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"
53 kmem_cache_t *obj_cache;
54 atomic_t obj_cache_count = ATOMIC_INIT(0);
56 static void lmv_activate_target(struct lmv_obd *lmv,
57 struct lmv_tgt_desc *tgt,
60 if (tgt->active == activate)
63 tgt->active = activate;
64 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
69 * -EINVAL : UUID can't be found in the LMV's target list
70 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
71 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
73 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
76 struct obd_device *obd;
77 struct lmv_tgt_desc *tgt;
81 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
82 lmv, uuid->uuid, activate);
84 spin_lock(&lmv->lmv_lock);
85 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
86 if (tgt->ltd_exp == NULL)
89 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
90 i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
92 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
96 if (i == lmv->desc.ld_tgt_count)
97 GOTO(out_lmv_lock, rc = -EINVAL);
99 obd = class_exp2obd(tgt->ltd_exp);
101 GOTO(out_lmv_lock, rc = -ENOTCONN);
103 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
104 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
105 obd->obd_type->typ_name, i);
106 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
108 if (tgt->active == activate) {
109 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
110 activate ? "" : "in");
111 GOTO(out_lmv_lock, rc);
114 CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
115 obd, activate ? "" : "in");
117 lmv_activate_target(lmv, tgt, activate);
122 spin_unlock(&lmv->lmv_lock);
126 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
127 int active, void *data)
130 struct obd_uuid *uuid;
132 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
133 CERROR("unexpected notification of %s %s!\n",
134 watched->obd_type->typ_name,
138 uuid = &watched->u.cli.cl_import->imp_target_uuid;
140 /* Set MDC as active before notifying the observer, so the observer can
141 * use the MDC normally.
143 rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
145 CERROR("%sactivation of %s failed: %d\n",
146 active ? "" : "de", uuid->uuid, rc);
150 if (obd->obd_observer)
151 /* Pass the notification up the chain. */
152 rc = obd_notify(obd->obd_observer, watched, active, data);
157 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
159 struct lprocfs_static_vars lvars;
163 lprocfs_init_vars(lmv, &lvars);
164 rc = lprocfs_obd_attach(dev, lvars.obd_vars);
167 struct proc_dir_entry *entry;
169 entry = create_proc_entry("target_obd_status", 0444,
170 dev->obd_proc_entry);
173 entry->proc_fops = &lmv_proc_target_fops;
180 int lmv_detach(struct obd_device *dev)
182 return lprocfs_obd_detach(dev);
185 /* this is fake connect function. Its purpose is to initialize lmv and say
186 * caller that everything is okay. Real connection will be performed later. */
187 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
188 struct obd_uuid *cluuid, unsigned long flags)
191 struct proc_dir_entry *lmv_proc_dir;
193 struct lmv_obd *lmv = &obd->u.lmv;
194 struct obd_export *exp;
198 rc = class_connect(conn, obd, cluuid);
200 CERROR("class_connection() returned %d\n", rc);
204 exp = class_conn2export(conn);
206 /* we don't want to actually do the underlying connections more than
207 * once, so keep track. */
209 if (lmv->refcount > 1) {
210 class_export_put(exp);
216 lmv->cluuid = *cluuid;
217 lmv->connect_flags = flags;
218 sema_init(&lmv->init_sem, 1);
221 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
223 if (IS_ERR(lmv_proc_dir)) {
224 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
225 obd->obd_type->typ_name, obd->obd_name);
231 * all real clients shouls perform actual connection rightaway, because
232 * it is possible, that LMV will not have opportunity to connect
233 * targets, as MDC stuff will bit called directly, for instance while
234 * reading ../mdc/../kbytesfree procfs file, etc.
