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 flags)
187 struct proc_dir_entry *lmv_proc_dir;
189 struct lmv_obd *lmv = &obd->u.lmv;
190 struct obd_export *exp;
194 rc = class_connect(conn, obd, cluuid);
196 CERROR("class_connection() returned %d\n", rc);
200 exp = class_conn2export(conn);
202 /* we don't want to actually do the underlying connections more than
203 * once, so keep track. */
205 if (lmv->refcount > 1) {
206 class_export_put(exp);
212 lmv->cluuid = *cluuid;
213 lmv->connect_flags = flags;
214 sema_init(&lmv->init_sem, 1);
217 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
219 if (IS_ERR(lmv_proc_dir)) {
220 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
221 obd->obd_type->typ_name, obd->obd_name);
227 * all real clients shouls perform actual connection rightaway, because
228 * it is possible, that LMV will not have opportunity to connect
229 * targets, as MDC stuff will bit called directly, for instance while
230 * reading ../mdc/../kbytesfree procfs file, etc.
232 if (flags & OBD_OPT_REAL_CLIENT)
233 rc = lmv_check_connect(obd);
237 lprocfs_remove(lmv_proc_dir);
243 void lmv_set_timeouts(struct obd_device *obd)
245 struct lmv_tgt_desc *tgts;
250 if (lmv->server_timeout == 0)
253 if (lmv->connected == 0)
256 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
257 if (tgts->ltd_exp == NULL)
260 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
261 "inter_mds", 0, NULL);
265 #define MAX_STRING_SIZE 128
267 /* performs a check if passed obd is connected. If no - connect it. */
268 int lmv_check_connect(struct obd_device *obd)
271 struct proc_dir_entry *lmv_proc_dir;
273 struct lmv_obd *lmv = &obd->u.lmv;
274 struct lmv_tgt_desc *tgts;
275 struct obd_uuid *cluuid;
276 struct obd_export *exp;
282 down(&lmv->init_sem);
283 if (lmv->connected) {
288 cluuid = &lmv->cluuid;
291 CDEBUG(D_OTHER, "time to connect %s to %s\n",
292 cluuid->uuid, obd->obd_name);
294 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
295 struct obd_device *tgt_obd;
296 struct lustre_handle conn = {0, };
297 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
299 LASSERT(tgts != NULL);
301 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
304 CERROR("Target %s not attached\n", tgts->uuid.uuid);
305 GOTO(out_disc, rc = -EINVAL);
308 /* for MDS: don't connect to yourself */
309 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
310 CDEBUG(D_OTHER, "don't connect back to %s\n",
312 tgts->ltd_exp = NULL;
316 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
317 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
318 tgts->uuid.uuid, obd->obd_uuid.uuid,
321 if (!tgt_obd->obd_set_up) {
322 CERROR("Target %s not set up\n", tgts->uuid.uuid);
323 GOTO(out_disc, rc = -EINVAL);
326 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
329 CERROR("Target %s connect error %d\n",
330 tgts->uuid.uuid, rc);
333 tgts->ltd_exp = class_conn2export(&conn);
335 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
336 lmv->max_cookiesize);
338 rc = obd_register_observer(tgt_obd, obd);
340 CERROR("Target %s register_observer error %d\n",
341 tgts->uuid.uuid, rc);
342 obd_disconnect(tgts->ltd_exp, 0);
346 lmv->desc.ld_active_tgt_count++;
349 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
350 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
351 atomic_read(&obd->obd_refcount));
354 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
356 struct obd_device *mdc_obd = class_conn2obd(&conn);
357 struct proc_dir_entry *mdc_symlink;
358 char name[MAX_STRING_SIZE + 1];
360 LASSERT(mdc_obd != NULL);
361 LASSERT(mdc_obd->obd_type != NULL);
362 LASSERT(mdc_obd->obd_type->typ_name != NULL);
363 name[MAX_STRING_SIZE] = '\0';
364 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
365 mdc_obd->obd_type->typ_name,
367 mdc_symlink = proc_symlink(mdc_obd->obd_name,
369 if (mdc_symlink == NULL) {
370 CERROR("could not register LMV target "
371 "/proc/fs/lustre/%s/%s/target_obds/%s.",
372 obd->obd_type->typ_name, obd->obd_name,
374 lprocfs_remove(lmv_proc_dir);
381 lmv_set_timeouts(obd);
382 class_export_put(exp);
389 struct obd_uuid uuid;
391 --lmv->desc.ld_active_tgt_count;
393 /* save for CERROR below; (we know it's terminated) */
395 rc2 = obd_disconnect(tgts->ltd_exp, 0);
397 CERROR("error: LMV target %s disconnect on MDC idx %d: "
398 "error %d\n", uuid.uuid, i, rc2);
400 class_disconnect(exp, 0);
405 static int lmv_disconnect(struct obd_export *exp, int flags)
407 struct obd_device *obd = class_exp2obd(exp);
408 struct lmv_obd *lmv = &obd->u.lmv;
411 struct proc_dir_entry *lmv_proc_dir;
419 /* Only disconnect the underlying layers on the final disconnect. */
421 if (lmv->refcount != 0)
