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, unsigned long 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 MDT 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), error %d\n", i,
635 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
640 memcpy(osfs, &temp, sizeof(temp));
642 osfs->os_bavail += temp.os_bavail;
643 osfs->os_blocks += temp.os_blocks;
644 osfs->os_ffree += temp.os_ffree;
645 osfs->os_files += temp.os_files;
651 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
653 struct obd_device *obd = exp->exp_obd;
654 struct lmv_obd *lmv = &obd->u.lmv;
658 rc = lmv_check_connect(obd);
662 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
668 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
669 __u64 valid, unsigned int ea_size,
670 struct ptlrpc_request **request)
672 struct obd_device *obd = exp->exp_obd;
673 struct lmv_obd *lmv = &obd->u.lmv;
674 int rc, i = id_group(id);
678 rc = lmv_check_connect(obd);
682 LASSERT(i < lmv->desc.ld_tgt_count);
684 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
689 obj = lmv_grab_obj(obd, id);
691 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
692 OLID4(id), obj ? "(splitted)" : "");
695 * if object is splitted, then we loop over all the slaves and gather
696 * size attribute. In ideal world we would have to gather also mds field
697 * from all slaves, as object is spread over the cluster and this is
698 * definitely interesting information and it is not good to loss it,
702 struct mds_body *body;
704 if (*request == NULL) {
709 body = lustre_msg_buf((*request)->rq_repmsg, 0,
711 LASSERT(body != NULL);
715 for (i = 0; i < obj->objcount; i++) {
717 if (lmv->tgts[i].ltd_exp == NULL) {
718 CWARN("%s: NULL export for %d\n",
723 /* skip master obj. */
724 if (id_equal_fid(&obj->id, &obj->objs[i].id))
727 body->size += obj->objs[i].size;
737 static int lmv_change_cbdata(struct obd_export *exp,
738 struct lustre_id *id,
742 struct obd_device *obd = exp->exp_obd;
743 struct lmv_obd *lmv = &obd->u.lmv;
747 rc = lmv_check_connect(obd);
751 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
752 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
754 rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
760 static int lmv_change_cbdata_name(struct obd_export *exp,
761 struct lustre_id *pid,
763 struct lustre_id *cid,
767 struct obd_device *obd = exp->exp_obd;
768 struct lmv_obd *lmv = &obd->u.lmv;
769 struct lustre_id rcid = *cid;
774 rc = lmv_check_connect(obd);
778 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
779 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
781 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
782 OLID4(pid), len, name, OLID4(cid));
784 /* this is default mds for directory name belongs to. */
786 obj = lmv_grab_obj(obd, pid);
788 /* directory is splitted. look for right mds for this name. */
789 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
790 rcid = obj->objs[mds].id;
791 mds = id_group(&rcid);
794 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
798 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
800 struct obd_device *obd = exp->exp_obd;
801 struct lmv_obd *lmv = &obd->u.lmv;
805 rc = lmv_check_connect(obd);
809 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
810 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
811 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
815 int lmv_close(struct obd_export *exp, struct obdo *obdo,
816 struct obd_client_handle *och,
817 struct ptlrpc_request **request)
819 struct obd_device *obd = exp->exp_obd;
820 struct lmv_obd *lmv = &obd->u.lmv;
821 int rc, i = obdo->o_mds;
824 rc = lmv_check_connect(obd);
828 LASSERT(i < lmv->desc.ld_tgt_count);
829 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
830 (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
831 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
835 int lmv_get_mea_and_update_object(struct obd_export *exp,
836 struct lustre_id *id)
838 struct obd_device *obd = exp->exp_obd;
839 struct lmv_obd *lmv = &obd->u.lmv;
840 struct ptlrpc_request *req = NULL;
847 mealen = MEA_SIZE_LMV(lmv);
849 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
851 /* time to update mea of parent id */
852 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
853 id, valid, mealen, &req);
855 CERROR("md_getattr() failed, error %d\n", rc);
859 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
861 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
866 GOTO(cleanup, rc = -ENODATA);
868 obj = lmv_create_obj(exp, id, md.mea);
873 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
877 ptlrpc_req_finished(req);
881 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
882 const void *data, int datalen, int mode, __u32 uid,
883 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
885 struct obd_device *obd = exp->exp_obd;
886 struct lmv_obd *lmv = &obd->u.lmv;
887 struct mds_body *body;
889 int rc, mds, loop = 0;
892 rc = lmv_check_connect(obd);
896 if (!lmv->desc.ld_active_tgt_count)
899 LASSERT(++loop <= 2);
900 obj = lmv_grab_obj(obd, &op_data->id1);
902 mds = raw_name2idx(obj->hashtype, obj->objcount,
903 op_data->name, op_data->namelen);
904 op_data->id1 = obj->objs[mds].id;
908 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
909 op_data->name, OLID4(&op_data->id1));
911 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
912 op_data, data, datalen, mode, uid, gid, rdev,
915 if (*request == NULL)
918 body = lustre_msg_buf((*request)->rq_repmsg, 0,
923 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
924 } else if (rc == -ERESTART) {
926 * directory got splitted. time to update local object and
927 * repeat the request with proper MDS.
