1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2003 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
26 # define EXPORT_SYMTAB
28 #define DEBUG_SUBSYSTEM S_LDLM
31 # include <libcfs/libcfs.h>
33 # include <liblustre.h>
36 #include <lustre_mds.h>
37 #include <lustre_dlm.h>
38 #include <lustre_net.h>
39 #include <lustre_sec.h>
40 #include "ldlm_internal.h"
43 /* @priority: if non-zero, move the selected to the list head
44 * @create: if zero, only search in existed connections
46 static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid,
47 int priority, int create)
49 struct ptlrpc_connection *ptlrpc_conn;
50 struct obd_import_conn *imp_conn = NULL, *item;
54 if (!create && !priority) {
55 CDEBUG(D_HA, "Nothing to do\n");
59 ptlrpc_conn = ptlrpc_uuid_to_connection(uuid);
61 CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid);
66 OBD_ALLOC(imp_conn, sizeof(*imp_conn));
68 GOTO(out_put, rc = -ENOMEM);
72 spin_lock(&imp->imp_lock);
73 list_for_each_entry(item, &imp->imp_conn_list, oic_item) {
74 if (obd_uuid_equals(uuid, &item->oic_uuid)) {
76 list_del(&item->oic_item);
77 list_add(&item->oic_item, &imp->imp_conn_list);
78 item->oic_last_attempt = 0;
80 CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n",
81 imp, imp->imp_obd->obd_name, uuid->uuid,
82 (priority ? ", moved to head" : ""));
83 spin_unlock(&imp->imp_lock);
84 GOTO(out_free, rc = 0);
89 imp_conn->oic_conn = ptlrpc_conn;
90 imp_conn->oic_uuid = *uuid;
91 item->oic_last_attempt = 0;
93 list_add(&imp_conn->oic_item, &imp->imp_conn_list);
95 list_add_tail(&imp_conn->oic_item, &imp->imp_conn_list);
96 CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n",
97 imp, imp->imp_obd->obd_name, uuid->uuid,
98 (priority ? "head" : "tail"));
100 spin_unlock(&imp->imp_lock);
101 GOTO(out_free, rc = -ENOENT);
104 spin_unlock(&imp->imp_lock);
108 OBD_FREE(imp_conn, sizeof(*imp_conn));
110 ptlrpc_put_connection(ptlrpc_conn);
114 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid)
116 return import_set_conn(imp, uuid, 1, 0);
119 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
122 return import_set_conn(imp, uuid, priority, 1);
125 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
127 struct obd_import_conn *imp_conn;
128 struct obd_export *dlmexp;
132 spin_lock(&imp->imp_lock);
133 if (list_empty(&imp->imp_conn_list)) {
134 LASSERT(!imp->imp_connection);
138 list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) {
139 if (!obd_uuid_equals(uuid, &imp_conn->oic_uuid))
141 LASSERT(imp_conn->oic_conn);
143 /* is current conn? */
144 if (imp_conn == imp->imp_conn_current) {
145 LASSERT(imp_conn->oic_conn == imp->imp_connection);
147 if (imp->imp_state != LUSTRE_IMP_CLOSED &&
148 imp->imp_state != LUSTRE_IMP_DISCON) {
149 CERROR("can't remove current connection\n");
150 GOTO(out, rc = -EBUSY);
153 ptlrpc_put_connection(imp->imp_connection);
154 imp->imp_connection = NULL;
156 dlmexp = class_conn2export(&imp->imp_dlm_handle);
157 if (dlmexp && dlmexp->exp_connection) {
158 LASSERT(dlmexp->exp_connection ==
160 ptlrpc_put_connection(dlmexp->exp_connection);
161 dlmexp->exp_connection = NULL;
165 list_del(&imp_conn->oic_item);
166 ptlrpc_put_connection(imp_conn->oic_conn);
167 OBD_FREE(imp_conn, sizeof(*imp_conn));
168 CDEBUG(D_HA, "imp %p@%s: remove connection %s\n",
169 imp, imp->imp_obd->obd_name, uuid->uuid);
174 spin_unlock(&imp->imp_lock);
176 CERROR("connection %s not found\n", uuid->uuid);
180 static void destroy_import(struct obd_import *imp)
182 /* drop security policy instance after all rpc finished/aborted
183 * to let all busy contexts be released. */
184 class_import_get(imp);
185 class_destroy_import(imp);
186 sptlrpc_import_sec_put(imp);
187 class_import_put(imp);
190 /* configure an RPC client OBD device
195 * 3 - inactive-on-startup
197 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
199 struct client_obd *cli = &obddev->u.cli;
200 struct obd_import *imp;
201 struct obd_uuid server_uuid;
202 int rq_portal, rp_portal, connect_op;
203 char *name = obddev->obd_type->typ_name;
207 /* In a more perfect world, we would hang a ptlrpc_client off of
208 * obd_type and just use the values from there. */
209 if (!strcmp(name, LUSTRE_OSC_NAME)) {
210 rq_portal = OST_REQUEST_PORTAL;
211 rp_portal = OSC_REPLY_PORTAL;
212 connect_op = OST_CONNECT;
213 } else if (!strcmp(name, LUSTRE_MDC_NAME)) {
214 rq_portal = MDS_REQUEST_PORTAL;
215 rp_portal = MDC_REPLY_PORTAL;
216 connect_op = MDS_CONNECT;
217 } else if (!strcmp(name, LUSTRE_MGC_NAME)) {
218 rq_portal = MGS_REQUEST_PORTAL;
219 rp_portal = MGC_REPLY_PORTAL;
220 connect_op = MGS_CONNECT;
222 CERROR("unknown client OBD type \"%s\", can't setup\n",
227 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
228 CERROR("requires a TARGET UUID\n");
232 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) {
233 CERROR("client UUID must be less than 38 characters\n");
237 if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
238 CERROR("setup requires a SERVER UUID\n");
242 if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) {
243 CERROR("target UUID must be less than 38 characters\n");
247 sema_init(&cli->cl_sem, 1);
248 sema_init(&cli->cl_mgc_sem, 1);
249 sptlrpc_rule_set_init(&cli->cl_sptlrpc_rset);
250 cli->cl_sec_part = LUSTRE_SP_ANY;
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_write_list);
265 CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list);
266 client_obd_list_lock_init(&cli->cl_loi_list_lock);
267 cli->cl_r_in_flight = 0;
268 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 cli->cl_checksum = 1;
281 atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
283 /* This value may be changed at connect time in
284 ptlrpc_connect_interpret. */
285 cli->cl_max_pages_per_rpc = min((int)PTLRPC_MAX_BRW_PAGES,
286 (int)(1024 * 1024 >> CFS_PAGE_SHIFT));
288 if (!strcmp(name, LUSTRE_MDC_NAME)) {
289 cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
290 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 128 /* MB */) {
291 cli->cl_max_rpcs_in_flight = 2;
292 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 256 /* MB */) {
293 cli->cl_max_rpcs_in_flight = 3;
294 } else if (num_physpages >> (20 - CFS_PAGE_SHIFT) <= 512 /* MB */) {
295 cli->cl_max_rpcs_in_flight = 4;
297 cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
302 CERROR("ldlm_get_ref failed: %d\n", rc);
306 ptlrpc_init_client(rq_portal, rp_portal, name,
307 &obddev->obd_ldlm_client);
309 imp = class_new_import(obddev);
311 GOTO(err_ldlm, rc = -ENOENT);
312 imp->imp_client = &obddev->obd_ldlm_client;
313 imp->imp_connect_op = connect_op;
314 imp->imp_initial_recov = 1;
315 imp->imp_initial_recov_bk = 0;
316 CFS_INIT_LIST_HEAD(&imp->imp_pinger_chain);
317 memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
318 LUSTRE_CFG_BUFLEN(lcfg, 1));
319 class_import_put(imp);
321 rc = client_import_add_conn(imp, &server_uuid, 1);
323 CERROR("can't add initial connection\n");
324 GOTO(err_import, rc);
327 cli->cl_import = imp;
328 /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
329 cli->cl_max_mds_easize = sizeof(struct lov_mds_md);