425 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
428 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
429 struct obd_device *mdc_obd;
431 if (lmv->tgts[i].ltd_exp == NULL)
434 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
438 struct proc_dir_entry *mdc_symlink;
440 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
442 lprocfs_remove(mdc_symlink);
444 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
445 obd->obd_type->typ_name, obd->obd_name,
450 if (obd->obd_no_recov) {
452 mdc_obd->obd_no_recov = 1;
454 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
455 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
456 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
458 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
459 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
461 if (lmv->tgts[i].active) {
462 CERROR("Target %s disconnect error %d\n",
463 lmv->tgts[i].uuid.uuid, rc);
468 lmv_activate_target(lmv, &lmv->tgts[i], 0);
469 lmv->tgts[i].ltd_exp = NULL;
474 lprocfs_remove(lmv_proc_dir);
476 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
477 obd->obd_type->typ_name, obd->obd_name);
482 /* this is the case when no real connection is established by
483 * lmv_check_connect(). */
485 class_export_put(exp);
486 rc = class_disconnect(exp, 0);
487 if (lmv->refcount == 0)
492 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
493 int len, void *karg, void *uarg)
495 struct obd_device *obddev = class_exp2obd(exp);
496 struct lmv_obd *lmv = &obddev->u.lmv;
497 int i, rc = 0, set = 0;
501 if (lmv->desc.ld_tgt_count == 0)
504 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
507 if (lmv->tgts[i].ltd_exp == NULL)
510 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
512 if (lmv->tgts[i].active) {
513 CERROR("error: iocontrol MDC %s on MDT"
514 "idx %d: err = %d\n",
515 lmv->tgts[i].uuid.uuid, i, err);
528 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
531 struct lmv_desc *desc;
532 struct obd_uuid *uuids;
533 struct lmv_tgt_desc *tgts;
534 struct obd_device *tgt_obd;
535 struct lustre_cfg *lcfg = buf;
536 struct lmv_obd *lmv = &obd->u.lmv;
539 if (lcfg->lcfg_inllen1 < 1) {
540 CERROR("LMV setup requires a descriptor\n");
544 if (lcfg->lcfg_inllen2 < 1) {
545 CERROR("LMV setup requires an OST UUID list\n");
549 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
550 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
551 CERROR("descriptor size wrong: %d > %d\n",
552 (int)sizeof(*desc), lcfg->lcfg_inllen1);
556 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
557 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
558 CERROR("UUID array size wrong: %u * %u != %u\n",
559 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
563 lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
564 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
565 if (lmv->tgts == NULL) {
566 CERROR("Out of memory\n");
571 spin_lock_init(&lmv->lmv_lock);
573 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
574 tgts->uuid = uuids[i];
576 lmv->max_cookiesize = 0;
578 lmv->max_easize = sizeof(struct lustre_id) *
579 desc->ld_tgt_count + sizeof(struct mea);
581 rc = lmv_setup_mgr(obd);
583 CERROR("Can't setup LMV object manager, "
585 OBD_FREE(lmv->tgts, lmv->tgts_size);
588 tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
591 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
598 static int lmv_cleanup(struct obd_device *obd, int flags)
600 struct lmv_obd *lmv = &obd->u.lmv;
603 lmv_cleanup_mgr(obd);
604 OBD_FREE(lmv->tgts, lmv->tgts_size);
609 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
610 unsigned long max_age)
612 struct lmv_obd *lmv = &obd->u.lmv;
613 struct obd_statfs temp;
617 rc = lmv_check_connect(obd);
621 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
622 if (lmv->tgts[i].ltd_exp == NULL) {
623 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
627 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
629 CERROR("can't stat MDS #%d (%s)\n", i,
630 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
634 memcpy(osfs, &temp, sizeof(temp));
636 osfs->os_bavail += temp.os_bavail;
637 osfs->os_blocks += temp.os_blocks;
638 osfs->os_ffree += temp.os_ffree;
639 osfs->os_files += temp.os_files;
645 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
647 struct obd_device *obd = exp->exp_obd;
648 struct lmv_obd *lmv = &obd->u.lmv;
652 rc = lmv_check_connect(obd);
656 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
662 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
663 unsigned long valid, unsigned int ea_size,
664 struct ptlrpc_request **request)
666 struct obd_device *obd = exp->exp_obd;
