1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 # define EXPORT_SYMTAB
25 #define DEBUG_SUBSYSTEM S_LMV
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/pagemap.h>
32 #include <asm/div64.h>
33 #include <linux/seq_file.h>
35 #include <liblustre.h>
37 #include <linux/ext2_fs.h>
39 #include <linux/obd_support.h>
40 #include <linux/lustre_lib.h>
41 #include <linux/lustre_net.h>
42 #include <linux/lustre_idl.h>
43 #include <linux/lustre_dlm.h>
44 #include <linux/lustre_mds.h>
45 #include <linux/obd_class.h>
46 #include <linux/obd_ost.h>
47 #include <linux/lprocfs_status.h>
48 #include <linux/lustre_fsfilt.h>
49 #include <linux/obd_lmv.h>
50 #include "lmv_internal.h"
54 * -EINVAL : UUID can't be found in the LMV's target list
55 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
56 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
58 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
61 struct obd_device *obd;
62 struct lmv_tgt_desc *tgt;
66 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
67 lmv, uuid->uuid, activate);
69 spin_lock(&lmv->lmv_lock);
70 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
71 if (tgt->ltd_exp == NULL)
74 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
75 i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
76 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
80 if (i == lmv->desc.ld_tgt_count)
81 GOTO(out, rc = -EINVAL);
83 obd = class_exp2obd(tgt->ltd_exp);
85 GOTO(out, rc = -ENOTCONN);
87 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
88 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
89 obd->obd_type->typ_name, i);
90 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
92 if (tgt->active == activate) {
93 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
94 activate ? "" : "in");
98 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd, activate ? "" : "in");
100 tgt->active = activate;
102 lmv->desc.ld_active_tgt_count++;
104 lmv->desc.ld_active_tgt_count--;
108 spin_unlock(&lmv->lmv_lock);
112 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
113 int active, void *data)
116 struct obd_uuid *uuid;
118 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
119 CERROR("unexpected notification of %s %s!\n",
120 watched->obd_type->typ_name,
124 uuid = &watched->u.cli.cl_import->imp_target_uuid;
126 /* Set MDC as active before notifying the observer, so the
127 * observer can use the MDC normally.
129 rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
131 CERROR("%sactivation of %s failed: %d\n",
132 active ? "" : "de", uuid->uuid, rc);
136 if (obd->obd_observer)
137 /* Pass the notification up the chain. */
138 rc = obd_notify(obd->obd_observer, watched, active, data);
143 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
145 struct lprocfs_static_vars lvars;
149 lprocfs_init_vars(lmv, &lvars);
150 rc = lprocfs_obd_attach(dev, lvars.obd_vars);
153 struct proc_dir_entry *entry;
155 entry = create_proc_entry("target_obd", 0444, dev->obd_proc_entry);
158 /* entry->proc_fops = &lmv_proc_target_fops; */
165 int lmv_detach(struct obd_device *dev)
167 return lprocfs_obd_detach(dev);
170 /* This is fake connect function. Its purpose is to initialize lmv and
171 * say caller that everything is okay. Real connection will be performed
173 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
174 struct obd_uuid *cluuid, unsigned long connect_flags)
176 struct lmv_obd *lmv = &obd->u.lmv;
177 struct obd_export *exp;
181 rc = class_connect(conn, obd, cluuid);
183 CERROR("class_connection() returned %d\n", rc);
187 exp = class_conn2export(conn);
188 /* We don't want to actually do the underlying connections more than
189 * once, so keep track. */
191 if (lmv->refcount > 1) {
192 class_export_put(exp);
196 lmv->cluuid = *cluuid;
197 lmv->connect_flags = connect_flags;
200 sema_init(&lmv->init_sem, 1);
205 void lmv_set_timeouts(struct obd_device *obd)
207 struct lmv_tgt_desc *tgts;
212 if (lmv->server_timeout == 0)
215 if (lmv->connected == 0)
218 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
219 if (tgts->ltd_exp == NULL)
221 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
222 "inter_mds", 0, NULL);
226 /* Performs a check if passed obd is connected. If no - connect it. */
227 int lmv_check_connect(struct obd_device *obd)
229 struct lmv_obd *lmv = &obd->u.lmv;
230 struct obd_uuid *cluuid;
231 struct lmv_tgt_desc *tgts;
232 struct obd_export *exp;
238 down(&lmv->init_sem);
239 if (lmv->connected) {
245 cluuid = &lmv->cluuid;
248 CDEBUG(D_OTHER, "time to connect %s to %s\n",
249 cluuid->uuid, obd->obd_name);
251 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
252 struct obd_device *tgt_obd;
253 struct obd_uuid lmv_osc_uuid = { "LMV_OSC_UUID" };
254 struct lustre_handle conn = {0, };
256 LASSERT(tgts != NULL);
258 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