330 cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
332 if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
333 if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) {
334 CDEBUG(D_HA, "marking %s %s->%s as inactive\n",
335 name, obddev->obd_name,
336 cli->cl_target_uuid.uuid);
337 spin_lock(&imp->imp_lock);
338 imp->imp_invalid = 1;
339 spin_unlock(&imp->imp_lock);
343 cli->cl_qchk_stat = CL_NOT_QUOTACHECKED;
348 class_destroy_import(imp);
356 int client_obd_cleanup(struct obd_device *obddev)
359 sptlrpc_rule_set_free(&obddev->u.cli.cl_sptlrpc_rset);
360 ldlm_put_ref(obddev->obd_force);
364 /* ->o_connect() method for client side (OSC and MDC and MGC) */
365 int client_connect_import(const struct lu_env *env,
366 struct lustre_handle *dlm_handle,
367 struct obd_device *obd, struct obd_uuid *cluuid,
368 struct obd_connect_data *data)
370 struct client_obd *cli = &obd->u.cli;
371 struct obd_import *imp = cli->cl_import;
372 struct obd_export *exp;
373 struct obd_connect_data *ocd;
374 struct ldlm_namespace *to_be_freed = NULL;
378 mutex_down(&cli->cl_sem);
379 rc = class_connect(dlm_handle, obd, cluuid);
383 cli->cl_conn_count++;
384 if (cli->cl_conn_count > 1)
386 exp = class_conn2export(dlm_handle);
388 if (obd->obd_namespace != NULL)
389 CERROR("already have namespace!\n");
390 obd->obd_namespace = ldlm_namespace_new(obd->obd_name,
391 LDLM_NAMESPACE_CLIENT,
392 LDLM_NAMESPACE_GREEDY);
393 if (obd->obd_namespace == NULL)
394 GOTO(out_disco, rc = -ENOMEM);
396 imp->imp_dlm_handle = *dlm_handle;
397 rc = ptlrpc_init_import(imp);
401 ocd = &imp->imp_connect_data;
404 imp->imp_connect_flags_orig = data->ocd_connect_flags;
407 rc = ptlrpc_connect_import(imp, NULL);
409 LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
412 LASSERT(exp->exp_connection);
415 LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
416 ocd->ocd_connect_flags, "old "LPX64", new "LPX64"\n",
417 data->ocd_connect_flags, ocd->ocd_connect_flags);
418 data->ocd_connect_flags = ocd->ocd_connect_flags;
421 ptlrpc_pinger_add_import(imp);
427 ldlm_namespace_free_prior(obd->obd_namespace);
428 to_be_freed = obd->obd_namespace;
429 obd->obd_namespace = NULL;
431 cli->cl_conn_count--;
432 class_disconnect(exp);
434 class_export_put(exp);
437 mutex_up(&cli->cl_sem);
439 ldlm_namespace_free_post(to_be_freed, 1);
444 int client_disconnect_export(struct obd_export *exp)
446 struct obd_device *obd = class_exp2obd(exp);
447 struct client_obd *cli;
448 struct obd_import *imp;
450 struct ldlm_namespace *to_be_freed = NULL;
454 CERROR("invalid export for disconnect: exp %p cookie "LPX64"\n",
455 exp, exp ? exp->exp_handle.h_cookie : -1);
460 imp = cli->cl_import;
462 mutex_down(&cli->cl_sem);
463 if (!cli->cl_conn_count) {
464 CERROR("disconnecting disconnected device (%s)\n",
466 GOTO(out_sem, rc = -EINVAL);
469 cli->cl_conn_count--;
470 if (cli->cl_conn_count)
471 GOTO(out_no_disconnect, rc = 0);
473 /* Mark import deactivated now, so we don't try to reconnect if any
474 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
475 * fully deactivate the import, or that would drop all requests. */
476 spin_lock(&imp->imp_lock);
477 imp->imp_deactive = 1;
478 spin_unlock(&imp->imp_lock);
480 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
481 * delete it regardless. (It's safe to delete an import that was
483 (void)ptlrpc_pinger_del_import(imp);
485 if (obd->obd_namespace != NULL) {
486 /* obd_force == local only */
487 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
488 obd->obd_force ? LDLM_FL_LOCAL_ONLY:0,
490 ldlm_namespace_free_prior(obd->obd_namespace);
491 to_be_freed = obd->obd_namespace;
494 rc = ptlrpc_disconnect_import(imp, 0);
496 ptlrpc_invalidate_import(imp);
497 /* set obd_namespace to NULL only after invalidate, because we can have
498 * some connect requests in flight, and his need store a connect flags
499 * in obd_namespace. bug 14260 */
500 obd->obd_namespace = NULL;
502 ptlrpc_free_rq_pool(imp->imp_rq_pool);
504 cli->cl_import = NULL;
508 err = class_disconnect(exp);
512 mutex_up(&cli->cl_sem);
514 ldlm_namespace_free_post(to_be_freed, obd->obd_force);
519 /* --------------------------------------------------------------------------
520 * from old lib/target.c
521 * -------------------------------------------------------------------------- */
523 int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
524 struct obd_uuid *cluuid, int initial_conn)
527 if (exp->exp_connection && exp->exp_imp_reverse && !initial_conn) {
528 struct lustre_handle *hdl;
529 hdl = &exp->exp_imp_reverse->imp_remote_handle;
530 /* Might be a re-connect after a partition. */
531 if (!memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) {
532 CWARN("%s: %s reconnecting\n", exp->exp_obd->obd_name,
534 conn->cookie = exp->exp_handle.h_cookie;
535 /* target_handle_connect() treats EALREADY and
536 * -EALREADY differently. EALREADY means we are
537 * doing a valid reconnect from the same client. */
540 CERROR("%s reconnecting from %s, "
541 "handle mismatch (ours "LPX64", theirs "
542 LPX64")\n", cluuid->uuid,
543 exp->exp_connection->c_remote_uuid.uuid,
544 hdl->cookie, conn->cookie);
545 memset(conn, 0, sizeof *conn);
546 /* target_handle_connect() treats EALREADY and
547 * -EALREADY differently. -EALREADY is an error
548 * (same UUID, different handle). */
553 conn->cookie = exp->exp_handle.h_cookie;
554 CDEBUG(D_HA, "connect export for UUID '%s' at %p, cookie "LPX64"\n",
555 cluuid->uuid, exp, conn->cookie);
559 void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data,
562 struct obd_export *exp = cb_data;
564 CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n",
565 obd->obd_name, exp->exp_client_uuid.uuid);
567 spin_lock(&exp->exp_lock);
568 exp->exp_need_sync = 0;
569 spin_unlock(&exp->exp_lock);
571 EXPORT_SYMBOL(target_client_add_cb);
573 int target_handle_connect(struct ptlrpc_request *req)
575 struct obd_device *target, *targref = NULL;
576 struct obd_export *export = NULL;
577 struct obd_import *revimp;
578 struct lustre_handle conn;
579 struct lustre_handle *tmp;
580 struct obd_uuid tgtuuid;
581 struct obd_uuid cluuid;
582 struct obd_uuid remote_uuid;
585 int initial_conn = 0;
586 struct obd_connect_data *data, *tmpdata;
589 OBD_RACE(OBD_FAIL_TGT_CONN_RACE);
591 str = req_capsule_client_get(&req->rq_pill, &RMF_TGTUUID);
593 DEBUG_REQ(D_ERROR, req, "bad target UUID for connect");
594 GOTO(out, rc = -EINVAL);
597 obd_str2uuid(&tgtuuid, str);
598 target = class_uuid2obd(&tgtuuid);
600 target = class_name2obd(str);
602 if (!target || target->obd_stopping || !target->obd_set_up) {
603 LCONSOLE_ERROR_MSG(0x137, "UUID '%s' is not available "
604 " for connect (%s)\n", str,
605 !target ? "no target" :
606 (target->obd_stopping ? "stopping" :
608 GOTO(out, rc = -ENODEV);
611 if (target->obd_no_conn) {
612 LCONSOLE_WARN("%s: temporarily refusing client connection "
613 "from %s\n", target->obd_name,
614 libcfs_nid2str(req->rq_peer.nid));
615 GOTO(out, rc = -EAGAIN);
618 /* Make sure the target isn't cleaned up while we're here. Yes,
619 there's still a race between the above check and our incref here.
