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 lmv_tgt_desc *tgt;
77 struct obd_device *obd;
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)
129 struct obd_uuid *uuid;
133 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
134 CERROR("unexpected notification of %s %s!\n",
135 watched->obd_type->typ_name,
139 uuid = &watched->u.cli.cl_import->imp_target_uuid;
141 /* Set MDC as active before notifying the observer, so the observer can
142 * use the MDC normally.
144 rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
146 CERROR("%sactivation of %s failed: %d\n",
147 active ? "" : "de", uuid->uuid, rc);
151 if (obd->obd_observer)
152 /* Pass the notification up the chain. */
153 rc = obd_notify(obd->obd_observer, watched, active, data);
158 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
160 struct lprocfs_static_vars lvars;
164 lprocfs_init_vars(lmv, &lvars);
165 rc = lprocfs_obd_attach(dev, lvars.obd_vars);
168 struct proc_dir_entry *entry;
170 entry = create_proc_entry("target_obd_status", 0444,
171 dev->obd_proc_entry);
174 entry->proc_fops = &lmv_proc_target_fops;
181 int lmv_detach(struct obd_device *dev)
183 return lprocfs_obd_detach(dev);
186 /* this is fake connect function. Its purpose is to initialize lmv and say
187 * caller that everything is okay. Real connection will be performed later. */
188 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
189 struct obd_uuid *cluuid, unsigned long flags)
192 struct proc_dir_entry *lmv_proc_dir;
194 struct lmv_obd *lmv = &obd->u.lmv;
195 struct obd_export *exp;
199 rc = class_connect(conn, obd, cluuid);
201 CERROR("class_connection() returned %d\n", rc);
205 exp = class_conn2export(conn);
207 /* we don't want to actually do the underlying connections more than
208 * once, so keep track. */
210 if (lmv->refcount > 1) {
211 class_export_put(exp);
217 lmv->cluuid = *cluuid;
218 lmv->connect_flags = flags;
219 sema_init(&lmv->init_sem, 1);
222 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
224 if (IS_ERR(lmv_proc_dir)) {
225 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
226 obd->obd_type->typ_name, obd->obd_name);
232 * all real clients shouls perform actual connection rightaway, because
233 * it is possible, that LMV will not have opportunity to connect
234 * targets, as MDC stuff will bit called directly, for instance while
235 * reading ../mdc/../kbytesfree procfs file, etc.
237 if (flags & OBD_OPT_REAL_CLIENT)
238 rc = lmv_check_connect(obd);
243 lprocfs_remove(lmv_proc_dir);
250 void lmv_set_timeouts(struct obd_device *obd)
252 struct lmv_tgt_desc *tgts;
257 if (lmv->server_timeout == 0)
260 if (lmv->connected == 0)
263 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
264 if (tgts->ltd_exp == NULL)
267 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
268 "inter_mds", 0, NULL);
272 #define MAX_STRING_SIZE 128
274 /* performs a check if passed obd is connected. If no - connect it. */
275 int lmv_check_connect(struct obd_device *obd)
278 struct proc_dir_entry *lmv_proc_dir;
280 struct lmv_obd *lmv = &obd->u.lmv;
281 struct lmv_tgt_desc *tgts;
282 struct obd_uuid *cluuid;
283 struct obd_export *exp;
290 down(&lmv->init_sem);
291 if (lmv->connected) {
296 cluuid = &lmv->cluuid;
299 CDEBUG(D_OTHER, "time to connect %s to %s\n",
300 cluuid->uuid, obd->obd_name);
302 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
303 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
304 struct lustre_handle conn = {0, };
305 struct obd_device *tgt_obd;
307 LASSERT(tgts != NULL);
309 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
312 CERROR("target %s not attached\n", tgts->uuid.uuid);
313 GOTO(out_disc, rc = -EINVAL);
316 /* for MDS: don't connect to yourself */
317 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
318 CDEBUG(D_OTHER, "don't connect back to %s\n",
320 tgts->ltd_exp = NULL;
324 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
325 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
326 tgts->uuid.uuid, obd->obd_uuid.uuid,
329 if (!tgt_obd->obd_set_up) {
330 CERROR("target %s not set up\n", tgts->uuid.uuid);
331 GOTO(out_disc, rc = -EINVAL);
334 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
337 CERROR("target %s connect error %d\n",
338 tgts->uuid.uuid, rc);
341 tgts->ltd_exp = class_conn2export(&conn);
343 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
344 lmv->max_cookiesize);
346 rc = obd_register_observer(tgt_obd, obd);
348 CERROR("target %s register_observer error %d\n",
