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);
303 class_export_put(exp);
308 struct obd_uuid uuid;
310 --lmv->desc.ld_active_tgt_count;
312 /* save for CERROR below; (we know it's terminated) */
314 rc2 = obd_disconnect(tgts->ltd_exp, 0);
316 CERROR("error: LMV target %s disconnect on MDT idx %d: "
317 "error %d\n", uuid.uuid, i, rc2);
319 class_disconnect(exp, 0);
323 static int lmv_disconnect(struct obd_export *exp, int flags)
325 struct obd_device *obd = class_exp2obd(exp);
326 struct lmv_obd *lmv = &obd->u.lmv;
333 /* Only disconnect the underlying layers on the final disconnect. */
335 if (lmv->refcount != 0)
338 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
339 if (lmv->tgts[i].ltd_exp == NULL)
342 if (obd->obd_no_recov) {
343 /* Pass it on to our clients.
344 * XXX This should be an argument to disconnect,
345 * XXX not a back-door flag on the OBD. Ah well.
347 struct obd_device *mdc_obd;
348 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
350 mdc_obd->obd_no_recov = 1;
353 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
354 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
355 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
357 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
359 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
361 if (lmv->tgts[i].active) {
362 CERROR("Target %s disconnect error %d\n",
363 lmv->tgts[i].uuid.uuid, rc);
367 if (lmv->tgts[i].active) {
368 lmv->desc.ld_active_tgt_count--;
369 lmv->tgts[i].active = 0;
371 lmv->tgts[i].ltd_exp = NULL;
375 /* this is the case when no real connection is established by
376 * lmv_check_connect(). */
378 class_export_put(exp);
379 rc = class_disconnect(exp, 0);
383 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
384 int len, void *karg, void *uarg)
386 struct obd_device *obddev = class_exp2obd(exp);
387 struct lmv_obd *lmv = &obddev->u.lmv;
388 int i, rc = 0, set = 0;
392 if (lmv->desc.ld_tgt_count == 0)
395 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
398 if (lmv->tgts[i].ltd_exp == NULL) {
399 CWARN("%s: NULL export for %d\n", obddev->obd_name, i);
403 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
405 if (lmv->tgts[i].active) {
406 CERROR("error: iocontrol MDC %s on MDT"
407 "idx %d: err = %d\n",
408 lmv->tgts[i].uuid.uuid, i, err);
421 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
424 struct lmv_desc *desc;
425 struct obd_uuid *uuids;
426 struct lmv_tgt_desc *tgts;
427 struct lustre_cfg *lcfg = buf;
428 struct lmv_obd *lmv = &obd->u.lmv;
431 if (lcfg->lcfg_inllen1 < 1) {
432 CERROR("LMV setup requires a descriptor\n");
436 if (lcfg->lcfg_inllen2 < 1) {
437 CERROR("LMV setup requires an OST UUID list\n");
441 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
442 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
443 CERROR("descriptor size wrong: %d > %d\n",
444 (int)sizeof(*desc), lcfg->lcfg_inllen1);
448 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
449 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
450 CERROR("UUID array size wrong: %u * %u != %u\n",
451 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
455 lmv->bufsize = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
456 OBD_ALLOC(lmv->tgts, lmv->bufsize);
457 if (lmv->tgts == NULL) {
458 CERROR("Out of memory\n");
463 spin_lock_init(&lmv->lmv_lock);
465 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
466 tgts->uuid = uuids[i];
468 lmv->max_cookiesize = 0;
470 lmv->max_easize = sizeof(struct ll_fid) *
471 desc->ld_tgt_count + sizeof(struct mea);
473 rc = lmv_setup_mgr(obd);
475 CERROR("Can't setup LMV object manager, "
477 OBD_FREE(lmv->tgts, lmv->bufsize);
483 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
484 unsigned long max_age)
486 struct lmv_obd *lmv = &obd->u.lmv;
487 struct obd_statfs temp;
491 rc = lmv_check_connect(obd);
495 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
496 if (lmv->tgts[i].ltd_exp == NULL) {
497 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
501 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
503 CERROR("can't stat MDS #%d (%s)\n", i,