667 struct lmv_obd *lmv = &obd->u.lmv;
668 int rc, i = id_group(id);
672 rc = lmv_check_connect(obd);
676 LASSERT(i < lmv->desc.ld_tgt_count);
678 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
683 obj = lmv_grab_obj(obd, id);
685 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
686 OLID4(id), obj ? "(splitted)" : "");
688 /* if object is splitted, then we loop over all the slaves and gather
689 * size attribute. In ideal world we would have to gather also mds field
690 * from all slaves, as object is spread over the cluster and this is
691 * definitely interesting information and it is not good to loss it,
694 struct mds_body *body;
696 if (*request == NULL) {
701 body = lustre_msg_buf((*request)->rq_repmsg, 0,
703 LASSERT(body != NULL);
707 for (i = 0; i < obj->objcount; i++) {
709 if (lmv->tgts[i].ltd_exp == NULL) {
710 CWARN("%s: NULL export for %d\n",
715 /* skip master obj. */
716 if (id_equal_fid(&obj->id, &obj->objs[i].id))
719 body->size += obj->objs[i].size;
729 static int lmv_change_cbdata(struct obd_export *exp,
730 struct lustre_id *id,
734 struct obd_device *obd = exp->exp_obd;
735 struct lmv_obd *lmv = &obd->u.lmv;
739 rc = lmv_check_connect(obd);
743 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
744 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
746 rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
752 static int lmv_change_cbdata_name(struct obd_export *exp,
753 struct lustre_id *pid,
755 struct lustre_id *cid,
759 struct obd_device *obd = exp->exp_obd;
760 struct lmv_obd *lmv = &obd->u.lmv;
761 struct lustre_id rcid = *cid;
766 rc = lmv_check_connect(obd);
770 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
771 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
773 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
774 OLID4(pid), len, name, OLID4(cid));
776 /* this is default mds for directory name belongs to. */
778 obj = lmv_grab_obj(obd, pid);
780 /* directory is splitted. look for right mds for this name. */
781 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
782 rcid = obj->objs[mds].id;
783 mds = id_group(&rcid);
786 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
790 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
792 struct obd_device *obd = exp->exp_obd;
793 struct lmv_obd *lmv = &obd->u.lmv;
797 rc = lmv_check_connect(obd);
801 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
802 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
803 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
807 int lmv_close(struct obd_export *exp, struct obdo *obdo,
808 struct obd_client_handle *och,
809 struct ptlrpc_request **request)
811 struct obd_device *obd = exp->exp_obd;
812 struct lmv_obd *lmv = &obd->u.lmv;
813 int rc, i = obdo->o_mds;
816 rc = lmv_check_connect(obd);
820 LASSERT(i < lmv->desc.ld_tgt_count);
821 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
822 (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
823 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
827 int lmv_get_mea_and_update_object(struct obd_export *exp,
828 struct lustre_id *id)
830 struct obd_device *obd = exp->exp_obd;
831 struct lmv_obd *lmv = &obd->u.lmv;
832 struct ptlrpc_request *req = NULL;
839 mealen = MEA_SIZE_LMV(lmv);
841 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
843 /* time to update mea of parent id */
844 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
845 id, valid, mealen, &req);
847 CERROR("md_getattr() failed, error %d\n", rc);
851 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
853 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
858 GOTO(cleanup, rc = -ENODATA);
860 obj = lmv_create_obj(exp, id, md.mea);
865 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
869 ptlrpc_req_finished(req);
873 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
874 const void *data, int datalen, int mode, __u32 uid,
875 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
877 struct obd_device *obd = exp->exp_obd;
878 struct lmv_obd *lmv = &obd->u.lmv;
879 struct mds_body *body;
881 int rc, mds, loop = 0;
884 rc = lmv_check_connect(obd);
888 if (!lmv->desc.ld_active_tgt_count)
891 LASSERT(++loop <= 2);
892 obj = lmv_grab_obj(obd, &op_data->id1);
894 mds = raw_name2idx(obj->hashtype, obj->objcount,
895 op_data->name, op_data->namelen);
896 op_data->id1 = obj->objs[mds].id;
900 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
901 op_data->name, OLID4(&op_data->id1));
903 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
904 op_data, data, datalen, mode, uid, gid, rdev, request);
906 if (*request == NULL)
909 body = lustre_msg_buf((*request)->rq_repmsg, 0,