349 tgts->uuid.uuid, rc);
350 obd_disconnect(tgts->ltd_exp, 0);
354 lmv->desc.ld_active_tgt_count++;
357 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
358 tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
359 atomic_read(&obd->obd_refcount));
362 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
364 struct obd_device *mdc_obd = class_conn2obd(&conn);
365 struct proc_dir_entry *mdc_symlink;
366 char name[MAX_STRING_SIZE + 1];
368 LASSERT(mdc_obd != NULL);
369 LASSERT(mdc_obd->obd_type != NULL);
370 LASSERT(mdc_obd->obd_type->typ_name != NULL);
371 name[MAX_STRING_SIZE] = '\0';
372 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
373 mdc_obd->obd_type->typ_name,
375 mdc_symlink = proc_symlink(mdc_obd->obd_name,
377 if (mdc_symlink == NULL) {
378 CERROR("could not register LMV target "
379 "/proc/fs/lustre/%s/%s/target_obds/%s.",
380 obd->obd_type->typ_name, obd->obd_name,
382 lprocfs_remove(lmv_proc_dir);
389 lmv_set_timeouts(obd);
390 class_export_put(exp);
397 struct obd_uuid uuid;
399 --lmv->desc.ld_active_tgt_count;
401 /* save for CERROR below; (we know it's terminated) */
403 rc2 = obd_disconnect(tgts->ltd_exp, 0);
405 CERROR("error: LMV target %s disconnect on MDC idx %d: "
406 "error %d\n", uuid.uuid, i, rc2);
408 class_disconnect(exp, 0);
413 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
415 struct obd_device *obd = class_exp2obd(exp);
416 struct lmv_obd *lmv = &obd->u.lmv;
419 struct proc_dir_entry *lmv_proc_dir;
427 /* Only disconnect the underlying layers on the final disconnect. */
429 if (lmv->refcount != 0)
433 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
436 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
437 struct obd_device *mdc_obd;
439 if (lmv->tgts[i].ltd_exp == NULL)
442 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
446 struct proc_dir_entry *mdc_symlink;
448 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
450 lprocfs_remove(mdc_symlink);
452 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
453 obd->obd_type->typ_name, obd->obd_name,
458 if (obd->obd_no_recov) {
460 mdc_obd->obd_no_recov = 1;
462 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
463 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
464 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
466 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
467 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
469 if (lmv->tgts[i].active) {
470 CERROR("Target %s disconnect error %d\n",
471 lmv->tgts[i].uuid.uuid, rc);
476 lmv_activate_target(lmv, &lmv->tgts[i], 0);
477 lmv->tgts[i].ltd_exp = NULL;
482 lprocfs_remove(lmv_proc_dir);
484 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
485 obd->obd_type->typ_name, obd->obd_name);
490 /* this is the case when no real connection is established by
491 * lmv_check_connect(). */
493 class_export_put(exp);
494 rc = class_disconnect(exp, 0);
495 if (lmv->refcount == 0)
500 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
501 int len, void *karg, void *uarg)
503 struct obd_device *obddev = class_exp2obd(exp);
504 struct lmv_obd *lmv = &obddev->u.lmv;
505 int i, rc = 0, set = 0;
508 if (lmv->desc.ld_tgt_count == 0)
511 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
514 if (lmv->tgts[i].ltd_exp == NULL)
517 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
519 if (lmv->tgts[i].active) {
520 CERROR("error: iocontrol MDC %s on MDT"
521 "idx %d: err = %d\n",
522 lmv->tgts[i].uuid.uuid, i, err);
535 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
538 struct lmv_desc *desc;
539 struct obd_uuid *uuids;
540 struct lmv_tgt_desc *tgts;
541 struct obd_device *tgt_obd;
542 struct lustre_cfg *lcfg = buf;
543 struct lmv_obd *lmv = &obd->u.lmv;
546 if (lcfg->lcfg_inllen1 < 1) {
547 CERROR("LMV setup requires a descriptor\n");
551 if (lcfg->lcfg_inllen2 < 1) {
552 CERROR("LMV setup requires an MDT UUID list\n");
556 desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
557 if (sizeof(*desc) > lcfg->lcfg_inllen1) {
558 CERROR("descriptor size wrong: %d > %d\n",
559 (int)sizeof(*desc), lcfg->lcfg_inllen1);
563 uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
564 if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
565 CERROR("UUID array size wrong: %u * %u != %u\n",
566 sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
570 lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
571 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
572 if (lmv->tgts == NULL) {
573 CERROR("Out of memory\n");