504 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
508 memcpy(osfs, &temp, sizeof(temp));
510 osfs->os_bavail += temp.os_bavail;
511 osfs->os_blocks += temp.os_blocks;
512 osfs->os_ffree += temp.os_ffree;
513 osfs->os_files += temp.os_files;
519 static int lmv_cleanup(struct obd_device *obd, int flags)
521 struct lmv_obd *lmv = &obd->u.lmv;
523 lmv_cleanup_mgr(obd);
524 OBD_FREE(lmv->tgts, lmv->bufsize);
528 static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid)
530 struct obd_device *obd = exp->exp_obd;
531 struct lmv_obd *lmv = &obd->u.lmv;
534 rc = lmv_check_connect(obd);
537 rc = md_getstatus(lmv->tgts[0].ltd_exp, fid);
542 static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid,
543 unsigned long valid, unsigned int ea_size,
544 struct ptlrpc_request **request)
546 struct obd_device *obd = exp->exp_obd;
547 struct lmv_obd *lmv = &obd->u.lmv;
548 int rc, i = fid->mds;
552 rc = lmv_check_connect(obd);
556 LASSERT(i < lmv->desc.ld_tgt_count);
558 rc = md_getattr(lmv->tgts[i].ltd_exp, fid, valid,
563 obj = lmv_grab_obj(obd, fid);
565 CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n",
566 (unsigned long)fid->mds, (unsigned long)fid->id,
567 (unsigned long)fid->generation, obj ? "(splitted)" : "");
569 /* if object is splitted, then we loop over all the slaves and gather
570 * size attribute. In ideal world we would have to gather also mds field
571 * from all slaves, as object is spread over the cluster and this is
572 * definitely interesting information and it is not good to loss it,
575 struct mds_body *body;
577 if (*request == NULL) {
582 body = lustre_msg_buf((*request)->rq_repmsg, 0,
584 LASSERT(body != NULL);
588 for (i = 0; i < obj->objcount; i++) {
590 if (lmv->tgts[i].ltd_exp == NULL) {
591 CWARN("%s: NULL export for %d\n",
596 /* skip master obj. */
597 if (fid_equal(&obj->fid, &obj->objs[i].fid))
600 body->size += obj->objs[i].size;
610 static int lmv_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
611 ldlm_iterator_t it, void *data)
613 struct obd_device *obd = exp->exp_obd;
614 struct lmv_obd *lmv = &obd->u.lmv;
618 rc = lmv_check_connect(obd);
622 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n", (unsigned long)fid->mds,
623 (unsigned long)fid->id, (unsigned long)fid->generation);
625 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
627 rc = md_change_cbdata(lmv->tgts[fid->mds].ltd_exp,
633 static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
634 char *name, int len, struct ll_fid *cfid,
635 ldlm_iterator_t it, void *data)
637 struct obd_device *obd = exp->exp_obd;
638 struct lmv_obd *lmv = &obd->u.lmv;
643 rc = lmv_check_connect(obd);
647 LASSERT(pfid->mds < lmv->desc.ld_tgt_count);
648 LASSERT(cfid->mds < lmv->desc.ld_tgt_count);
650 CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n",
651 (unsigned long)pfid->mds, (unsigned long)pfid->id,
652 (unsigned long)pfid->generation, len, name,
653 (unsigned long)cfid->mds, (unsigned long)cfid->id,
654 (unsigned long)cfid->generation);
656 /* this is default mds for directory name belongs to. */
658 obj = lmv_grab_obj(obd, pfid);
660 /* directory is splitted. look for right mds for this name. */
661 mds = raw_name2idx(obj->objcount, name, len);
662 mds = obj->objs[mds].fid.mds;
665 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cfid, it, data);
669 static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
671 struct obd_device *obd = exp->exp_obd;
672 struct lmv_obd *lmv = &obd->u.lmv;
675 rc = lmv_check_connect(obd);
678 CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n", (unsigned long) fid->mds,
679 (unsigned long) fid->id, (unsigned long) fid->generation);
680 LASSERT(fid->mds < lmv->desc.ld_tgt_count);
681 rc = md_valid_attrs(lmv->tgts[fid->mds].ltd_exp, fid);
685 int lmv_close(struct obd_export *exp, struct obdo *obdo,
686 struct obd_client_handle *och,
687 struct ptlrpc_request **request)
689 struct obd_device *obd = exp->exp_obd;
690 struct lmv_obd *lmv = &obd->u.lmv;
691 int rc, i = obdo->o_mds;
693 rc = lmv_check_connect(obd);
696 LASSERT(i < lmv->desc.ld_tgt_count);