911 LASSERT(body != NULL);
913 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
915 /* LASSERT(body->valid & OBD_MD_MDS ||
916 body->mds == id_group(&op_data->id1));*/
917 } else if (rc == -ERESTART) {
918 /* directory got splitted. time to update local object and
919 * repeat the request with proper MDS */
920 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
922 ptlrpc_req_finished(*request);
929 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
931 struct obd_device *obd = exp->exp_obd;
932 struct lmv_obd *lmv = &obd->u.lmv;
935 rc = lmv_check_connect(obd);
939 /* FIXME: choose right MDC here */
940 CWARN("this method isn't implemented yet\n");
941 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
945 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
946 struct lookup_intent *it, int lockmode,
947 struct mdc_op_data *data, struct lustre_handle *lockh,
948 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
949 ldlm_blocking_callback cb_blocking, void *cb_data)
951 struct obd_device *obd = exp->exp_obd;
952 struct lmv_obd *lmv = &obd->u.lmv;
953 struct mea *mea = data->mea1;
954 struct mdc_op_data data2;
958 LASSERT(mea != NULL);
959 for (i = 0; i < mea->mea_count; i++) {
960 memset(&data2, 0, sizeof(data2));
961 data2.id1 = mea->mea_ids[i];
962 mds = id_group(&data2.id1);
964 if (lmv->tgts[mds].ltd_exp == NULL)
967 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
968 lockmode, &data2, lockh + i, lmm, lmmsize,
969 cb_compl, cb_blocking, cb_data);
971 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
972 OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
975 if (it->d.lustre.it_data) {
976 struct ptlrpc_request *req;
977 req = (struct ptlrpc_request *) it->d.lustre.it_data;
978 ptlrpc_req_finished(req);
981 if (it->d.lustre.it_status)
982 GOTO(cleanup, rc = it->d.lustre.it_status);
987 /* drop all taken locks */
990 ldlm_lock_decref(lockh + i, lockmode);
996 int lmv_enqueue(struct obd_export *exp, int lock_type,
997 struct lookup_intent *it, int lock_mode,
998 struct mdc_op_data *data, struct lustre_handle *lockh,
999 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1000 ldlm_blocking_callback cb_blocking, void *cb_data)
1002 struct obd_device *obd = exp->exp_obd;
1003 struct lmv_obd *lmv = &obd->u.lmv;
1004 struct lmv_obj *obj;
1008 rc = lmv_check_connect(obd);
1012 if (data->mea1 && it->it_op == IT_UNLINK) {
1013 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1014 data, lockh, lmm, lmmsize,
1015 cb_compl, cb_blocking, cb_data);
1019 if (data->namelen) {
1020 obj = lmv_grab_obj(obd, &data->id1);
1022 /* directory is splitted. look for right mds for this
1024 mds = raw_name2idx(obj->hashtype, obj->objcount,
1025 (char *)data->name, data->namelen);
1026 data->id1 = obj->objs[mds].id;
1030 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1033 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1034 lock_type, it, lock_mode, data, lockh, lmm,
1035 lmmsize, cb_compl, cb_blocking, cb_data);
1039 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1040 char *filename, int namelen, unsigned long valid,
1041 unsigned int ea_size, struct ptlrpc_request **request)
1043 int rc, mds = id_group(id), loop = 0;
1044 struct obd_device *obd = exp->exp_obd;
1045 struct lmv_obd *lmv = &obd->u.lmv;
1046 struct lustre_id rid = *id;
1047 struct mds_body *body;
1048 struct lmv_obj *obj;
1052 rc = lmv_check_connect(obd);
1056 LASSERT(++loop <= 2);
1057 obj = lmv_grab_obj(obd, id);
1059 /* directory is splitted. look for right mds for this name */
1060 mds = raw_name2idx(obj->hashtype, obj->objcount,
1061 filename, namelen - 1);
1062 rid = obj->objs[mds].id;
1066 CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1067 namelen, filename, OLID4(id), OLID4(&rid));
1072 * here should be applied OBD_MD_FID to ->valid, because otherwise,
1073 * mds_getattr_lock() will not fetch fid component of lustre_id and
1074 * thus, next call to md_getattr_lock() will be performed to wrong mds.
1076 if (!(old_valid & OBD_MD_FID))
1077 valid |= OBD_MD_FID;
1079 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1080 &rid, filename, namelen, valid,
1084 * this could be cross-node reference. in this case all we have
1085 * right now is lustre_id triple. we'd like to find other
1088 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1089 LASSERT(body != NULL);
1090 if (body->valid & OBD_MD_MDS) {
1091 struct ptlrpc_request *req = NULL;
1094 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1097 * turning OBD_MD_FID fetching off, as we already have
1098 * full lustre_id and do need to fetch fid component
1099 * again. This will help to make thing slightly faster.