620 Really, class_uuid2obd should take the ref. */
621 targref = class_incref(target);
624 str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID);
626 DEBUG_REQ(D_ERROR, req, "bad client UUID for connect");
627 GOTO(out, rc = -EINVAL);
630 obd_str2uuid(&cluuid, str);
632 /* XXX extract a nettype and format accordingly */
633 switch (sizeof(lnet_nid_t)) {
634 /* NB the casts only avoid compiler warnings */
636 snprintf(remote_uuid.uuid, sizeof remote_uuid,
637 "NET_"LPX64"_UUID", (__u64)req->rq_peer.nid);
640 snprintf(remote_uuid.uuid, sizeof remote_uuid,
641 "NET_%x_UUID", (__u32)req->rq_peer.nid);
647 tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
649 GOTO(out, rc = -EPROTO);
653 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
655 GOTO(out, rc = -EPROTO);
657 rc = req_capsule_server_pack(&req->rq_pill);
661 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
663 DEBUG_REQ(D_WARNING, req, "Refusing old (unversioned) "
664 "libclient connection attempt");
665 GOTO(out, rc = -EPROTO);
666 } else if (data->ocd_version < LUSTRE_VERSION_CODE -
667 LUSTRE_VERSION_ALLOWED_OFFSET ||
668 data->ocd_version > LUSTRE_VERSION_CODE +
669 LUSTRE_VERSION_ALLOWED_OFFSET) {
670 DEBUG_REQ(D_WARNING, req, "Refusing %s (%d.%d.%d.%d) "
671 "libclient connection attempt",
672 data->ocd_version < LUSTRE_VERSION_CODE ?
674 OBD_OCD_VERSION_MAJOR(data->ocd_version),
675 OBD_OCD_VERSION_MINOR(data->ocd_version),
676 OBD_OCD_VERSION_PATCH(data->ocd_version),
677 OBD_OCD_VERSION_FIX(data->ocd_version));
678 data = req_capsule_server_sized_get(&req->rq_pill,
680 offsetof(typeof(*data), ocd_version) +
681 sizeof(data->ocd_version));
683 data->ocd_connect_flags = OBD_CONNECT_VERSION;
684 data->ocd_version = LUSTRE_VERSION_CODE;
686 GOTO(out, rc = -EPROTO);
690 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL)
693 /* lctl gets a backstage, all-access pass. */
694 if (obd_uuid_equals(&cluuid, &target->obd_uuid))
695 goto dont_check_exports;
697 spin_lock(&target->obd_dev_lock);
698 export = lustre_hash_get_object_by_key(target->obd_uuid_hash_body, &cluuid);
700 if (export != NULL && export->exp_connecting) { /* bug 9635, et. al. */
701 CWARN("%s: exp %p already connecting\n",
702 export->exp_obd->obd_name, export);
703 class_export_put(export);
706 } else if (export != NULL && export->exp_connection != NULL &&
707 req->rq_peer.nid != export->exp_connection->c_peer.nid) {
708 /* make darn sure this is coming from the same peer
709 * if the UUIDs matched */
710 CWARN("%s: cookie %s seen on new NID %s when "
711 "existing NID %s is already connected\n",
712 target->obd_name, cluuid.uuid,
713 libcfs_nid2str(req->rq_peer.nid),
714 libcfs_nid2str(export->exp_connection->c_peer.nid));
715 class_export_put(export);
718 } else if (export != NULL) {
719 spin_lock(&export->exp_lock);
720 export->exp_connecting = 1;
721 spin_unlock(&export->exp_lock);
722 class_export_put(export);
723 spin_unlock(&target->obd_dev_lock);
724 LASSERT(export->exp_obd == target);
726 rc = target_handle_reconnect(&conn, export, &cluuid, initial_conn);
729 /* If we found an export, we already unlocked. */
731 spin_unlock(&target->obd_dev_lock);
732 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout);
733 } else if (req->rq_export == NULL &&
734 atomic_read(&export->exp_rpc_count) > 0) {
735 CWARN("%s: refuse connection from %s/%s to 0x%p/%d\n",
736 target->obd_name, cluuid.uuid,
737 libcfs_nid2str(req->rq_peer.nid),
738 export, atomic_read(&export->exp_refcount));
739 GOTO(out, rc = -EBUSY);
740 } else if (req->rq_export != NULL &&
741 (atomic_read(&export->exp_rpc_count) > 1)) {
742 CWARN("%s: refuse reconnection from %s@%s to 0x%p/%d\n",
743 target->obd_name, cluuid.uuid,
744 libcfs_nid2str(req->rq_peer.nid),
745 export, atomic_read(&export->exp_rpc_count));
746 GOTO(out, rc = -EBUSY);
747 } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 &&
749 CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; "
750 "cookies not random?\n", target->obd_name,
751 libcfs_nid2str(req->rq_peer.nid), cluuid.uuid);
752 GOTO(out, rc = -EALREADY);
754 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
755 if (req->rq_export == NULL && initial_conn)
756 export->exp_last_request_time =
757 max(export->exp_last_request_time,
758 (time_t)CURRENT_SECONDS);
761 /* We want to handle EALREADY but *not* -EALREADY from
762 * target_handle_reconnect(), return reconnection state in a flag */
763 if (rc == EALREADY) {
764 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
769 /* Tell the client if we're in recovery. */
770 /* If this is the first client, start the recovery timer */
771 CWARN("%s: connection from %s@%s %st"LPU64" exp %p cur %ld last %ld\n",
772 target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
773 target->obd_recovering ? "recovering/" : "", data->ocd_transno,
774 export, (long)CURRENT_SECONDS,
775 export ? (long)export->exp_last_request_time : 0);
778 if (target->obd_recovering) {
779 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
780 target_start_recovery_timer(target);
783 /* Tell the client if we support replayable requests */
784 if (target->obd_replayable)
785 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_REPLAYABLE);
787 if (export == NULL) {
788 if (target->obd_recovering) {
791 t = cfs_timer_deadline(&target->obd_recovery_timer);
792 t = cfs_time_sub(t, cfs_time_current());
793 CERROR("%s: denying connection for new client %s (%s): "
794 "%d clients in recovery for %lds\n",
796 libcfs_nid2str(req->rq_peer.nid), cluuid.uuid,
797 target->obd_recoverable_clients,
798 cfs_duration_sec(t));
802 rc = obd_connect(req->rq_svc_thread->t_env,
803 &conn, target, &cluuid, data);
806 rc = obd_reconnect(req->rq_svc_thread->t_env,
807 export, target, &cluuid, data);
811 /* Return only the parts of obd_connect_data that we understand, so the
812 * client knows that we don't understand the rest. */
814 tmpdata = req_capsule_server_get(&req->rq_pill,
816 //data->ocd_connect_flags &= OBD_CONNECT_SUPPORTED;
820 /* If all else goes well, this is our RPC return code. */
823 lustre_msg_set_handle(req->rq_repmsg, &conn);
825 /* ownership of this export ref transfers to the request AFTER we
826 * drop any previous reference the request had, but we don't want
827 * that to go to zero before we get our new export reference. */
828 export = class_conn2export(&conn);
830 DEBUG_REQ(D_ERROR, req, "Missing export!");
831 GOTO(out, rc = -ENODEV);
834 /* If the client and the server are the same node, we will already
835 * have an export that really points to the client's DLM export,
836 * because we have a shared handles table.