697 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds,
698 (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation);
699 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
703 int lmv_get_mea_and_update_object(struct obd_export *exp, struct ll_fid *fid)
705 struct obd_device *obd = exp->exp_obd;
706 struct lmv_obd *lmv = &obd->u.lmv;
707 struct ptlrpc_request *req = NULL;
714 mealen = MEA_SIZE_LMV(lmv);
716 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
718 /* time to update mea of parent fid */
719 rc = md_getattr(lmv->tgts[fid->mds].ltd_exp, fid,
720 valid, mealen, &req);
722 CERROR("md_getattr() failed, error %d\n", rc);
726 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
728 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
733 GOTO(cleanup, rc = -ENODATA);
735 obj = lmv_create_obj(exp, fid, md.mea);
740 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
744 ptlrpc_req_finished(req);
748 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
749 const void *data, int datalen, int mode, __u32 uid,
750 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
752 struct obd_device *obd = exp->exp_obd;
753 struct lmv_obd *lmv = &obd->u.lmv;
754 struct mds_body *body;
756 int rc, mds, loop = 0;
759 rc = lmv_check_connect(obd);
763 if (!lmv->desc.ld_active_tgt_count)
766 LASSERT(++loop <= 2);
767 obj = lmv_grab_obj(obd, &op_data->fid1);
769 mds = raw_name2idx(obj->objcount, op_data->name,
771 op_data->fid1 = obj->objs[mds].fid;
775 CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu\n", op_data->namelen,
776 op_data->name, (unsigned long)op_data->fid1.mds,
777 (unsigned long)op_data->fid1.id,
778 (unsigned long)op_data->fid1.generation);
780 rc = md_create(lmv->tgts[op_data->fid1.mds].ltd_exp, op_data, data,
781 datalen, mode, uid, gid, rdev, request);
783 if (*request == NULL)
786 body = lustre_msg_buf((*request)->rq_repmsg, 0,
788 LASSERT(body != NULL);
790 CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, "
791 "mds = %d\n", (unsigned long)body->fid1.id,
792 (unsigned long)body->fid1.generation, op_data->fid1.mds);
794 LASSERT(body->valid & OBD_MD_MDS ||
795 body->mds == op_data->fid1.mds);
796 } else if (rc == -ERESTART) {
797 /* directory got splitted. time to update local object and
798 * repeat the request with proper MDS */
799 rc = lmv_get_mea_and_update_object(exp, &op_data->fid1);
801 ptlrpc_req_finished(*request);
808 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
810 struct obd_device *obd = exp->exp_obd;
811 struct lmv_obd *lmv = &obd->u.lmv;
814 rc = lmv_check_connect(obd);
818 /* FIXME: choose right MDC here */
819 CWARN("this method isn't implemented yet\n");
820 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
824 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
825 struct lookup_intent *it, int lockmode,
826 struct mdc_op_data *data, struct lustre_handle *lockh,
827 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
828 ldlm_blocking_callback cb_blocking, void *cb_data)
830 struct obd_device *obd = exp->exp_obd;
831 struct lmv_obd *lmv = &obd->u.lmv;
832 struct mea *mea = data->mea1;
833 struct mdc_op_data data2;
837 LASSERT(mea != NULL);
838 for (i = 0; i < mea->mea_count; i++) {
839 memset(&data2, 0, sizeof(data2));
840 data2.fid1 = mea->mea_fids[i];
841 mds = data2.fid1.mds;
843 if (lmv->tgts[mds].ltd_exp == NULL)
846 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, lockmode,
847 &data2, lockh + i, lmm, lmmsize, cb_completion,
848 cb_blocking, cb_data);
850 CDEBUG(D_OTHER, "take lock on slave %lu/%lu/%lu -> %d/%d\n",
851 (unsigned long)mea->mea_fids[i].mds,
852 (unsigned long)mea->mea_fids[i].id,
853 (unsigned long)mea->mea_fids[i].generation,
854 rc, it->d.lustre.it_status);
857 if (it->d.lustre.it_data) {
858 struct ptlrpc_request *req;
859 req = (struct ptlrpc_request *) it->d.lustre.it_data;
860 ptlrpc_req_finished(req);
863 if (it->d.lustre.it_status)
864 GOTO(cleanup, rc = it->d.lustre.it_status);
869 /* drop all taken locks */
872 ldlm_lock_decref(lockh + i, lockmode);
878 int lmv_enqueue(struct obd_export *exp, int lock_type,
879 struct lookup_intent *it, int lock_mode,