929 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
931 ptlrpc_req_finished(*request);
938 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
940 struct obd_device *obd = exp->exp_obd;
941 struct lmv_obd *lmv = &obd->u.lmv;
944 rc = lmv_check_connect(obd);
948 /* FIXME: choose right MDC here */
949 CWARN("this method isn't implemented yet\n");
950 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
954 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
955 struct lookup_intent *it, int lockmode,
956 struct mdc_op_data *data, struct lustre_handle *lockh,
957 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
958 ldlm_blocking_callback cb_blocking, void *cb_data)
960 struct obd_device *obd = exp->exp_obd;
961 struct lmv_obd *lmv = &obd->u.lmv;
962 struct mea *mea = data->mea1;
963 struct mdc_op_data data2;
967 LASSERT(mea != NULL);
968 for (i = 0; i < mea->mea_count; i++) {
969 memset(&data2, 0, sizeof(data2));
970 data2.id1 = mea->mea_ids[i];
971 mds = id_group(&data2.id1);
973 if (lmv->tgts[mds].ltd_exp == NULL)
976 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
977 lockmode, &data2, lockh + i, lmm, lmmsize,
978 cb_compl, cb_blocking, cb_data);
980 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
981 OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
984 if (it->d.lustre.it_data) {
985 struct ptlrpc_request *req;
986 req = (struct ptlrpc_request *) it->d.lustre.it_data;
987 ptlrpc_req_finished(req);
990 if (it->d.lustre.it_status)
991 GOTO(cleanup, rc = it->d.lustre.it_status);
996 /* drop all taken locks */
999 ldlm_lock_decref(lockh + i, lockmode);
1000 lockh[i].cookie = 0;
1005 int lmv_enqueue(struct obd_export *exp, int lock_type,
1006 struct lookup_intent *it, int lock_mode,
1007 struct mdc_op_data *data, struct lustre_handle *lockh,
1008 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1009 ldlm_blocking_callback cb_blocking, void *cb_data)
1011 struct obd_device *obd = exp->exp_obd;
1012 struct lmv_obd *lmv = &obd->u.lmv;
1013 struct lmv_obj *obj;
1017 rc = lmv_check_connect(obd);
1021 if (data->mea1 && it->it_op == IT_UNLINK) {
1022 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1023 data, lockh, lmm, lmmsize,
1024 cb_compl, cb_blocking, cb_data);
1028 if (data->namelen) {
1029 obj = lmv_grab_obj(obd, &data->id1);
1031 /* directory is splitted. look for right mds for this
1033 mds = raw_name2idx(obj->hashtype, obj->objcount,
1034 (char *)data->name, data->namelen);
1035 data->id1 = obj->objs[mds].id;
1039 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1042 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1043 lock_type, it, lock_mode, data, lockh, lmm,
1044 lmmsize, cb_compl, cb_blocking, cb_data);
1048 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1049 char *filename, int namelen, __u64 valid,
1050 unsigned int ea_size, struct ptlrpc_request **request)
1052 int rc, mds = id_group(id), loop = 0;
1053 struct obd_device *obd = exp->exp_obd;
1054 struct lmv_obd *lmv = &obd->u.lmv;
1055 struct lustre_id rid = *id;
1056 struct mds_body *body;
1057 struct lmv_obj *obj;
1061 rc = lmv_check_connect(obd);
1065 LASSERT(++loop <= 2);
1066 obj = lmv_grab_obj(obd, id);
1068 /* directory is splitted. look for right mds for this name */
1069 mds = raw_name2idx(obj->hashtype, obj->objcount,
1070 filename, namelen - 1);
1071 rid = obj->objs[mds].id;
1075 CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1076 namelen, filename, OLID4(id), OLID4(&rid));
1081 * here should be applied OBD_MD_FID to ->valid, because otherwise,
1082 * mds_getattr_lock() will not fetch fid component of lustre_id and
1083 * thus, next call to md_getattr_lock() will be performed to wrong mds.
