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,
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);
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)
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;
203 void lmv_set_timeouts(struct obd_device *obd)
205 struct lmv_tgt_desc *tgts;
210 if (lmv->server_timeout == 0)
213 if (lmv->connected == 0)
216 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
217 if (tgts->ltd_exp == NULL)
219 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
220 "inter_mds", 0, NULL);
224 /* Performs a check if passed obd is connected. If no - connect it. */
225 int lmv_check_connect(struct obd_device *obd) {
226 struct lmv_obd *lmv = &obd->u.lmv;
227 struct obd_uuid *cluuid;
228 struct lmv_tgt_desc *tgts;
229 struct obd_export *exp;
236 cluuid = &lmv->cluuid;
239 CDEBUG(D_OTHER, "time to connect %s to %s\n",
240 cluuid->uuid, obd->obd_name);
242 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
243 struct obd_device *tgt_obd;
244 struct obd_uuid lmv_osc_uuid = { "LMV_OSC_UUID" };
245 struct lustre_handle conn = {0, };
247 LASSERT(tgts != NULL);
249 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
252 CERROR("Target %s not attached\n", tgts->uuid.uuid);
253 GOTO(out_disc, rc = -EINVAL);
256 /* for MDS: don't connect to yourself */
257 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
258 CDEBUG(D_OTHER, "don't connect back to %s\n",
260 tgts->ltd_exp = NULL;
264 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
265 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
266 tgts->uuid.uuid, obd->obd_uuid.uuid,
269 if (!tgt_obd->obd_set_up) {
270 CERROR("Target %s not set up\n", tgts->uuid.uuid);
271 GOTO(out_disc, rc = -EINVAL);
274 rc = obd_connect(&conn, tgt_obd, &lmv_osc_uuid);
276 CERROR("Target %s connect error %d\n",
277 tgts->uuid.uuid, rc);
280 tgts->ltd_exp = class_conn2export(&conn);
282 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
283 lmv->max_cookiesize);
285 rc = obd_register_observer(tgt_obd, obd);
287 CERROR("Target %s register_observer error %d\n",
288 tgts->uuid.uuid, rc);
289 obd_disconnect(tgts->ltd_exp, 0);
293 lmv->desc.ld_active_tgt_count++;
296 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
297 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
298 atomic_read(&obd->obd_refcount));
301 lmv_set_timeouts(obd);
302 class_export_put(exp);
307 struct obd_uuid uuid;
309 --lmv->desc.ld_active_tgt_count;
311 /* save for CERROR below; (we know it's terminated) */
313 rc2 = obd_disconnect(tgts->ltd_exp, 0);
315 CERROR("error: LMV target %s disconnect on MDT idx %d: "
316 "error %d\n", uuid.uuid, i, rc2);
318 class_disconnect(exp, 0);
322 static int lmv_disconnect(struct obd_export *exp, int flags)
324 struct obd_device *obd = class_exp2obd(exp);
325 struct lmv_obd *lmv = &obd->u.lmv;
332 /* Only disconnect the underlying layers on the final disconnect. */
334 if (lmv->refcount != 0)
337 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
338 if (lmv->tgts[i].ltd_exp == NULL)
341 if (obd->obd_no_recov) {
342 /* Pass it on to our clients.
343 * XXX This should be an argument to disconnect,
344 * XXX not a back-door flag on the OBD. Ah well.
346 struct obd_device *mdc_obd;
347 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
349 mdc_obd->obd_no_recov = 1;
352 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
353 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
354 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
356 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
358 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
360 if (lmv->tgts[i].active) {
361 CERROR("Target %s disconnect error %d\n",
362 lmv->tgts[i].uuid.uuid, rc);
366 if (lmv->tgts[i].active) {
367 lmv->desc.ld_active_tgt_count--;
368 lmv->tgts[i].active = 0;
370 lmv->tgts[i].ltd_exp = NULL;
374 /* this is the case when no real connection is established by
375 * lmv_check_connect(). */
377 class_export_put(exp);
378 rc = class_disconnect(exp, 0);
379 if (lmv->refcount == 0)
384 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
385 int len, void *karg, void *uarg)
387 struct obd_device *obddev = class_exp2obd(exp);
388 struct lmv_obd *lmv = &obddev->u.lmv;
389 int i, rc = 0, set = 0;
393 if (lmv->desc.ld_tgt_count == 0)
396 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
399 if (lmv->tgts[i].ltd_exp == NULL) {
400 CWARN("%s: NULL export for %d\n", obddev->obd_name, i);