578 spin_lock_init(&lmv->lmv_lock);
580 for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
581 tgts->uuid = uuids[i];
583 lmv->max_cookiesize = 0;
585 lmv->max_easize = sizeof(struct lustre_id) *
586 desc->ld_tgt_count + sizeof(struct mea);
588 rc = lmv_setup_mgr(obd);
590 CERROR("Can't setup LMV object manager, "
592 OBD_FREE(lmv->tgts, lmv->tgts_size);
595 tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
598 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
605 static int lmv_cleanup(struct obd_device *obd, int flags)
607 struct lmv_obd *lmv = &obd->u.lmv;
610 lmv_cleanup_mgr(obd);
611 OBD_FREE(lmv->tgts, lmv->tgts_size);
616 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
617 unsigned long max_age)
619 struct lmv_obd *lmv = &obd->u.lmv;
620 struct obd_statfs *temp;
624 rc = lmv_check_connect(obd);
628 OBD_ALLOC(temp, sizeof(*temp));
632 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
633 if (lmv->tgts[i].ltd_exp == NULL) {
634 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
638 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
640 CERROR("can't stat MDS #%d (%s), error %d\n", i,
641 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
643 GOTO(out_free_temp, rc);
646 memcpy(osfs, temp, sizeof(*temp));
648 osfs->os_bavail += temp->os_bavail;
649 osfs->os_blocks += temp->os_blocks;
650 osfs->os_ffree += temp->os_ffree;
651 osfs->os_files += temp->os_files;
657 OBD_FREE(temp, sizeof(*temp));
661 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
663 struct obd_device *obd = exp->exp_obd;
664 struct lmv_obd *lmv = &obd->u.lmv;
668 rc = lmv_check_connect(obd);
672 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
678 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
679 __u64 valid, unsigned int ea_size,
680 struct ptlrpc_request **request)
682 struct obd_device *obd = exp->exp_obd;
683 struct lmv_obd *lmv = &obd->u.lmv;
684 int rc, i = id_group(id);
688 rc = lmv_check_connect(obd);
692 LASSERT(i < lmv->desc.ld_tgt_count);
694 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
699 obj = lmv_grab_obj(obd, id);
701 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
702 OLID4(id), obj ? "(splitted)" : "");
705 * if object is splitted, then we loop over all the slaves and gather
706 * size attribute. In ideal world we would have to gather also mds field
707 * from all slaves, as object is spread over the cluster and this is
708 * definitely interesting information and it is not good to loss it,
712 struct mds_body *body;
714 if (*request == NULL) {
719 body = lustre_msg_buf((*request)->rq_repmsg, 0,
721 LASSERT(body != NULL);
725 for (i = 0; i < obj->objcount; i++) {
727 if (lmv->tgts[i].ltd_exp == NULL) {
728 CWARN("%s: NULL export for %d\n",
733 /* skip master obj. */
734 if (id_equal_fid(&obj->id, &obj->objs[i].id))
737 body->size += obj->objs[i].size;
747 static int lmv_change_cbdata(struct obd_export *exp,
748 struct lustre_id *id,
752 struct obd_device *obd = exp->exp_obd;
753 struct lmv_obd *lmv = &obd->u.lmv;
757 rc = lmv_check_connect(obd);
761 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
762 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
764 rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
770 static int lmv_change_cbdata_name(struct obd_export *exp,
771 struct lustre_id *pid,
773 struct lustre_id *cid,
777 struct obd_device *obd = exp->exp_obd;
778 struct lmv_obd *lmv = &obd->u.lmv;
779 struct lustre_id rcid = *cid;
784 rc = lmv_check_connect(obd);
788 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
789 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
791 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
792 OLID4(pid), len, name, OLID4(cid));
794 /* this is default mds for directory name belongs to. */
796 obj = lmv_grab_obj(obd, pid);
798 /* directory is splitted. look for right mds for this name. */
799 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
800 rcid = obj->objs[mds].id;
801 mds = id_group(&rcid);
804 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
808 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
810 struct obd_device *obd = exp->exp_obd;
811 struct lmv_obd *lmv = &obd->u.lmv;
815 rc = lmv_check_connect(obd);
819 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
820 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
821 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
825 int lmv_close(struct obd_export *exp, struct obdo *obdo,
826 struct obd_client_handle *och,