1085 if (!(old_valid & OBD_MD_FID))
1086 valid |= OBD_MD_FID;
1088 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1089 &rid, filename, namelen, valid,
1093 * this could be cross-node reference. in this case all we have
1094 * right now is lustre_id triple. we'd like to find other
1097 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1098 LASSERT(body != NULL);
1099 if (body->valid & OBD_MD_MDS) {
1100 struct ptlrpc_request *req = NULL;
1103 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1106 * turning OBD_MD_FID fetching off, as we already have
1107 * full lustre_id and do need to fetch fid component
1108 * again. This will help to make thing slightly faster.
1110 if (!(old_valid & OBD_MD_FID))
1111 valid &= ~OBD_MD_FID;
1113 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1114 &rid, NULL, 1, valid, ea_size, &req);
1115 ptlrpc_req_finished(*request);
1118 } else if (rc == -ERESTART) {
1119 /* directory got splitted. time to update local object and
1120 * repeat the request with proper MDS */
1121 rc = lmv_get_mea_and_update_object(exp, &rid);
1123 ptlrpc_req_finished(*request);
1131 * llite passes id of an target inode in data->id1 and id of directory in
1134 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1135 struct ptlrpc_request **request)
1137 struct obd_device *obd = exp->exp_obd;
1138 struct lmv_obd *lmv = &obd->u.lmv;
1139 struct lmv_obj *obj;
1143 rc = lmv_check_connect(obd);
1147 if (data->namelen != 0) {
1148 /* usual link request */
1149 obj = lmv_grab_obj(obd, &data->id1);
1151 rc = raw_name2idx(obj->hashtype, obj->objcount,
1152 data->name, data->namelen);
1153 data->id1 = obj->objs[rc].id;
1157 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1158 OLID4(&data->id2), data->namelen, data->name,
1161 /* request from MDS to acquire i_links for inode by id1 */
1162 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1166 rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1171 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1172 const char *old, int oldlen, const char *new, int newlen,
1173 struct ptlrpc_request **request)
1175 struct obd_device *obd = exp->exp_obd;
1176 struct lmv_obd *lmv = &obd->u.lmv;
1177 struct lmv_obj *obj;
1181 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1182 oldlen, old, OLID4(&data->id1), newlen, new, OLID4(&data->id2));
1184 rc = lmv_check_connect(obd);
1190 * MDS with old dir entry is asking another MDS to create name
1194 "create %*s(%d/%d) in "DLID4" pointing "
1195 "to "DLID4"\n", newlen, new, oldlen, newlen,
1196 OLID4(&data->id2), OLID4(&data->id1));
1198 mds = id_group(&data->id2);
1201 * target directory can be splitted, sowe should forward request
1204 obj = lmv_grab_obj(obd, &data->id2);
1206 mds = raw_name2idx(obj->hashtype, obj->objcount,
1207 (char *)new, newlen);
1208 data->id2 = obj->objs[mds].id;
1209 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1216 obj = lmv_grab_obj(obd, &data->id1);
1219 * directory is already splitted, so we have to forward request
1222 mds = raw_name2idx(obj->hashtype, obj->objcount,
1223 (char *)old, oldlen);
1224 data->id1 = obj->objs[mds].id;
1225 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1230 obj = lmv_grab_obj(obd, &data->id2);
1233 * directory is already splitted, so we have to forward request
1236 mds = raw_name2idx(obj->hashtype, obj->objcount,
1237 (char *)new, newlen);
1239 data->id2 = obj->objs[mds].id;
1240 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1245 mds = id_group(&data->id1);
1248 if (id_group(&data->id1) != id_group(&data->id2)) {
1249 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1250 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1254 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1255 new, newlen, request);