404 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
406 if (lmv->tgts[i].active) {
407 CERROR("error: iocontrol MDC %s on MDT"
408 "idx %d: err = %d\n",
409 lmv->tgts[i].uuid.uuid, i, err);
422 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
425 struct lmv_desc *desc;
426 struct obd_uuid *uuids;
427 struct lmv_tgt_desc *tgts;
428 struct lustre_cfg *lcfg = buf;
429 struct lmv_obd *lmv = &obd->u.lmv;
432 if (lcfg->lcfg_inllen1 < 1) {
433 CERROR("LMV setup requires a descriptor\n");
437 if (lcfg->lcfg_inllen2 < 1) {
438 CERROR("LMV setup requires an OST UUID list\n");
442 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
443 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
444 CERROR("descriptor size wrong: %d > %d\n",
445 (int)sizeof(*desc), lcfg->lcfg_inllen1);
449 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
450 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
451 CERROR("UUID array size wrong: %u * %u != %u\n",
452 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
456 lmv->bufsize = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
457 OBD_ALLOC(lmv->tgts, lmv->bufsize);
458 if (lmv->tgts == NULL) {
459 CERROR("Out of memory\n");
464 spin_lock_init(&lmv->lmv_lock);
466 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
467 tgts->uuid = uuids[i];
469 lmv->max_cookiesize = 0;
471 lmv->max_easize = sizeof(struct ll_fid) *
472 desc->ld_tgt_count + sizeof(struct mea);
474 rc = lmv_setup_mgr(obd);
476 CERROR("Can't setup LMV object manager, "
478 OBD_FREE(lmv->tgts, lmv->bufsize);
484 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
485 unsigned long max_age)
487 struct lmv_obd *lmv = &obd->u.lmv;
488 struct obd_statfs temp;
492 rc = lmv_check_connect(obd);
496 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
497 if (lmv->tgts[i].ltd_exp == NULL) {
498 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
502 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
504 CERROR("can't stat MDS #%d (%s)\n", i,
505 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
509 memcpy(osfs, &temp, sizeof(temp));
511 osfs->os_bavail += temp.os_bavail;
512 osfs->os_blocks += temp.os_blocks;
513 osfs->os_ffree += temp.os_ffree;
514 osfs->os_files += temp.os_files;
520 static int lmv_cleanup(struct obd_device *obd, int flags)
522 struct lmv_obd *lmv = &obd->u.lmv;
524 lmv_cleanup_mgr(obd);
525 OBD_FREE(lmv->tgts, lmv->bufsize);
529 static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid)
531 struct obd_device *obd = exp->exp_obd;
532 struct lmv_obd *lmv = &obd->u.lmv;
535 rc = lmv_check_connect(obd);
538 rc = md_getstatus(lmv->tgts[0].ltd_exp, fid);
543 static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid,
544 unsigned long valid, unsigned int ea_size,
545 struct ptlrpc_request **request)
547 struct obd_device *obd = exp->exp_obd;
548 struct lmv_obd *lmv = &obd->u.lmv;
549 int rc, i = fid->mds;
553 rc = lmv_check_connect(obd);
557 LASSERT(i < lmv->desc.ld_tgt_count);
559 rc = md_getattr(lmv->tgts[i].ltd_exp, fid, valid,
564 obj = lmv_grab_obj(obd, fid);
566 CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n",
567 (unsigned long)fid->mds, (unsigned long)fid->id,
568 (unsigned long)fid->generation, obj ? "(splitted)" : "");
570 /* if object is splitted, then we loop over all the slaves and gather
571 * size attribute. In ideal world we would have to gather also mds field
572 * from all slaves, as object is spread over the cluster and this is
573 * definitely interesting information and it is not good to loss it,
576 struct mds_body *body;
578 if (*request == NULL) {
583 body = lustre_msg_buf((*request)->rq_repmsg, 0,
585 LASSERT(body != NULL);
589 for (i = 0; i < obj->objcount; i++) {
591 if (lmv->tgts[i].ltd_exp == NULL) {
592 CWARN("%s: NULL export for %d\n",
597 /* skip master obj. */
598 if (fid_equal(&obj->fid, &obj->objs[i].fid))
601 body->size += obj->objs[i].size;
611 static int lmv_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
612 ldlm_iterator_t it, void *data)
614 struct obd_device *obd = exp->exp_obd;
615 struct lmv_obd *lmv = &obd->u.lmv;
619 rc = lmv_check_connect(obd);
623 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n", (unsigned long)fid->mds,
624 (unsigned long)fid->id, (unsigned long)fid->generation);
626 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
628 rc = md_change_cbdata(lmv->tgts[fid->mds].ltd_exp,
634 static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