236 if (flags & OBD_OPT_REAL_CLIENT)
237 rc = lmv_check_connect(obd);
242 lprocfs_remove(lmv_proc_dir);
249 void lmv_set_timeouts(struct obd_device *obd)
251 struct lmv_tgt_desc *tgts;
256 if (lmv->server_timeout == 0)
259 if (lmv->connected == 0)
262 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
263 if (tgts->ltd_exp == NULL)
266 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
267 "inter_mds", 0, NULL);
271 #define MAX_STRING_SIZE 128
273 /* performs a check if passed obd is connected. If no - connect it. */
274 int lmv_check_connect(struct obd_device *obd)
277 struct proc_dir_entry *lmv_proc_dir;
279 struct lmv_obd *lmv = &obd->u.lmv;
280 struct lmv_tgt_desc *tgts;
281 struct obd_uuid *cluuid;
282 struct obd_export *exp;
288 down(&lmv->init_sem);
289 if (lmv->connected) {
294 cluuid = &lmv->cluuid;
297 CDEBUG(D_OTHER, "time to connect %s to %s\n",
298 cluuid->uuid, obd->obd_name);
300 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
301 struct obd_device *tgt_obd;
302 struct lustre_handle conn = {0, };
303 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
305 LASSERT(tgts != NULL);
307 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
310 CERROR("Target %s not attached\n", tgts->uuid.uuid);
311 GOTO(out_disc, rc = -EINVAL);
314 /* for MDS: don't connect to yourself */
315 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
316 CDEBUG(D_OTHER, "don't connect back to %s\n",
318 tgts->ltd_exp = NULL;
322 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
323 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
324 tgts->uuid.uuid, obd->obd_uuid.uuid,
327 if (!tgt_obd->obd_set_up) {
328 CERROR("Target %s not set up\n", tgts->uuid.uuid);
329 GOTO(out_disc, rc = -EINVAL);
332 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
335 CERROR("Target %s connect error %d\n",
336 tgts->uuid.uuid, rc);
339 tgts->ltd_exp = class_conn2export(&conn);
341 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
342 lmv->max_cookiesize);
344 rc = obd_register_observer(tgt_obd, obd);
346 CERROR("Target %s register_observer error %d\n",
347 tgts->uuid.uuid, rc);
348 obd_disconnect(tgts->ltd_exp, 0);
352 lmv->desc.ld_active_tgt_count++;
355 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
356 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
357 atomic_read(&obd->obd_refcount));
360 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
362 struct obd_device *mdc_obd = class_conn2obd(&conn);
363 struct proc_dir_entry *mdc_symlink;
364 char name[MAX_STRING_SIZE + 1];
366 LASSERT(mdc_obd != NULL);
367 LASSERT(mdc_obd->obd_type != NULL);
368 LASSERT(mdc_obd->obd_type->typ_name != NULL);
369 name[MAX_STRING_SIZE] = '\0';
370 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
371 mdc_obd->obd_type->typ_name,
373 mdc_symlink = proc_symlink(mdc_obd->obd_name,
375 if (mdc_symlink == NULL) {
376 CERROR("could not register LMV target "
377 "/proc/fs/lustre/%s/%s/target_obds/%s.",
378 obd->obd_type->typ_name, obd->obd_name,
380 lprocfs_remove(lmv_proc_dir);
387 lmv_set_timeouts(obd);
388 class_export_put(exp);
395 struct obd_uuid uuid;
397 --lmv->desc.ld_active_tgt_count;
399 /* save for CERROR below; (we know it's terminated) */
401 rc2 = obd_disconnect(tgts->ltd_exp, 0);
403 CERROR("error: LMV target %s disconnect on MDC idx %d: "
404 "error %d\n", uuid.uuid, i, rc2);
406 class_disconnect(exp, 0);
411 static int lmv_disconnect(struct obd_export *exp, int flags)
413 struct obd_device *obd = class_exp2obd(exp);
414 struct lmv_obd *lmv = &obd->u.lmv;
417 struct proc_dir_entry *lmv_proc_dir;
425 /* Only disconnect the underlying layers on the final disconnect. */
427 if (lmv->refcount != 0)
431 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
434 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
435 struct obd_device *mdc_obd;
437 if (lmv->tgts[i].ltd_exp == NULL)
440 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
444 struct proc_dir_entry *mdc_symlink;
446 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
448 lprocfs_remove(mdc_symlink);
450 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
451 obd->obd_type->typ_name, obd->obd_name,
456 if (obd->obd_no_recov) {
458 mdc_obd->obd_no_recov = 1;
460 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
461 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
462 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
464 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