261 CERROR("Target %s not attached\n", tgts->uuid.uuid);
262 GOTO(out_disc, rc = -EINVAL);
265 /* for MDS: don't connect to yourself */
266 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
267 CDEBUG(D_OTHER, "don't connect back to %s\n",
269 tgts->ltd_exp = NULL;
273 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
274 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
275 tgts->uuid.uuid, obd->obd_uuid.uuid,
278 if (!tgt_obd->obd_set_up) {
279 CERROR("Target %s not set up\n", tgts->uuid.uuid);
280 GOTO(out_disc, rc = -EINVAL);
283 rc = obd_connect(&conn, tgt_obd, &lmv_osc_uuid, lmv->connect_flags);
285 CERROR("Target %s connect error %d\n",
286 tgts->uuid.uuid, rc);
289 tgts->ltd_exp = class_conn2export(&conn);
291 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
292 lmv->max_cookiesize);
294 rc = obd_register_observer(tgt_obd, obd);
296 CERROR("Target %s register_observer error %d\n",
297 tgts->uuid.uuid, rc);
298 obd_disconnect(tgts->ltd_exp, 0);
302 lmv->desc.ld_active_tgt_count++;
305 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
306 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
307 atomic_read(&obd->obd_refcount));
310 lmv_set_timeouts(obd);
311 class_export_put(exp);
317 struct obd_uuid uuid;
319 --lmv->desc.ld_active_tgt_count;
321 /* save for CERROR below; (we know it's terminated) */
323 rc2 = obd_disconnect(tgts->ltd_exp, 0);
325 CERROR("error: LMV target %s disconnect on MDT idx %d: "
326 "error %d\n", uuid.uuid, i, rc2);
328 class_disconnect(exp, 0);
333 static int lmv_disconnect(struct obd_export *exp, int flags)
335 struct obd_device *obd = class_exp2obd(exp);
336 struct lmv_obd *lmv = &obd->u.lmv;
343 /* Only disconnect the underlying layers on the final disconnect. */
345 if (lmv->refcount != 0)
348 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
349 if (lmv->tgts[i].ltd_exp == NULL)
352 if (obd->obd_no_recov) {
353 /* Pass it on to our clients.
354 * XXX This should be an argument to disconnect,
355 * XXX not a back-door flag on the OBD. Ah well.
357 struct obd_device *mdc_obd;
358 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
360 mdc_obd->obd_no_recov = 1;
363 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
364 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
365 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
367 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
369 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
371 if (lmv->tgts[i].active) {
372 CERROR("Target %s disconnect error %d\n",
373 lmv->tgts[i].uuid.uuid, rc);
377 if (lmv->tgts[i].active) {
378 lmv->desc.ld_active_tgt_count--;
379 lmv->tgts[i].active = 0;
381 lmv->tgts[i].ltd_exp = NULL;
385 /* this is the case when no real connection is established by
386 * lmv_check_connect(). */
388 class_export_put(exp);
389 rc = class_disconnect(exp, 0);
390 if (lmv->refcount == 0)
395 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
396 int len, void *karg, void *uarg)
398 struct obd_device *obddev = class_exp2obd(exp);
399 struct lmv_obd *lmv = &obddev->u.lmv;
400 int i, rc = 0, set = 0;
404 if (lmv->desc.ld_tgt_count == 0)
407 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
410 if (lmv->tgts[i].ltd_exp == NULL) {
411 CWARN("%s: NULL export for %d\n", obddev->obd_name, i);
415 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
417 if (lmv->tgts[i].active) {
418 CERROR("error: iocontrol MDC %s on MDT"
419 "idx %d: err = %d\n",
420 lmv->tgts[i].uuid.uuid, i, err);
433 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
436 struct lmv_desc *desc;
437 struct obd_uuid *uuids;
438 struct lmv_tgt_desc *tgts;
439 struct obd_device *tgt_obd;
440 struct lustre_cfg *lcfg = buf;
441 struct lmv_obd *lmv = &obd->u.lmv;
444 if (lcfg->lcfg_inllen1 < 1) {
445 CERROR("LMV setup requires a descriptor\n");
449 if (lcfg->lcfg_inllen2 < 1) {
450 CERROR("LMV setup requires an OST UUID list\n");
454 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
455 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
456 CERROR("descriptor size wrong: %d > %d\n",
457 (int)sizeof(*desc), lcfg->lcfg_inllen1);
461 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
462 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
463 CERROR("UUID array size wrong: %u * %u != %u\n",
464 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
468 lmv->bufsize = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
469 OBD_ALLOC(lmv->tgts, lmv->bufsize);
470 if (lmv->tgts == NULL) {
471 CERROR("Out of memory\n");
476 spin_lock_init(&lmv->lmv_lock);
478 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
479 tgts->uuid = uuids[i];
481 lmv->max_cookiesize = 0;
483 lmv->max_easize = sizeof(struct ll_fid) *
484 desc->ld_tgt_count + sizeof(struct mea);
486 rc = lmv_setup_mgr(obd);
488 CERROR("Can't setup LMV object manager, "