635 char *name, int len, struct ll_fid *cfid,
636 ldlm_iterator_t it, void *data)
638 struct obd_device *obd = exp->exp_obd;
639 struct lmv_obd *lmv = &obd->u.lmv;
644 rc = lmv_check_connect(obd);
648 LASSERT(pfid->mds < lmv->desc.ld_tgt_count);
649 LASSERT(cfid->mds < lmv->desc.ld_tgt_count);
651 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n",
652 (unsigned long)pfid->mds, (unsigned long)pfid->id,
653 (unsigned long)pfid->generation, len, name,
654 (unsigned long)cfid->mds, (unsigned long)cfid->id,
655 (unsigned long)cfid->generation);
657 /* this is default mds for directory name belongs to. */
659 obj = lmv_grab_obj(obd, pfid);
661 /* directory is splitted. look for right mds for this name. */
662 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
663 mds = obj->objs[mds].fid.mds;
666 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cfid, it, data);
670 static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
672 struct obd_device *obd = exp->exp_obd;
673 struct lmv_obd *lmv = &obd->u.lmv;
676 rc = lmv_check_connect(obd);
679 CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n", (unsigned long) fid->mds,
680 (unsigned long) fid->id, (unsigned long) fid->generation);
681 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
682 rc = md_valid_attrs(lmv->tgts[fid->mds].ltd_exp, fid);
686 int lmv_close(struct obd_export *exp, struct obdo *obdo,
687 struct obd_client_handle *och,
688 struct ptlrpc_request **request)
690 struct obd_device *obd = exp->exp_obd;
691 struct lmv_obd *lmv = &obd->u.lmv;
692 int rc, i = obdo->o_mds;
694 rc = lmv_check_connect(obd);
697 LASSERT(i < lmv->desc.ld_tgt_count);
698 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds,
699 (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation);
700 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
704 int lmv_get_mea_and_update_object(struct obd_export *exp, struct ll_fid *fid)
706 struct obd_device *obd = exp->exp_obd;
707 struct lmv_obd *lmv = &obd->u.lmv;
708 struct ptlrpc_request *req = NULL;
715 mealen = MEA_SIZE_LMV(lmv);
717 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
719 /* time to update mea of parent fid */
720 rc = md_getattr(lmv->tgts[fid->mds].ltd_exp, fid,
721 valid, mealen, &req);
723 CERROR("md_getattr() failed, error %d\n", rc);
727 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
729 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
734 GOTO(cleanup, rc = -ENODATA);
736 obj = lmv_create_obj(exp, fid, md.mea);
741 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
745 ptlrpc_req_finished(req);
749 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
750 const void *data, int datalen, int mode, __u32 uid,
751 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
753 struct obd_device *obd = exp->exp_obd;
754 struct lmv_obd *lmv = &obd->u.lmv;
755 struct mds_body *body;
757 int rc, mds, loop = 0;
760 rc = lmv_check_connect(obd);
764 if (!lmv->desc.ld_active_tgt_count)
767 LASSERT(++loop <= 2);
768 obj = lmv_grab_obj(obd, &op_data->fid1);
770 mds = raw_name2idx(obj->hashtype, obj->objcount, op_data->name,
772 op_data->fid1 = obj->objs[mds].fid;
776 CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu\n", op_data->namelen,
777 op_data->name, (unsigned long)op_data->fid1.mds,
778 (unsigned long)op_data->fid1.id,
779 (unsigned long)op_data->fid1.generation);
781 rc = md_create(lmv->tgts[op_data->fid1.mds].ltd_exp, op_data, data,
782 datalen, mode, uid, gid, rdev, request);
784 if (*request == NULL)
787 body = lustre_msg_buf((*request)->rq_repmsg, 0,
789 LASSERT(body != NULL);
791 CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, "
792 "mds = %d\n", (unsigned long)body->fid1.id,
793 (unsigned long)body->fid1.generation, op_data->fid1.mds);
795 LASSERT(body->valid & OBD_MD_MDS ||
796 body->mds == op_data->fid1.mds);
797 } else if (rc == -ERESTART) {
798 /* directory got splitted. time to update local object and
799 * repeat the request with proper MDS */
800 rc = lmv_get_mea_and_update_object(exp, &op_data->fid1);
802 ptlrpc_req_finished(*request);
809 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
811 struct obd_device *obd = exp->exp_obd;
812 struct lmv_obd *lmv = &obd->u.lmv;
815 rc = lmv_check_connect(obd);
819 /* FIXME: choose right MDC here */
820 CWARN("this method isn't implemented yet\n");
821 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
825 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
