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"
52 static void lmv_activate_target(struct lmv_obd *lmv,
53 struct lmv_tgt_desc *tgt,
56 if (tgt->active == activate)
59 tgt->active = activate;
60 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
65 * -EINVAL : UUID can't be found in the LMV's target list
66 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
67 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
69 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
72 struct obd_device *obd;
73 struct lmv_tgt_desc *tgt;
77 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
78 lmv, uuid->uuid, activate);
80 spin_lock(&lmv->lmv_lock);
81 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
82 if (tgt->ltd_exp == NULL)
85 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
86 i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
88 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
92 if (i == lmv->desc.ld_tgt_count)
93 GOTO(out_lmv_lock, rc = -EINVAL);
95 obd = class_exp2obd(tgt->ltd_exp);
97 GOTO(out_lmv_lock, rc = -ENOTCONN);
99 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
100 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
101 obd->obd_type->typ_name, i);
102 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
104 if (tgt->active == activate) {
105 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
106 activate ? "" : "in");
107 GOTO(out_lmv_lock, rc);
110 CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
111 obd, activate ? "" : "in");
113 lmv_activate_target(lmv, tgt, activate);
118 spin_unlock(&lmv->lmv_lock);
122 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
123 int active, void *data)
126 struct obd_uuid *uuid;
128 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
129 CERROR("unexpected notification of %s %s!\n",
130 watched->obd_type->typ_name,
134 uuid = &watched->u.cli.cl_import->imp_target_uuid;
136 /* Set MDC as active before notifying the observer, so the observer can
137 * use the MDC normally.
139 rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
141 CERROR("%sactivation of %s failed: %d\n",
142 active ? "" : "de", uuid->uuid, rc);
146 if (obd->obd_observer)
147 /* Pass the notification up the chain. */
148 rc = obd_notify(obd->obd_observer, watched, active, data);
153 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
155 struct lprocfs_static_vars lvars;
159 lprocfs_init_vars(lmv, &lvars);
160 rc = lprocfs_obd_attach(dev, lvars.obd_vars);
163 struct proc_dir_entry *entry;
165 entry = create_proc_entry("target_obd_status", 0444,
166 dev->obd_proc_entry);
169 entry->proc_fops = &lmv_proc_target_fops;
176 int lmv_detach(struct obd_device *dev)
178 return lprocfs_obd_detach(dev);
181 /* this is fake connect function. Its purpose is to initialize lmv and say
182 * caller that everything is okay. Real connection will be performed later. */
183 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
184 struct obd_uuid *cluuid, unsigned long connect_flags)
186 struct lmv_obd *lmv = &obd->u.lmv;
187 struct obd_export *exp;
188 struct proc_dir_entry *lmv_proc_dir;
192 rc = class_connect(conn, obd, cluuid);
194 CERROR("class_connection() returned %d\n", rc);
198 exp = class_conn2export(conn);
200 /* we don't want to actually do the underlying connections more than
201 * once, so keep track. */
203 if (lmv->refcount > 1) {
204 class_export_put(exp);
210 lmv->cluuid = *cluuid;
211 sema_init(&lmv->init_sem, 1);
212 lmv->connect_flags = connect_flags;
214 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
216 if (IS_ERR(lmv_proc_dir)) {
217 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
218 obd->obd_type->typ_name, obd->obd_name);
226 void lmv_set_timeouts(struct obd_device *obd)
228 struct lmv_tgt_desc *tgts;
233 if (lmv->server_timeout == 0)
236 if (lmv->connected == 0)
239 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
240 if (tgts->ltd_exp == NULL)
243 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
244 "inter_mds", 0, NULL);
248 #define MAX_STRING_SIZE 128
250 /* performs a check if passed obd is connected. If no - connect it. */
251 int lmv_check_connect(struct obd_device *obd)
254 struct proc_dir_entry *lmv_proc_dir;
256 struct lmv_obd *lmv = &obd->u.lmv;
257 struct lmv_tgt_desc *tgts;
258 struct obd_uuid *cluuid;
259 struct obd_export *exp;
265 down(&lmv->init_sem);
266 if (lmv->connected) {
271 cluuid = &lmv->cluuid;
274 CDEBUG(D_OTHER, "time to connect %s to %s\n",
275 cluuid->uuid, obd->obd_name);
277 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
278 struct obd_device *tgt_obd;
279 struct lustre_handle conn = {0, };
280 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
282 LASSERT(tgts != NULL);
284 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
287 CERROR("Target %s not attached\n", tgts->uuid.uuid);
288 GOTO(out_disc, rc = -EINVAL);
291 /* for MDS: don't connect to yourself */
292 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
293 CDEBUG(D_OTHER, "don't connect back to %s\n",
295 tgts->ltd_exp = NULL;
299 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
300 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