1259 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1260 struct iattr *iattr, void *ea, int ealen, void *ea2,
1261 int ea2len, struct ptlrpc_request **request)
1263 struct obd_device *obd = exp->exp_obd;
1264 struct lmv_obd *lmv = &obd->u.lmv;
1265 struct ptlrpc_request *req;
1266 struct mds_body *body;
1267 struct lmv_obj *obj;
1271 rc = lmv_check_connect(obd);
1275 obj = lmv_grab_obj(obd, &data->id1);
1277 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1278 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1281 for (i = 0; i < obj->objcount; i++) {
1282 data->id1 = obj->objs[i].id;
1284 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1285 data, iattr, ea, ealen, ea2, ea2len, &req);
1287 if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1289 * this is master object and this request should
1290 * be returned back to llite.
1294 ptlrpc_req_finished(req);
1302 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1303 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1304 data, iattr, ea, ealen, ea2, ea2len, request);
1306 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1308 LASSERT(body != NULL);
1309 LASSERT(id_group(&body->id1) == id_group(&data->id1));
1315 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1316 struct ptlrpc_request **request)
1318 struct obd_device *obd = exp->exp_obd;
1319 struct lmv_obd *lmv = &obd->u.lmv;
1323 rc = lmv_check_connect(obd);
1327 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1332 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1333 struct ldlm_lock_desc *desc,
1334 void *data, int flag)
1336 struct lustre_handle lockh;
1337 struct lmv_obj *obj;
1342 case LDLM_CB_BLOCKING:
1343 ldlm_lock2handle(lock, &lockh);
1344 rc = ldlm_cli_cancel(&lockh);
1346 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1350 case LDLM_CB_CANCELING:
1351 /* time to drop cached attrs for dirobj */
1352 obj = lock->l_ast_data;
1354 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1355 ", master "DLID4"\n",
1356 lock->l_resource->lr_name.name[3] == 1 ?
1357 "LOOKUP" : "UPDATE",
1358 lock->l_resource->lr_name.name[0],
1359 lock->l_resource->lr_name.name[1],
1370 void lmv_remove_dots(struct page *page)
1372 char *kaddr = page_address(page);
1373 unsigned limit = PAGE_CACHE_SIZE;
1374 unsigned offs, rec_len;
1375 struct ext2_dir_entry_2 *p;
1377 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1378 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1379 rec_len = le16_to_cpu(p->rec_len);
1381 if ((p->name_len == 1 && p->name[0] == '.') ||
1382 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1387 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1388 __u64 offset, struct page *page,
1389 struct ptlrpc_request **request)
1391 struct obd_device *obd = exp->exp_obd;
1392 struct lmv_obd *lmv = &obd->u.lmv;
1393 struct lustre_id rid = *id;
1394 struct lmv_obj *obj;
1398 #warning "we need well-desgined readdir() implementation"
1399 rc = lmv_check_connect(obd);
1403 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1404 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1405 offset, OLID4(&rid));
1407 obj = lmv_grab_obj(obd, id);
1411 /* find dirobj containing page with requested offset. */
1412 for (i = 0; i < obj->objcount; i++) {
1413 if (offset < obj->objs[i].size)
1415 offset -= obj->objs[i].size;
1417 rid = obj->objs[i].id;
1419 lmv_unlock_obj(obj);
1422 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1423 OLID4(&rid), (unsigned long)offset);
1425 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1426 offset, page, request);
1428 if (rc == 0 && !id_equal_fid(&rid, id))
1429 /* this page isn't from master object. To avoid "." and ".."