827 struct ptlrpc_request **request)
829 struct obd_device *obd = exp->exp_obd;
830 struct lmv_obd *lmv = &obd->u.lmv;
831 int rc, i = obdo->o_mds;
834 rc = lmv_check_connect(obd);
838 LASSERT(i < lmv->desc.ld_tgt_count);
839 CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
840 (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
841 rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
845 int lmv_get_mea_and_update_object(struct obd_export *exp,
846 struct lustre_id *id)
848 struct obd_device *obd = exp->exp_obd;
849 struct lmv_obd *lmv = &obd->u.lmv;
850 struct ptlrpc_request *req = NULL;
858 mealen = MEA_SIZE_LMV(lmv);
860 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
862 /* time to update mea of parent id */
863 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
864 id, valid, mealen, &req);
866 CERROR("md_getattr() failed, error %d\n", rc);
870 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
872 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
877 GOTO(cleanup, rc = -ENODATA);
879 obj = lmv_create_obj(exp, id, md.mea);
884 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
889 ptlrpc_req_finished(req);
893 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
894 const void *data, int datalen, int mode, __u32 uid,
895 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
897 struct obd_device *obd = exp->exp_obd;
898 struct lmv_obd *lmv = &obd->u.lmv;
899 struct mds_body *body;
901 int rc, mds, loop = 0;
904 rc = lmv_check_connect(obd);
908 if (!lmv->desc.ld_active_tgt_count)
911 LASSERT(++loop <= 2);
912 obj = lmv_grab_obj(obd, &op_data->id1);
914 mds = raw_name2idx(obj->hashtype, obj->objcount,
915 op_data->name, op_data->namelen);
916 op_data->id1 = obj->objs[mds].id;
920 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
921 op_data->name, OLID4(&op_data->id1));
923 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
924 op_data, data, datalen, mode, uid, gid, rdev,
927 if (*request == NULL)
930 body = lustre_msg_buf((*request)->rq_repmsg, 0,
935 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
936 } else if (rc == -ERESTART) {
938 * directory got splitted. time to update local object and
939 * repeat the request with proper MDS.
941 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
943 ptlrpc_req_finished(*request);
950 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
952 struct obd_device *obd = exp->exp_obd;
953 struct lmv_obd *lmv = &obd->u.lmv;
957 rc = lmv_check_connect(obd);
961 /* FIXME: choose right MDC here */
962 CWARN("this method isn't implemented yet\n");
963 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
967 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
968 struct lookup_intent *it, int lockmode,
969 struct mdc_op_data *data, struct lustre_handle *lockh,
970 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
971 ldlm_blocking_callback cb_blocking, void *cb_data)
973 struct obd_device *obd = exp->exp_obd;
974 struct lmv_obd *lmv = &obd->u.lmv;
975 struct mea *mea = data->mea1;
976 struct mdc_op_data *data2;
980 OBD_ALLOC(data2, sizeof(*data2));
984 LASSERT(mea != NULL);
985 for (i = 0; i < mea->mea_count; i++) {
986 memset(data2, 0, sizeof(*data2));
987 data2->id1 = mea->mea_ids[i];
988 mds = id_group(&data2->id1);
990 if (lmv->tgts[mds].ltd_exp == NULL)
993 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
994 lockmode, data2, lockh + i, lmm, lmmsize,
995 cb_compl, cb_blocking, cb_data);
997 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
998 OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1001 if (it->d.lustre.it_data) {
1002 struct ptlrpc_request *req;
1003 req = (struct ptlrpc_request *)it->d.lustre.it_data;
1004 ptlrpc_req_finished(req);
1007 if (it->d.lustre.it_status)
1008 GOTO(cleanup, rc = it->d.lustre.it_status);
1011 OBD_FREE(data2, sizeof(*data2));
1014 OBD_FREE(data2, sizeof(*data2));
1016 /* drop all taken locks */
1018 if (lockh[i].cookie)
1019 ldlm_lock_decref(lockh + i, lockmode);
1020 lockh[i].cookie = 0;
1025 int lmv_enqueue(struct obd_export *exp, int lock_type,
1026 struct lookup_intent *it, int lock_mode,
1027 struct mdc_op_data *data, struct lustre_handle *lockh,
1028 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1029 ldlm_blocking_callback cb_blocking, void *cb_data)
1031 struct obd_device *obd = exp->exp_obd;
1032 struct lmv_obd *lmv = &obd->u.lmv;
1033 struct lmv_obj *obj;
1037 rc = lmv_check_connect(obd);
1041 if (data->mea1 && it->it_op == IT_UNLINK) {