301 tgts->uuid.uuid, obd->obd_uuid.uuid,
304 if (!tgt_obd->obd_set_up) {
305 CERROR("Target %s not set up\n", tgts->uuid.uuid);
306 GOTO(out_disc, rc = -EINVAL);
309 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
312 CERROR("Target %s connect error %d\n",
313 tgts->uuid.uuid, rc);
316 tgts->ltd_exp = class_conn2export(&conn);
318 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
319 lmv->max_cookiesize);
321 rc = obd_register_observer(tgt_obd, obd);
323 CERROR("Target %s register_observer error %d\n",
324 tgts->uuid.uuid, rc);
325 obd_disconnect(tgts->ltd_exp, 0);
329 lmv->desc.ld_active_tgt_count++;
332 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
333 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
334 atomic_read(&obd->obd_refcount));
337 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
339 struct obd_device *mdc_obd = class_conn2obd(&conn);
340 struct proc_dir_entry *mdc_symlink;
341 char name[MAX_STRING_SIZE + 1];
343 LASSERT(mdc_obd != NULL);
344 LASSERT(mdc_obd->obd_type != NULL);
345 LASSERT(mdc_obd->obd_type->typ_name != NULL);
346 name[MAX_STRING_SIZE] = '\0';
347 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
348 mdc_obd->obd_type->typ_name,
350 mdc_symlink = proc_symlink(mdc_obd->obd_name,
352 if (mdc_symlink == NULL) {
353 CERROR("could not register LMV target "
354 "/proc/fs/lustre/%s/%s/target_obds/%s.",
355 obd->obd_type->typ_name, obd->obd_name,
357 lprocfs_remove(lmv_proc_dir);
364 lmv_set_timeouts(obd);
365 class_export_put(exp);
372 struct obd_uuid uuid;
374 --lmv->desc.ld_active_tgt_count;
376 /* save for CERROR below; (we know it's terminated) */
378 rc2 = obd_disconnect(tgts->ltd_exp, 0);
380 CERROR("error: LMV target %s disconnect on MDC idx %d: "
381 "error %d\n", uuid.uuid, i, rc2);
383 class_disconnect(exp, 0);
388 static int lmv_disconnect(struct obd_export *exp, int flags)
390 struct obd_device *obd = class_exp2obd(exp);
391 struct lmv_obd *lmv = &obd->u.lmv;
394 struct proc_dir_entry *lmv_proc_dir;
402 /* Only disconnect the underlying layers on the final disconnect. */
404 if (lmv->refcount != 0)
408 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
411 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
412 struct obd_device *mdc_obd;
414 if (lmv->tgts[i].ltd_exp == NULL)
417 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
421 struct proc_dir_entry *mdc_symlink;
423 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
425 lprocfs_remove(mdc_symlink);
427 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
428 obd->obd_type->typ_name, obd->obd_name,
433 if (obd->obd_no_recov) {
435 mdc_obd->obd_no_recov = 1;
437 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
438 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
439 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
441 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
442 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
444 if (lmv->tgts[i].active) {
445 CERROR("Target %s disconnect error %d\n",
446 lmv->tgts[i].uuid.uuid, rc);
451 lmv_activate_target(lmv, &lmv->tgts[i], 0);
452 lmv->tgts[i].ltd_exp = NULL;
457 lprocfs_remove(lmv_proc_dir);
459 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
460 obd->obd_type->typ_name, obd->obd_name);
465 /* this is the case when no real connection is established by
466 * lmv_check_connect(). */
468 class_export_put(exp);
469 rc = class_disconnect(exp, 0);
470 if (lmv->refcount == 0)
475 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
476 int len, void *karg, void *uarg)
478 struct obd_device *obddev = class_exp2obd(exp);
479 struct lmv_obd *lmv = &obddev->u.lmv;
480 int i, rc = 0, set = 0;
484 if (lmv->desc.ld_tgt_count == 0)
487 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
490 if (lmv->tgts[i].ltd_exp == NULL)
493 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
495 if (lmv->tgts[i].active) {
496 CERROR("error: iocontrol MDC %s on MDT"
497 "idx %d: err = %d\n",
498 lmv->tgts[i].uuid.uuid, i, err);
511 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
514 struct lmv_desc *desc;
515 struct obd_uuid *uuids;
516 struct lmv_tgt_desc *tgts;
517 struct obd_device *tgt_obd;
518 struct lustre_cfg *lcfg = buf;
519 struct lmv_obd *lmv = &obd->u.lmv;
522 if (lcfg->lcfg_inllen1 < 1) {
523 CERROR("LMV setup requires a descriptor\n");
527 if (lcfg->lcfg_inllen2 < 1) {
528 CERROR("LMV setup requires an OST UUID list\n");
532 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
533 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
534 CERROR("descriptor size wrong: %d > %d\n",
535 (int)sizeof(*desc), lcfg->lcfg_inllen1);
539 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
540 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
541 CERROR("UUID array size wrong: %u * %u != %u\n",
542 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
546 lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
547 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
548 if (lmv->tgts == NULL) {
549 CERROR("Out of memory\n");
554 spin_lock_init(&lmv->lmv_lock);
556 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
557 tgts->uuid = uuids[i];