838 * XXX this will go away when shaver stops sending the "connect" handle
839 * in the real "remote handle" field of the request --phik 24 Apr 2003
841 if (req->rq_export != NULL)
842 class_export_put(req->rq_export);
844 req->rq_export = export;
846 spin_lock(&export->exp_lock);
848 lustre_msg_set_conn_cnt(req->rq_repmsg, export->exp_conn_cnt + 1);
849 } else if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
850 spin_unlock(&export->exp_lock);
851 CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n",
852 cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
853 export->exp_conn_cnt,
854 lustre_msg_get_conn_cnt(req->rq_reqmsg));
856 GOTO(out, rc = -EALREADY);
858 export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
860 /* request from liblustre? Don't evict it for not pinging. */
861 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
862 export->exp_libclient = 1;
863 spin_unlock(&export->exp_lock);
865 spin_lock(&target->obd_dev_lock);
866 list_del_init(&export->exp_obd_chain_timed);
867 spin_unlock(&target->obd_dev_lock);
869 spin_unlock(&export->exp_lock);
872 if (export->exp_connection != NULL)
873 ptlrpc_put_connection(export->exp_connection);
874 export->exp_connection = ptlrpc_get_connection(req->rq_peer,
878 spin_lock(&target->obd_dev_lock);
879 /* Export might be hashed already, e.g. if this is reconnect */
880 if (hlist_unhashed(&export->exp_nid_hash))
881 lustre_hash_additem(export->exp_obd->obd_nid_hash_body,
882 &export->exp_connection->c_peer.nid,
883 &export->exp_nid_hash);
884 spin_unlock(&target->obd_dev_lock);
886 spin_lock_bh(&target->obd_processing_task_lock);
887 if (target->obd_recovering && !export->exp_in_recovery) {
888 spin_lock(&export->exp_lock);
889 export->exp_in_recovery = 1;
890 export->exp_req_replay_needed = 1;
891 export->exp_lock_replay_needed = 1;
892 spin_unlock(&export->exp_lock);
893 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
894 && (data->ocd_transno == 0))
895 CWARN("Connect with zero transno!\n");
897 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
898 && data->ocd_transno < target->obd_next_recovery_transno)
899 target->obd_next_recovery_transno = data->ocd_transno;
900 target->obd_connected_clients++;
901 /* each connected client is counted as recoverable */
902 target->obd_recoverable_clients++;
903 atomic_inc(&target->obd_req_replay_clients);
904 atomic_inc(&target->obd_lock_replay_clients);
905 if (target->obd_connected_clients ==
906 target->obd_max_recoverable_clients)
907 wake_up(&target->obd_next_transno_waitq);
909 spin_unlock_bh(&target->obd_processing_task_lock);
910 tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
913 if (export->exp_imp_reverse != NULL) {
914 /* destroyed import can be still referenced in ctxt */
915 obd_set_info_async(export, strlen(KEY_REVIMP_UPD),
916 KEY_REVIMP_UPD, 0, NULL, NULL);
918 /* in some recovery senarios, previous ctx init rpc handled
919 * in sptlrpc_target_export_check() might be used to install
920 * a reverse ctx in this reverse import, and later OBD_CONNECT
921 * using the same gss ctx could reach here and following new
922 * reverse import. note all reverse ctx in new/old import are
923 * actually based on the same gss ctx. so we invalidate ctx
924 * here before destroy import, otherwise flush old import will
925 * lead to remote reverse ctx be destroied, thus the reverse
926 * ctx of new import will lost its peer.
927 * there might be a better way to deal with this???
929 sptlrpc_import_inval_all_ctx(export->exp_imp_reverse);
931 destroy_import(export->exp_imp_reverse);
934 /* for the rest part, we return -ENOTCONN in case of errors
935 * in order to let client initialize connection again.
937 revimp = export->exp_imp_reverse = class_new_import(target);
939 CERROR("fail to alloc new reverse import.\n");
940 GOTO(out, rc = -ENOTCONN);
943 revimp->imp_connection = ptlrpc_connection_addref(export->exp_connection);
944 revimp->imp_client = &export->exp_obd->obd_ldlm_client;
945 revimp->imp_remote_handle = conn;
946 revimp->imp_dlm_fake = 1;
947 revimp->imp_state = LUSTRE_IMP_FULL;
949 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_NEXT_VER) {
950 revimp->imp_msg_magic = LUSTRE_MSG_MAGIC_V2;
951 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_NEXT_VER);
954 rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx,
955 req->rq_flvr.sf_rpc);
957 CERROR("Failed to get sec for reverse import: %d\n", rc);
958 export->exp_imp_reverse = NULL;
959 class_destroy_import(revimp);
962 class_import_put(revimp);
965 spin_lock(&export->exp_lock);
966 export->exp_connecting = 0;
967 spin_unlock(&export->exp_lock);
970 class_decref(targref);
976 int target_handle_disconnect(struct ptlrpc_request *req)
981 rc = req_capsule_server_pack(&req->rq_pill);
985 /* keep the rq_export around so we can send the reply */
986 req->rq_status = obd_disconnect(class_export_get(req->rq_export));
991 void target_destroy_export(struct obd_export *exp)
993 /* exports created from last_rcvd data, and "fake"
994 exports created by lctl don't have an import */
995 if (exp->exp_imp_reverse != NULL)
996 destroy_import(exp->exp_imp_reverse);
998 /* We cancel locks at disconnect time, but this will catch any locks
999 * granted in a race with recovery-induced disconnect. */
1000 if (exp->exp_obd->obd_namespace != NULL)
1001 ldlm_cancel_locks_for_export(exp);
1005 * Recovery functions
1008 struct ptlrpc_request *ptlrpc_clone_req( struct ptlrpc_request *orig_req)
1010 struct ptlrpc_request *copy_req;
1011 struct lustre_msg *copy_reqmsg;
1012 struct ptlrpc_user_desc *udesc = NULL;
1014 OBD_ALLOC_PTR(copy_req);
1017 OBD_ALLOC(copy_reqmsg, orig_req->rq_reqlen);
1019 OBD_FREE_PTR(copy_req);
1023 if (orig_req->rq_user_desc) {
1024 int ngroups = orig_req->rq_user_desc->pud_ngroups;
1026 OBD_ALLOC(udesc, sptlrpc_user_desc_size(ngroups));
1028 OBD_FREE(copy_reqmsg, orig_req->rq_reqlen);
1029 OBD_FREE_PTR(copy_req);
1032 memcpy(udesc, orig_req->rq_user_desc,
1033 sptlrpc_user_desc_size(ngroups));
1036 *copy_req = *orig_req;
1037 memcpy(copy_reqmsg, orig_req->rq_reqmsg, orig_req->rq_reqlen);
1038 copy_req->rq_reqmsg = copy_reqmsg;
1039 copy_req->rq_user_desc = udesc;
1041 class_export_get(copy_req->rq_export);
1042 CFS_INIT_LIST_HEAD(©_req->rq_list);
1043 sptlrpc_svc_ctx_addref(copy_req);
1045 if (copy_req->rq_reply_state) {
1046 /* the copied req takes over the reply state */
1047 orig_req->rq_reply_state = NULL;
1048 /* to catch further access */
1049 orig_req->rq_repmsg = NULL;
1050 orig_req->rq_replen = 0;
1056 void ptlrpc_free_clone( struct ptlrpc_request *req)
1058 if (req->rq_reply_state) {
1059 ptlrpc_rs_decref(req->rq_reply_state);
1060 req->rq_reply_state = NULL;
1063 sptlrpc_svc_ctx_decref(req);
1064 class_export_put(req->rq_export);
1065 list_del(&req->rq_list);
1067 if (req->rq_user_desc) {
1068 int ngroups = req->rq_user_desc->pud_ngroups;
1069 OBD_FREE(req->rq_user_desc, sptlrpc_user_desc_size(ngroups));
1071 OBD_FREE(req->rq_reqmsg, req->rq_reqlen);
1076 static void target_finish_recovery(struct obd_device *obd)
1079 LCONSOLE_INFO("%s: sending delayed replies to recovered clients\n",
1082 ldlm_reprocess_all_ns(obd->obd_namespace);
1084 /* when recovery finished, cleanup orphans on mds and ost */
1085 if (OBT(obd) && OBP(obd, postrecov)) {
1086 int rc = OBP(obd, postrecov)(obd);
1087 LCONSOLE_WARN("%s: recovery %s: rc %d\n", obd->obd_name,
1088 rc < 0 ? "failed" : "complete", rc);
1091 obd->obd_recovery_end = CURRENT_SECONDS;
1095 static void abort_req_replay_queue(struct obd_device *obd)
1097 struct ptlrpc_request *req, *n;
1099 list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) {
1100 DEBUG_REQ(D_WARNING, req, "aborted:");
1101 req->rq_status = -ENOTCONN;
1102 if (ptlrpc_error(req)) {
1103 DEBUG_REQ(D_ERROR, req,
1104 "failed abort_req_reply; skipping");
1106 ptlrpc_free_clone(req);
1110 static void abort_lock_replay_queue(struct obd_device *obd)
1112 struct ptlrpc_request *req, *n;
1114 list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){
1115 DEBUG_REQ(D_ERROR, req, "aborted:");
1116 req->rq_status = -ENOTCONN;
1117 if (ptlrpc_error(req)) {
1118 DEBUG_REQ(D_ERROR, req,
1119 "failed abort_lock_reply; skipping");
1121 ptlrpc_free_clone(req);