1101 if (!(old_valid & OBD_MD_FID))
1102 valid &= ~OBD_MD_FID;
1104 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1105 &rid, NULL, 1, valid, ea_size, &req);
1106 ptlrpc_req_finished(*request);
1109 } else if (rc == -ERESTART) {
1110 /* directory got splitted. time to update local object and
1111 * repeat the request with proper MDS */
1112 rc = lmv_get_mea_and_update_object(exp, &rid);
1114 ptlrpc_req_finished(*request);
1122 * llite passes id of an target inode in data->id1 and id of directory in
1125 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1126 struct ptlrpc_request **request)
1128 struct obd_device *obd = exp->exp_obd;
1129 struct lmv_obd *lmv = &obd->u.lmv;
1130 struct lmv_obj *obj;
1134 rc = lmv_check_connect(obd);
1138 if (data->namelen != 0) {
1139 /* usual link request */
1140 obj = lmv_grab_obj(obd, &data->id1);
1142 rc = raw_name2idx(obj->hashtype, obj->objcount,
1143 data->name, data->namelen);
1144 data->id1 = obj->objs[rc].id;
1148 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1149 OLID4(&data->id2), data->namelen, data->name,
1152 /* request from MDS to acquire i_links for inode by id1 */
1153 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1157 rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1162 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1163 const char *old, int oldlen, const char *new, int newlen,
1164 struct ptlrpc_request **request)
1166 struct obd_device *obd = exp->exp_obd;
1167 struct lmv_obd *lmv = &obd->u.lmv;
1168 struct lmv_obj *obj;
1172 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1173 oldlen, old, OLID4(&data->id1), newlen, new, OLID4(&data->id2));
1175 rc = lmv_check_connect(obd);
1181 * MDS with old dir entry is asking another MDS to create name
1185 "create %*s(%d/%d) in "DLID4" pointing "
1186 "to "DLID4"\n", newlen, new, oldlen, newlen,
1187 OLID4(&data->id2), OLID4(&data->id1));
1189 mds = id_group(&data->id2);
1192 * target directory can be splitted, sowe should forward request
1195 obj = lmv_grab_obj(obd, &data->id2);
1197 mds = raw_name2idx(obj->hashtype, obj->objcount,
1198 (char *)new, newlen);
1199 data->id2 = obj->objs[mds].id;
1200 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1207 obj = lmv_grab_obj(obd, &data->id1);
1210 * directory is already splitted, so we have to forward request
1213 mds = raw_name2idx(obj->hashtype, obj->objcount,
1214 (char *)old, oldlen);
1215 data->id1 = obj->objs[mds].id;
1216 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1221 obj = lmv_grab_obj(obd, &data->id2);
1224 * directory is already splitted, so we have to forward request
1227 mds = raw_name2idx(obj->hashtype, obj->objcount,
1228 (char *)new, newlen);
1230 data->id2 = obj->objs[mds].id;
1231 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1236 mds = id_group(&data->id1);
1239 if (id_group(&data->id1) != id_group(&data->id2)) {
1240 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1241 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1245 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1246 new, newlen, request);
1250 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1251 struct iattr *iattr, void *ea, int ealen, void *ea2,
1252 int ea2len, struct ptlrpc_request **request)
1254 struct obd_device *obd = exp->exp_obd;
1255 struct lmv_obd *lmv = &obd->u.lmv;
1256 struct ptlrpc_request *req;
1257 struct mds_body *body;
1258 struct lmv_obj *obj;
1262 rc = lmv_check_connect(obd);
1266 obj = lmv_grab_obj(obd, &data->id1);
1268 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1269 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1272 for (i = 0; i < obj->objcount; i++) {
1273 data->id1 = obj->objs[i].id;
1275 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1276 data, iattr, ea, ealen, ea2, ea2len, &req);
1278 if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1280 * this is master object and this request should
1281 * be returned back to llite.