490 OBD_FREE(lmv->tgts, lmv->bufsize);
493 tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
496 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
500 rc = obd_llog_init(obd, &obd->obd_llogs, tgt_obd, 0, NULL);
502 CERROR("failed to setup llogging subsystems\n");
508 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
509 unsigned long max_age)
511 struct lmv_obd *lmv = &obd->u.lmv;
512 struct obd_statfs temp;
516 rc = lmv_check_connect(obd);
520 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
521 if (lmv->tgts[i].ltd_exp == NULL) {
522 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
526 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
528 CERROR("can't stat MDS #%d (%s)\n", i,
529 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
533 memcpy(osfs, &temp, sizeof(temp));
535 osfs->os_bavail += temp.os_bavail;
536 osfs->os_blocks += temp.os_blocks;
537 osfs->os_ffree += temp.os_ffree;
538 osfs->os_files += temp.os_files;
544 static int lmv_cleanup(struct obd_device *obd, int flags)
546 struct lmv_obd *lmv = &obd->u.lmv;
548 lmv_cleanup_mgr(obd);
549 OBD_FREE(lmv->tgts, lmv->bufsize);
553 static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid)
555 struct obd_device *obd = exp->exp_obd;
556 struct lmv_obd *lmv = &obd->u.lmv;
559 rc = lmv_check_connect(obd);
562 rc = md_getstatus(lmv->tgts[0].ltd_exp, fid);
567 static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid,
568 unsigned long valid, unsigned int ea_size,
569 struct ptlrpc_request **request)
571 struct obd_device *obd = exp->exp_obd;
572 struct lmv_obd *lmv = &obd->u.lmv;
573 int rc, i = fid->mds;
577 rc = lmv_check_connect(obd);
581 LASSERT(i < lmv->desc.ld_tgt_count);
583 rc = md_getattr(lmv->tgts[i].ltd_exp, fid, valid,
588 obj = lmv_grab_obj(obd, fid);
590 CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n",
591 (unsigned long)fid->mds, (unsigned long)fid->id,
592 (unsigned long)fid->generation, obj ? "(splitted)" : "");
594 /* if object is splitted, then we loop over all the slaves and gather
595 * size attribute. In ideal world we would have to gather also mds field
596 * from all slaves, as object is spread over the cluster and this is
597 * definitely interesting information and it is not good to loss it,
600 struct mds_body *body;
602 if (*request == NULL) {
607 body = lustre_msg_buf((*request)->rq_repmsg, 0,
609 LASSERT(body != NULL);
613 for (i = 0; i < obj->objcount; i++) {
615 if (lmv->tgts[i].ltd_exp == NULL) {
616 CWARN("%s: NULL export for %d\n",
621 /* skip master obj. */
622 if (fid_equal(&obj->fid, &obj->objs[i].fid))
625 body->size += obj->objs[i].size;
635 static int lmv_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
636 ldlm_iterator_t it, void *data)
638 struct obd_device *obd = exp->exp_obd;
639 struct lmv_obd *lmv = &obd->u.lmv;
643 rc = lmv_check_connect(obd);
647 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n", (unsigned long)fid->mds,
648 (unsigned long)fid->id, (unsigned long)fid->generation);
650 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
652 rc = md_change_cbdata(lmv->tgts[fid->mds].ltd_exp,
658 static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
659 char *name, int len, struct ll_fid *cfid,
660 ldlm_iterator_t it, void *data)
662 struct obd_device *obd = exp->exp_obd;
663 struct lmv_obd *lmv = &obd->u.lmv;
668 rc = lmv_check_connect(obd);
672 LASSERT(pfid->mds < lmv->desc.ld_tgt_count);
673 LASSERT(cfid->mds < lmv->desc.ld_tgt_count);
675 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n",
676 (unsigned long)pfid->mds, (unsigned long)pfid->id,
677 (unsigned long)pfid->generation, len, name,
678 (unsigned long)cfid->mds, (unsigned long)cfid->id,
679 (unsigned long)cfid->generation);
681 /* this is default mds for directory name belongs to. */
683 obj = lmv_grab_obj(obd, pfid);
685 /* directory is splitted. look for right mds for this name. */
686 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
687 mds = obj->objs[mds].fid.mds;
690 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cfid, it, data);
694 static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
696 struct obd_device *obd = exp->exp_obd;
697 struct lmv_obd *lmv = &obd->u.lmv;
700 rc = lmv_check_connect(obd);
703 CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n", (unsigned long) fid->mds,
704 (unsigned long) fid->id, (unsigned long) fid->generation);
705 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
706 rc = md_valid_attrs(lmv->tgts[fid->mds].ltd_exp, fid);
710 int lmv_close(struct obd_export *exp, struct obdo *obdo,
711 struct obd_client_handle *och,
712 struct ptlrpc_request **request)
714 struct obd_device *obd = exp->exp_obd;
715 struct lmv_obd *lmv = &obd->u.lmv;
716 int rc, i = obdo->o_mds;
718 rc = lmv_check_connect(obd);
721 LASSERT(i < lmv->desc.ld_tgt_count);