826 struct lookup_intent *it, int lockmode,
827 struct mdc_op_data *data, struct lustre_handle *lockh,
828 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
829 ldlm_blocking_callback cb_blocking, void *cb_data)
831 struct obd_device *obd = exp->exp_obd;
832 struct lmv_obd *lmv = &obd->u.lmv;
833 struct mea *mea = data->mea1;
834 struct mdc_op_data data2;
838 LASSERT(mea != NULL);
839 for (i = 0; i < mea->mea_count; i++) {
840 memset(&data2, 0, sizeof(data2));
841 data2.fid1 = mea->mea_fids[i];
842 mds = data2.fid1.mds;
844 if (lmv->tgts[mds].ltd_exp == NULL)
847 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, lockmode,
848 &data2, lockh + i, lmm, lmmsize, cb_completion,
849 cb_blocking, cb_data);
851 CDEBUG(D_OTHER, "take lock on slave %lu/%lu/%lu -> %d/%d\n",
852 (unsigned long)mea->mea_fids[i].mds,
853 (unsigned long)mea->mea_fids[i].id,
854 (unsigned long)mea->mea_fids[i].generation,
855 rc, it->d.lustre.it_status);
858 if (it->d.lustre.it_data) {
859 struct ptlrpc_request *req;
860 req = (struct ptlrpc_request *) it->d.lustre.it_data;
861 ptlrpc_req_finished(req);
864 if (it->d.lustre.it_status)
865 GOTO(cleanup, rc = it->d.lustre.it_status);
870 /* drop all taken locks */
873 ldlm_lock_decref(lockh + i, lockmode);
879 int lmv_enqueue(struct obd_export *exp, int lock_type,
880 struct lookup_intent *it, int lock_mode,
881 struct mdc_op_data *data, struct lustre_handle *lockh,
882 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
883 ldlm_blocking_callback cb_blocking, void *cb_data)
885 struct obd_device *obd = exp->exp_obd;
886 struct lmv_obd *lmv = &obd->u.lmv;
891 rc = lmv_check_connect(obd);
895 if (it->it_op == IT_UNLINK) {
896 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
897 data, lockh, lmm, lmmsize,
898 cb_completion, cb_blocking, cb_data);
903 obj = lmv_grab_obj(obd, &data->fid1);
905 /* directory is splitted. look for right mds for this
907 mds = raw_name2idx(obj->hashtype, obj->objcount,
908 (char *)data->name, data->namelen);
909 data->fid1 = obj->objs[mds].fid;
913 CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n", LL_IT2STR(it),
914 (unsigned long)data->fid1.id, (unsigned long)data->fid1.generation);
916 rc = md_enqueue(lmv->tgts[data->fid1.mds].ltd_exp, lock_type, it,
917 lock_mode, data, lockh, lmm, lmmsize, cb_completion,
918 cb_blocking, cb_data);
923 int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid,
924 char *filename, int namelen, unsigned long valid,
925 unsigned int ea_size, struct ptlrpc_request **request)
927 struct obd_device *obd = exp->exp_obd;
928 struct lmv_obd *lmv = &obd->u.lmv;
929 struct ll_fid rfid = *fid;
930 int rc, mds = fid->mds, loop = 0;
931 struct mds_body *body;
934 rc = lmv_check_connect(obd);
938 LASSERT(++loop <= 2);
939 obj = lmv_grab_obj(obd, fid);
941 /* directory is splitted. look for right mds for this name */
942 mds = raw_name2idx(obj->hashtype, obj->objcount, filename, namelen - 1);
943 rfid = obj->objs[mds].fid;
947 CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu -> %lu/%lu/%lu\n",
948 namelen, filename, (unsigned long)fid->mds,
949 (unsigned long)fid->id, (unsigned long)fid->generation,
950 (unsigned long)rfid.mds, (unsigned long)rfid.id,
951 (unsigned long)rfid.generation);
953 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid, filename,
954 namelen, valid, ea_size, request);
956 /* this could be cross-node reference. in this case all we have
957 * right now is mds/ino/generation triple. we'd like to find
958 * other attributes */
959 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
960 LASSERT(body != NULL);
961 if (body->valid & OBD_MD_MDS) {
962 struct ptlrpc_request *req = NULL;
964 CDEBUG(D_OTHER, "request attrs for %lu/%lu/%lu\n",
965 (unsigned long) rfid.mds,
966 (unsigned long) rfid.id,
967 (unsigned long) rfid.generation);
968 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid,
969 NULL, 1, valid, ea_size, &req);
970 ptlrpc_req_finished(*request);
973 } else if (rc == -ERESTART) {
974 /* directory got splitted. time to update local object and
975 * repeat the request with proper MDS */
976 rc = lmv_get_mea_and_update_object(exp, &rfid);
978 ptlrpc_req_finished(*request);
987 * llite passes fid of an target inode in data->fid1 and fid of directory in
990 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
991 struct ptlrpc_request **request)
993 struct obd_device *obd = exp->exp_obd;
994 struct lmv_obd *lmv = &obd->u.lmv;