1285 ptlrpc_req_finished(req);
1293 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1294 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1295 data, iattr, ea, ealen, ea2, ea2len, request);
1297 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1299 LASSERT(body != NULL);
1300 LASSERT(id_group(&body->id1) == id_group(&data->id1));
1306 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1307 struct ptlrpc_request **request)
1309 struct obd_device *obd = exp->exp_obd;
1310 struct lmv_obd *lmv = &obd->u.lmv;
1314 rc = lmv_check_connect(obd);
1318 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1323 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1324 struct ldlm_lock_desc *desc,
1325 void *data, int flag)
1327 struct lustre_handle lockh;
1328 struct lmv_obj *obj;
1333 case LDLM_CB_BLOCKING:
1334 ldlm_lock2handle(lock, &lockh);
1335 rc = ldlm_cli_cancel(&lockh);
1337 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1341 case LDLM_CB_CANCELING:
1342 /* time to drop cached attrs for dirobj */
1343 obj = lock->l_ast_data;
1345 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1346 ", master "DLID4"\n",
1347 lock->l_resource->lr_name.name[3] == 1 ?
1348 "LOOKUP" : "UPDATE",
1349 lock->l_resource->lr_name.name[0],
1350 lock->l_resource->lr_name.name[1],
1361 void lmv_remove_dots(struct page *page)
1363 char *kaddr = page_address(page);
1364 unsigned limit = PAGE_CACHE_SIZE;
1365 unsigned offs, rec_len;
1366 struct ext2_dir_entry_2 *p;
1368 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1369 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1370 rec_len = le16_to_cpu(p->rec_len);
1372 if ((p->name_len == 1 && p->name[0] == '.') ||
1373 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1378 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1379 __u64 offset, struct page *page,
1380 struct ptlrpc_request **request)
1382 struct obd_device *obd = exp->exp_obd;
1383 struct lmv_obd *lmv = &obd->u.lmv;
1384 struct lustre_id rid = *id;
1385 struct lmv_obj *obj;
1389 #warning "we need well-desgined readdir() implementation"
1390 rc = lmv_check_connect(obd);
1394 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1395 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1396 offset, OLID4(&rid));
1398 obj = lmv_grab_obj(obd, id);
1402 /* find dirobj containing page with requested offset. */
1403 for (i = 0; i < obj->objcount; i++) {
1404 if (offset < obj->objs[i].size)
1406 offset -= obj->objs[i].size;
1408 rid = obj->objs[i].id;
1410 lmv_unlock_obj(obj);
1413 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1414 OLID4(&rid), (unsigned long)offset);
1416 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1417 offset, page, request);
1419 if (rc == 0 && !id_equal_fid(&rid, id))
1420 /* this page isn't from master object. To avoid "." and ".."
1421 * duplication in directory, we have to remove them from all
1423 lmv_remove_dots(page);
1428 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1429 struct ptlrpc_request **req)
1431 struct obd_device *obd = exp->exp_obd;
1432 struct lmv_obd *lmv = &obd->u.lmv;
1433 struct mea *mea = data->mea1;
1434 struct mdc_op_data data2;
1438 LASSERT(mea != NULL);
1439 for (i = 0; i < mea->mea_count; i++) {
1440 memset(&data2, 0, sizeof(data2));
1441 data2.id1 = mea->mea_ids[i];
1442 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1444 mds = id_group(&data2.id1);
1446 if (lmv->tgts[mds].ltd_exp == NULL)
1449 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1450 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1451 OLID4(&mea->mea_ids[i]), rc);
1453 ptlrpc_req_finished(*req);
1462 int lmv_delete_object(struct obd_export *exp, struct lustre_id *id)
1466 if (!lmv_delete_obj(exp, id)) {
1467 CDEBUG(D_OTHER, "object "DLID4" is not found.\n",
1474 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1475 struct ptlrpc_request **request)
1477 struct obd_device *obd = exp->exp_obd;
1478 struct lmv_obd *lmv = &obd->u.lmv;
1482 rc = lmv_check_connect(obd);
1486 if (data->namelen == 0 && data->mea1 != NULL) {
1487 /* mds asks to remove slave objects */
1488 rc = lmv_unlink_slaves(exp, data, request);
1490 } else if (data->namelen != 0) {
1491 struct lmv_obj *obj;
1493 obj = lmv_grab_obj(obd, &data->id1);
1495 i = raw_name2idx(obj->hashtype, obj->objcount,
1496 data->name, data->namelen);
1497 data->id1 = obj->objs[i].id;
1500 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1501 data->namelen, data->name, OLID4(&data->id1),
1504 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1507 rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1512 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1513 char *name, int len)
1515 struct obd_device *obd = exp->exp_obd;
1516 struct lmv_obd *lmv = &obd->u.lmv;
1520 rc = lmv_check_connect(obd);
1522 RETURN(ERR_PTR(rc));
1523 obd = lmv->tgts[0].ltd_exp->exp_obd;
1528 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1530 struct obd_device *obd = exp->exp_obd;
1531 struct lmv_obd *lmv = &obd->u.lmv;
1532 int i, rc = 0, change = 0;
1535 if (lmv->max_easize < easize) {
1536 lmv->max_easize = easize;
1539 if (lmv->max_cookiesize < cookiesize) {
1540 lmv->max_cookiesize = cookiesize;
1546 if (lmv->connected == 0)
1549 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1550 if (lmv->tgts[i].ltd_exp == NULL) {
1551 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1555 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1557 CERROR("obd_init_ea_size() failed on MDT target %d, "
1558 "error %d.\n", i, rc);
1565 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1566 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1568 struct obd_device *obd = exp->exp_obd;
1569 struct lmv_obd *lmv = &obd->u.lmv;
1570 struct lov_stripe_md obj_md;
1571 struct lov_stripe_md *obj_mdp = &obj_md;
1575 LASSERT(ea == NULL);
1576 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1578 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1583 int lmv_getready(struct obd_export *exp)
1585 struct obd_device *obd = exp->exp_obd;
1589 rc = lmv_check_connect(obd);
1594 * to be called from MDS only. @oa should have correct store cookie and o_fid
1595 * values for "master" object, as it will be used.