559 lmv->max_cookiesize = 0;
561 lmv->max_easize = sizeof(struct lustre_id) *
562 desc->ld_tgt_count + sizeof(struct mea);
564 rc = lmv_setup_mgr(obd);
566 CERROR("Can't setup LMV object manager, "
568 OBD_FREE(lmv->tgts, lmv->tgts_size);
571 tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
574 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
581 static int lmv_cleanup(struct obd_device *obd, int flags)
583 struct lmv_obd *lmv = &obd->u.lmv;
586 lmv_cleanup_mgr(obd);
587 OBD_FREE(lmv->tgts, lmv->tgts_size);
592 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
593 unsigned long max_age)
595 struct lmv_obd *lmv = &obd->u.lmv;
596 struct obd_statfs temp;
600 rc = lmv_check_connect(obd);
604 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
605 if (lmv->tgts[i].ltd_exp == NULL) {
606 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
610 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
612 CERROR("can't stat MDS #%d (%s)\n", i,
613 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
617 memcpy(osfs, &temp, sizeof(temp));
619 osfs->os_bavail += temp.os_bavail;
620 osfs->os_blocks += temp.os_blocks;
621 osfs->os_ffree += temp.os_ffree;
622 osfs->os_files += temp.os_files;
628 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
630 struct obd_device *obd = exp->exp_obd;
631 struct lmv_obd *lmv = &obd->u.lmv;
635 rc = lmv_check_connect(obd);
639 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
645 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
646 unsigned long valid, unsigned int ea_size,
647 struct ptlrpc_request **request)
649 struct obd_device *obd = exp->exp_obd;
650 struct lmv_obd *lmv = &obd->u.lmv;
651 int rc, i = id_group(id);
655 rc = lmv_check_connect(obd);
659 LASSERT(i < lmv->desc.ld_tgt_count);
661 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
666 obj = lmv_grab_obj(obd, id);
668 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
669 OLID4(id), obj ? "(splitted)" : "");
671 /* if object is splitted, then we loop over all the slaves and gather
672 * size attribute. In ideal world we would have to gather also mds field
673 * from all slaves, as object is spread over the cluster and this is
674 * definitely interesting information and it is not good to loss it,
677 struct mds_body *body;
679 if (*request == NULL) {
684 body = lustre_msg_buf((*request)->rq_repmsg, 0,
686 LASSERT(body != NULL);
690 for (i = 0; i < obj->objcount; i++) {
692 if (lmv->tgts[i].ltd_exp == NULL) {
693 CWARN("%s: NULL export for %d\n",
698 /* skip master obj. */
699 if (lmv_id_equal(&obj->id, &obj->objs[i].id))
702 body->size += obj->objs[i].size;
712 static int lmv_change_cbdata(struct obd_export *exp,
713 struct lustre_id *id,
717 struct obd_device *obd = exp->exp_obd;
718 struct lmv_obd *lmv = &obd->u.lmv;
722 rc = lmv_check_connect(obd);
726 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
727 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
729 rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
735 static int lmv_change_cbdata_name(struct obd_export *exp,
736 struct lustre_id *pid,
738 struct lustre_id *cid,
739 ldlm_iterator_t it, void *data)
741 struct obd_device *obd = exp->exp_obd;
742 struct lmv_obd *lmv = &obd->u.lmv;
747 rc = lmv_check_connect(obd);
751 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
752 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
754 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
755 OLID4(pid), len, name, OLID4(cid));
757 /* this is default mds for directory name belongs to. */
759 obj = lmv_grab_obj(obd, pid);
761 /* directory is splitted. look for right mds for this name. */
762 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
763 mds = id_group(&obj->objs[mds].id);
766 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cid, it, data);
770 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
772 struct obd_device *obd = exp->exp_obd;
773 struct lmv_obd *lmv = &obd->u.lmv;
777 rc = lmv_check_connect(obd);
781 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
782 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
783 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
787 int lmv_close(struct obd_export *exp, struct obdo *obdo,
788 struct obd_client_handle *och,
789 struct ptlrpc_request **request)
791 struct obd_device *obd = exp->exp_obd;
792 struct lmv_obd *lmv = &obd->u.lmv;
793 int rc, i = obdo->o_mds;
795 rc = lmv_check_connect(obd);
798 LASSERT(i < lmv->desc.ld_tgt_count);
799 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
800 (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
801 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
805 int lmv_get_mea_and_update_object(struct obd_export *exp,
806 struct lustre_id *id)
808 struct obd_device *obd = exp->exp_obd;
809 struct lmv_obd *lmv = &obd->u.lmv;
810 struct ptlrpc_request *req = NULL;
817 mealen = MEA_SIZE_LMV(lmv);
819 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
821 /* time to update mea of parent id */
822 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
823 id, valid, mealen, &req);
825 CERROR("md_getattr() failed, error %d\n", rc);