465 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
467 if (lmv->tgts[i].active) {
468 CERROR("Target %s disconnect error %d\n",
469 lmv->tgts[i].uuid.uuid, rc);
474 lmv_activate_target(lmv, &lmv->tgts[i], 0);
475 lmv->tgts[i].ltd_exp = NULL;
480 lprocfs_remove(lmv_proc_dir);
482 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
483 obd->obd_type->typ_name, obd->obd_name);
488 /* this is the case when no real connection is established by
489 * lmv_check_connect(). */
491 class_export_put(exp);
492 rc = class_disconnect(exp, 0);
493 if (lmv->refcount == 0)
498 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
499 int len, void *karg, void *uarg)
501 struct obd_device *obddev = class_exp2obd(exp);
502 struct lmv_obd *lmv = &obddev->u.lmv;
503 int i, rc = 0, set = 0;
506 if (lmv->desc.ld_tgt_count == 0)
509 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
512 if (lmv->tgts[i].ltd_exp == NULL)
515 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
517 if (lmv->tgts[i].active) {
518 CERROR("error: iocontrol MDC %s on MDT"
519 "idx %d: err = %d\n",
520 lmv->tgts[i].uuid.uuid, i, err);
533 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
536 struct lmv_desc *desc;
537 struct obd_uuid *uuids;
538 struct lmv_tgt_desc *tgts;
539 struct obd_device *tgt_obd;
540 struct lustre_cfg *lcfg = buf;
541 struct lmv_obd *lmv = &obd->u.lmv;
544 if (lcfg->lcfg_inllen1 < 1) {
545 CERROR("LMV setup requires a descriptor\n");
549 if (lcfg->lcfg_inllen2 < 1) {
550 CERROR("LMV setup requires an OST UUID list\n");
554 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
555 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
556 CERROR("descriptor size wrong: %d > %d\n",
557 (int)sizeof(*desc), lcfg->lcfg_inllen1);
561 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
562 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
563 CERROR("UUID array size wrong: %u * %u != %u\n",
564 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
568 lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
569 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
570 if (lmv->tgts == NULL) {
571 CERROR("Out of memory\n");
576 spin_lock_init(&lmv->lmv_lock);
578 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
579 tgts->uuid = uuids[i];
581 lmv->max_cookiesize = 0;
583 lmv->max_easize = sizeof(struct lustre_id) *
584 desc->ld_tgt_count + sizeof(struct mea);
586 rc = lmv_setup_mgr(obd);
588 CERROR("Can't setup LMV object manager, "
590 OBD_FREE(lmv->tgts, lmv->tgts_size);
593 tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
596 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
603 static int lmv_cleanup(struct obd_device *obd, int flags)
605 struct lmv_obd *lmv = &obd->u.lmv;
608 lmv_cleanup_mgr(obd);
609 OBD_FREE(lmv->tgts, lmv->tgts_size);
614 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
615 unsigned long max_age)
617 struct lmv_obd *lmv = &obd->u.lmv;
618 struct obd_statfs temp;
622 rc = lmv_check_connect(obd);
626 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
627 if (lmv->tgts[i].ltd_exp == NULL) {
628 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
632 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
634 CERROR("can't stat MDS #%d (%s)\n", i,
635 lmv->tgts[i].ltd_exp->exp_obd->obd_name);
639 memcpy(osfs, &temp, sizeof(temp));
641 osfs->os_bavail += temp.os_bavail;
642 osfs->os_blocks += temp.os_blocks;
643 osfs->os_ffree += temp.os_ffree;
644 osfs->os_files += temp.os_files;
650 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
652 struct obd_device *obd = exp->exp_obd;
653 struct lmv_obd *lmv = &obd->u.lmv;
657 rc = lmv_check_connect(obd);
661 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
667 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
668 unsigned long valid, unsigned int ea_size,
669 struct ptlrpc_request **request)
671 struct obd_device *obd = exp->exp_obd;
672 struct lmv_obd *lmv = &obd->u.lmv;
673 int rc, i = id_group(id);
677 rc = lmv_check_connect(obd);
681 LASSERT(i < lmv->desc.ld_tgt_count);
683 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
688 obj = lmv_grab_obj(obd, id);
690 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
691 OLID4(id), obj ? "(splitted)" : "");
693 /* if object is splitted, then we loop over all the slaves and gather
694 * size attribute. In ideal world we would have to gather also mds field