1597 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1598 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1600 struct obd_device *obd = exp->exp_obd;
1601 struct lmv_obd *lmv = &obd->u.lmv;
1602 struct lustre_id mid;
1607 rc = lmv_check_connect(obd);
1611 LASSERT(oa != NULL);
1614 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1616 CERROR("Can't create object, rc = %d\n", rc);
1621 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1623 CERROR("obd_alloc_diskmd() failed, error %d\n",
1634 * here we should take care about splitted dir, so store cookie and fid
1635 * for "master" object should already be allocated and passed in @oa.
1637 LASSERT(oa->o_id != 0);
1638 LASSERT(oa->o_fid != 0);
1640 /* save "master" object id */
1643 mea = (struct mea *)*ea;
1644 mea->mea_master = -1;
1645 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1647 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1648 mea->mea_count = lmv->desc.ld_tgt_count;
1650 for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1651 struct lov_stripe_md obj_md;
1652 struct lov_stripe_md *obj_mdp = &obj_md;
1654 if (lmv->tgts[i].ltd_exp == NULL) {
1655 /* this is "master" MDS */
1656 mea->mea_master = i;
1657 mea->mea_ids[c] = mid;
1663 * "master" MDS should always be part of stripped dir,
1666 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1669 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1670 OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1672 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1674 CERROR("obd_create() failed on MDT target %d, "
1675 "error %d\n", c, rc);
1679 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1680 i, oa->o_id, oa->o_generation);
1684 * here, when object is created (or it is master and was passed
1685 * from caller) on desired MDS we save its fid to local mea_ids.
1690 * store cookie should be defined here for both cases (master
1691 * object and not master), because master is already created.
1695 /* fill mea by store cookie and fid */
1696 obdo2id(&mea->mea_ids[c], oa);
1699 LASSERT(c == mea->mea_count);
1701 CDEBUG(D_OTHER, "%d dirobjects created\n",
1702 (int)mea->mea_count);
1707 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1708 struct obd_device *tgt, int count,
1709 struct llog_catid *logid)
1711 struct llog_ctxt *ctxt;
1715 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1718 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1719 ctxt->loc_imp = tgt->u.cli.cl_import;
1725 static int lmv_llog_finish(struct obd_device *obd,
1726 struct obd_llogs *llogs, int count)
1731 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1735 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1736 void *key, __u32 *vallen, void *val)
1738 struct obd_device *obd;
1739 struct lmv_obd *lmv;
1743 obd = class_exp2obd(exp);
1745 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1746 exp->exp_handle.h_cookie);
1751 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1752 __u32 *mdsize = val;
1753 *vallen = sizeof(__u32);
1754 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1755 + sizeof(struct mea);
1757 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1758 struct obd_uuid *cluuid = &lmv->cluuid;
1759 struct lmv_tgt_desc *tgts;
1760 __u32 *mdsnum = val;
1763 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1764 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1765 *vallen = sizeof(__u32);
1771 } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1772 /* getting rootid from first MDS. */
1773 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1776 } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1777 struct lmv_desc *desc_ret = val;
1778 *desc_ret = lmv->desc;
1782 CDEBUG(D_IOCTL, "invalid key\n");
1786 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1787 void *key, obd_count vallen, void *val)
1789 struct obd_device *obd;