829 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
831 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
836 GOTO(cleanup, rc = -ENODATA);
838 obj = lmv_create_obj(exp, id, md.mea);
843 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
847 ptlrpc_req_finished(req);
851 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
852 const void *data, int datalen, int mode, __u32 uid,
853 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
855 struct obd_device *obd = exp->exp_obd;
856 struct lmv_obd *lmv = &obd->u.lmv;
857 struct mds_body *body;
859 int rc, mds, loop = 0;
862 rc = lmv_check_connect(obd);
866 if (!lmv->desc.ld_active_tgt_count)
869 LASSERT(++loop <= 2);
870 obj = lmv_grab_obj(obd, &op_data->id1);
872 mds = raw_name2idx(obj->hashtype, obj->objcount,
873 op_data->name, op_data->namelen);
874 op_data->id1 = obj->objs[mds].id;
878 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
879 op_data->name, OLID4(&op_data->id1));
881 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
882 op_data, data, datalen, mode, uid, gid, rdev, request);
884 if (*request == NULL)
887 body = lustre_msg_buf((*request)->rq_repmsg, 0,
889 LASSERT(body != NULL);
891 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
893 /* LASSERT(body->valid & OBD_MD_MDS ||
894 body->mds == id_group(&op_data->id1));*/
895 } else if (rc == -ERESTART) {
896 /* directory got splitted. time to update local object and
897 * repeat the request with proper MDS */
898 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
900 ptlrpc_req_finished(*request);
907 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
909 struct obd_device *obd = exp->exp_obd;
910 struct lmv_obd *lmv = &obd->u.lmv;
913 rc = lmv_check_connect(obd);
917 /* FIXME: choose right MDC here */
918 CWARN("this method isn't implemented yet\n");
919 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
923 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
924 struct lookup_intent *it, int lockmode,
925 struct mdc_op_data *data, struct lustre_handle *lockh,
926 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
927 ldlm_blocking_callback cb_blocking, void *cb_data)
929 struct obd_device *obd = exp->exp_obd;
930 struct lmv_obd *lmv = &obd->u.lmv;
931 struct mea *mea = data->mea1;
932 struct mdc_op_data data2;
936 LASSERT(mea != NULL);
937 for (i = 0; i < mea->mea_count; i++) {
938 memset(&data2, 0, sizeof(data2));
939 data2.id1 = mea->mea_ids[i];
940 mds = id_group(&data2.id1);
942 if (lmv->tgts[mds].ltd_exp == NULL)
945 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
946 lockmode, &data2, lockh + i, lmm, lmmsize,
947 cb_compl, cb_blocking, cb_data);
949 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
950 OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
953 if (it->d.lustre.it_data) {
954 struct ptlrpc_request *req;
955 req = (struct ptlrpc_request *) it->d.lustre.it_data;
956 ptlrpc_req_finished(req);
959 if (it->d.lustre.it_status)
960 GOTO(cleanup, rc = it->d.lustre.it_status);
965 /* drop all taken locks */
968 ldlm_lock_decref(lockh + i, lockmode);
974 int lmv_enqueue(struct obd_export *exp, int lock_type,
975 struct lookup_intent *it, int lock_mode,
976 struct mdc_op_data *data, struct lustre_handle *lockh,
977 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
978 ldlm_blocking_callback cb_blocking, void *cb_data)
980 struct obd_device *obd = exp->exp_obd;
981 struct lmv_obd *lmv = &obd->u.lmv;
986 rc = lmv_check_connect(obd);
990 if (it->it_op == IT_UNLINK) {
991 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
992 data, lockh, lmm, lmmsize,
993 cb_compl, cb_blocking, cb_data);
998 obj = lmv_grab_obj(obd, &data->id1);
1000 /* directory is splitted. look for right mds for this
1002 mds = raw_name2idx(obj->hashtype, obj->objcount,
1003 (char *)data->name, data->namelen);
1004 data->id1 = obj->objs[mds].id;
1008 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1011 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1012 lock_type, it, lock_mode, data, lockh, lmm,
1013 lmmsize, cb_compl, cb_blocking, cb_data);
1018 int lmv_getattr_name(struct obd_export *exp, struct lustre_id *id,
1019 char *filename, int namelen, unsigned long valid,
1020 unsigned int ea_size, struct ptlrpc_request **request)
1022 int rc, mds = id_group(id), loop = 0;
1023 struct obd_device *obd = exp->exp_obd;
1024 struct lmv_obd *lmv = &obd->u.lmv;
1025 struct lustre_id rid = *id;
1026 struct mds_body *body;
1027 struct lmv_obj *obj;
1029 rc = lmv_check_connect(obd);
1033 LASSERT(++loop <= 2);
1034 obj = lmv_grab_obj(obd, id);
1036 /* directory is splitted. look for right mds for this name */
1037 mds = raw_name2idx(obj->hashtype, obj->objcount,
1038 filename, namelen - 1);
1039 rid = obj->objs[mds].id;
1043 CDEBUG(D_OTHER, "getattr_name for %*s on "DLID4" -> "DLID4"\n",
1044 namelen, filename, OLID4(id), OLID4(&rid));
1046 rc = md_getattr_name(lmv->tgts[id_group(&rid)].ltd_exp,
1047 &rid, filename, namelen, valid, ea_size, request);
1049 /* this could be cross-node reference. in this case all we have
1050 * right now is mds/ino/generation triple. we'd like to find