722 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds,
723 (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation);
724 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
728 int lmv_get_mea_and_update_object(struct obd_export *exp, struct ll_fid *fid)
730 struct obd_device *obd = exp->exp_obd;
731 struct lmv_obd *lmv = &obd->u.lmv;
732 struct ptlrpc_request *req = NULL;
739 mealen = MEA_SIZE_LMV(lmv);
741 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
743 /* time to update mea of parent fid */
744 rc = md_getattr(lmv->tgts[fid->mds].ltd_exp, fid,
745 valid, mealen, &req);
747 CERROR("md_getattr() failed, error %d\n", rc);
751 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
753 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
758 GOTO(cleanup, rc = -ENODATA);
760 obj = lmv_create_obj(exp, fid, md.mea);
765 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
769 ptlrpc_req_finished(req);
773 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
774 const void *data, int datalen, int mode, __u32 uid,
775 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
777 struct obd_device *obd = exp->exp_obd;
778 struct lmv_obd *lmv = &obd->u.lmv;
779 struct mds_body *body;
781 int rc, mds, loop = 0;
784 rc = lmv_check_connect(obd);
788 if (!lmv->desc.ld_active_tgt_count)
791 LASSERT(++loop <= 2);
792 obj = lmv_grab_obj(obd, &op_data->fid1);
794 mds = raw_name2idx(obj->hashtype, obj->objcount, op_data->name,
796 op_data->fid1 = obj->objs[mds].fid;
800 CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu\n", op_data->namelen,
801 op_data->name, (unsigned long)op_data->fid1.mds,
802 (unsigned long)op_data->fid1.id,
803 (unsigned long)op_data->fid1.generation);
805 rc = md_create(lmv->tgts[op_data->fid1.mds].ltd_exp, op_data, data,
806 datalen, mode, uid, gid, rdev, request);
808 if (*request == NULL)
811 body = lustre_msg_buf((*request)->rq_repmsg, 0,
813 LASSERT(body != NULL);
815 CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, "
816 "mds = %d\n", (unsigned long)body->fid1.id,
817 (unsigned long)body->fid1.generation, op_data->fid1.mds);
819 LASSERT(body->valid & OBD_MD_MDS ||
820 body->mds == op_data->fid1.mds);
821 } else if (rc == -ERESTART) {
822 /* directory got splitted. time to update local object and
823 * repeat the request with proper MDS */
824 rc = lmv_get_mea_and_update_object(exp, &op_data->fid1);
826 ptlrpc_req_finished(*request);
833 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
835 struct obd_device *obd = exp->exp_obd;
836 struct lmv_obd *lmv = &obd->u.lmv;
839 rc = lmv_check_connect(obd);
843 /* FIXME: choose right MDC here */
844 CWARN("this method isn't implemented yet\n");
845 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
849 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
850 struct lookup_intent *it, int lockmode,
851 struct mdc_op_data *data, struct lustre_handle *lockh,
852 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
853 ldlm_blocking_callback cb_blocking, void *cb_data)
855 struct obd_device *obd = exp->exp_obd;
856 struct lmv_obd *lmv = &obd->u.lmv;
857 struct mea *mea = data->mea1;
858 struct mdc_op_data data2;
862 LASSERT(mea != NULL);
863 for (i = 0; i < mea->mea_count; i++) {
864 memset(&data2, 0, sizeof(data2));
865 data2.fid1 = mea->mea_fids[i];
866 mds = data2.fid1.mds;
868 if (lmv->tgts[mds].ltd_exp == NULL)
871 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, lockmode,
872 &data2, lockh + i, lmm, lmmsize, cb_completion,
873 cb_blocking, cb_data);
875 CDEBUG(D_OTHER, "take lock on slave %lu/%lu/%lu -> %d/%d\n",
876 (unsigned long)mea->mea_fids[i].mds,
877 (unsigned long)mea->mea_fids[i].id,
878 (unsigned long)mea->mea_fids[i].generation,
879 rc, it->d.lustre.it_status);
882 if (it->d.lustre.it_data) {
883 struct ptlrpc_request *req;
884 req = (struct ptlrpc_request *) it->d.lustre.it_data;
885 ptlrpc_req_finished(req);
888 if (it->d.lustre.it_status)
889 GOTO(cleanup, rc = it->d.lustre.it_status);
894 /* drop all taken locks */
897 ldlm_lock_decref(lockh + i, lockmode);
903 int lmv_enqueue(struct obd_export *exp, int lock_type,
904 struct lookup_intent *it, int lock_mode,
905 struct mdc_op_data *data, struct lustre_handle *lockh,
906 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
907 ldlm_blocking_callback cb_blocking, void *cb_data)
909 struct obd_device *obd = exp->exp_obd;
910 struct lmv_obd *lmv = &obd->u.lmv;
915 rc = lmv_check_connect(obd);
919 if (it->it_op == IT_UNLINK) {
920 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
921 data, lockh, lmm, lmmsize,
922 cb_completion, cb_blocking, cb_data);
927 obj = lmv_grab_obj(obd, &data->fid1);
929 /* directory is splitted. look for right mds for this
931 mds = raw_name2idx(obj->hashtype, obj->objcount,
932 (char *)data->name, data->namelen);
933 data->fid1 = obj->objs[mds].fid;