999 rc = lmv_check_connect(obd);
1003 if (data->namelen != 0) {
1004 /* usual link request */
1005 obj = lmv_grab_obj(obd, &data->fid1);
1007 rc = raw_name2idx(obj->hashtype, obj->objcount, data->name,
1009 data->fid1 = obj->objs[rc].fid;
1013 CDEBUG(D_OTHER,"link %lu/%lu/%lu:%*s to %lu/%lu/%lu mds %lu\n",
1014 (unsigned long)data->fid2.mds,
1015 (unsigned long)data->fid2.id,
1016 (unsigned long)data->fid2.generation,
1017 data->namelen, data->name,
1018 (unsigned long)data->fid1.mds,
1019 (unsigned long)data->fid1.id,
1020 (unsigned long)data->fid1.generation,
1021 (unsigned long)data->fid1.mds);
1023 /* request from MDS to acquire i_links for inode by fid1 */
1024 CDEBUG(D_OTHER, "inc i_nlinks for %lu/%lu/%lu\n",
1025 (unsigned long)data->fid1.mds,
1026 (unsigned long)data->fid1.id,
1027 (unsigned long)data->fid1.generation);
1030 rc = md_link(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1034 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1035 const char *old, int oldlen, const char *new, int newlen,
1036 struct ptlrpc_request **request)
1038 struct obd_device *obd = exp->exp_obd;
1039 struct lmv_obd *lmv = &obd->u.lmv;
1040 struct lmv_obj *obj;
1044 CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n",
1045 oldlen, old, (unsigned long)data->fid1.mds,
1046 (unsigned long)data->fid1.id,
1047 (unsigned long)data->fid1.generation,
1048 newlen, new, (unsigned long) data->fid2.mds,
1049 (unsigned long) data->fid2.id,
1050 (unsigned long) data->fid2.generation);
1052 if (!fid_equal(&data->fid1, &data->fid2))
1053 CWARN("cross-node rename %lu/%lu/%lu:%*s to %lu/%lu/%lu:%*s\n",
1054 (unsigned long)data->fid1.mds,
1055 (unsigned long)data->fid1.id,
1056 (unsigned long)data->fid1.generation, oldlen, old,
1057 (unsigned long)data->fid2.mds,
1058 (unsigned long)data->fid2.id,
1059 (unsigned long)data->fid2.generation, newlen, new);
1061 rc = lmv_check_connect(obd);
1066 /* MDS with old dir entry is asking another MDS to create name
1069 "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n",
1070 newlen, new, oldlen, newlen,
1071 (unsigned long)data->fid2.mds,
1072 (unsigned long)data->fid2.id,
1073 (unsigned long)data->fid2.generation,
1074 (unsigned long)data->fid1.mds,
1075 (unsigned long)data->fid1.id,
1076 (unsigned long)data->fid1.generation);
1077 mds = data->fid2.mds;
1081 obj = lmv_grab_obj(obd, &data->fid1);
1083 /* directory is already splitted, so we have to forward request
1084 * to the right MDS */
1085 mds = raw_name2idx(obj->hashtype, obj->objcount, (char *)old, oldlen);
1086 data->fid1 = obj->objs[mds].fid;
1087 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1088 (unsigned long)obj->objs[mds].fid.mds,
1089 (unsigned long)obj->objs[mds].fid.id,
1090 (unsigned long)obj->objs[mds].fid.generation);
1094 obj = lmv_grab_obj(obd, &data->fid2);
1096 /* directory is already splitted, so we have to forward request
1097 * to the right MDS */
1098 mds = raw_name2idx(obj->hashtype, obj->objcount, (char *)new, newlen);
1099 data->fid2 = obj->objs[mds].fid;
1100 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1101 (unsigned long)obj->objs[mds].fid.mds,
1102 (unsigned long)obj->objs[mds].fid.id,
1103 (unsigned long)obj->objs[mds].fid.generation);
1107 mds = data->fid1.mds;
1110 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1111 new, newlen, request);
1115 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1116 struct iattr *iattr, void *ea, int ealen, void *ea2,
1117 int ea2len, struct ptlrpc_request **request)
1119 struct obd_device *obd = exp->exp_obd;
1120 struct lmv_obd *lmv = &obd->u.lmv;
1121 struct ptlrpc_request *req;
1122 struct mds_body *body;
1123 struct lmv_obj *obj;
1127 rc = lmv_check_connect(obd);
1131 obj = lmv_grab_obj(obd, &data->fid1);
1133 CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n",
1134 (unsigned long)data->fid1.mds, (unsigned long)data->fid1.id,
1135 (unsigned long)data->fid1.generation, iattr->ia_valid,
1136 obj ? ", splitted" : "");
1139 for (i = 0; i < obj->objcount; i++) {
1140 data->fid1 = obj->objs[i].fid;
1142 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1143 iattr, ea, ealen, ea2, ea2len, &req);
1145 if (fid_equal(&obj->fid, &obj->objs[i].fid)) {
1146 /* this is master object and this request should
1147 * be returned back to llite */
1150 ptlrpc_req_finished(req);
1158 LASSERT(data->fid1.mds < lmv->desc.ld_tgt_count);