1051 * other attributes */
1052 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1053 LASSERT(body != NULL);
1054 if (body->valid & OBD_MD_MDS) {
1055 struct ptlrpc_request *req = NULL;
1058 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1060 rc = md_getattr_name(lmv->tgts[id_group(&rid)].ltd_exp,
1061 &rid, NULL, 1, valid, ea_size, &req);
1062 ptlrpc_req_finished(*request);
1065 } else if (rc == -ERESTART) {
1066 /* directory got splitted. time to update local object and
1067 * repeat the request with proper MDS */
1068 rc = lmv_get_mea_and_update_object(exp, &rid);
1070 ptlrpc_req_finished(*request);
1078 * llite passes id of an target inode in data->id1 and id of directory in
1081 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1082 struct ptlrpc_request **request)
1084 struct obd_device *obd = exp->exp_obd;
1085 struct lmv_obd *lmv = &obd->u.lmv;
1086 struct lmv_obj *obj;
1090 rc = lmv_check_connect(obd);
1094 if (data->namelen != 0) {
1095 /* usual link request */
1096 obj = lmv_grab_obj(obd, &data->id1);
1098 rc = raw_name2idx(obj->hashtype, obj->objcount,
1099 data->name, data->namelen);
1100 data->id1 = obj->objs[rc].id;
1104 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1105 OLID4(&data->id2), data->namelen, data->name,
1108 /* request from MDS to acquire i_links for inode by id1 */
1109 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1113 rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1118 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1119 const char *old, int oldlen, const char *new, int newlen,
1120 struct ptlrpc_request **request)
1122 struct obd_device *obd = exp->exp_obd;
1123 struct lmv_obd *lmv = &obd->u.lmv;
1124 struct lmv_obj *obj;
1128 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1129 oldlen, old, OLID4(&data->id1), newlen, new, OLID4(&data->id2));
1131 if (!lmv_id_equal(&data->id1, &data->id2)) {
1132 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1133 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1137 rc = lmv_check_connect(obd);
1142 /* MDS with old dir entry is asking another MDS to create name
1145 "create %*s(%d/%d) in "DLID4" pointing "
1146 "to "DLID4"\n", newlen, new, oldlen, newlen,
1147 OLID4(&data->id2), OLID4(&data->id1));
1148 mds = id_group(&data->id2);
1152 obj = lmv_grab_obj(obd, &data->id1);
1154 /* directory is already splitted, so we have to forward request
1155 * to the right MDS */
1156 mds = raw_name2idx(obj->hashtype, obj->objcount,
1157 (char *)old, oldlen);
1158 data->id1 = obj->objs[mds].id;
1159 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1160 OLID4(&obj->objs[mds].id));
1164 obj = lmv_grab_obj(obd, &data->id2);
1166 /* directory is already splitted, so we have to forward request
1167 * to the right MDS */
1168 mds = raw_name2idx(obj->hashtype, obj->objcount,
1169 (char *)new, newlen);
1171 data->id2 = obj->objs[mds].id;
1172 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1173 OLID4(&obj->objs[mds].id));
1177 mds = id_group(&data->id1);
1180 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1181 new, newlen, request);
1185 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1186 struct iattr *iattr, void *ea, int ealen, void *ea2,
1187 int ea2len, struct ptlrpc_request **request)
1189 struct obd_device *obd = exp->exp_obd;
1190 struct lmv_obd *lmv = &obd->u.lmv;
1191 struct ptlrpc_request *req;
1192 struct mds_body *body;
1193 struct lmv_obj *obj;
1197 rc = lmv_check_connect(obd);
1201 obj = lmv_grab_obj(obd, &data->id1);
1203 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1204 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1207 for (i = 0; i < obj->objcount; i++) {
1208 data->id1 = obj->objs[i].id;
1210 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1211 data, iattr, ea, ealen, ea2, ea2len, &req);
1213 if (lmv_id_equal(&obj->id, &obj->objs[i].id)) {
1214 /* this is master object and this request should
1215 * be returned back to llite */
1218 ptlrpc_req_finished(req);
1226 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1227 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1228 data, iattr, ea, ealen, ea2, ea2len, request);
1230 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1232 LASSERT(body != NULL);
1233 // LASSERT(body->mds == id_group(&data->id1));
1239 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1240 struct ptlrpc_request **request)
1242 struct obd_device *obd = exp->exp_obd;
1243 struct lmv_obd *lmv = &obd->u.lmv;
1247 rc = lmv_check_connect(obd);
1251 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1256 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1257 struct ldlm_lock_desc *desc,
1258 void *data, int flag)
1260 struct lustre_handle lockh;
1261 struct lmv_obj *obj;
1266 case LDLM_CB_BLOCKING:
1267 ldlm_lock2handle(lock, &lockh);
1268 rc = ldlm_cli_cancel(&lockh);
1270 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1274 case LDLM_CB_CANCELING:
1275 /* time to drop cached attrs for dirobj */
1276 obj = lock->l_ast_data;
1278 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1279 ", master "DLID4"\n",
1280 lock->l_resource->lr_name.name[3] == 1 ?