937 CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n", LL_IT2STR(it),
938 (unsigned long)data->fid1.id, (unsigned long)data->fid1.generation);
940 rc = md_enqueue(lmv->tgts[data->fid1.mds].ltd_exp, lock_type, it,
941 lock_mode, data, lockh, lmm, lmmsize, cb_completion,
942 cb_blocking, cb_data);
947 int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid,
948 char *filename, int namelen, unsigned long valid,
949 unsigned int ea_size, struct ptlrpc_request **request)
951 struct obd_device *obd = exp->exp_obd;
952 struct lmv_obd *lmv = &obd->u.lmv;
953 struct ll_fid rfid = *fid;
954 int rc, mds = fid->mds, loop = 0;
955 struct mds_body *body;
958 rc = lmv_check_connect(obd);
962 LASSERT(++loop <= 2);
963 obj = lmv_grab_obj(obd, fid);
965 /* directory is splitted. look for right mds for this name */
966 mds = raw_name2idx(obj->hashtype, obj->objcount, filename, namelen - 1);
967 rfid = obj->objs[mds].fid;
971 CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu -> %lu/%lu/%lu\n",
972 namelen, filename, (unsigned long)fid->mds,
973 (unsigned long)fid->id, (unsigned long)fid->generation,
974 (unsigned long)rfid.mds, (unsigned long)rfid.id,
975 (unsigned long)rfid.generation);
977 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid, filename,
978 namelen, valid, ea_size, request);
980 /* this could be cross-node reference. in this case all we have
981 * right now is mds/ino/generation triple. we'd like to find
982 * other attributes */
983 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
984 LASSERT(body != NULL);
985 if (body->valid & OBD_MD_MDS) {
986 struct ptlrpc_request *req = NULL;
988 CDEBUG(D_OTHER, "request attrs for %lu/%lu/%lu\n",
989 (unsigned long) rfid.mds,
990 (unsigned long) rfid.id,
991 (unsigned long) rfid.generation);
992 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid,
993 NULL, 1, valid, ea_size, &req);
994 ptlrpc_req_finished(*request);
997 } else if (rc == -ERESTART) {
998 /* directory got splitted. time to update local object and
999 * repeat the request with proper MDS */
1000 rc = lmv_get_mea_and_update_object(exp, &rfid);
1002 ptlrpc_req_finished(*request);
1011 * llite passes fid of an target inode in data->fid1 and fid of directory in
1014 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1015 struct ptlrpc_request **request)
1017 struct obd_device *obd = exp->exp_obd;
1018 struct lmv_obd *lmv = &obd->u.lmv;
1019 struct lmv_obj *obj;
1023 rc = lmv_check_connect(obd);
1027 if (data->namelen != 0) {
1028 /* usual link request */
1029 obj = lmv_grab_obj(obd, &data->fid1);
1031 rc = raw_name2idx(obj->hashtype, obj->objcount, data->name,
1033 data->fid1 = obj->objs[rc].fid;
1037 CDEBUG(D_OTHER,"link %lu/%lu/%lu:%*s to %lu/%lu/%lu mds %lu\n",
1038 (unsigned long)data->fid2.mds,
1039 (unsigned long)data->fid2.id,
1040 (unsigned long)data->fid2.generation,
1041 data->namelen, data->name,
1042 (unsigned long)data->fid1.mds,
1043 (unsigned long)data->fid1.id,
1044 (unsigned long)data->fid1.generation,
1045 (unsigned long)data->fid1.mds);
1047 /* request from MDS to acquire i_links for inode by fid1 */
1048 CDEBUG(D_OTHER, "inc i_nlinks for %lu/%lu/%lu\n",
1049 (unsigned long)data->fid1.mds,
1050 (unsigned long)data->fid1.id,
1051 (unsigned long)data->fid1.generation);
1054 rc = md_link(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1058 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1059 const char *old, int oldlen, const char *new, int newlen,
1060 struct ptlrpc_request **request)
1062 struct obd_device *obd = exp->exp_obd;
1063 struct lmv_obd *lmv = &obd->u.lmv;
1064 struct lmv_obj *obj;
1068 CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n",
1069 oldlen, old, (unsigned long)data->fid1.mds,
1070 (unsigned long)data->fid1.id,
1071 (unsigned long)data->fid1.generation,
1072 newlen, new, (unsigned long) data->fid2.mds,
1073 (unsigned long) data->fid2.id,
1074 (unsigned long) data->fid2.generation);
1076 if (!fid_equal(&data->fid1, &data->fid2))
1077 CWARN("cross-node rename %lu/%lu/%lu:%*s to %lu/%lu/%lu:%*s\n",
1078 (unsigned long)data->fid1.mds,
1079 (unsigned long)data->fid1.id,
1080 (unsigned long)data->fid1.generation, oldlen, old,
1081 (unsigned long)data->fid2.mds,
1082 (unsigned long)data->fid2.id,
1083 (unsigned long)data->fid2.generation, newlen, new);
1085 rc = lmv_check_connect(obd);
1090 /* MDS with old dir entry is asking another MDS to create name
1093 "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n",
1094 newlen, new, oldlen, newlen,
1095 (unsigned long)data->fid2.mds,
1096 (unsigned long)data->fid2.id,
1097 (unsigned long)data->fid2.generation,
1098 (unsigned long)data->fid1.mds,
1099 (unsigned long)data->fid1.id,
1100 (unsigned long)data->fid1.generation);
1101 mds = data->fid2.mds;
1105 obj = lmv_grab_obj(obd, &data->fid1);
1107 /* directory is already splitted, so we have to forward request