1042 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1043 data, lockh, lmm, lmmsize,
1044 cb_compl, cb_blocking, cb_data);
1048 if (data->namelen) {
1049 obj = lmv_grab_obj(obd, &data->id1);
1051 /* directory is splitted. look for right mds for this
1053 mds = raw_name2idx(obj->hashtype, obj->objcount,
1054 (char *)data->name, data->namelen);
1055 data->id1 = obj->objs[mds].id;
1059 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1062 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1063 lock_type, it, lock_mode, data, lockh, lmm,
1064 lmmsize, cb_compl, cb_blocking, cb_data);
1068 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1069 char *filename, int namelen, __u64 valid,
1070 unsigned int ea_size, struct ptlrpc_request **request)
1072 int rc, mds = id_group(id), loop = 0;
1073 struct obd_device *obd = exp->exp_obd;
1074 struct lmv_obd *lmv = &obd->u.lmv;
1075 struct lustre_id rid = *id;
1076 struct mds_body *body;
1077 struct lmv_obj *obj;
1081 rc = lmv_check_connect(obd);
1085 LASSERT(++loop <= 2);
1086 obj = lmv_grab_obj(obd, id);
1088 /* directory is splitted. look for right mds for this name */
1089 mds = raw_name2idx(obj->hashtype, obj->objcount,
1090 filename, namelen - 1);
1091 rid = obj->objs[mds].id;
1095 CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1096 namelen, filename, OLID4(id), OLID4(&rid));
1101 * here should be applied OBD_MD_FID to ->valid, because otherwise,
1102 * mds_getattr_lock() will not fetch fid component of lustre_id and
1103 * thus, next call to md_getattr_lock() will be performed to wrong mds.
1105 if (!(old_valid & OBD_MD_FID))
1106 valid |= OBD_MD_FID;
1108 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1109 &rid, filename, namelen, valid,
1113 * this could be cross-node reference. in this case all we have
1114 * right now is lustre_id triple. we'd like to find other
1117 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1118 LASSERT(body != NULL);
1119 if (body->valid & OBD_MD_MDS) {
1120 struct ptlrpc_request *req = NULL;
1123 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1126 * turning OBD_MD_FID fetching off, as we already have
1127 * full lustre_id and do need to fetch fid component
1128 * again. This will help to make thing slightly faster.
1130 if (!(old_valid & OBD_MD_FID))
1131 valid &= ~OBD_MD_FID;
1133 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1134 &rid, NULL, 1, valid, ea_size, &req);
1135 ptlrpc_req_finished(*request);
1138 } else if (rc == -ERESTART) {
1139 /* directory got splitted. time to update local object and
1140 * repeat the request with proper MDS */
1141 rc = lmv_get_mea_and_update_object(exp, &rid);
1143 ptlrpc_req_finished(*request);
1151 * llite passes id of an target inode in data->id1 and id of directory in
1154 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1155 struct ptlrpc_request **request)
1157 struct obd_device *obd = exp->exp_obd;
1158 struct lmv_obd *lmv = &obd->u.lmv;
1159 struct lmv_obj *obj;
1163 rc = lmv_check_connect(obd);
1167 if (data->namelen != 0) {
1168 /* usual link request */
1169 obj = lmv_grab_obj(obd, &data->id1);
1171 rc = raw_name2idx(obj->hashtype, obj->objcount,
1172 data->name, data->namelen);
1173 data->id1 = obj->objs[rc].id;
1177 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1178 OLID4(&data->id2), data->namelen, data->name,
1181 /* request from MDS to acquire i_links for inode by id1 */
1182 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1186 rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1191 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1192 const char *old, int oldlen, const char *new, int newlen,
1193 struct ptlrpc_request **request)
1195 struct obd_device *obd = exp->exp_obd;
1196 struct lmv_obd *lmv = &obd->u.lmv;
1197 struct lmv_obj *obj;
1201 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1202 oldlen, old, OLID4(&data->id1), newlen, new,
1205 rc = lmv_check_connect(obd);
1211 * MDS with old dir entry is asking another MDS to create name
1215 "create %*s(%d/%d) in "DLID4" pointing "
1216 "to "DLID4"\n", newlen, new, oldlen, newlen,
1217 OLID4(&data->id2), OLID4(&data->id1));
1219 mds = id_group(&data->id2);
1222 * target directory can be splitted, sowe should forward request
1225 obj = lmv_grab_obj(obd, &data->id2);
1227 mds = raw_name2idx(obj->hashtype, obj->objcount,
1228 (char *)new, newlen);
1229 data->id2 = obj->objs[mds].id;
1230 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1237 obj = lmv_grab_obj(obd, &data->id1);