695 * from all slaves, as object is spread over the cluster and this is
696 * definitely interesting information and it is not good to loss it,
699 struct mds_body *body;
701 if (*request == NULL) {
706 body = lustre_msg_buf((*request)->rq_repmsg, 0,
708 LASSERT(body != NULL);
712 for (i = 0; i < obj->objcount; i++) {
714 if (lmv->tgts[i].ltd_exp == NULL) {
715 CWARN("%s: NULL export for %d\n",
720 /* skip master obj. */
721 if (id_equal_fid(&obj->id, &obj->objs[i].id))
724 body->size += obj->objs[i].size;
734 static int lmv_change_cbdata(struct obd_export *exp,
735 struct lustre_id *id,
739 struct obd_device *obd = exp->exp_obd;
740 struct lmv_obd *lmv = &obd->u.lmv;
744 rc = lmv_check_connect(obd);
748 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
749 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
751 rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
757 static int lmv_change_cbdata_name(struct obd_export *exp,
758 struct lustre_id *pid,
760 struct lustre_id *cid,
764 struct obd_device *obd = exp->exp_obd;
765 struct lmv_obd *lmv = &obd->u.lmv;
766 struct lustre_id rcid = *cid;
771 rc = lmv_check_connect(obd);
775 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
776 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
778 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
779 OLID4(pid), len, name, OLID4(cid));
781 /* this is default mds for directory name belongs to. */
783 obj = lmv_grab_obj(obd, pid);
785 /* directory is splitted. look for right mds for this name. */
786 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
787 rcid = obj->objs[mds].id;
788 mds = id_group(&rcid);
791 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
795 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
797 struct obd_device *obd = exp->exp_obd;
798 struct lmv_obd *lmv = &obd->u.lmv;
802 rc = lmv_check_connect(obd);
806 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
807 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
808 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
812 int lmv_close(struct obd_export *exp, struct obdo *obdo,
813 struct obd_client_handle *och,
814 struct ptlrpc_request **request)
816 struct obd_device *obd = exp->exp_obd;
817 struct lmv_obd *lmv = &obd->u.lmv;
818 int rc, i = obdo->o_mds;
821 rc = lmv_check_connect(obd);
825 LASSERT(i < lmv->desc.ld_tgt_count);
826 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
827 (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
828 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
832 int lmv_get_mea_and_update_object(struct obd_export *exp,
833 struct lustre_id *id)
835 struct obd_device *obd = exp->exp_obd;
836 struct lmv_obd *lmv = &obd->u.lmv;
837 struct ptlrpc_request *req = NULL;
844 mealen = MEA_SIZE_LMV(lmv);
846 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
848 /* time to update mea of parent id */
849 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
850 id, valid, mealen, &req);
852 CERROR("md_getattr() failed, error %d\n", rc);
856 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
858 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
863 GOTO(cleanup, rc = -ENODATA);
865 obj = lmv_create_obj(exp, id, md.mea);
870 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
874 ptlrpc_req_finished(req);
878 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
879 const void *data, int datalen, int mode, __u32 uid,
880 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
882 struct obd_device *obd = exp->exp_obd;
883 struct lmv_obd *lmv = &obd->u.lmv;
884 struct mds_body *body;
886 int rc, mds, loop = 0;
889 rc = lmv_check_connect(obd);
893 if (!lmv->desc.ld_active_tgt_count)
896 LASSERT(++loop <= 2);
897 obj = lmv_grab_obj(obd, &op_data->id1);
899 mds = raw_name2idx(obj->hashtype, obj->objcount,
900 op_data->name, op_data->namelen);
901 op_data->id1 = obj->objs[mds].id;
905 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
906 op_data->name, OLID4(&op_data->id1));
908 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
909 op_data, data, datalen, mode, uid, gid, rdev, request);
911 if (*request == NULL)
914 body = lustre_msg_buf((*request)->rq_repmsg, 0,
916 LASSERT(body != NULL);
918 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
920 /* LASSERT(body->valid & OBD_MD_MDS ||
921 body->mds == id_group(&op_data->id1));*/
922 } else if (rc == -ERESTART) {
923 /* directory got splitted. time to update local object and