1126 /* Called from a cleanup function if the device is being cleaned up
1127 forcefully. The exports should all have been disconnected already,
1128 the only thing left to do is
1129 - clear the recovery flags
1131 - free queued requests and replies, but don't send replies
1132 Because the obd_stopping flag is set, no new requests should be received.
1135 void target_cleanup_recovery(struct obd_device *obd)
1137 struct ptlrpc_request *req, *n;
1140 LASSERT(obd->obd_stopping);
1142 spin_lock_bh(&obd->obd_processing_task_lock);
1143 if (!obd->obd_recovering) {
1144 spin_unlock_bh(&obd->obd_processing_task_lock);
1148 obd->obd_recovering = obd->obd_abort_recovery = 0;
1149 target_cancel_recovery_timer(obd);
1150 spin_unlock_bh(&obd->obd_processing_task_lock);
1152 list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) {
1153 LASSERT (req->rq_reply_state == 0);
1154 ptlrpc_free_clone(req);
1156 list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){
1157 LASSERT (req->rq_reply_state == 0);
1158 ptlrpc_free_clone(req);
1160 list_for_each_entry_safe(req, n, &obd->obd_final_req_queue, rq_list) {
1161 LASSERT (req->rq_reply_state == 0);
1162 ptlrpc_free_clone(req);
1168 static void target_recovery_expired(unsigned long castmeharder)
1170 struct obd_device *obd = (struct obd_device *)castmeharder;
1171 CERROR("%s: recovery timed out, aborting\n", obd->obd_name);
1172 spin_lock_bh(&obd->obd_processing_task_lock);
1173 if (obd->obd_recovering)
1174 obd->obd_abort_recovery = 1;
1175 cfs_waitq_signal(&obd->obd_next_transno_waitq);
1176 spin_unlock_bh(&obd->obd_processing_task_lock);
1180 /* obd_processing_task_lock should be held */
1181 void target_cancel_recovery_timer(struct obd_device *obd)
1183 CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
1184 cfs_timer_disarm(&obd->obd_recovery_timer);
1187 static void reset_recovery_timer(struct obd_device *obd)
1189 time_t timeout_shift = OBD_RECOVERY_TIMEOUT;
1190 spin_lock_bh(&obd->obd_processing_task_lock);
1191 if (!obd->obd_recovering) {
1192 spin_unlock_bh(&obd->obd_processing_task_lock);
1195 if (cfs_time_current_sec() + OBD_RECOVERY_TIMEOUT >
1196 obd->obd_recovery_start + obd->obd_recovery_max_time)
1197 timeout_shift = obd->obd_recovery_start +
1198 obd->obd_recovery_max_time - cfs_time_current_sec();
1199 cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(timeout_shift));
1200 spin_unlock_bh(&obd->obd_processing_task_lock);
1201 CDEBUG(D_HA, "%s: timer will expire in %u seconds\n", obd->obd_name,
1202 (unsigned int)timeout_shift);
1203 /* Only used for lprocfs_status */
1204 obd->obd_recovery_end = CURRENT_SECONDS + timeout_shift;
1208 /* Only start it the first time called */
1209 void target_start_recovery_timer(struct obd_device *obd)
1211 spin_lock_bh(&obd->obd_processing_task_lock);
1212 if (obd->obd_recovery_handler
1213 || timer_pending((struct timer_list *)&obd->obd_recovery_timer)) {
1214 spin_unlock_bh(&obd->obd_processing_task_lock);
1217 CWARN("%s: starting recovery timer (%us)\n", obd->obd_name,
1218 OBD_RECOVERY_TIMEOUT);
1219 cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd);
1220 spin_unlock_bh(&obd->obd_processing_task_lock);
1222 reset_recovery_timer(obd);
1226 static int check_for_next_transno(struct obd_device *obd)
1228 struct ptlrpc_request *req = NULL;
1229 int wake_up = 0, connected, completed, queue_len, max;
1230 __u64 next_transno, req_transno;
1232 spin_lock_bh(&obd->obd_processing_task_lock);
1234 if (!list_empty(&obd->obd_req_replay_queue)) {
1235 req = list_entry(obd->obd_req_replay_queue.next,
1236 struct ptlrpc_request, rq_list);
1237 req_transno = lustre_msg_get_transno(req->rq_reqmsg);
1242 max = obd->obd_max_recoverable_clients;
1243 connected = obd->obd_connected_clients;
1244 completed = connected - obd->obd_recoverable_clients;
1245 queue_len = obd->obd_requests_queued_for_recovery;
1246 next_transno = obd->obd_next_recovery_transno;
1248 CDEBUG(D_HA, "max: %d, connected: %d, completed: %d, queue_len: %d, "
1249 "req_transno: "LPU64", next_transno: "LPU64"\n",
1250 max, connected, completed, queue_len, req_transno, next_transno);
1252 if (obd->obd_abort_recovery) {
1253 CDEBUG(D_HA, "waking for aborted recovery\n");
1255 } else if (atomic_read(&obd->obd_req_replay_clients) == 0) {
1256 CDEBUG(D_HA, "waking for completed recovery\n");
1258 } else if (req_transno == next_transno) {
1259 CDEBUG(D_HA, "waking for next ("LPD64")\n", next_transno);
1261 } else if (queue_len + completed == max) {
1262 /* handle gaps occured due to lost reply. It is allowed gaps
1263 * because all clients are connected and there will be resend
1264 * for missed transaction */
1265 LASSERTF(req_transno >= next_transno,
1266 "req_transno: "LPU64", next_transno: "LPU64"\n",
1267 req_transno, next_transno);
1269 CDEBUG(req_transno > obd->obd_last_committed ? D_ERROR : D_HA,
1270 "waking for skipped transno (skip: "LPD64
1271 ", ql: %d, comp: %d, conn: %d, next: "LPD64")\n",
1272 next_transno, queue_len, completed, connected, req_transno);
1273 obd->obd_next_recovery_transno = req_transno;
1275 } else if (queue_len == atomic_read(&obd->obd_req_replay_clients)) {
1276 /* some clients haven't connected in time, but we can try
1277 * to replay requests that demand on already committed ones
1278 * also, we can replay first non-committed transation */
1279 LASSERT(req_transno != 0);
1280 if (req_transno == obd->obd_last_committed + 1) {
1281 obd->obd_next_recovery_transno = req_transno;
1282 } else if (req_transno > obd->obd_last_committed) {
1283 /* can't continue recovery: have no needed transno */
1284 obd->obd_abort_recovery = 1;
1285 CDEBUG(D_ERROR, "abort due to missed clients. max: %d, "
1286 "connected: %d, completed: %d, queue_len: %d, "
1287 "req_transno: "LPU64", next_transno: "LPU64"\n",
1288 max, connected, completed, queue_len,
1289 req_transno, next_transno);
1294 spin_unlock_bh(&obd->obd_processing_task_lock);
1298 static struct ptlrpc_request *target_next_replay_req(struct obd_device *obd)
1300 struct l_wait_info lwi = { 0 };
1301 struct ptlrpc_request *req;
1303 CDEBUG(D_HA, "Waiting for transno "LPD64"\n",
1304 obd->obd_next_recovery_transno);
1305 l_wait_event(obd->obd_next_transno_waitq,
1306 check_for_next_transno(obd), &lwi);
1308 spin_lock_bh(&obd->obd_processing_task_lock);
1309 if (obd->obd_abort_recovery) {
1311 } else if (!list_empty(&obd->obd_req_replay_queue)) {
1312 req = list_entry(obd->obd_req_replay_queue.next,
1313 struct ptlrpc_request, rq_list);
1314 list_del_init(&req->rq_list);
1315 obd->obd_requests_queued_for_recovery--;
1319 spin_unlock_bh(&obd->obd_processing_task_lock);
1323 static int check_for_next_lock(struct obd_device *obd)
1325 struct ptlrpc_request *req = NULL;
1328 spin_lock_bh(&obd->obd_processing_task_lock);
1329 if (!list_empty(&obd->obd_lock_replay_queue)) {
1330 req = list_entry(obd->obd_lock_replay_queue.next,
1331 struct ptlrpc_request, rq_list);
1332 CDEBUG(D_HA, "waking for next lock\n");