1240 * directory is already splitted, so we have to forward request
1243 mds = raw_name2idx(obj->hashtype, obj->objcount,
1244 (char *)old, oldlen);
1245 data->id1 = obj->objs[mds].id;
1246 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1251 obj = lmv_grab_obj(obd, &data->id2);
1254 * directory is already splitted, so we have to forward request
1257 mds = raw_name2idx(obj->hashtype, obj->objcount,
1258 (char *)new, newlen);
1260 data->id2 = obj->objs[mds].id;
1261 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1266 mds = id_group(&data->id1);
1269 if (id_group(&data->id1) != id_group(&data->id2)) {
1270 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1271 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1275 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1276 new, newlen, request);
1280 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1281 struct iattr *iattr, void *ea, int ealen, void *ea2,
1282 int ea2len, struct ptlrpc_request **request)
1284 struct obd_device *obd = exp->exp_obd;
1285 struct lmv_obd *lmv = &obd->u.lmv;
1286 struct ptlrpc_request *req;
1287 struct mds_body *body;
1288 struct lmv_obj *obj;
1292 rc = lmv_check_connect(obd);
1296 obj = lmv_grab_obj(obd, &data->id1);
1298 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1299 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1302 for (i = 0; i < obj->objcount; i++) {
1303 data->id1 = obj->objs[i].id;
1305 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1306 data, iattr, ea, ealen, ea2, ea2len, &req);
1308 if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1310 * this is master object and this request should
1311 * be returned back to llite.
1315 ptlrpc_req_finished(req);
1323 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1324 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1325 data, iattr, ea, ealen, ea2, ea2len, request);
1327 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1329 LASSERT(body != NULL);
1330 LASSERT(id_group(&body->id1) == id_group(&data->id1));
1336 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1337 struct ptlrpc_request **request)
1339 struct obd_device *obd = exp->exp_obd;
1340 struct lmv_obd *lmv = &obd->u.lmv;
1344 rc = lmv_check_connect(obd);
1348 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1353 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1354 struct ldlm_lock_desc *desc,
1355 void *data, int flag)
1357 struct lustre_handle lockh;
1358 struct lmv_obj *obj;
1363 case LDLM_CB_BLOCKING:
1364 ldlm_lock2handle(lock, &lockh);
1365 rc = ldlm_cli_cancel(&lockh);
1367 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1371 case LDLM_CB_CANCELING:
1372 /* time to drop cached attrs for dirobj */
1373 obj = lock->l_ast_data;
1375 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1376 ", master "DLID4"\n",
1377 lock->l_resource->lr_name.name[3] == 1 ?
1378 "LOOKUP" : "UPDATE",
1379 lock->l_resource->lr_name.name[0],
1380 lock->l_resource->lr_name.name[1],
1391 void lmv_remove_dots(struct page *page)
1393 unsigned limit = PAGE_CACHE_SIZE;
1394 char *kaddr = page_address(page);
1395 struct ext2_dir_entry_2 *p;
1396 unsigned offs, rec_len;
1398 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1399 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1400 rec_len = le16_to_cpu(p->rec_len);
1402 if ((p->name_len == 1 && p->name[0] == '.') ||
1403 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1408 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1409 __u64 offset, struct page *page,
1410 struct ptlrpc_request **request)
1412 struct obd_device *obd = exp->exp_obd;
1413 struct lmv_obd *lmv = &obd->u.lmv;
1414 struct lustre_id rid = *id;
1415 struct lmv_obj *obj;
1419 #warning "we need well-desgined readdir() implementation"
1420 rc = lmv_check_connect(obd);
1424 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1425 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1426 offset, OLID4(&rid));
1428 obj = lmv_grab_obj(obd, id);
1432 /* find dirobj containing page with requested offset. */
1433 for (i = 0; i < obj->objcount; i++) {
1434 if (offset < obj->objs[i].size)
1436 offset -= obj->objs[i].size;
1438 rid = obj->objs[i].id;
1440 lmv_unlock_obj(obj);
1443 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1444 OLID4(&rid), (unsigned long)offset);
1446 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1447 offset, page, request);
1449 if (rc == 0 && !id_equal_fid(&rid, id))
1450 /* this page isn't from master object. To avoid "." and ".."