1108 * to the right MDS */
1109 mds = raw_name2idx(obj->hashtype, obj->objcount, (char *)old, oldlen);
1110 data->fid1 = obj->objs[mds].fid;
1111 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1112 (unsigned long)obj->objs[mds].fid.mds,
1113 (unsigned long)obj->objs[mds].fid.id,
1114 (unsigned long)obj->objs[mds].fid.generation);
1118 obj = lmv_grab_obj(obd, &data->fid2);
1120 /* directory is already splitted, so we have to forward request
1121 * to the right MDS */
1122 mds = raw_name2idx(obj->hashtype, obj->objcount, (char *)new, newlen);
1123 data->fid2 = obj->objs[mds].fid;
1124 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1125 (unsigned long)obj->objs[mds].fid.mds,
1126 (unsigned long)obj->objs[mds].fid.id,
1127 (unsigned long)obj->objs[mds].fid.generation);
1131 mds = data->fid1.mds;
1134 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1135 new, newlen, request);
1139 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1140 struct iattr *iattr, void *ea, int ealen, void *ea2,
1141 int ea2len, struct ptlrpc_request **request)
1143 struct obd_device *obd = exp->exp_obd;
1144 struct lmv_obd *lmv = &obd->u.lmv;
1145 struct ptlrpc_request *req;
1146 struct mds_body *body;
1147 struct lmv_obj *obj;
1151 rc = lmv_check_connect(obd);
1155 obj = lmv_grab_obj(obd, &data->fid1);
1157 CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n",
1158 (unsigned long)data->fid1.mds, (unsigned long)data->fid1.id,
1159 (unsigned long)data->fid1.generation, iattr->ia_valid,
1160 obj ? ", splitted" : "");
1163 for (i = 0; i < obj->objcount; i++) {
1164 data->fid1 = obj->objs[i].fid;
1166 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1167 iattr, ea, ealen, ea2, ea2len, &req);
1169 if (fid_equal(&obj->fid, &obj->objs[i].fid)) {
1170 /* this is master object and this request should
1171 * be returned back to llite */
1174 ptlrpc_req_finished(req);
1182 LASSERT(data->fid1.mds < lmv->desc.ld_tgt_count);
1183 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1184 iattr, ea, ealen, ea2, ea2len, request);
1186 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1188 LASSERT(body != NULL);
1189 LASSERT(body->mds == data->fid1.mds);
1195 int lmv_sync(struct obd_export *exp, struct ll_fid *fid,
1196 struct ptlrpc_request **request)
1198 struct obd_device *obd = exp->exp_obd;
1199 struct lmv_obd *lmv = &obd->u.lmv;
1203 rc = lmv_check_connect(obd);
1207 rc = md_sync(lmv->tgts[fid->mds].ltd_exp, fid, request);
1211 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1212 void *data, int flag)
1214 struct lustre_handle lockh;
1215 struct lmv_obj *obj;
1220 case LDLM_CB_BLOCKING:
1221 ldlm_lock2handle(lock, &lockh);
1222 rc = ldlm_cli_cancel(&lockh);
1224 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1228 case LDLM_CB_CANCELING:
1229 /* time to drop cached attrs for dirobj */
1230 obj = lock->l_ast_data;
1232 CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n",
1233 lock->l_resource->lr_name.name[3] == 1 ? "LOOKUP" : "UPDATE",
1234 (unsigned long)lock->l_resource->lr_name.name[0],
1235 (unsigned long)lock->l_resource->lr_name.name[1],
1236 (unsigned long)obj->fid.mds, (unsigned long)obj->fid.id,
1237 (unsigned long)obj->fid.generation);
1247 void lmv_remove_dots(struct page *page)
1249 char *kaddr = page_address(page);
1250 unsigned limit = PAGE_CACHE_SIZE;
1251 unsigned offs, rec_len;
1252 struct ext2_dir_entry_2 *p;
1254 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1255 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1256 rec_len = le16_to_cpu(p->rec_len);
1258 if ((p->name_len == 1 && p->name[0] == '.') ||
1259 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1264 int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
1265 __u64 offset, struct page *page,
1266 struct ptlrpc_request **request)
1268 struct obd_device *obd = exp->exp_obd;
1269 struct lmv_obd *lmv = &obd->u.lmv;
1270 struct ll_fid rfid = *mdc_fid;
1271 struct lmv_obj *obj;
1275 rc = lmv_check_connect(obd);
1279 LASSERT(mdc_fid->mds < lmv->desc.ld_tgt_count);
1280 CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n",
1281 offset, (unsigned long) rfid.mds,
1282 (unsigned long) rfid.id,
1283 (unsigned long) rfid.generation);
1285 obj = lmv_grab_obj(obd, mdc_fid);
1289 /* find dirobj containing page with requested offset. */
1290 for (i = 0; i < obj->objcount; i++) {
1291 if (offset < obj->objs[i].size)
1293 offset -= obj->objs[i].size;
1295 rfid = obj->objs[i].fid;
1297 lmv_unlock_obj(obj);
1300 CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n",
1301 (unsigned long)rfid.mds, (unsigned long)rfid.id,
1302 (unsigned long)rfid.generation, (unsigned long)offset);
1304 rc = md_readpage(lmv->tgts[rfid.mds].ltd_exp, &rfid, offset,
1307 if (rc == 0 && !fid_equal(&rfid, mdc_fid))
1308 /* this page isn't from master object. To avoid "." and ".."