1334 } else if (atomic_read(&obd->obd_lock_replay_clients) == 0) {
1335 CDEBUG(D_HA, "waking for completed lock replay\n");
1337 } else if (obd->obd_abort_recovery) {
1338 CDEBUG(D_HA, "waking for aborted recovery\n");
1341 spin_unlock_bh(&obd->obd_processing_task_lock);
1346 static struct ptlrpc_request *target_next_replay_lock(struct obd_device *obd)
1348 struct l_wait_info lwi = { 0 };
1349 struct ptlrpc_request *req;
1351 CDEBUG(D_HA, "Waiting for lock\n");
1352 l_wait_event(obd->obd_next_transno_waitq,
1353 check_for_next_lock(obd), &lwi);
1355 spin_lock_bh(&obd->obd_processing_task_lock);
1356 if (obd->obd_abort_recovery) {
1358 } else if (!list_empty(&obd->obd_lock_replay_queue)) {
1359 req = list_entry(obd->obd_lock_replay_queue.next,
1360 struct ptlrpc_request, rq_list);
1361 list_del_init(&req->rq_list);
1365 spin_unlock_bh(&obd->obd_processing_task_lock);
1369 static struct ptlrpc_request *target_next_final_ping(struct obd_device *obd)
1371 struct ptlrpc_request *req;
1373 spin_lock_bh(&obd->obd_processing_task_lock);
1374 if (!list_empty(&obd->obd_final_req_queue)) {
1375 req = list_entry(obd->obd_final_req_queue.next,
1376 struct ptlrpc_request, rq_list);
1377 list_del_init(&req->rq_list);
1381 spin_unlock_bh(&obd->obd_processing_task_lock);
1385 static inline int req_replay_done(struct obd_export *exp)
1387 return (exp->exp_req_replay_needed == 0);
1390 static inline int lock_replay_done(struct obd_export *exp)
1392 return (exp->exp_lock_replay_needed == 0);
1395 static inline int connect_done(struct obd_export *exp)
1397 return (exp->exp_in_recovery != 0);
1400 static int check_for_clients(struct obd_device *obd)
1402 if (obd->obd_abort_recovery)
1404 LASSERT(obd->obd_connected_clients <= obd->obd_max_recoverable_clients);
1405 if (obd->obd_no_conn == 0 &&
1406 obd->obd_connected_clients == obd->obd_max_recoverable_clients)
1411 static int handle_recovery_req(struct ptlrpc_thread *thread,
1412 struct ptlrpc_request *req,
1413 svc_handler_t handler)
1418 rc = lu_context_init(&req->rq_session, LCT_SESSION);
1420 CERROR("Failure to initialize session: %d\n", rc);
1423 req->rq_session.lc_thread = thread;
1424 lu_context_enter(&req->rq_session);
1425 req->rq_svc_thread = thread;
1426 req->rq_svc_thread->t_env->le_ses = &req->rq_session;
1428 /* thread context */
1429 lu_context_enter(&thread->t_env->le_ctx);
1431 lu_context_exit(&thread->t_env->le_ctx);
1433 lu_context_exit(&req->rq_session);
1434 lu_context_fini(&req->rq_session);
1435 /* don't reset timer for final stage */
1436 if (!req_replay_done(req->rq_export) ||
1437 !lock_replay_done(req->rq_export))
1438 reset_recovery_timer(class_exp2obd(req->rq_export));
1439 ptlrpc_free_clone(req);
1443 static int target_recovery_thread(void *arg)
1445 struct obd_device *obd = arg;
1446 struct ptlrpc_request *req;
1447 struct target_recovery_data *trd = &obd->obd_recovery_data;
1448 struct l_wait_info lwi = { 0 };
1449 unsigned long delta;
1450 unsigned long flags;
1452 struct ptlrpc_thread fake_svc_thread, *thread = &fake_svc_thread;
1453 __u32 recov_ctx_tags = LCT_MD_THREAD;
1457 cfs_daemonize("tgt_recov");
1459 SIGNAL_MASK_LOCK(current, flags);
1460 sigfillset(¤t->blocked);
1462 SIGNAL_MASK_UNLOCK(current, flags);
1464 rc = lu_context_init(&env.le_ctx, recov_ctx_tags);
1468 thread->t_env = &env;
1469 env.le_ctx.lc_thread = thread;
1471 CERROR("%s: started recovery thread pid %d\n", obd->obd_name,
1473 trd->trd_processing_task = current->pid;
1475 obd->obd_recovering = 1;
1476 complete(&trd->trd_starting);
1478 /* first of all, we have to know the first transno to replay */
1479 obd->obd_abort_recovery = 0;
1480 l_wait_event(obd->obd_next_transno_waitq,
1481 check_for_clients(obd), &lwi);
1483 spin_lock_bh(&obd->obd_processing_task_lock);
1484 target_cancel_recovery_timer(obd);
1485 spin_unlock_bh(&obd->obd_processing_task_lock);
1487 /* If some clients haven't connected in time, evict them */
1488 if (obd->obd_abort_recovery) {
1489 CWARN("Some clients haven't connect in time (%d/%d),"
1490 "evict them\n", obd->obd_connected_clients,
1491 obd->obd_max_recoverable_clients);
1492 obd->obd_abort_recovery = obd->obd_stopping;
1493 class_disconnect_stale_exports(obd, connect_done);
1495 /* next stage: replay requests */
1497 obd->obd_req_replaying = 1;
1498 CDEBUG(D_INFO, "1: request replay stage - %d clients from t"LPU64"\n",
1499 atomic_read(&obd->obd_req_replay_clients),
1500 obd->obd_next_recovery_transno);
1501 while ((req = target_next_replay_req(obd))) {
1502 LASSERT(trd->trd_processing_task == current->pid);
1503 DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s",
1504 lustre_msg_get_transno(req->rq_reqmsg),
1505 libcfs_nid2str(req->rq_peer.nid));
1506 handle_recovery_req(thread, req,
1507 trd->trd_recovery_handler);
1508 obd->obd_replayed_requests++;
1509 spin_lock_bh(&obd->obd_processing_task_lock);
1510 obd->obd_next_recovery_transno++;
1511 spin_unlock_bh(&obd->obd_processing_task_lock);
1514 spin_lock_bh(&obd->obd_processing_task_lock);
1515 target_cancel_recovery_timer(obd);
1516 spin_unlock_bh(&obd->obd_processing_task_lock);
1518 /* If some clients haven't replayed requests in time, evict them */
1519 if (obd->obd_abort_recovery) {
1520 CDEBUG(D_ERROR, "req replay timed out, aborting ...\n");
1521 obd->obd_abort_recovery = obd->obd_stopping;
1522 class_disconnect_stale_exports(obd, req_replay_done);
1523 abort_req_replay_queue(obd);
1525 /* The second stage: replay locks */
1526 CDEBUG(D_INFO, "2: lock replay stage - %d clients\n",
1527 atomic_read(&obd->obd_lock_replay_clients));
1528 while ((req = target_next_replay_lock(obd))) {
1529 LASSERT(trd->trd_processing_task == current->pid);
1530 DEBUG_REQ(D_HA|D_WARNING, req, "processing lock from %s: ",
1531 libcfs_nid2str(req->rq_peer.nid));
1532 handle_recovery_req(thread, req,
1533 trd->trd_recovery_handler);
1534 obd->obd_replayed_locks++;
1537 spin_lock_bh(&obd->obd_processing_task_lock);
1538 target_cancel_recovery_timer(obd);
1539 spin_unlock_bh(&obd->obd_processing_task_lock);
1540 /* If some clients haven't replayed requests in time, evict them */
1541 if (obd->obd_abort_recovery) {
1543 CERROR("lock replay timed out, aborting ...\n");
1544 obd->obd_abort_recovery = obd->obd_stopping;
1545 stale = class_disconnect_stale_exports(obd, lock_replay_done);
1546 abort_lock_replay_queue(obd);
1549 /* We drop recoverying flag to forward all new requests
1550 * to regular mds_handle() since now */
1551 spin_lock_bh(&obd->obd_processing_task_lock);
1552 obd->obd_recovering = obd->obd_abort_recovery = 0;
1553 spin_unlock_bh(&obd->obd_processing_task_lock);
1554 /* The third stage: reply on final pings */
1555 CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n");
1556 while ((req = target_next_final_ping(obd))) {
1557 LASSERT(trd->trd_processing_task == current->pid);
1558 DEBUG_REQ(D_HA, req, "processing final ping from %s: ",
1559 libcfs_nid2str(req->rq_peer.nid));