1451 * duplication in directory, we have to remove them from all
1453 lmv_remove_dots(page);
1458 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1459 struct ptlrpc_request **req)
1461 struct obd_device *obd = exp->exp_obd;
1462 struct lmv_obd *lmv = &obd->u.lmv;
1463 struct mea *mea = data->mea1;
1464 struct mdc_op_data *data2;
1468 OBD_ALLOC(data2, sizeof(*data2));
1472 LASSERT(mea != NULL);
1473 for (i = 0; i < mea->mea_count; i++) {
1474 memset(data2, 0, sizeof(*data2));
1475 data2->id1 = mea->mea_ids[i];
1476 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1478 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1481 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1484 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1485 OLID4(&mea->mea_ids[i]), rc);
1488 ptlrpc_req_finished(*req);
1494 OBD_FREE(data2, sizeof(*data2));
1498 int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1503 if (lmv_delete_obj(exp, id)) {
1504 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1510 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1511 struct ptlrpc_request **request)
1513 struct obd_device *obd = exp->exp_obd;
1514 struct lmv_obd *lmv = &obd->u.lmv;
1518 rc = lmv_check_connect(obd);
1522 if (data->namelen == 0 && data->mea1 != NULL) {
1523 /* mds asks to remove slave objects */
1524 rc = lmv_unlink_slaves(exp, data, request);
1528 if (data->namelen != 0) {
1529 struct lmv_obj *obj;
1531 obj = lmv_grab_obj(obd, &data->id1);
1533 i = raw_name2idx(obj->hashtype, obj->objcount,
1534 data->name, data->namelen);
1535 data->id1 = obj->objs[i].id;
1538 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1539 data->namelen, data->name, OLID4(&data->id1),
1542 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1545 rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1550 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1551 char *name, int len)
1553 struct obd_device *obd = exp->exp_obd;
1554 struct lmv_obd *lmv = &obd->u.lmv;
1558 rc = lmv_check_connect(obd);
1560 RETURN(ERR_PTR(rc));
1561 obd = lmv->tgts[0].ltd_exp->exp_obd;
1567 int lmv_init_ea_size(struct obd_export *exp, int easize,
1570 struct obd_device *obd = exp->exp_obd;
1571 struct lmv_obd *lmv = &obd->u.lmv;
1572 int i, rc = 0, change = 0;
1575 if (lmv->max_easize < easize) {
1576 lmv->max_easize = easize;
1579 if (lmv->max_cookiesize < cookiesize) {
1580 lmv->max_cookiesize = cookiesize;
1586 if (lmv->connected == 0)
1589 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1590 if (lmv->tgts[i].ltd_exp == NULL) {
1591 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1595 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1597 CERROR("obd_init_ea_size() failed on MDT target %d, "
1598 "error %d.\n", i, rc);
1605 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1606 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1608 struct obd_device *obd = exp->exp_obd;
1609 struct lmv_obd *lmv = &obd->u.lmv;
1610 struct lov_stripe_md obj_md;
1611 struct lov_stripe_md *obj_mdp = &obj_md;
1615 LASSERT(ea == NULL);
1616 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1618 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
1624 int lmv_getready(struct obd_export *exp)
1626 struct obd_device *obd = exp->exp_obd;
1630 rc = lmv_check_connect(obd);
1635 * to be called from MDS only. @oa should have correct store cookie and o_fid
1636 * values for "master" object, as it will be used.
1638 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1639 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1641 struct obd_device *obd = exp->exp_obd;
1642 struct lmv_obd *lmv = &obd->u.lmv;
1643 struct lustre_id mid;
1648 rc = lmv_check_connect(obd);
1652 LASSERT(oa != NULL);
1655 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1657 CERROR("Can't create object, rc = %d\n", rc);
1662 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1664 CERROR("obd_alloc_diskmd() failed, error %d\n",
1675 * here we should take care about splitted dir, so store cookie and fid
1676 * for "master" object should already be allocated and passed in @oa.