1281 "LOOKUP" : "UPDATE",
1282 lock->l_resource->lr_name.name[0],
1283 lock->l_resource->lr_name.name[1],
1294 void lmv_remove_dots(struct page *page)
1296 char *kaddr = page_address(page);
1297 unsigned limit = PAGE_CACHE_SIZE;
1298 unsigned offs, rec_len;
1299 struct ext2_dir_entry_2 *p;
1301 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1302 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1303 rec_len = le16_to_cpu(p->rec_len);
1305 if ((p->name_len == 1 && p->name[0] == '.') ||
1306 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1311 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1312 __u64 offset, struct page *page,
1313 struct ptlrpc_request **request)
1315 struct obd_device *obd = exp->exp_obd;
1316 struct lmv_obd *lmv = &obd->u.lmv;
1317 struct lustre_id rid = *id;
1318 struct lmv_obj *obj;
1322 #warning "we need well-desgined readdir() implementation"
1323 rc = lmv_check_connect(obd);
1327 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1328 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1329 offset, OLID4(&rid));
1331 obj = lmv_grab_obj(obd, id);
1335 /* find dirobj containing page with requested offset. */
1336 for (i = 0; i < obj->objcount; i++) {
1337 if (offset < obj->objs[i].size)
1339 offset -= obj->objs[i].size;
1341 rid = obj->objs[i].id;
1343 lmv_unlock_obj(obj);
1346 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1347 OLID4(&rid), (unsigned long)offset);
1349 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1350 offset, page, request);
1352 if (rc == 0 && !lmv_id_equal(&rid, id))
1353 /* this page isn't from master object. To avoid "." and ".."
1354 * duplication in directory, we have to remove them from all
1356 lmv_remove_dots(page);
1361 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1362 struct ptlrpc_request **req)
1364 struct obd_device *obd = exp->exp_obd;
1365 struct lmv_obd *lmv = &obd->u.lmv;
1366 struct mea *mea = data->mea1;
1367 struct mdc_op_data data2;
1371 LASSERT(mea != NULL);
1372 for (i = 0; i < mea->mea_count; i++) {
1373 memset(&data2, 0, sizeof(data2));
1374 data2.id1 = mea->mea_ids[i];
1375 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1377 mds = id_group(&data2.id1);
1379 if (lmv->tgts[mds].ltd_exp == NULL)
1382 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1383 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1384 OLID4(&mea->mea_ids[i]), rc);
1386 ptlrpc_req_finished(*req);
1395 int lmv_delete_object(struct obd_export *exp, struct lustre_id *id)
1399 if (!lmv_delete_obj(exp, id)) {
1400 CDEBUG(D_OTHER, "object "DLID4" is not found.\n",
1407 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1408 struct ptlrpc_request **request)
1410 struct obd_device *obd = exp->exp_obd;
1411 struct lmv_obd *lmv = &obd->u.lmv;
1415 rc = lmv_check_connect(obd);
1419 if (data->namelen == 0 && data->mea1 != NULL) {
1420 /* mds asks to remove slave objects */
1421 rc = lmv_unlink_slaves(exp, data, request);
1423 } else if (data->namelen != 0) {
1424 struct lmv_obj *obj;
1426 obj = lmv_grab_obj(obd, &data->id1);
1428 i = raw_name2idx(obj->hashtype, obj->objcount,
1429 data->name, data->namelen);
1430 data->id1 = obj->objs[i].id;
1433 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1434 data->namelen, data->name, OLID4(&data->id1),
1437 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1440 rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1445 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1446 char *name, int len)
1448 struct obd_device *obd = exp->exp_obd;
1449 struct lmv_obd *lmv = &obd->u.lmv;
1453 rc = lmv_check_connect(obd);
1455 RETURN(ERR_PTR(rc));
1456 obd = lmv->tgts[0].ltd_exp->exp_obd;
1461 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1463 struct obd_device *obd = exp->exp_obd;
1464 struct lmv_obd *lmv = &obd->u.lmv;
1465 int i, rc = 0, change = 0;
1468 if (lmv->max_easize < easize) {
1469 lmv->max_easize = easize;
1472 if (lmv->max_cookiesize < cookiesize) {
1473 lmv->max_cookiesize = cookiesize;
1479 if (lmv->connected == 0)
1482 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1483 if (lmv->tgts[i].ltd_exp == NULL) {
1484 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1488 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1490 CERROR("obd_init_ea_size() failed on MDT target %d, "
1491 "error %d.\n", i, rc);
1498 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1499 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1501 struct obd_device *obd = exp->exp_obd;
1502 struct lmv_obd *lmv = &obd->u.lmv;
1503 struct lov_stripe_md obj_md;
1504 struct lov_stripe_md *obj_mdp = &obj_md;
1508 LASSERT(ea == NULL);
1509 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1511 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1516 int lmv_getready(struct obd_export *exp)
1518 struct obd_device *obd = exp->exp_obd;
1522 rc = lmv_check_connect(obd);
1527 * to be called from MDS only.
