_committed_to_req);
2002 int target_handle_qc_callback(struct ptlrpc_request *req)
2004 struct obd_quotactl *oqctl;
2005 struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
2007 oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
2008 lustre_swab_obd_quotactl);
2009 if (oqctl == NULL) {
2010 CERROR("Can't unpack obd_quatactl\n");
2014 cli->cl_qchk_stat = oqctl->qc_stat;
2019 #ifdef HAVE_QUOTA_SUPPORT
2020 int target_handle_dqacq_callback(struct ptlrpc_request *req)
2023 struct obd_device *obd = req->rq_export->exp_obd;
2024 struct obd_device *master_obd;
2025 struct lustre_quota_ctxt *qctxt;
2026 struct qunit_data *qdata = NULL;
2028 int repsize[2] = { sizeof(struct ptlrpc_body), 0 };
2031 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_DROP_QUOTA_REQ))
2034 repsize[1] = quota_get_qunit_data_size(req->rq_export->
2037 rc = lustre_pack_reply(req, 2, repsize, NULL);
2041 LASSERT(req->rq_export);
2043 /* there are three forms of qunit(historic causes), so we need to
2044 * adjust qunits from slaves to the same form here */
2045 OBD_ALLOC(qdata, sizeof(struct qunit_data));
2048 rc = quota_get_qdata(req, qdata, QUOTA_REQUEST, QUOTA_EXPORT);
2050 CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc);
2054 /* we use the observer */
2055 if (!obd->obd_observer || !obd->obd_observer->obd_observer) {
2056 CERROR("Can't find the observer, it is recovering\n");
2057 req->rq_status = -EAGAIN;
2058 GOTO(send_reply, rc = -EAGAIN);
2061 master_obd = obd->obd_observer->obd_observer;
2062 qctxt = &master_obd->u.obt.obt_qctxt;
2064 if (!qctxt->lqc_setup) {
2065 /* quota_type has not been processed yet, return EAGAIN
2066 * until we know whether or not quotas are supposed to
2068 CDEBUG(D_QUOTA, "quota_type not processed yet, return "
2070 req->rq_status = -EAGAIN;
2071 rc = ptlrpc_reply(req);
2075 LASSERT(qctxt->lqc_handler);
2076 rc = qctxt->lqc_handler(master_obd, qdata,
2077 lustre_msg_get_opc(req->rq_reqmsg));
2078 if (rc && rc != -EDQUOT)
2079 CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR,
2080 "dqacq failed! (rc:%d)\n", rc);
2081 req->rq_status = rc;
2083 /* there are three forms of qunit(historic causes), so we need to
2084 * adjust the same form to different forms slaves needed */
2085 rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT);
2087 CDEBUG(D_ERROR, "Can't pack qunit_data(rc: %d)\n", rc);
2091 /* Block the quota req. b=14840 */
2092 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout);
2094 rc = ptlrpc_reply(req);
2096 OBD_FREE(qdata, sizeof(struct qunit_data));
2100 #endif /* !__KERNEL__ */
2102 #endif /* HAVE_QUOTA_SUPPORT */
2104 ldlm_mode_t lck_compat_array[] = {
2105 [LCK_EX] LCK_COMPAT_EX,
2106 [LCK_PW] LCK_COMPAT_PW,
2107 [LCK_PR] LCK_COMPAT_PR,
2108 [LCK_CW] LCK_COMPAT_CW,
2109 [LCK_CR] LCK_COMPAT_CR,
2110 [LCK_NL] LCK_COMPAT_NL,
2111 [LCK_GROUP] LCK_COMPAT_GROUP