880 struct mdc_op_data *data, struct lustre_handle *lockh,
881 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
882 ldlm_blocking_callback cb_blocking, void *cb_data)
884 struct obd_device *obd = exp->exp_obd;
885 struct lmv_obd *lmv = &obd->u.lmv;
890 rc = lmv_check_connect(obd);
894 if (it->it_op == IT_UNLINK) {
895 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
896 data, lockh, lmm, lmmsize,
897 cb_completion, cb_blocking, cb_data);
902 obj = lmv_grab_obj(obd, &data->fid1);
904 /* directory is splitted. look for right mds for this
906 mds = raw_name2idx(obj->objcount, (char *)data->name,
908 data->fid1 = obj->objs[mds].fid;
912 CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n", LL_IT2STR(it),
913 (unsigned long)data->fid1.id, (unsigned long)data->fid1.generation);
915 rc = md_enqueue(lmv->tgts[data->fid1.mds].ltd_exp, lock_type, it,
916 lock_mode, data, lockh, lmm, lmmsize, cb_completion,
917 cb_blocking, cb_data);
922 int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid,
923 char *filename, int namelen, unsigned long valid,
924 unsigned int ea_size, struct ptlrpc_request **request)
926 struct obd_device *obd = exp->exp_obd;
927 struct lmv_obd *lmv = &obd->u.lmv;
928 struct ll_fid rfid = *fid;
929 int rc, mds = fid->mds, loop = 0;
930 struct mds_body *body;
933 rc = lmv_check_connect(obd);
937 LASSERT(++loop <= 2);
938 obj = lmv_grab_obj(obd, fid);
940 /* directory is splitted. look for right mds for this name */
941 mds = raw_name2idx(obj->objcount, filename, namelen - 1);
942 rfid = obj->objs[mds].fid;
946 CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu -> %lu/%lu/%lu\n",
947 namelen, filename, (unsigned long)fid->mds,
948 (unsigned long)fid->id, (unsigned long)fid->generation,
949 (unsigned long)rfid.mds, (unsigned long)rfid.id,
950 (unsigned long)rfid.generation);
952 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid, filename,
953 namelen, valid, ea_size, request);
955 /* this could be cross-node reference. in this case all we have
956 * right now is mds/ino/generation triple. we'd like to find
957 * other attributes */
958 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
959 LASSERT(body != NULL);
960 if (body->valid & OBD_MD_MDS) {
961 struct ptlrpc_request *req = NULL;
963 CDEBUG(D_OTHER, "request attrs for %lu/%lu/%lu\n",
964 (unsigned long) rfid.mds,
965 (unsigned long) rfid.id,
966 (unsigned long) rfid.generation);
967 rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid,
968 NULL, 1, valid, ea_size, &req);
969 ptlrpc_req_finished(*request);
972 } else if (rc == -ERESTART) {
973 /* directory got splitted. time to update local object and
974 * repeat the request with proper MDS */
975 rc = lmv_get_mea_and_update_object(exp, &rfid);
977 ptlrpc_req_finished(*request);
986 * llite passes fid of an target inode in data->fid1 and fid of directory in
989 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
990 struct ptlrpc_request **request)
992 struct obd_device *obd = exp->exp_obd;
993 struct lmv_obd *lmv = &obd->u.lmv;
998 rc = lmv_check_connect(obd);
1002 if (data->namelen != 0) {
1003 /* usual link request */
1004 obj = lmv_grab_obj(obd, &data->fid1);
1006 rc = raw_name2idx(obj->objcount, data->name,
1008 data->fid1 = obj->objs[rc].fid;
1012 CDEBUG(D_OTHER,"link %lu/%lu/%lu:%*s to %lu/%lu/%lu mds %lu\n",
1013 (unsigned long)data->fid2.mds,
1014 (unsigned long)data->fid2.id,
1015 (unsigned long)data->fid2.generation,
1016 data->namelen, data->name,
1017 (unsigned long)data->fid1.mds,
1018 (unsigned long)data->fid1.id,
1019 (unsigned long)data->fid1.generation,
1020 (unsigned long)data->fid1.mds);
1022 /* request from MDS to acquire i_links for inode by fid1 */
1023 CDEBUG(D_OTHER, "inc i_nlinks for %lu/%lu/%lu\n",
1024 (unsigned long)data->fid1.mds,
1025 (unsigned long)data->fid1.id,
1026 (unsigned long)data->fid1.generation);
1029 rc = md_link(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1033 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1034 const char *old, int oldlen, const char *new, int newlen,
1035 struct ptlrpc_request **request)