1529 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1530 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1532 struct obd_device *obd = exp->exp_obd;
1533 struct lmv_obd *lmv = &obd->u.lmv;
1534 struct lustre_id mid;
1540 rc = lmv_check_connect(obd);
1544 LASSERT(oa != NULL);
1547 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1549 CERROR("Can't create object, rc = %d\n", rc);
1554 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1556 CERROR("obd_alloc_diskmd() failed, error %d\n",
1567 id_ino(&mid) = oa->o_id;
1568 id_fid(&mid) = oa->o_fid;
1569 id_gen(&mid) = oa->o_generation;
1571 mea = (struct mea *)*ea;
1572 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1573 mea->mea_count = lmv->desc.ld_tgt_count;
1575 mea->mea_master = -1;
1576 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1578 lcount = lmv->desc.ld_tgt_count;
1579 for (i = 0, c = 0; c < mea->mea_count && i < lcount; i++) {
1580 struct lov_stripe_md obj_md;
1581 struct lov_stripe_md *obj_mdp = &obj_md;
1583 if (lmv->tgts[i].ltd_exp == NULL) {
1584 /* this is master MDS */
1585 mea->mea_master = i;
1586 id_group(&mea->mea_ids[c]) = i;
1587 id_ino(&mea->mea_ids[c]) = id_ino(&mid);
1588 id_gen(&mea->mea_ids[c]) = id_gen(&mid);
1589 id_fid(&mea->mea_ids[c]) = id_fid(&mid);
1594 /* "master" MDS should always be part of stripped dir, so scan
1596 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1599 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1600 OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1602 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1604 CERROR("obd_create() failed on MDT target %d, "
1605 "error %d\n", c, rc);
1609 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1610 i, oa->o_id, oa->o_generation);
1612 /* here after object is created on desired MDS we save its fid
1613 * to local mea_ids. */
1616 id_group(&mea->mea_ids[c]) = i;
1617 id_ino(&mea->mea_ids[c]) = oa->o_id;
1618 id_fid(&mea->mea_ids[c]) = oa->o_fid;
1619 id_gen(&mea->mea_ids[c]) = oa->o_generation;
1622 LASSERT(c == mea->mea_count);
1623 CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1628 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1629 struct obd_device *tgt, int count,
1630 struct llog_catid *logid)
1632 struct llog_ctxt *ctxt;
1636 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1639 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1640 ctxt->loc_imp = tgt->u.cli.cl_import;
1646 static int lmv_llog_finish(struct obd_device *obd,
1647 struct obd_llogs *llogs, int count)
1652 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1656 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1657 void *key, __u32 *vallen, void *val)
1659 struct obd_device *obd;
1660 struct lmv_obd *lmv;
1664 obd = class_exp2obd(exp);
1666 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1667 exp->exp_handle.h_cookie);
1672 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1673 __u32 *mdsize = val;
1674 *vallen = sizeof(__u32);
1675 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1676 + sizeof(struct mea);
1678 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1679 struct obd_uuid *cluuid = &lmv->cluuid;
1680 struct lmv_tgt_desc *tgts;
1681 __u32 *mdsnum = val;
1684 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1685 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1686 *vallen = sizeof(__u32);
1692 } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1693 /* getting rootid from first MDS. */
1694 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1697 } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1698 struct lmv_desc *desc_ret = val;
1699 *desc_ret = lmv->desc;
1703 CDEBUG(D_IOCTL, "invalid key\n");
1707 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1708 void *key, obd_count vallen, void *val)
1710 struct obd_device *obd;
1711 struct lmv_obd *lmv;