1159 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1160 iattr, ea, ealen, ea2, ea2len, request);
1162 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1164 LASSERT(body != NULL);
1165 LASSERT(body->mds == data->fid1.mds);
1171 int lmv_sync(struct obd_export *exp, struct ll_fid *fid,
1172 struct ptlrpc_request **request)
1174 struct obd_device *obd = exp->exp_obd;
1175 struct lmv_obd *lmv = &obd->u.lmv;
1179 rc = lmv_check_connect(obd);
1183 rc = md_sync(lmv->tgts[fid->mds].ltd_exp, fid, request);
1187 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1188 void *data, int flag)
1190 struct lustre_handle lockh;
1191 struct lmv_obj *obj;
1196 case LDLM_CB_BLOCKING:
1197 ldlm_lock2handle(lock, &lockh);
1198 rc = ldlm_cli_cancel(&lockh);
1200 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1204 case LDLM_CB_CANCELING:
1205 /* time to drop cached attrs for dirobj */
1206 obj = lock->l_ast_data;
1208 CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n",
1209 lock->l_resource->lr_name.name[3] == 1 ? "LOOKUP" : "UPDATE",
1210 (unsigned long)lock->l_resource->lr_name.name[0],
1211 (unsigned long)lock->l_resource->lr_name.name[1],
1212 (unsigned long)obj->fid.mds, (unsigned long)obj->fid.id,
1213 (unsigned long)obj->fid.generation);
1223 void lmv_remove_dots(struct page *page)
1225 char *kaddr = page_address(page);
1226 unsigned limit = PAGE_CACHE_SIZE;
1227 unsigned offs, rec_len;
1228 struct ext2_dir_entry_2 *p;
1230 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1231 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1232 rec_len = le16_to_cpu(p->rec_len);
1234 if ((p->name_len == 1 && p->name[0] == '.') ||
1235 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1240 int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
1241 __u64 offset, struct page *page,
1242 struct ptlrpc_request **request)
1244 struct obd_device *obd = exp->exp_obd;
1245 struct lmv_obd *lmv = &obd->u.lmv;
1246 struct ll_fid rfid = *mdc_fid;
1247 struct lmv_obj *obj;
1251 rc = lmv_check_connect(obd);
1255 LASSERT(mdc_fid->mds < lmv->desc.ld_tgt_count);
1256 CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n",
1257 offset, (unsigned long) rfid.mds,
1258 (unsigned long) rfid.id,
1259 (unsigned long) rfid.generation);
1261 obj = lmv_grab_obj(obd, mdc_fid);
1265 /* find dirobj containing page with requested offset. */
1266 for (i = 0; i < obj->objcount; i++) {
1267 if (offset < obj->objs[i].size)
1269 offset -= obj->objs[i].size;
1271 rfid = obj->objs[i].fid;
1273 lmv_unlock_obj(obj);
1276 CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n",
1277 (unsigned long)rfid.mds, (unsigned long)rfid.id,
1278 (unsigned long)rfid.generation, (unsigned long)offset);
1280 rc = md_readpage(lmv->tgts[rfid.mds].ltd_exp, &rfid, offset,
1283 if (rc == 0 && !fid_equal(&rfid, mdc_fid))
1284 /* this page isn't from master object. To avoid "." and ".."
1285 * duplication in directory, we have to remove them from all
1287 lmv_remove_dots(page);
1292 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1293 struct ptlrpc_request **req)
1295 struct obd_device *obd = exp->exp_obd;
1296 struct lmv_obd *lmv = &obd->u.lmv;
1297 struct mea *mea = data->mea1;
1298 struct mdc_op_data data2;
1302 LASSERT(mea != NULL);
1303 for (i = 0; i < mea->mea_count; i++) {
1304 memset(&data2, 0, sizeof(data2));
1305 data2.fid1 = mea->mea_fids[i];
1306 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1307 mds = data2.fid1.mds;
1309 if (lmv->tgts[mds].ltd_exp == NULL)
1312 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1313 CDEBUG(D_OTHER, "unlink slave %lu/%lu/%lu -> %d\n",
1314 (unsigned long) mea->mea_fids[i].mds,
1315 (unsigned long) mea->mea_fids[i].id,
1316 (unsigned long) mea->mea_fids[i].generation, rc);
1318 ptlrpc_req_finished(*req);
1327 int lmv_delete_object(struct obd_export *exp, struct ll_fid *fid)
1331 if (!lmv_delete_obj(exp, fid)) {
1332 CDEBUG(D_OTHER, "Object %lu/%lu/%lu is not found.\n",
1333 (unsigned long)fid->mds, (unsigned long)fid->id,
1334 (unsigned long)fid->generation);
1340 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1341 struct ptlrpc_request **request)
1343 struct obd_device *obd = exp->exp_obd;
1344 struct lmv_obd *lmv = &obd->u.lmv;
1348 rc = lmv_check_connect(obd);
1352 if (data->namelen == 0 && data->mea1 != NULL) {
1353 /* mds asks to remove slave objects */
1354 rc = lmv_unlink_slaves(exp, data, request);
1356 } else if (data->namelen != 0) {
1357 struct lmv_obj *obj;