1037 struct obd_device *obd = exp->exp_obd;
1038 struct lmv_obd *lmv = &obd->u.lmv;
1039 struct lmv_obj *obj;
1043 CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n",
1044 oldlen, old, (unsigned long)data->fid1.mds,
1045 (unsigned long)data->fid1.id,
1046 (unsigned long)data->fid1.generation,
1047 newlen, new, (unsigned long) data->fid2.mds,
1048 (unsigned long) data->fid2.id,
1049 (unsigned long) data->fid2.generation);
1051 if (!fid_equal(&data->fid1, &data->fid2))
1052 CWARN("cross-node rename %lu/%lu/%lu:%*s to %lu/%lu/%lu:%*s\n",
1053 (unsigned long)data->fid1.mds,
1054 (unsigned long)data->fid1.id,
1055 (unsigned long)data->fid1.generation, oldlen, old,
1056 (unsigned long)data->fid2.mds,
1057 (unsigned long)data->fid2.id,
1058 (unsigned long)data->fid2.generation, newlen, new);
1060 rc = lmv_check_connect(obd);
1065 /* MDS with old dir entry is asking another MDS to create name
1068 "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n",
1069 newlen, new, oldlen, newlen,
1070 (unsigned long)data->fid2.mds,
1071 (unsigned long)data->fid2.id,
1072 (unsigned long)data->fid2.generation,
1073 (unsigned long)data->fid1.mds,
1074 (unsigned long)data->fid1.id,
1075 (unsigned long)data->fid1.generation);
1076 mds = data->fid2.mds;
1080 obj = lmv_grab_obj(obd, &data->fid1);
1082 /* directory is already splitted, so we have to forward request
1083 * to the right MDS */
1084 mds = raw_name2idx(obj->objcount, (char *)old, oldlen);
1085 data->fid1 = obj->objs[mds].fid;
1086 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1087 (unsigned long)obj->objs[mds].fid.mds,
1088 (unsigned long)obj->objs[mds].fid.id,
1089 (unsigned long)obj->objs[mds].fid.generation);
1093 obj = lmv_grab_obj(obd, &data->fid2);
1095 /* directory is already splitted, so we have to forward request
1096 * to the right MDS */
1097 mds = raw_name2idx(obj->objcount, (char *)new, newlen);
1098 data->fid2 = obj->objs[mds].fid;
1099 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1100 (unsigned long)obj->objs[mds].fid.mds,
1101 (unsigned long)obj->objs[mds].fid.id,
1102 (unsigned long)obj->objs[mds].fid.generation);
1106 mds = data->fid1.mds;
1109 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1110 new, newlen, request);
1114 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1115 struct iattr *iattr, void *ea, int ealen, void *ea2,
1116 int ea2len, struct ptlrpc_request **request)
1118 struct obd_device *obd = exp->exp_obd;
1119 struct lmv_obd *lmv = &obd->u.lmv;
1120 struct ptlrpc_request *req;
1121 struct mds_body *body;
1122 struct lmv_obj *obj;
1126 rc = lmv_check_connect(obd);
1130 obj = lmv_grab_obj(obd, &data->fid1);
1132 CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n",
1133 (unsigned long)data->fid1.mds, (unsigned long)data->fid1.id,
1134 (unsigned long)data->fid1.generation, iattr->ia_valid,
1135 obj ? ", splitted" : "");
1138 for (i = 0; i < obj->objcount; i++) {
1139 data->fid1 = obj->objs[i].fid;
1141 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1142 iattr, ea, ealen, ea2, ea2len, &req);
1144 if (fid_equal(&obj->fid, &obj->objs[i].fid)) {
1145 /* this is master object and this request should
1146 * be returned back to llite */
1149 ptlrpc_req_finished(req);
1157 LASSERT(data->fid1.mds < lmv->desc.ld_tgt_count);
1158 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1159 iattr, ea, ealen, ea2, ea2len, request);
1161 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1163 LASSERT(body != NULL);
1164 LASSERT(body->mds == data->fid1.mds);
1170 int lmv_sync(struct obd_export *exp, struct ll_fid *fid,
1171 struct ptlrpc_request **request)
1173 struct obd_device *obd = exp->exp_obd;
1174 struct lmv_obd *lmv = &obd->u.lmv;
1178 rc = lmv_check_connect(obd);
1182 rc = md_sync(lmv->tgts[fid->mds].ltd_exp, fid, request);
1186 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1187 void *data, int flag)
1189 struct lustre_handle lockh;
1190 struct lmv_obj *obj;
1195 case LDLM_CB_BLOCKING:
1196 ldlm_lock2handle(lock, &lockh);
1197 rc = ldlm_cli_cancel(&lockh);
1199 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1203 case LDLM_CB_CANCELING:
1204 /* time to drop cached attrs for dirobj */
1205 obj = lock->l_ast_data;
1207 CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n",
1208 lock->l_resource->lr_name.name[3] == 1 ? "LOOKUP" : "UPDATE",
1209 (unsigned long)lock->l_resource->lr_name.name[0],
1210 (unsigned long)lock->l_resource->lr_name.name[1],
1211 (unsigned long)obj->fid.mds, (unsigned long)obj->fid.id,
1212 (unsigned long)obj->fid.generation);
1222 void lmv_remove_dots(struct page *page)
1224 char *kaddr = page_address(page);
1225 unsigned limit = PAGE_CACHE_SIZE;
1226 unsigned offs, rec_len;
1227 struct ext2_dir_entry_2 *p;
1229 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1230 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1231 rec_len = le16_to_cpu(p->rec_len);
1233 if ((p->name_len == 1 && p->name[0] == '.') ||
1234 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1239 int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
1240 __u64 offset, struct page *page,
1241 struct ptlrpc_request **request)
1243 struct obd_device *obd = exp->exp_obd;
1244 struct lmv_obd *lmv = &obd->u.lmv;
1245 struct ll_fid rfid = *mdc_fid;
1246 struct lmv_obj *obj;
1250 rc = lmv_check_connect(obd);
1254 LASSERT(mdc_fid->mds < lmv->desc.ld_tgt_count);
1255 CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n",
1256 offset, (unsigned long) rfid.mds,
1257 (unsigned long) rfid.id,
1258 (unsigned long) rfid.generation);
1260 obj = lmv_grab_obj(obd, mdc_fid);
1264 /* find dirobj containing page with requested offset. */
1265 for (i = 0; i < obj->objcount; i++) {
1266 if (offset < obj->objs[i].size)
1268 offset -= obj->objs[i].size;
1270 rfid = obj->objs[i].fid;
1272 lmv_unlock_obj(obj);
1275 CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n",
1276 (unsigned long)rfid.mds, (unsigned long)rfid.id,
1277 (unsigned long)rfid.generation, (unsigned long)offset);
1279 rc = md_readpage(lmv->tgts[rfid.mds].ltd_exp, &rfid, offset,
1282 if (rc == 0 && !fid_equal(&rfid, mdc_fid))
1283 /* this page isn't from master object. To avoid "." and ".."
1284 * duplication in directory, we have to remove them from all
1286 lmv_remove_dots(page);
1291 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1292 struct ptlrpc_request **req)
1294 struct obd_device *obd = exp->exp_obd;
1295 struct lmv_obd *lmv = &obd->u.lmv;
1296 struct mea *mea = data->mea1;
1297 struct mdc_op_data data2;
1301 LASSERT(mea != NULL);
1302 for (i = 0; i < mea->mea_count; i++) {
1303 memset(&data2, 0, sizeof(data2));
1304 data2.fid1 = mea->mea_fids[i];
1305 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1306 mds = data2.fid1.mds;
1308 if (lmv->tgts[mds].ltd_exp == NULL)
1311 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1312 CDEBUG(D_OTHER, "unlink slave %lu/%lu/%lu -> %d\n",
1313 (unsigned long) mea->mea_fids[i].mds,
1314 (unsigned long) mea->mea_fids[i].id,
1315 (unsigned long) mea->mea_fids[i].generation, rc);
1317 ptlrpc_req_finished(*req);
1326 int lmv_delete_object(struct obd_export *exp, struct ll_fid *fid)
1330 if (!lmv_delete_obj(exp, fid)) {
1331 CDEBUG(D_OTHER, "Object %lu/%lu/%lu is not found.\n",
1332 (unsigned long)fid->mds, (unsigned long)fid->id,
1333 (unsigned long)fid->generation);
1339 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1340 struct ptlrpc_request **request)
1342 struct obd_device *obd = exp->exp_obd;
1343 struct lmv_obd *lmv = &obd->u.lmv;
1347 rc = lmv_check_connect(obd);
1351 if (data->namelen == 0 && data->mea1 != NULL) {
1352 /* mds asks to remove slave objects */
1353 rc = lmv_unlink_slaves(exp, data, request);
1355 } else if (data->namelen != 0) {
1356 struct lmv_obj *obj;
1358 obj = lmv_grab_obj(obd, &data->fid1);
1360 i = raw_name2idx(obj->objcount, data->name,
1362 data->fid1 = obj->objs[i].fid;
1365 CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n",
1366 data->namelen, data->name,
1367 (unsigned long) data->fid1.mds,
1368 (unsigned long) data->fid1.id,
1369 (unsigned long) data->fid1.generation, i);
1371 CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n",
1372 (unsigned long) data->fid1.mds,
1373 (unsigned long) data->fid1.id,