924 * repeat the request with proper MDS */
925 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
927 ptlrpc_req_finished(*request);
934 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
936 struct obd_device *obd = exp->exp_obd;
937 struct lmv_obd *lmv = &obd->u.lmv;
940 rc = lmv_check_connect(obd);
944 /* FIXME: choose right MDC here */
945 CWARN("this method isn't implemented yet\n");
946 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
950 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
951 struct lookup_intent *it, int lockmode,
952 struct mdc_op_data *data, struct lustre_handle *lockh,
953 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
954 ldlm_blocking_callback cb_blocking, void *cb_data)
956 struct obd_device *obd = exp->exp_obd;
957 struct lmv_obd *lmv = &obd->u.lmv;
958 struct mea *mea = data->mea1;
959 struct mdc_op_data data2;
963 LASSERT(mea != NULL);
964 for (i = 0; i < mea->mea_count; i++) {
965 memset(&data2, 0, sizeof(data2));
966 data2.id1 = mea->mea_ids[i];
967 mds = id_group(&data2.id1);
969 if (lmv->tgts[mds].ltd_exp == NULL)
972 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
973 lockmode, &data2, lockh + i, lmm, lmmsize,
974 cb_compl, cb_blocking, cb_data);
976 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
977 OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
980 if (it->d.lustre.it_data) {
981 struct ptlrpc_request *req;
982 req = (struct ptlrpc_request *) it->d.lustre.it_data;
983 ptlrpc_req_finished(req);
986 if (it->d.lustre.it_status)
987 GOTO(cleanup, rc = it->d.lustre.it_status);
992 /* drop all taken locks */
995 ldlm_lock_decref(lockh + i, lockmode);
1001 int lmv_enqueue(struct obd_export *exp, int lock_type,
1002 struct lookup_intent *it, int lock_mode,
1003 struct mdc_op_data *data, struct lustre_handle *lockh,
1004 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1005 ldlm_blocking_callback cb_blocking, void *cb_data)
1007 struct obd_device *obd = exp->exp_obd;
1008 struct lmv_obd *lmv = &obd->u.lmv;
1009 struct lmv_obj *obj;
1013 rc = lmv_check_connect(obd);
1017 if (data->mea1 && it->it_op == IT_UNLINK) {
1018 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1019 data, lockh, lmm, lmmsize,
1020 cb_compl, cb_blocking, cb_data);
1024 if (data->namelen) {
1025 obj = lmv_grab_obj(obd, &data->id1);
1027 /* directory is splitted. look for right mds for this
1029 mds = raw_name2idx(obj->hashtype, obj->objcount,
1030 (char *)data->name, data->namelen);
1031 data->id1 = obj->objs[mds].id;
1035 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1038 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1039 lock_type, it, lock_mode, data, lockh, lmm,
1040 lmmsize, cb_compl, cb_blocking, cb_data);
1044 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1045 char *filename, int namelen, unsigned long valid,
1046 unsigned int ea_size, struct ptlrpc_request **request)
1048 int rc, mds = id_group(id), loop = 0;
1049 struct obd_device *obd = exp->exp_obd;
1050 struct lmv_obd *lmv = &obd->u.lmv;
1051 struct lustre_id rid = *id;
1052 struct mds_body *body;
1053 struct lmv_obj *obj;
1057 rc = lmv_check_connect(obd);
1061 LASSERT(++loop <= 2);
1062 obj = lmv_grab_obj(obd, id);
1064 /* directory is splitted. look for right mds for this name */
1065 mds = raw_name2idx(obj->hashtype, obj->objcount,
1066 filename, namelen - 1);
1067 rid = obj->objs[mds].id;
1071 CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1072 namelen, filename, OLID4(id), OLID4(&rid));
1077 * here should be applied OBD_MD_FID to ->valid, because otherwise,
1078 * mds_getattr_lock() will not fetch fid component of lustre_id and
1079 * thus, next call to md_getattr_lock() will be performed to wrong mds.
1081 if (!(old_valid & OBD_MD_FID))
1082 valid |= OBD_MD_FID;
1084 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1085 &rid, filename, namelen, valid,
1089 * this could be cross-node reference. in this case all we have
1090 * right now is lustre_id triple. we'd like to find other
1093 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1094 LASSERT(body != NULL);
1095 if (body->valid & OBD_MD_MDS) {
1096 struct ptlrpc_request *req = NULL;
1099 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1102 * turning OBD_MD_FID fetching off, as we already have
1103 * full lustre_id and do need to fetch fid component
1104 * again. This will help to make thing slightly faster.