1430 * duplication in directory, we have to remove them from all
1432 lmv_remove_dots(page);
1437 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1438 struct ptlrpc_request **req)
1440 struct obd_device *obd = exp->exp_obd;
1441 struct lmv_obd *lmv = &obd->u.lmv;
1442 struct mea *mea = data->mea1;
1443 struct mdc_op_data data2;
1447 LASSERT(mea != NULL);
1448 for (i = 0; i < mea->mea_count; i++) {
1449 memset(&data2, 0, sizeof(data2));
1450 data2.id1 = mea->mea_ids[i];
1451 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1453 if (lmv->tgts[id_group(&data2.id1)].ltd_exp == NULL)
1456 rc = md_unlink(lmv->tgts[id_group(&data2.id1)].ltd_exp,
1459 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1460 OLID4(&mea->mea_ids[i]), rc);
1463 ptlrpc_req_finished(*req);
1472 int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1477 if (lmv_delete_obj(exp, id)) {
1478 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1484 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1485 struct ptlrpc_request **request)
1487 struct obd_device *obd = exp->exp_obd;
1488 struct lmv_obd *lmv = &obd->u.lmv;
1492 rc = lmv_check_connect(obd);
1496 if (data->namelen == 0 && data->mea1 != NULL) {
1497 /* mds asks to remove slave objects */
1498 rc = lmv_unlink_slaves(exp, data, request);
1502 if (data->namelen != 0) {
1503 struct lmv_obj *obj;
1505 obj = lmv_grab_obj(obd, &data->id1);
1507 i = raw_name2idx(obj->hashtype, obj->objcount,
1508 data->name, data->namelen);
1509 data->id1 = obj->objs[i].id;
1512 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1513 data->namelen, data->name, OLID4(&data->id1),
1516 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1519 rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1524 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1525 char *name, int len)
1527 struct obd_device *obd = exp->exp_obd;
1528 struct lmv_obd *lmv = &obd->u.lmv;
1532 rc = lmv_check_connect(obd);
1534 RETURN(ERR_PTR(rc));
1535 obd = lmv->tgts[0].ltd_exp->exp_obd;
1541 int lmv_init_ea_size(struct obd_export *exp, int easize,
1544 struct obd_device *obd = exp->exp_obd;
1545 struct lmv_obd *lmv = &obd->u.lmv;
1546 int i, rc = 0, change = 0;
1549 if (lmv->max_easize < easize) {
1550 lmv->max_easize = easize;
1553 if (lmv->max_cookiesize < cookiesize) {
1554 lmv->max_cookiesize = cookiesize;
1560 if (lmv->connected == 0)
1563 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1564 if (lmv->tgts[i].ltd_exp == NULL) {
1565 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1569 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1571 CERROR("obd_init_ea_size() failed on MDT target %d, "
1572 "error %d.\n", i, rc);
1579 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1580 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1582 struct obd_device *obd = exp->exp_obd;
1583 struct lmv_obd *lmv = &obd->u.lmv;
1584 struct lov_stripe_md obj_md;
1585 struct lov_stripe_md *obj_mdp = &obj_md;
1589 LASSERT(ea == NULL);
1590 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1592 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
1598 int lmv_getready(struct obd_export *exp)
1600 struct obd_device *obd = exp->exp_obd;
1604 rc = lmv_check_connect(obd);
1609 * to be called from MDS only. @oa should have correct store cookie and o_fid
1610 * values for "master" object, as it will be used.
1612 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1613 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1615 struct obd_device *obd = exp->exp_obd;
1616 struct lmv_obd *lmv = &obd->u.lmv;
1617 struct lustre_id mid;
1622 rc = lmv_check_connect(obd);
1626 LASSERT(oa != NULL);
1629 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1631 CERROR("Can't create object, rc = %d\n", rc);
1636 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1638 CERROR("obd_alloc_diskmd() failed, error %d\n",
1649 * here we should take care about splitted dir, so store cookie and fid
1650 * for "master" object should already be allocated and passed in @oa.