1359 obj = lmv_grab_obj(obd, &data->fid1);
1361 i = raw_name2idx(obj->hashtype, obj->objcount, data->name,
1363 data->fid1 = obj->objs[i].fid;
1366 CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n",
1367 data->namelen, data->name,
1368 (unsigned long) data->fid1.mds,
1369 (unsigned long) data->fid1.id,
1370 (unsigned long) data->fid1.generation, i);
1372 CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n",
1373 (unsigned long) data->fid1.mds,
1374 (unsigned long) data->fid1.id,
1375 (unsigned long) data->fid1.generation);
1377 rc = md_unlink(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1381 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1382 char *name, int len)
1384 struct obd_device *obd = exp->exp_obd;
1385 struct lmv_obd *lmv = &obd->u.lmv;
1389 rc = lmv_check_connect(obd);
1391 RETURN(ERR_PTR(rc));
1392 #warning "we need well-desgined readdir() implementation to remove this mess"
1393 obd = lmv->tgts[0].ltd_exp->exp_obd;
1398 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1400 struct obd_device *obd = exp->exp_obd;
1401 struct lmv_obd *lmv = &obd->u.lmv;
1402 int i, rc = 0, change = 0;
1405 if (lmv->max_easize < easize) {
1406 lmv->max_easize = easize;
1409 if (lmv->max_cookiesize < cookiesize) {
1410 lmv->max_cookiesize = cookiesize;
1416 if (lmv->connected == 0)
1419 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1420 if (lmv->tgts[i].ltd_exp == NULL) {
1421 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1425 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1427 CERROR("obd_init_ea_size() failed on MDT target %d, "
1428 "error %d.\n", i, rc);
1435 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1436 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1438 struct obd_device *obd = exp->exp_obd;
1439 struct lmv_obd *lmv = &obd->u.lmv;
1440 struct lov_stripe_md obj_md;
1441 struct lov_stripe_md *obj_mdp = &obj_md;
1445 rc = lmv_check_connect(obd);
1449 LASSERT(ea == NULL);
1450 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1452 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1458 * to be called from MDS only
1460 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1461 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1463 struct obd_device *obd = exp->exp_obd;
1464 struct lmv_obd *lmv = &obd->u.lmv;
1471 rc = lmv_check_connect(obd);
1475 LASSERT(oa != NULL);
1478 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1483 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1485 CERROR("obd_alloc_diskmd() failed, error %d\n",
1496 mfid.generation = oa->o_generation;
1498 mea = (struct mea *)*ea;
1499 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1500 mea->mea_count = lmv->desc.ld_tgt_count;
1501 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1503 mea->mea_master = -1;
1504 lcount = lmv->desc.ld_tgt_count;
1505 for (i = 0, c = 0; c < mea->mea_count && i < lcount; i++) {
1506 struct lov_stripe_md obj_md;
1507 struct lov_stripe_md *obj_mdp = &obj_md;
1509 if (lmv->tgts[i].ltd_exp == NULL) {
1510 /* this is master MDS */
1511 mea->mea_fids[c].id = mfid.id;
1512 mea->mea_fids[c].generation = mfid.generation;
1513 mea->mea_fids[c].mds = i;
1514 mea->mea_master = i;
1519 /* "master" MDS should always be part of stripped dir, so scan
1521 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1524 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE
1525 | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1527 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1529 CERROR("obd_create() failed on MDT target %d, "
1530 "error %d\n", c, rc);
1534 mea->mea_fids[c].id = oa->o_id;
1535 mea->mea_fids[c].generation = oa->o_generation;
1536 mea->mea_fids[c].mds = i;
1538 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1539 i, oa->o_id, oa->o_generation);
1541 LASSERT(c == mea->mea_count);
1542 CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1547 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1548 void *key, __u32 *vallen, void *val)
1550 struct obd_device *obd;
1551 struct lmv_obd *lmv;
1554 obd = class_exp2obd(exp);
1556 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1557 exp->exp_handle.h_cookie);
1562 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1563 __u32 *mdsize = val;
1564 *vallen = sizeof(__u32);
1565 *mdsize = sizeof(struct ll_fid) * lmv->desc.ld_tgt_count
1566 + sizeof(struct mea);
1568 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1569 struct obd_uuid *cluuid = &lmv->cluuid;