1374 (unsigned long) data->fid1.generation);
1376 rc = md_unlink(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1380 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1381 char *name, int len)
1383 struct obd_device *obd = exp->exp_obd;
1384 struct lmv_obd *lmv = &obd->u.lmv;
1388 rc = lmv_check_connect(obd);
1390 RETURN(ERR_PTR(rc));
1391 #warning "we need well-desgined readdir() implementation to remove this mess"
1392 obd = lmv->tgts[0].ltd_exp->exp_obd;
1397 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1399 struct obd_device *obd = exp->exp_obd;
1400 struct lmv_obd *lmv = &obd->u.lmv;
1401 int i, rc = 0, change = 0;
1404 if (lmv->max_easize < easize) {
1405 lmv->max_easize = easize;
1408 if (lmv->max_cookiesize < cookiesize) {
1409 lmv->max_cookiesize = cookiesize;
1415 if (lmv->connected == 0)
1418 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1419 if (lmv->tgts[i].ltd_exp == NULL) {
1420 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1424 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1426 CERROR("obd_init_ea_size() failed on MDT target %d, "
1427 "error %d.\n", i, rc);
1434 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1435 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1437 struct obd_device *obd = exp->exp_obd;
1438 struct lmv_obd *lmv = &obd->u.lmv;
1439 struct lov_stripe_md obj_md;
1440 struct lov_stripe_md *obj_mdp = &obj_md;
1444 rc = lmv_check_connect(obd);
1448 LASSERT(ea == NULL);
1449 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1451 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1457 * to be called from MDS only
1459 int lmv_obd_create(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;
1470 rc = lmv_check_connect(obd);
1474 LASSERT(oa != NULL);
1477 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1482 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1484 CERROR("obd_alloc_diskmd() failed, error %d\n",
1495 mfid.generation = oa->o_generation;
1497 mea = (struct mea *)*ea;
1498 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1499 mea->mea_count = lmv->desc.ld_tgt_count;
1501 mea->mea_master = -1;
1502 lcount = lmv->desc.ld_tgt_count;
1503 for (i = 0, c = 0; c < mea->mea_count && i < lcount; i++) {
1504 struct lov_stripe_md obj_md;
1505 struct lov_stripe_md *obj_mdp = &obj_md;
1507 if (lmv->tgts[i].ltd_exp == NULL) {
1508 /* this is master MDS */
1509 mea->mea_fids[c].id = mfid.id;
1510 mea->mea_fids[c].generation = mfid.generation;
1511 mea->mea_fids[c].mds = i;
1512 mea->mea_master = i;
1517 /* "master" MDS should always be part of stripped dir, so scan
1519 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1522 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE
1523 | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1525 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1527 CERROR("obd_create() failed on MDT target %d, "
1528 "error %d\n", c, rc);
1532 mea->mea_fids[c].id = oa->o_id;
1533 mea->mea_fids[c].generation = oa->o_generation;
1534 mea->mea_fids[c].mds = i;
1536 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1537 i, oa->o_id, oa->o_generation);
1539 LASSERT(c == mea->mea_count);
1540 CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1545 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1546 void *key, __u32 *vallen, void *val)
1548 struct obd_device *obd;
1549 struct lmv_obd *lmv;
1552 obd = class_exp2obd(exp);
1554 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1555 exp->exp_handle.h_cookie);
1560 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1561 __u32 *mdsize = val;
1562 *vallen = sizeof(__u32);
1563 *mdsize = sizeof(struct ll_fid) * lmv->desc.ld_tgt_count
1564 + sizeof(struct mea);
1566 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1567 struct obd_uuid *cluuid = &lmv->cluuid;
1568 struct lmv_tgt_desc *tgts;
1569 __u32 *mdsnum = val;
1572 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1573 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1574 *vallen = sizeof(__u32);
1582 CDEBUG(D_IOCTL, "invalid key\n");
1586 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1587 void *key, obd_count vallen, void *val)