1714 obd = class_exp2obd(exp);
1716 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1717 exp->exp_handle.h_cookie);
1722 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1723 lmv->server_timeout = 1;
1724 lmv_set_timeouts(obd);
1731 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1732 struct lov_stripe_md *lsm)
1734 struct obd_device *obd = class_exp2obd(exp);
1735 struct lmv_obd *lmv = &obd->u.lmv;
1739 mea_size = sizeof(struct lustre_id) *
1740 lmv->desc.ld_tgt_count + sizeof(struct mea);
1744 if (*lmmp && !lsm) {
1745 OBD_FREE(*lmmp, mea_size);
1750 if (*lmmp == NULL) {
1751 OBD_ALLOC(*lmmp, mea_size);
1759 #warning "MEA packing/convertation must be here! -bzzz"
1760 memcpy(*lmmp, lsm, mea_size);
1764 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1765 struct lov_mds_md *disk_src, int mdsize)
1767 struct obd_device *obd = class_exp2obd(exp);
1768 struct lmv_obd *lmv = &obd->u.lmv;
1769 struct mea **tmea = (struct mea **) mem_tgt;
1770 struct mea *mea = (void *) disk_src;
1774 mea_size = sizeof(struct lustre_id) *
1775 lmv->desc.ld_tgt_count + sizeof(struct mea);
1776 if (mem_tgt == NULL)
1779 if (*mem_tgt != NULL && disk_src == NULL) {
1780 OBD_FREE(*tmea, mea_size);
1784 LASSERT(mea_size == mdsize);
1786 OBD_ALLOC(*tmea, mea_size);
1793 #warning "MEA unpacking/convertation must be here! -bzzz"
1794 memcpy(*tmea, mea, mdsize);
1798 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1799 struct lov_stripe_md *ea, obd_count oa_bufs,
1800 struct brw_page *pgarr, struct obd_trans_info *oti)
1802 struct obd_device *obd = exp->exp_obd;
1803 struct lmv_obd *lmv = &obd->u.lmv;
1804 struct mea *mea = (struct mea *) ea;
1807 LASSERT(oa != NULL);
1808 LASSERT(ea != NULL);
1809 LASSERT(pgarr != NULL);
1810 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1812 oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1813 oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1814 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1816 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1817 oa, NULL, oa_bufs, pgarr, oti);
1821 struct obd_ops lmv_obd_ops = {
1822 .o_owner = THIS_MODULE,
1823 .o_attach = lmv_attach,
1824 .o_detach = lmv_detach,
1825 .o_setup = lmv_setup,
1826 .o_cleanup = lmv_cleanup,
1827 .o_connect = lmv_connect,
1828 .o_disconnect = lmv_disconnect,
1829 .o_statfs = lmv_statfs,
1830 .o_llog_init = lmv_llog_init,
1831 .o_llog_finish = lmv_llog_finish,
1832 .o_get_info = lmv_get_info,
1833 .o_set_info = lmv_set_info,
1834 .o_create = lmv_obd_create,
1835 .o_packmd = lmv_packmd,
1836 .o_unpackmd = lmv_unpackmd,
1838 .o_init_ea_size = lmv_init_ea_size,
1839 .o_notify = lmv_notify,
1840 .o_iocontrol = lmv_iocontrol,
1841 .o_getready = lmv_getready,
1844 struct md_ops lmv_md_ops = {
1845 .m_getstatus = lmv_getstatus,
1846 .m_getattr = lmv_getattr,
1847 .m_change_cbdata = lmv_change_cbdata,
1848 .m_change_cbdata_name = lmv_change_cbdata_name,
1849 .m_close = lmv_close,
1850 .m_create = lmv_create,
1851 .m_done_writing = lmv_done_writing,
1852 .m_enqueue = lmv_enqueue,
1853 .m_getattr_name = lmv_getattr_name,
1854 .m_intent_lock = lmv_intent_lock,
1856 .m_rename = lmv_rename,
1857 .m_setattr = lmv_setattr,
1859 .m_readpage = lmv_readpage,
1860 .m_unlink = lmv_unlink,
1861 .m_get_real_obd = lmv_get_real_obd,
1862 .m_valid_attrs = lmv_valid_attrs,
1863 .m_delete_object = lmv_delete_object,
1866 int __init lmv_init(void)
1868 struct lprocfs_static_vars lvars;
1871 lprocfs_init_vars(lmv, &lvars);
1872 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1873 lvars.module_vars, OBD_LMV_DEVICENAME);
1878 static void lmv_exit(void)
1880 class_unregister_type(OBD_LMV_DEVICENAME);
1883 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1884 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1885 MODULE_LICENSE("GPL");
1887 module_init(lmv_init);
1888 module_exit(lmv_exit);