1309 * duplication in directory, we have to remove them from all
1311 lmv_remove_dots(page);
1316 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1317 struct ptlrpc_request **req)
1319 struct obd_device *obd = exp->exp_obd;
1320 struct lmv_obd *lmv = &obd->u.lmv;
1321 struct mea *mea = data->mea1;
1322 struct mdc_op_data data2;
1326 LASSERT(mea != NULL);
1327 for (i = 0; i < mea->mea_count; i++) {
1328 memset(&data2, 0, sizeof(data2));
1329 data2.fid1 = mea->mea_fids[i];
1330 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1331 mds = data2.fid1.mds;
1333 if (lmv->tgts[mds].ltd_exp == NULL)
1336 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1337 CDEBUG(D_OTHER, "unlink slave %lu/%lu/%lu -> %d\n",
1338 (unsigned long) mea->mea_fids[i].mds,
1339 (unsigned long) mea->mea_fids[i].id,
1340 (unsigned long) mea->mea_fids[i].generation, rc);
1342 ptlrpc_req_finished(*req);
1351 int lmv_delete_object(struct obd_export *exp, struct ll_fid *fid)
1355 if (!lmv_delete_obj(exp, fid)) {
1356 CDEBUG(D_OTHER, "Object %lu/%lu/%lu is not found.\n",
1357 (unsigned long)fid->mds, (unsigned long)fid->id,
1358 (unsigned long)fid->generation);
1364 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1365 struct ptlrpc_request **request)
1367 struct obd_device *obd = exp->exp_obd;
1368 struct lmv_obd *lmv = &obd->u.lmv;
1372 rc = lmv_check_connect(obd);
1376 if (data->namelen == 0 && data->mea1 != NULL) {
1377 /* mds asks to remove slave objects */
1378 rc = lmv_unlink_slaves(exp, data, request);
1380 } else if (data->namelen != 0) {
1381 struct lmv_obj *obj;
1383 obj = lmv_grab_obj(obd, &data->fid1);
1385 i = raw_name2idx(obj->hashtype, obj->objcount, data->name,
1387 data->fid1 = obj->objs[i].fid;
1390 CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n",
1391 data->namelen, data->name,
1392 (unsigned long) data->fid1.mds,
1393 (unsigned long) data->fid1.id,
1394 (unsigned long) data->fid1.generation, i);
1396 CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n",
1397 (unsigned long) data->fid1.mds,
1398 (unsigned long) data->fid1.id,
1399 (unsigned long) data->fid1.generation);
1401 rc = md_unlink(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1405 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1406 char *name, int len)
1408 struct obd_device *obd = exp->exp_obd;
1409 struct lmv_obd *lmv = &obd->u.lmv;
1413 rc = lmv_check_connect(obd);
1415 RETURN(ERR_PTR(rc));
1416 #warning "we need well-desgined readdir() implementation to remove this mess"
1417 obd = lmv->tgts[0].ltd_exp->exp_obd;
1422 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1424 struct obd_device *obd = exp->exp_obd;
1425 struct lmv_obd *lmv = &obd->u.lmv;
1426 int i, rc = 0, change = 0;
1429 if (lmv->max_easize < easize) {
1430 lmv->max_easize = easize;
1433 if (lmv->max_cookiesize < cookiesize) {
1434 lmv->max_cookiesize = cookiesize;
1440 if (lmv->connected == 0)
1443 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1444 if (lmv->tgts[i].ltd_exp == NULL) {
1445 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1449 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1451 CERROR("obd_init_ea_size() failed on MDT target %d, "
1452 "error %d.\n", i, rc);
1459 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1460 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1462 struct obd_device *obd = exp->exp_obd;
1463 struct lmv_obd *lmv = &obd->u.lmv;
1464 struct lov_stripe_md obj_md;
1465 struct lov_stripe_md *obj_mdp = &obj_md;
1469 rc = lmv_check_connect(obd);
1473 LASSERT(ea == NULL);
1474 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1476 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1482 * to be called from MDS only
1484 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1485 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1487 struct obd_device *obd = exp->exp_obd;
1488 struct lmv_obd *lmv = &obd->u.lmv;
1495 rc = lmv_check_connect(obd);
1499 LASSERT(oa != NULL);
1502 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1507 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1509 CERROR("obd_alloc_diskmd() failed, error %d\n",
1520 mfid.generation = oa->o_generation;
1522 mea = (struct mea *)*ea;
1523 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1524 mea->mea_count = lmv->desc.ld_tgt_count;
1525 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1527 mea->mea_master = -1;
1528 lcount = lmv->desc.ld_tgt_count;
1529 for (i = 0, c = 0; c < mea->mea_count && i < lcount; i++) {
1530 struct lov_stripe_md obj_md;
1531 struct lov_stripe_md *obj_mdp = &obj_md;
1533 if (lmv->tgts[i].ltd_exp == NULL) {
1534 /* this is master MDS */
1535 mea->mea_fids[c].id = mfid.id;
1536 mea->mea_fids[c].generation = mfid.generation;
1537 mea->mea_fids[c].mds = i;
1538 mea->mea_master = i;
1543 /* "master" MDS should always be part of stripped dir, so scan
1545 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1548 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE
1549 | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1551 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1553 CERROR("obd_create() failed on MDT target %d, "
1554 "error %d\n", c, rc);
1558 mea->mea_fids[c].id = oa->o_id;
1559 mea->mea_fids[c].generation = oa->o_generation;
1560 mea->mea_fids[c].mds = i;
1562 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1563 i, oa->o_id, oa->o_generation);