1106 if (!(old_valid & OBD_MD_FID))
1107 valid &= ~OBD_MD_FID;
1109 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1110 &rid, NULL, 1, valid, ea_size, &req);
1111 ptlrpc_req_finished(*request);
1114 } else if (rc == -ERESTART) {
1115 /* directory got splitted. time to update local object and
1116 * repeat the request with proper MDS */
1117 rc = lmv_get_mea_and_update_object(exp, &rid);
1119 ptlrpc_req_finished(*request);
1127 * llite passes id of an target inode in data->id1 and id of directory in
1130 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1131 struct ptlrpc_request **request)
1133 struct obd_device *obd = exp->exp_obd;
1134 struct lmv_obd *lmv = &obd->u.lmv;
1135 struct lmv_obj *obj;
1139 rc = lmv_check_connect(obd);
1143 if (data->namelen != 0) {
1144 /* usual link request */
1145 obj = lmv_grab_obj(obd, &data->id1);
1147 rc = raw_name2idx(obj->hashtype, obj->objcount,
1148 data->name, data->namelen);
1149 data->id1 = obj->objs[rc].id;
1153 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1154 OLID4(&data->id2), data->namelen, data->name,
1157 /* request from MDS to acquire i_links for inode by id1 */
1158 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1162 rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1167 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1168 const char *old, int oldlen, const char *new, int newlen,
1169 struct ptlrpc_request **request)
1171 struct obd_device *obd = exp->exp_obd;
1172 struct lmv_obd *lmv = &obd->u.lmv;
1173 struct lmv_obj *obj;
1177 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1178 oldlen, old, OLID4(&data->id1), newlen, new, OLID4(&data->id2));
1180 rc = lmv_check_connect(obd);
1186 * MDS with old dir entry is asking another MDS to create name
1190 "create %*s(%d/%d) in "DLID4" pointing "
1191 "to "DLID4"\n", newlen, new, oldlen, newlen,
1192 OLID4(&data->id2), OLID4(&data->id1));
1194 mds = id_group(&data->id2);
1197 * target directory can be splitted, sowe should forward request
1200 obj = lmv_grab_obj(obd, &data->id2);
1202 mds = raw_name2idx(obj->hashtype, obj->objcount,
1203 (char *)new, newlen);
1204 data->id2 = obj->objs[mds].id;
1205 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1212 obj = lmv_grab_obj(obd, &data->id1);
1215 * directory is already splitted, so we have to forward request
1218 mds = raw_name2idx(obj->hashtype, obj->objcount,
1219 (char *)old, oldlen);
1220 data->id1 = obj->objs[mds].id;
1221 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1226 obj = lmv_grab_obj(obd, &data->id2);
1229 * directory is already splitted, so we have to forward request
1232 mds = raw_name2idx(obj->hashtype, obj->objcount,
1233 (char *)new, newlen);
1235 data->id2 = obj->objs[mds].id;
1236 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1241 mds = id_group(&data->id1);
1244 if (id_group(&data->id1) != id_group(&data->id2)) {
1245 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1246 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1250 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1251 new, newlen, request);
1255 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1256 struct iattr *iattr, void *ea, int ealen, void *ea2,
1257 int ea2len, struct ptlrpc_request **request)
1259 struct obd_device *obd = exp->exp_obd;
1260 struct lmv_obd *lmv = &obd->u.lmv;
1261 struct ptlrpc_request *req;
1262 struct mds_body *body;
1263 struct lmv_obj *obj;
1267 rc = lmv_check_connect(obd);
1271 obj = lmv_grab_obj(obd, &data->id1);
1273 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1274 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1277 for (i = 0; i < obj->objcount; i++) {
1278 data->id1 = obj->objs[i].id;
1280 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1281 data, iattr, ea, ealen, ea2, ea2len, &req);
1283 if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1285 * this is master object and this request should
1286 * be returned back to llite.