1652 LASSERT(oa->o_id != 0);
1653 LASSERT(oa->o_fid != 0);
1655 /* save "master" object id */
1658 mea = (struct mea *)*ea;
1659 mea->mea_master = -1;
1660 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1662 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1663 mea->mea_count = lmv->desc.ld_tgt_count;
1665 for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1666 struct lov_stripe_md obj_md;
1667 struct lov_stripe_md *obj_mdp = &obj_md;
1669 if (lmv->tgts[i].ltd_exp == NULL) {
1670 /* this is "master" MDS */
1671 mea->mea_master = i;
1672 mea->mea_ids[c] = mid;
1678 * "master" MDS should always be part of stripped dir,
1681 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1684 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1685 OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1687 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1689 CERROR("obd_create() failed on MDT target %d, "
1690 "error %d\n", c, rc);
1694 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1695 i, oa->o_id, oa->o_generation);
1699 * here, when object is created (or it is master and was passed
1700 * from caller) on desired MDS we save its fid to local mea_ids.
1705 * store cookie should be defined here for both cases (master
1706 * object and not master), because master is already created.
1710 /* fill mea by store cookie and fid */
1711 obdo2id(&mea->mea_ids[c], oa);
1714 LASSERT(c == mea->mea_count);
1716 CDEBUG(D_OTHER, "%d dirobjects created\n",
1717 (int)mea->mea_count);
1722 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1723 struct obd_device *tgt, int count,
1724 struct llog_catid *logid)
1726 struct llog_ctxt *ctxt;
1730 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1733 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1734 ctxt->loc_imp = tgt->u.cli.cl_import;
1740 static int lmv_llog_finish(struct obd_device *obd,
1741 struct obd_llogs *llogs, int count)
1746 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1750 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1751 void *key, __u32 *vallen, void *val)
1753 struct obd_device *obd;
1754 struct lmv_obd *lmv;
1758 obd = class_exp2obd(exp);
1760 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1761 exp->exp_handle.h_cookie);
1766 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1767 __u32 *mdsize = val;
1768 *vallen = sizeof(__u32);
1769 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1770 + sizeof(struct mea);
1772 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1773 struct obd_uuid *cluuid = &lmv->cluuid;
1774 struct lmv_tgt_desc *tgts;
1775 __u32 *mdsnum = val;
1778 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1779 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1780 *vallen = sizeof(__u32);
1786 } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1787 /* getting rootid from first MDS. */
1788 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1791 } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1792 struct lmv_desc *desc_ret = val;
1793 *desc_ret = lmv->desc;
1797 CDEBUG(D_IOCTL, "invalid key\n");
1801 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1802 void *key, obd_count vallen, void *val)
1804 struct obd_device *obd;
1805 struct lmv_obd *lmv;
1808 obd = class_exp2obd(exp);
1810 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1811 exp->exp_handle.h_cookie);
1816 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1817 lmv->server_timeout = 1;
1818 lmv_set_timeouts(obd);
1825 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1826 struct lov_stripe_md *lsm)
1828 struct obd_device *obd = class_exp2obd(exp);
1829 struct lmv_obd *lmv = &obd->u.lmv;
1830 struct mea *meap, *lsmp;
1834 mea_size = (sizeof(struct lustre_id) *
1835 lmv->desc.ld_tgt_count) + sizeof(struct mea);
1839 if (*lmmp && !lsm) {
1840 OBD_FREE(*lmmp, mea_size);
1845 if (*lmmp == NULL) {
1846 OBD_ALLOC(*lmmp, mea_size);
1854 lsmp = (struct mea *)lsm;
1855 meap = (struct mea *)*lmmp;