1565 LASSERT(c == mea->mea_count);
1566 CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1571 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1572 struct obd_device *tgt, int count,
1573 struct llog_catid *logid)
1575 struct llog_ctxt *ctxt;
1579 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1582 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1583 ctxt->loc_imp = tgt->u.cli.cl_import;
1589 static int lmv_llog_finish(struct obd_device *obd,
1590 struct obd_llogs *llogs, int count)
1595 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1599 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1600 void *key, __u32 *vallen, void *val)
1602 struct obd_device *obd;
1603 struct lmv_obd *lmv;
1606 obd = class_exp2obd(exp);
1608 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1609 exp->exp_handle.h_cookie);
1614 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1615 __u32 *mdsize = val;
1616 *vallen = sizeof(__u32);
1617 *mdsize = sizeof(struct ll_fid) * lmv->desc.ld_tgt_count
1618 + sizeof(struct mea);
1620 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1621 struct obd_uuid *cluuid = &lmv->cluuid;
1622 struct lmv_tgt_desc *tgts;
1623 __u32 *mdsnum = val;
1626 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1627 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1628 *vallen = sizeof(__u32);
1636 CDEBUG(D_IOCTL, "invalid key\n");
1640 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1641 void *key, obd_count vallen, void *val)
1643 struct obd_device *obd;
1644 struct lmv_obd *lmv;
1647 obd = class_exp2obd(exp);
1649 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1650 exp->exp_handle.h_cookie);
1655 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1656 lmv->server_timeout = 1;
1657 lmv_set_timeouts(obd);
1664 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1665 struct lov_stripe_md *lsm)
1667 struct obd_device *obd = class_exp2obd(exp);
1668 struct lmv_obd *lmv = &obd->u.lmv;
1672 mea_size = sizeof(struct ll_fid) *
1673 lmv->desc.ld_tgt_count + sizeof(struct mea);
1677 if (*lmmp && !lsm) {
1678 OBD_FREE(*lmmp, mea_size);
1683 if (*lmmp == NULL) {
1684 OBD_ALLOC(*lmmp, mea_size);
1692 #warning "MEA packing/convertation must be here! -bzzz"
1693 memcpy(*lmmp, lsm, mea_size);
1697 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1698 struct lov_mds_md *disk_src, int mdsize)
1700 struct obd_device *obd = class_exp2obd(exp);
1701 struct lmv_obd *lmv = &obd->u.lmv;
1702 struct mea **tmea = (struct mea **) mem_tgt;
1703 struct mea *mea = (void *) disk_src;
1707 mea_size = sizeof(struct ll_fid) *
1708 lmv->desc.ld_tgt_count + sizeof(struct mea);
1709 if (mem_tgt == NULL)
1712 if (*mem_tgt != NULL && disk_src == NULL) {
1713 OBD_FREE(*tmea, mea_size);
1717 LASSERT(mea_size == mdsize);
1719 OBD_ALLOC(*tmea, mea_size);
1726 #warning "MEA unpacking/convertation must be here! -bzzz"
1727 memcpy(*tmea, mea, mdsize);
1731 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1732 struct lov_stripe_md *ea, obd_count oa_bufs,
1733 struct brw_page *pgarr, struct obd_trans_info *oti)
1735 struct obd_device *obd = exp->exp_obd;
1736 struct lmv_obd *lmv = &obd->u.lmv;
1737 struct mea *mea = (struct mea *) ea;
1740 LASSERT(oa != NULL);
1741 LASSERT(ea != NULL);
1742 LASSERT(pgarr != NULL);
1743 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1745 oa->o_gr = mea->mea_fids[oa->o_mds].generation;
1746 oa->o_id = mea->mea_fids[oa->o_mds].id;
1747 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1748 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, oa,
1749 NULL, oa_bufs, pgarr, oti);
1753 struct obd_ops lmv_obd_ops = {
1754 .o_owner = THIS_MODULE,
1755 .o_attach = lmv_attach,
1756 .o_detach = lmv_detach,
1757 .o_setup = lmv_setup,
1758 .o_cleanup = lmv_cleanup,
1759 .o_connect = lmv_connect,
1760 .o_disconnect = lmv_disconnect,
1761 .o_statfs = lmv_statfs,
1762 .o_llog_init = lmv_llog_init,
1763 .o_llog_finish = lmv_llog_finish,
1764 .o_get_info = lmv_get_info,
1765 .o_set_info = lmv_set_info,
1766 .o_create = lmv_obd_create,
1767 .o_packmd = lmv_packmd,
1768 .o_unpackmd = lmv_unpackmd,
1770 .o_init_ea_size = lmv_init_ea_size,
1771 .o_notify = lmv_notify,
1772 .o_iocontrol = lmv_iocontrol,
1775 struct md_ops lmv_md_ops = {
1776 .m_getstatus = lmv_getstatus,
1777 .m_getattr = lmv_getattr,
1778 .m_change_cbdata = lmv_change_cbdata,
1779 .m_change_cbdata_name = lmv_change_cbdata_name,
1780 .m_close = lmv_close,
1781 .m_create = lmv_create,
1782 .m_done_writing = lmv_done_writing,
1783 .m_enqueue = lmv_enqueue,
1784 .m_getattr_name = lmv_getattr_name,
1785 .m_intent_lock = lmv_intent_lock,
1787 .m_rename = lmv_rename,
1788 .m_setattr = lmv_setattr,
1790 .m_readpage = lmv_readpage,
1791 .m_unlink = lmv_unlink,
1792 .m_get_real_obd = lmv_get_real_obd,
1793 .m_valid_attrs = lmv_valid_attrs,
1794 .m_delete_object = lmv_delete_object,
1797 int __init lmv_init(void)
1799 struct lprocfs_static_vars lvars;
1802 lprocfs_init_vars(lmv, &lvars);
1803 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1804 lvars.module_vars, OBD_LMV_DEVICENAME);
1809 static void lmv_exit(void)
1811 class_unregister_type(OBD_LMV_DEVICENAME);
1814 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1815 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1816 MODULE_LICENSE("GPL");
1818 module_init(lmv_init);
1819 module_exit(lmv_exit);