1290 ptlrpc_req_finished(req);
1298 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1299 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1300 data, iattr, ea, ealen, ea2, ea2len, request);
1302 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1304 LASSERT(body != NULL);
1305 LASSERT(id_group(&body->id1) == id_group(&data->id1));
1311 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1312 struct ptlrpc_request **request)
1314 struct obd_device *obd = exp->exp_obd;
1315 struct lmv_obd *lmv = &obd->u.lmv;
1319 rc = lmv_check_connect(obd);
1323 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1328 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1329 struct ldlm_lock_desc *desc,
1330 void *data, int flag)
1332 struct lustre_handle lockh;
1333 struct lmv_obj *obj;
1338 case LDLM_CB_BLOCKING:
1339 ldlm_lock2handle(lock, &lockh);
1340 rc = ldlm_cli_cancel(&lockh);
1342 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1346 case LDLM_CB_CANCELING:
1347 /* time to drop cached attrs for dirobj */
1348 obj = lock->l_ast_data;
1350 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1351 ", master "DLID4"\n",
1352 lock->l_resource->lr_name.name[3] == 1 ?
1353 "LOOKUP" : "UPDATE",
1354 lock->l_resource->lr_name.name[0],
1355 lock->l_resource->lr_name.name[1],
1366 void lmv_remove_dots(struct page *page)
1368 char *kaddr = page_address(page);
1369 unsigned limit = PAGE_CACHE_SIZE;
1370 unsigned offs, rec_len;
1371 struct ext2_dir_entry_2 *p;
1373 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1374 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1375 rec_len = le16_to_cpu(p->rec_len);
1377 if ((p->name_len == 1 && p->name[0] == '.') ||
1378 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1383 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1384 __u64 offset, struct page *page,
1385 struct ptlrpc_request **request)
1387 struct obd_device *obd = exp->exp_obd;
1388 struct lmv_obd *lmv = &obd->u.lmv;
1389 struct lustre_id rid = *id;
1390 struct lmv_obj *obj;
1394 #warning "we need well-desgined readdir() implementation"
1395 rc = lmv_check_connect(obd);
1399 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1400 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1401 offset, OLID4(&rid));
1403 obj = lmv_grab_obj(obd, id);
1407 /* find dirobj containing page with requested offset. */
1408 for (i = 0; i < obj->objcount; i++) {
1409 if (offset < obj->objs[i].size)
1411 offset -= obj->objs[i].size;
1413 rid = obj->objs[i].id;
1415 lmv_unlock_obj(obj);
1418 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1419 OLID4(&rid), (unsigned long)offset);
1421 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1422 offset, page, request);
1424 if (rc == 0 && !id_equal_fid(&rid, id))
1425 /* this page isn't from master object. To avoid "." and ".."
1426 * duplication in directory, we have to remove them from all
1428 lmv_remove_dots(page);
1433 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1434 struct ptlrpc_request **req)
1436 struct obd_device *obd = exp->exp_obd;
1437 struct lmv_obd *lmv = &obd->u.lmv;
1438 struct mea *mea = data->mea1;
1439 struct mdc_op_data data2;
1443 LASSERT(mea != NULL);
1444 for (i = 0; i < mea->mea_count; i++) {
1445 memset(&data2, 0, sizeof(data2));
1446 data2.id1 = mea->mea_ids[i];
1447 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1449 mds = id_group(&data2.id1);
1451 if (lmv->tgts[mds].ltd_exp == NULL)
1454 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1455 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1456 OLID4(&mea->mea_ids[i]), rc);
1458 ptlrpc_req_finished(*req);
1467 int lmv_put_inode(struct obd_export *exp, struct lustre_id *id)
1470 lmv_delete_obj(exp, id);
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_put_inode = lmv_put_inode,
1945 int __init lmv_init(void)
1947 struct lprocfs_static_vars lvars;
1950 obj_cache = kmem_cache_create("lmv_objects",
1951 sizeof(struct lmv_obj),
1954 CERROR("error allocating lmv objects cache\n");
1958 lprocfs_init_vars(lmv, &lvars);
1959 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1961 OBD_LMV_DEVICENAME);
1966 static void lmv_exit(void)
1968 class_unregister_type(OBD_LMV_DEVICENAME);
1970 LASSERTF(kmem_cache_destroy(obj_cache) == 0,
1971 "can't free lmv objects cache, %d object(s)"
1972 "still in use\n", atomic_read(&obj_cache_count));
1975 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1976 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1977 MODULE_LICENSE("GPL");
1979 module_init(lmv_init);
1980 module_exit(lmv_exit);