1570 struct lmv_tgt_desc *tgts;
1571 __u32 *mdsnum = val;
1574 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1575 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1576 *vallen = sizeof(__u32);
1584 CDEBUG(D_IOCTL, "invalid key\n");
1588 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1589 void *key, obd_count vallen, void *val)
1591 struct obd_device *obd;
1592 struct lmv_obd *lmv;
1595 obd = class_exp2obd(exp);
1597 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1598 exp->exp_handle.h_cookie);
1603 if (keylen >= strlen("client") && strcmp(key, "client") == 0) {
1604 struct lmv_tgt_desc *tgts;
1607 rc = lmv_check_connect(obd);
1611 for (i = 0, tgts = lmv->tgts;
1612 i < lmv->desc.ld_tgt_count; i++, tgts++) {
1613 rc = obd_set_info(tgts->ltd_exp, keylen, key, vallen, val);
1618 } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1619 lmv->server_timeout = 1;
1620 lmv_set_timeouts(obd);
1627 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1628 struct lov_stripe_md *lsm)
1630 struct obd_device *obd = class_exp2obd(exp);
1631 struct lmv_obd *lmv = &obd->u.lmv;
1635 mea_size = sizeof(struct ll_fid) *
1636 lmv->desc.ld_tgt_count + sizeof(struct mea);
1640 if (*lmmp && !lsm) {
1641 OBD_FREE(*lmmp, mea_size);
1646 if (*lmmp == NULL) {
1647 OBD_ALLOC(*lmmp, mea_size);
1655 #warning "MEA packing/convertation must be here! -bzzz"
1656 memcpy(*lmmp, lsm, mea_size);
1660 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1661 struct lov_mds_md *disk_src, int mdsize)
1663 struct obd_device *obd = class_exp2obd(exp);
1664 struct lmv_obd *lmv = &obd->u.lmv;
1665 struct mea **tmea = (struct mea **) mem_tgt;
1666 struct mea *mea = (void *) disk_src;
1670 mea_size = sizeof(struct ll_fid) *
1671 lmv->desc.ld_tgt_count + sizeof(struct mea);
1672 if (mem_tgt == NULL)
1675 if (*mem_tgt != NULL && disk_src == NULL) {
1676 OBD_FREE(*tmea, mea_size);
1680 LASSERT(mea_size == mdsize);
1682 OBD_ALLOC(*tmea, mea_size);
1689 #warning "MEA unpacking/convertation must be here! -bzzz"
1690 memcpy(*tmea, mea, mdsize);
1694 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1695 struct lov_stripe_md *ea, obd_count oa_bufs,
1696 struct brw_page *pgarr, struct obd_trans_info *oti)
1698 struct obd_device *obd = exp->exp_obd;
1699 struct lmv_obd *lmv = &obd->u.lmv;
1700 struct mea *mea = (struct mea *) ea;
1703 LASSERT(oa != NULL);
1704 LASSERT(ea != NULL);
1705 LASSERT(pgarr != NULL);
1706 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1708 oa->o_gr = mea->mea_fids[oa->o_mds].generation;
1709 oa->o_id = mea->mea_fids[oa->o_mds].id;
1710 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1711 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, oa,
1712 NULL, oa_bufs, pgarr, oti);
1716 struct obd_ops lmv_obd_ops = {
1717 .o_owner = THIS_MODULE,
1718 .o_attach = lmv_attach,
1719 .o_detach = lmv_detach,
1720 .o_setup = lmv_setup,
1721 .o_cleanup = lmv_cleanup,
1722 .o_connect = lmv_connect,
1723 .o_disconnect = lmv_disconnect,
1724 .o_statfs = lmv_statfs,
1725 .o_get_info = lmv_get_info,
1726 .o_set_info = lmv_set_info,
1727 .o_create = lmv_obd_create,
1728 .o_packmd = lmv_packmd,
1729 .o_unpackmd = lmv_unpackmd,
1731 .o_init_ea_size = lmv_init_ea_size,
1732 .o_notify = lmv_notify,
1733 .o_iocontrol = lmv_iocontrol,
1736 struct md_ops lmv_md_ops = {
1737 .m_getstatus = lmv_getstatus,
1738 .m_getattr = lmv_getattr,
1739 .m_change_cbdata = lmv_change_cbdata,
1740 .m_change_cbdata_name = lmv_change_cbdata_name,
1741 .m_close = lmv_close,
1742 .m_create = lmv_create,
1743 .m_done_writing = lmv_done_writing,
1744 .m_enqueue = lmv_enqueue,
1745 .m_getattr_name = lmv_getattr_name,
1746 .m_intent_lock = lmv_intent_lock,
1748 .m_rename = lmv_rename,
1749 .m_setattr = lmv_setattr,
1751 .m_readpage = lmv_readpage,
1752 .m_unlink = lmv_unlink,
1753 .m_get_real_obd = lmv_get_real_obd,
1754 .m_valid_attrs = lmv_valid_attrs,
1755 .m_delete_object = lmv_delete_object,
1758 int __init lmv_init(void)
1760 struct lprocfs_static_vars lvars;
1763 lprocfs_init_vars(lmv, &lvars);
1764 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1765 lvars.module_vars, OBD_LMV_DEVICENAME);
1770 static void lmv_exit(void)
1772 class_unregister_type(OBD_LMV_DEVICENAME);
1775 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1776 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1777 MODULE_LICENSE("GPL");
1779 module_init(lmv_init);
1780 module_exit(lmv_exit);