1790 struct lmv_obd *lmv;
1793 obd = class_exp2obd(exp);
1795 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1796 exp->exp_handle.h_cookie);
1801 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1802 lmv->server_timeout = 1;
1803 lmv_set_timeouts(obd);
1810 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1811 struct lov_stripe_md *lsm)
1813 struct obd_device *obd = class_exp2obd(exp);
1814 struct lmv_obd *lmv = &obd->u.lmv;
1818 mea_size = sizeof(struct lustre_id) *
1819 lmv->desc.ld_tgt_count + sizeof(struct mea);
1823 if (*lmmp && !lsm) {
1824 OBD_FREE(*lmmp, mea_size);
1829 if (*lmmp == NULL) {
1830 OBD_ALLOC(*lmmp, mea_size);
1838 #warning "MEA packing/convertation must be here! -bzzz"
1839 memcpy(*lmmp, lsm, mea_size);
1843 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1844 struct lov_mds_md *disk_src, int mdsize)
1846 struct obd_device *obd = class_exp2obd(exp);
1847 struct lmv_obd *lmv = &obd->u.lmv;
1848 struct mea **tmea = (struct mea **) mem_tgt;
1849 struct mea *mea = (void *) disk_src;
1853 mea_size = sizeof(struct lustre_id) *
1854 lmv->desc.ld_tgt_count + sizeof(struct mea);
1855 if (mem_tgt == NULL)
1858 if (*mem_tgt != NULL && disk_src == NULL) {
1859 OBD_FREE(*tmea, mea_size);
1863 LASSERT(mea_size == mdsize);
1865 OBD_ALLOC(*tmea, mea_size);
1872 #warning "MEA unpacking/convertation must be here! -bzzz"
1873 memcpy(*tmea, mea, mdsize);
1877 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1878 struct lov_stripe_md *ea, obd_count oa_bufs,
1879 struct brw_page *pgarr, struct obd_trans_info *oti)
1881 struct obd_device *obd = exp->exp_obd;
1882 struct lmv_obd *lmv = &obd->u.lmv;
1883 struct mea *mea = (struct mea *) ea;
1886 LASSERT(oa != NULL);
1887 LASSERT(ea != NULL);
1888 LASSERT(pgarr != NULL);
1889 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1891 oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1892 oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1893 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1895 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1896 oa, NULL, oa_bufs, pgarr, oti);
1900 struct obd_ops lmv_obd_ops = {
1901 .o_owner = THIS_MODULE,
1902 .o_attach = lmv_attach,
1903 .o_detach = lmv_detach,
1904 .o_setup = lmv_setup,
1905 .o_cleanup = lmv_cleanup,
1906 .o_connect = lmv_connect,
1907 .o_disconnect = lmv_disconnect,
1908 .o_statfs = lmv_statfs,
1909 .o_llog_init = lmv_llog_init,
1910 .o_llog_finish = lmv_llog_finish,
1911 .o_get_info = lmv_get_info,
1912 .o_set_info = lmv_set_info,
1913 .o_create = lmv_obd_create,
1914 .o_packmd = lmv_packmd,
1915 .o_unpackmd = lmv_unpackmd,
1917 .o_init_ea_size = lmv_init_ea_size,
1918 .o_notify = lmv_notify,
1919 .o_iocontrol = lmv_iocontrol,
1920 .o_getready = lmv_getready,
1923 struct md_ops lmv_md_ops = {
1924 .m_getstatus = lmv_getstatus,
1925 .m_getattr = lmv_getattr,
1926 .m_change_cbdata = lmv_change_cbdata,
1927 .m_change_cbdata_name = lmv_change_cbdata_name,
1928 .m_close = lmv_close,
1929 .m_create = lmv_create,
1930 .m_done_writing = lmv_done_writing,
1931 .m_enqueue = lmv_enqueue,
1932 .m_getattr_lock = lmv_getattr_lock,
1933 .m_intent_lock = lmv_intent_lock,
1935 .m_rename = lmv_rename,
1936 .m_setattr = lmv_setattr,
1938 .m_readpage = lmv_readpage,
1939 .m_unlink = lmv_unlink,
1940 .m_get_real_obd = lmv_get_real_obd,
1941 .m_valid_attrs = lmv_valid_attrs,
1942 .m_delete_object = lmv_delete_object,
1945 int __init lmv_init(void)
1947 struct lprocfs_static_vars lvars;
1950 lprocfs_init_vars(lmv, &lvars);
1951 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1952 lvars.module_vars, OBD_LMV_DEVICENAME);
1957 static void lmv_exit(void)
1959 class_unregister_type(OBD_LMV_DEVICENAME);
1962 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1963 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1964 MODULE_LICENSE("GPL");
1966 module_init(lmv_init);
1967 module_exit(lmv_exit);