1589 struct obd_device *obd;
1590 struct lmv_obd *lmv;
1593 obd = class_exp2obd(exp);
1595 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1596 exp->exp_handle.h_cookie);
1601 if (keylen >= strlen("client") && strcmp(key, "client") == 0) {
1602 struct lmv_tgt_desc *tgts;
1605 rc = lmv_check_connect(obd);
1609 for (i = 0, tgts = lmv->tgts;
1610 i < lmv->desc.ld_tgt_count; i++, tgts++) {
1611 rc = obd_set_info(tgts->ltd_exp, keylen, key, vallen, val);
1616 } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1617 lmv->server_timeout = 1;
1618 lmv_set_timeouts(obd);
1625 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1626 struct lov_stripe_md *lsm)
1628 struct obd_device *obd = class_exp2obd(exp);
1629 struct lmv_obd *lmv = &obd->u.lmv;
1633 mea_size = sizeof(struct ll_fid) *
1634 lmv->desc.ld_tgt_count + sizeof(struct mea);
1638 if (*lmmp && !lsm) {
1639 OBD_FREE(*lmmp, mea_size);
1644 if (*lmmp == NULL) {
1645 OBD_ALLOC(*lmmp, mea_size);
1653 #warning "MEA packing/convertation must be here! -bzzz"
1654 memcpy(*lmmp, lsm, mea_size);
1658 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1659 struct lov_mds_md *disk_src, int mdsize)
1661 struct obd_device *obd = class_exp2obd(exp);
1662 struct lmv_obd *lmv = &obd->u.lmv;
1663 struct mea **tmea = (struct mea **) mem_tgt;
1664 struct mea *mea = (void *) disk_src;
1668 mea_size = sizeof(struct ll_fid) *
1669 lmv->desc.ld_tgt_count + sizeof(struct mea);
1670 if (mem_tgt == NULL)
1673 if (*mem_tgt != NULL && disk_src == NULL) {
1674 OBD_FREE(*tmea, mea_size);
1678 LASSERT(mea_size == mdsize);
1680 OBD_ALLOC(*tmea, mea_size);
1687 #warning "MEA unpacking/convertation must be here! -bzzz"
1688 memcpy(*tmea, mea, mdsize);
1692 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1693 struct lov_stripe_md *ea, obd_count oa_bufs,
1694 struct brw_page *pgarr, struct obd_trans_info *oti)
1696 struct obd_device *obd = exp->exp_obd;
1697 struct lmv_obd *lmv = &obd->u.lmv;
1698 struct mea *mea = (struct mea *) ea;
1701 LASSERT(oa != NULL);
1702 LASSERT(ea != NULL);
1703 LASSERT(pgarr != NULL);
1704 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1706 oa->o_gr = mea->mea_fids[oa->o_mds].generation;
1707 oa->o_id = mea->mea_fids[oa->o_mds].id;
1708 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1709 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, oa,
1710 NULL, oa_bufs, pgarr, oti);
1714 struct obd_ops lmv_obd_ops = {
1715 .o_owner = THIS_MODULE,
1716 .o_attach = lmv_attach,
1717 .o_detach = lmv_detach,
1718 .o_setup = lmv_setup,
1719 .o_cleanup = lmv_cleanup,
1720 .o_connect = lmv_connect,
1721 .o_disconnect = lmv_disconnect,
1722 .o_statfs = lmv_statfs,
1723 .o_get_info = lmv_get_info,
1724 .o_set_info = lmv_set_info,
1725 .o_create = lmv_obd_create,
1726 .o_packmd = lmv_packmd,
1727 .o_unpackmd = lmv_unpackmd,
1729 .o_init_ea_size = lmv_init_ea_size,
1730 .o_notify = lmv_notify,
1731 .o_iocontrol = lmv_iocontrol,
1734 struct md_ops lmv_md_ops = {
1735 .m_getstatus = lmv_getstatus,
1736 .m_getattr = lmv_getattr,
1737 .m_change_cbdata = lmv_change_cbdata,
1738 .m_change_cbdata_name = lmv_change_cbdata_name,
1739 .m_close = lmv_close,
1740 .m_create = lmv_create,
1741 .m_done_writing = lmv_done_writing,
1742 .m_enqueue = lmv_enqueue,
1743 .m_getattr_name = lmv_getattr_name,
1744 .m_intent_lock = lmv_intent_lock,
1746 .m_rename = lmv_rename,
1747 .m_setattr = lmv_setattr,
1749 .m_readpage = lmv_readpage,
1750 .m_unlink = lmv_unlink,
1751 .m_get_real_obd = lmv_get_real_obd,
1752 .m_valid_attrs = lmv_valid_attrs,
1753 .m_delete_object = lmv_delete_object,
1756 int __init lmv_init(void)
1758 struct lprocfs_static_vars lvars;
1761 lprocfs_init_vars(lmv, &lvars);
1762 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1763 lvars.module_vars, OBD_LMV_DEVICENAME);
1768 static void lmv_exit(void)
1770 class_unregister_type(OBD_LMV_DEVICENAME);
1773 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1774 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1775 MODULE_LICENSE("GPL");
1777 module_init(lmv_init);
1778 module_exit(lmv_exit);