1560 handle_recovery_req(thread, req,
1561 trd->trd_recovery_handler);
1564 delta = (jiffies - delta) / HZ;
1565 CDEBUG(D_INFO,"4: recovery completed in %lus - %d/%d reqs/locks\n",
1566 delta, obd->obd_replayed_requests, obd->obd_replayed_locks);
1567 LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0);
1568 LASSERT(atomic_read(&obd->obd_lock_replay_clients) == 0);
1569 if (delta > obd_timeout * 2) {
1570 CWARN("too long recovery - read logs\n");
1571 libcfs_debug_dumplog();
1574 target_finish_recovery(obd);
1576 lu_context_fini(&env.le_ctx);
1577 trd->trd_processing_task = 0;
1578 complete(&trd->trd_finishing);
1582 int target_start_recovery_thread(struct obd_device *obd, svc_handler_t handler)
1585 struct target_recovery_data *trd = &obd->obd_recovery_data;
1587 memset(trd, 0, sizeof(*trd));
1588 init_completion(&trd->trd_starting);
1589 init_completion(&trd->trd_finishing);
1590 trd->trd_recovery_handler = handler;
1592 if (kernel_thread(target_recovery_thread, obd, 0) > 0) {
1593 wait_for_completion(&trd->trd_starting);
1594 LASSERT(obd->obd_recovering != 0);
1601 void target_stop_recovery_thread(struct obd_device *obd)
1603 spin_lock_bh(&obd->obd_processing_task_lock);
1604 if (obd->obd_recovery_data.trd_processing_task > 0) {
1605 struct target_recovery_data *trd = &obd->obd_recovery_data;
1606 CERROR("%s: Aborting recovery\n", obd->obd_name);
1607 obd->obd_abort_recovery = 1;
1608 wake_up(&obd->obd_next_transno_waitq);
1609 spin_unlock_bh(&obd->obd_processing_task_lock);
1610 wait_for_completion(&trd->trd_finishing);
1612 spin_unlock_bh(&obd->obd_processing_task_lock);
1616 void target_recovery_fini(struct obd_device *obd)
1618 class_disconnect_exports(obd);
1619 target_stop_recovery_thread(obd);
1620 target_cleanup_recovery(obd);
1622 EXPORT_SYMBOL(target_recovery_fini);
1624 void target_recovery_init(struct obd_device *obd, svc_handler_t handler)
1626 if (obd->obd_max_recoverable_clients == 0)
1629 CWARN("RECOVERY: service %s, %d recoverable clients, "
1630 "last_transno "LPU64"\n", obd->obd_name,
1631 obd->obd_max_recoverable_clients, obd->obd_last_committed);
1632 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
1633 target_start_recovery_thread(obd, handler);
1634 obd->obd_recovery_start = CURRENT_SECONDS;
1635 /* Only used for lprocfs_status */
1636 obd->obd_recovery_end = obd->obd_recovery_start + OBD_RECOVERY_TIMEOUT;
1637 /* bz13079: this should be set to desired value for ost but not for mds */
1638 obd->obd_recovery_max_time = OBD_RECOVERY_MAX_TIME;
1640 EXPORT_SYMBOL(target_recovery_init);
1644 int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req)
1646 struct obd_export *exp = req->rq_export;
1647 LASSERT(exp != NULL);
1648 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
1649 /* client declares he's ready to replay locks */
1650 spin_lock_bh(&obd->obd_processing_task_lock);
1651 if (exp->exp_req_replay_needed) {
1652 LASSERT(atomic_read(&obd->obd_req_replay_clients) > 0);
1653 spin_lock(&exp->exp_lock);
1654 exp->exp_req_replay_needed = 0;
1655 spin_unlock(&exp->exp_lock);
1656 atomic_dec(&obd->obd_req_replay_clients);
1657 LASSERT(obd->obd_recoverable_clients > 0);
1658 obd->obd_recoverable_clients--;
1659 if (atomic_read(&obd->obd_req_replay_clients) == 0)
1660 CDEBUG(D_HA, "all clients have replayed reqs\n");
1661 wake_up(&obd->obd_next_transno_waitq);
1663 spin_unlock_bh(&obd->obd_processing_task_lock);
1665 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
1666 /* client declares he's ready to complete recovery
1667 * so, we put the request on th final queue */
1668 spin_lock_bh(&obd->obd_processing_task_lock);
1669 if (exp->exp_lock_replay_needed) {
1670 LASSERT(atomic_read(&obd->obd_lock_replay_clients) > 0);
1671 spin_lock(&exp->exp_lock);
1672 exp->exp_lock_replay_needed = 0;
1673 spin_unlock(&exp->exp_lock);
1674 atomic_dec(&obd->obd_lock_replay_clients);
1675 if (atomic_read(&obd->obd_lock_replay_clients) == 0)
1676 CDEBUG(D_HA, "all clients have replayed locks\n");
1677 wake_up(&obd->obd_next_transno_waitq);
1679 spin_unlock_bh(&obd->obd_processing_task_lock);
1685 int target_queue_recovery_request(struct ptlrpc_request *req,
1686 struct obd_device *obd)
1688 struct list_head *tmp;
1690 __u64 transno = lustre_msg_get_transno(req->rq_reqmsg);
1694 if (obd->obd_recovery_data.trd_processing_task == cfs_curproc_pid()) {
1695 /* Processing the queue right now, don't re-add. */
1699 target_process_req_flags(obd, req);
1701 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
1702 /* client declares he's ready to complete recovery
1703 * so, we put the request on th final queue */
1704 req = ptlrpc_clone_req(req);
1707 DEBUG_REQ(D_HA, req, "queue final req");
1708 spin_lock_bh(&obd->obd_processing_task_lock);
1709 if (obd->obd_recovering)
1710 list_add_tail(&req->rq_list, &obd->obd_final_req_queue);
1712 spin_unlock_bh(&obd->obd_processing_task_lock);
1713 ptlrpc_free_clone(req);
1714 if (obd->obd_stopping) {
1720 spin_unlock_bh(&obd->obd_processing_task_lock);
1723 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
1724 /* client declares he's ready to replay locks */
1725 req = ptlrpc_clone_req(req);
1728 DEBUG_REQ(D_HA, req, "queue lock replay req");
1729 spin_lock_bh(&obd->obd_processing_task_lock);
1730 LASSERT(obd->obd_recovering);
1731 /* usually due to recovery abort */
1732 if (!req->rq_export->exp_in_recovery) {
1733 spin_unlock_bh(&obd->obd_processing_task_lock);
1734 ptlrpc_free_clone(req);
1737 LASSERT(req->rq_export->exp_lock_replay_needed);
1738 list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue);
1739 spin_unlock_bh(&obd->obd_processing_task_lock);
1740 wake_up(&obd->obd_next_transno_waitq);
1744 /* CAVEAT EMPTOR: The incoming request message has been swabbed
1745 * (i.e. buflens etc are in my own byte order), but type-dependent
1746 * buffers (eg mds_body, ost_body etc) have NOT been swabbed. */
1749 CFS_INIT_LIST_HEAD(&req->rq_list);
1750 DEBUG_REQ(D_HA, req, "not queueing");
1754 spin_lock_bh(&obd->obd_processing_task_lock);
1756 /* If we're processing the queue, we want don't want to queue this
1759 * Also, if this request has a transno less than the one we're waiting
1760 * for, we should process it now. It could (and currently always will)
1761 * be an open request for a descriptor that was opened some time ago.
1763 * Also, a resent, replayed request that has already been
1764 * handled will pass through here and be processed immediately.
1766 CWARN("Next recovery transno: "LPU64", current: "LPU64", replaying: %i\n",
1767 obd->obd_next_recovery_transno, transno, obd->obd_req_replaying);
1768 if (transno < obd->obd_next_recovery_transno && obd->obd_req_replaying) {
1769 /* Processing the queue right now, don't re-add. */
1770 LASSERT(list_empty(&req->rq_list));
1771 spin_unlock_bh(&obd->obd_processing_task_lock);
1774 spin_unlock_bh(&obd->obd_processing_task_lock);
1776 /* A resent, replayed request that is still on the queue; just drop it.