1857 meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
1858 meap->mea_count = cpu_to_le32(lsmp->mea_count);
1859 meap->mea_master = cpu_to_le32(lsmp->mea_master);
1861 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1862 meap->mea_ids[i] = meap->mea_ids[i];
1863 id_cpu_to_le(&meap->mea_ids[i]);
1869 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1870 struct lov_mds_md *disk_src, int mdsize)
1872 struct obd_device *obd = class_exp2obd(exp);
1873 struct mea **tmea = (struct mea **)mem_tgt;
1874 struct mea *mea = (struct mea *)disk_src;
1875 struct lmv_obd *lmv = &obd->u.lmv;
1879 mea_size = sizeof(struct lustre_id) *
1880 lmv->desc.ld_tgt_count + sizeof(struct mea);
1881 if (mem_tgt == NULL)
1884 if (*mem_tgt != NULL && disk_src == NULL) {
1885 OBD_FREE(*tmea, mea_size);
1889 LASSERT(mea_size == mdsize);
1891 OBD_ALLOC(*tmea, mea_size);
1898 (*tmea)->mea_magic = le32_to_cpu(mea->mea_magic);
1899 (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
1900 (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
1902 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1903 (*tmea)->mea_ids[i] = mea->mea_ids[i];
1904 id_le_to_cpu(&(*tmea)->mea_ids[i]);
1910 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1911 struct lov_stripe_md *ea, obd_count oa_bufs,
1912 struct brw_page *pgarr, struct obd_trans_info *oti)
1914 struct obd_device *obd = exp->exp_obd;
1915 struct lmv_obd *lmv = &obd->u.lmv;
1916 struct mea *mea = (struct mea *) ea;
1919 LASSERT(oa != NULL);
1920 LASSERT(ea != NULL);
1921 LASSERT(pgarr != NULL);
1922 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1924 oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1925 oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1926 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1928 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1929 oa, NULL, oa_bufs, pgarr, oti);
1933 struct obd_ops lmv_obd_ops = {
1934 .o_owner = THIS_MODULE,
1935 .o_attach = lmv_attach,
1936 .o_detach = lmv_detach,
1937 .o_setup = lmv_setup,
1938 .o_cleanup = lmv_cleanup,
1939 .o_connect = lmv_connect,
1940 .o_disconnect = lmv_disconnect,
1941 .o_statfs = lmv_statfs,
1942 .o_llog_init = lmv_llog_init,
1943 .o_llog_finish = lmv_llog_finish,
1944 .o_get_info = lmv_get_info,
1945 .o_set_info = lmv_set_info,
1946 .o_create = lmv_obd_create,
1947 .o_packmd = lmv_packmd,
1948 .o_unpackmd = lmv_unpackmd,
1950 .o_init_ea_size = lmv_init_ea_size,
1951 .o_notify = lmv_notify,
1952 .o_iocontrol = lmv_iocontrol,
1953 .o_getready = lmv_getready,
1956 struct md_ops lmv_md_ops = {
1957 .m_getstatus = lmv_getstatus,
1958 .m_getattr = lmv_getattr,
1959 .m_change_cbdata = lmv_change_cbdata,
1960 .m_change_cbdata_name = lmv_change_cbdata_name,
1961 .m_close = lmv_close,
1962 .m_create = lmv_create,
1963 .m_done_writing = lmv_done_writing,
1964 .m_enqueue = lmv_enqueue,
1965 .m_getattr_lock = lmv_getattr_lock,
1966 .m_intent_lock = lmv_intent_lock,
1968 .m_rename = lmv_rename,
1969 .m_setattr = lmv_setattr,
1971 .m_readpage = lmv_readpage,
1972 .m_unlink = lmv_unlink,
1973 .m_get_real_obd = lmv_get_real_obd,
1974 .m_valid_attrs = lmv_valid_attrs,
1975 .m_delete_inode = lmv_delete_inode,
1978 int __init lmv_init(void)
1980 struct lprocfs_static_vars lvars;
1983 obj_cache = kmem_cache_create("lmv_objects",
1984 sizeof(struct lmv_obj),
1987 CERROR("error allocating lmv objects cache\n");
1991 lprocfs_init_vars(lmv, &lvars);
1992 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1994 OBD_LMV_DEVICENAME);
1996 kmem_cache_destroy(obj_cache);
2002 static void lmv_exit(void)
2004 class_unregister_type(OBD_LMV_DEVICENAME);
2006 LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2007 "can't free lmv objects cache, %d object(s)"
2008 "still in use\n", atomic_read(&obj_cache_count));
2011 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2012 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2013 MODULE_LICENSE("GPL");
2015 module_init(lmv_init);
2016 module_exit(lmv_exit);