1777 The queued request will handle this. */
1778 if ((lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT|MSG_REPLAY)) ==
1779 (MSG_RESENT | MSG_REPLAY)) {
1780 DEBUG_REQ(D_ERROR, req, "dropping resent queued req");
1784 req = ptlrpc_clone_req(req);
1788 spin_lock_bh(&obd->obd_processing_task_lock);
1789 LASSERT(obd->obd_recovering);
1790 if (!req->rq_export->exp_in_recovery) {
1791 spin_unlock_bh(&obd->obd_processing_task_lock);
1792 ptlrpc_free_clone(req);
1795 LASSERT(req->rq_export->exp_req_replay_needed);
1798 list_for_each(tmp, &obd->obd_req_replay_queue) {
1799 struct ptlrpc_request *reqiter =
1800 list_entry(tmp, struct ptlrpc_request, rq_list);
1802 if (lustre_msg_get_transno(reqiter->rq_reqmsg) > transno) {
1803 list_add_tail(&req->rq_list, &reqiter->rq_list);
1810 list_add_tail(&req->rq_list, &obd->obd_req_replay_queue);
1812 obd->obd_requests_queued_for_recovery++;
1813 wake_up(&obd->obd_next_transno_waitq);
1814 spin_unlock_bh(&obd->obd_processing_task_lock);
1819 struct obd_device * target_req2obd(struct ptlrpc_request *req)
1821 return req->rq_export->exp_obd;
1824 static inline struct ldlm_pool *ldlm_exp2pl(struct obd_export *exp)
1826 LASSERT(exp != NULL);
1827 return &exp->exp_obd->obd_namespace->ns_pool;
1830 int target_pack_pool_reply(struct ptlrpc_request *req)
1832 struct ldlm_pool *pl;
1835 if (!req->rq_export || !req->rq_export->exp_obd ||
1836 !exp_connect_lru_resize(req->rq_export)) {
1837 lustre_msg_set_slv(req->rq_repmsg, 0);
1838 lustre_msg_set_limit(req->rq_repmsg, 0);
1842 pl = ldlm_exp2pl(req->rq_export);
1844 spin_lock(&pl->pl_lock);
1845 LASSERT(ldlm_pool_get_slv(pl) != 0 && ldlm_pool_get_limit(pl) != 0);
1846 lustre_msg_set_slv(req->rq_repmsg, ldlm_pool_get_slv(pl));
1847 lustre_msg_set_limit(req->rq_repmsg, ldlm_pool_get_limit(pl));
1848 spin_unlock(&pl->pl_lock);
1853 int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id)
1855 if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) {
1856 DEBUG_REQ(D_ERROR, req, "dropping reply");
1861 DEBUG_REQ(D_ERROR, req, "processing error (%d)", rc);
1862 req->rq_status = rc;
1863 return (ptlrpc_error(req));
1865 DEBUG_REQ(D_NET, req, "sending reply");
1868 return (ptlrpc_send_reply(req, 1));
1871 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
1874 struct ptlrpc_reply_state *rs;
1875 struct obd_device *obd;
1876 struct obd_export *exp;
1877 struct ptlrpc_service *svc;
1879 if (req->rq_no_reply)
1882 svc = req->rq_rqbd->rqbd_service;
1883 rs = req->rq_reply_state;
1884 if (rs == NULL || !rs->rs_difficult) {
1886 target_send_reply_msg (req, rc, fail_id);
1890 /* must be an export if locks saved */
1891 LASSERT (req->rq_export != NULL);
1892 /* req/reply consistent */
1893 LASSERT (rs->rs_service == svc);
1896 LASSERT (!rs->rs_scheduled);
1897 LASSERT (!rs->rs_scheduled_ever);
1898 LASSERT (!rs->rs_handled);
1899 LASSERT (!rs->rs_on_net);
1900 LASSERT (rs->rs_export == NULL);
1901 LASSERT (list_empty(&rs->rs_obd_list));
1902 LASSERT (list_empty(&rs->rs_exp_list));
1904 exp = class_export_get (req->rq_export);
1907 /* disable reply scheduling onto srv_reply_queue while I'm setting up */
1908 rs->rs_scheduled = 1;
1910 rs->rs_xid = req->rq_xid;
1911 rs->rs_transno = req->rq_transno;
1912 rs->rs_export = exp;
1914 spin_lock(&obd->obd_uncommitted_replies_lock);
1916 if (rs->rs_transno > obd->obd_last_committed) {
1917 /* not committed already */
1918 list_add_tail (&rs->rs_obd_list,
1919 &obd->obd_uncommitted_replies);
1922 spin_unlock (&obd->obd_uncommitted_replies_lock);
1923 spin_lock (&exp->exp_lock);
1925 list_add_tail (&rs->rs_exp_list, &exp->exp_outstanding_replies);
1927 spin_unlock(&exp->exp_lock);
1929 netrc = target_send_reply_msg (req, rc, fail_id);
1931 spin_lock(&svc->srv_lock);
1933 svc->srv_n_difficult_replies++;
1936 /* error sending: reply is off the net. Also we need +1
1937 * reply ref until ptlrpc_server_handle_reply() is done
1938 * with the reply state (if the send was successful, there
1939 * would have been +1 ref for the net, which
1940 * reply_out_callback leaves alone) */
1942 ptlrpc_rs_addref(rs);
1943 atomic_inc (&svc->srv_outstanding_replies);
1946 if (!rs->rs_on_net || /* some notifier */
1947 list_empty(&rs->rs_exp_list) || /* completed already */
1948 list_empty(&rs->rs_obd_list)) {
1949 list_add_tail (&rs->rs_list, &svc->srv_reply_queue);
1950 cfs_waitq_signal (&svc->srv_waitq);
1952 list_add (&rs->rs_list, &svc->srv_active_replies);
1953 rs->rs_scheduled = 0; /* allow notifier to schedule */
1956 spin_unlock(&svc->srv_lock);
1959 int target_handle_ping(struct ptlrpc_request *req)
1961 obd_ping(req->rq_export);
1962 return req_capsule_server_pack(&req->rq_pill);
1965 void target_committed_to_req(struct ptlrpc_request *req)
1967 struct obd_device *obd;
1969 if (req == NULL || req->rq_export == NULL)
1972 obd = req->rq_export->exp_obd;
1976 if (!obd->obd_no_transno && req->rq_repmsg != NULL)
1977 lustre_msg_set_last_committed(req->rq_repmsg,
1978 obd->obd_last_committed);
1980 DEBUG_REQ(D_IOCTL, req, "not sending last_committed update");
1982 CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n",
1983 obd->obd_last_committed, req->rq_transno, req->rq_xid);
1986 EXPORT_SYMBOL(target_committed_to_req);
1988 #ifdef HAVE_QUOTA_SUPPORT
1989 int target_handle_qc_callback(struct ptlrpc_request *req)
1991 struct obd_quotactl *oqctl;
1992 struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
1994 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1998 cli->cl_qchk_stat = oqctl->qc_stat;
2003 int target_handle_dqacq_callback(struct ptlrpc_request *req)
2006 struct obd_device *obd = req->rq_export->exp_obd;
2007 struct obd_device *master_obd;
2008 struct lustre_quota_ctxt *qctxt;
2009 struct qunit_data *qdata;
2011 struct qunit_data_old *qdata_old;
2015 rc = req_capsule_server_pack(&req->rq_pill);
2017 CERROR("packing reply failed!: rc = %d\n", rc);
2021 LASSERT(req->rq_export);
2023 /* fixed for bug10707 */
2024 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) &&
2025 !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) {
2026 CDEBUG(D_QUOTA, "qd_count is 64bit!\n");
2027 rep = req_capsule_server_get(&req->rq_pill,
2030 qdata = req_capsule_client_swab_get(&req->rq_pill,
2032 (void*)lustre_swab_qdata);
2034 CDEBUG(D_QUOTA, "qd_count is 32bit!\n");
2035 rep = req_capsule_server_get(&req->rq_pill, &RMF_QUNIT_DATA);
2037 qdata_old = req_capsule_client_swab_get(&req->rq_pill,
2039 (void*)lustre_swab_qdata_old);
2040 qdata = lustre_quota_old_to_new(qdata_old);
2046 /* we use the observer */
2047 LASSERT(obd->obd_observer && obd->obd_observer->obd_observer);
2048 master_obd = obd->obd_observer->obd_observer;
2049 qctxt = &master_obd->u.obt.obt_qctxt;
2051 LASSERT(qctxt->lqc_handler);
2052 rc = qctxt->lqc_handler(master_obd, qdata,
2053 lustre_msg_get_opc(req->rq_reqmsg));
2054 if (rc && rc != -EDQUOT)
2055 CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR,
2056 "dqacq failed! (rc:%d)\n", rc);
2058 /* the qd_count might be changed in lqc_handler */
2059 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) &&
2060 !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) {
2061 memcpy(rep, qdata, sizeof(*qdata));
2063 qdata_old = lustre_quota_new_to_old(qdata);
2064 memcpy(rep, qdata_old, sizeof(*qdata_old));
2066 req->rq_status = rc;
2067 rc = ptlrpc_reply(req);
2072 #endif /* !__KERNEL__ */
2074 #endif /* HAVE_QUOTA_SUPPORT */
2076 ldlm_mode_t lck_compat_array[] = {
2077 [LCK_EX] LCK_COMPAT_EX,
2078 [LCK_PW] LCK_COMPAT_PW,
2079 [LCK_PR] LCK_COMPAT_PR,
2080 [LCK_CW] LCK_COMPAT_CW,
2081 [LCK_CR] LCK_COMPAT_CR,
2082 [LCK_NL] LCK_COMPAT_NL,
2083 [LCK_GROUP] LCK_COMPAT_GROUP