1678 LASSERT(oa->o_id != 0);
1679 LASSERT(oa->o_fid != 0);
1681 /* save "master" object id */
1684 mea = (struct mea *)*ea;
1685 mea->mea_master = -1;
1686 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1688 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1689 mea->mea_count = lmv->desc.ld_tgt_count;
1691 for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1692 struct lov_stripe_md obj_md;
1693 struct lov_stripe_md *obj_mdp = &obj_md;
1695 if (lmv->tgts[i].ltd_exp == NULL) {
1696 /* this is "master" MDS */
1697 mea->mea_master = i;
1698 mea->mea_ids[c] = mid;
1704 * "master" MDS should always be part of stripped dir,
1707 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1710 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1711 OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1713 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1715 CERROR("obd_create() failed on MDT target %d, "
1716 "error %d\n", c, rc);
1720 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1721 i, oa->o_id, oa->o_generation);
1725 * here, when object is created (or it is master and was passed
1726 * from caller) on desired MDS we save its fid to local mea_ids.
1731 * store cookie should be defined here for both cases (master
1732 * object and not master), because master is already created.
1736 /* fill mea by store cookie and fid */
1737 obdo2id(&mea->mea_ids[c], oa);
1740 LASSERT(c == mea->mea_count);
1742 CDEBUG(D_OTHER, "%d dirobjects created\n",
1743 (int)mea->mea_count);
1748 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1749 struct obd_device *tgt, int count,
1750 struct llog_catid *logid)
1752 struct llog_ctxt *ctxt;
1756 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1759 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1760 ctxt->loc_imp = tgt->u.cli.cl_import;
1766 static int lmv_llog_finish(struct obd_device *obd,
1767 struct obd_llogs *llogs, int count)
1772 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1776 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1777 void *key, __u32 *vallen, void *val)
1779 struct obd_device *obd;
1780 struct lmv_obd *lmv;
1784 obd = class_exp2obd(exp);
1786 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1787 exp->exp_handle.h_cookie);
1792 if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1793 __u32 *mdsize = val;
1794 *vallen = sizeof(__u32);
1795 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1796 + sizeof(struct mea);
1798 } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1799 struct obd_uuid *cluuid = &lmv->cluuid;
1800 struct lmv_tgt_desc *tgts;
1801 __u32 *mdsnum = val;
1804 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1805 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1806 *vallen = sizeof(__u32);
1812 } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1813 /* getting rootid from first MDS. */
1814 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1817 } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1818 struct lmv_desc *desc_ret = val;
1819 *desc_ret = lmv->desc;
1823 CDEBUG(D_IOCTL, "invalid key\n");
1827 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1828 void *key, obd_count vallen, void *val)
1830 struct obd_device *obd;
1831 struct lmv_obd *lmv;
1834 obd = class_exp2obd(exp);
1836 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1837 exp->exp_handle.h_cookie);
1842 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1843 lmv->server_timeout = 1;
1844 lmv_set_timeouts(obd);
1851 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1852 struct lov_stripe_md *lsm)
1854 struct obd_device *obd = class_exp2obd(exp);
1855 struct lmv_obd *lmv = &obd->u.lmv;
1856 struct mea *meap, *lsmp;
1860 mea_size = (sizeof(struct lustre_id) *
1861 lmv->desc.ld_tgt_count) + sizeof(struct mea);
1865 if (*lmmp && !lsm) {
1866 OBD_FREE(*lmmp, mea_size);
1871 if (*lmmp == NULL) {
1872 OBD_ALLOC(*lmmp, mea_size);
1880 lsmp = (struct mea *)lsm;
1881 meap = (struct mea *)*lmmp;
1883 meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
1884 meap->mea_count = cpu_to_le32(lsmp->mea_count);
1885 meap->mea_master = cpu_to_le32(lsmp->mea_master);
1887 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1888 meap->mea_ids[i] = meap->mea_ids[i];
1889 id_cpu_to_le(&meap->mea_ids[i]);
1895 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1896 struct lov_mds_md *disk_src, int mdsize)
1898 struct obd_device *obd = class_exp2obd(exp);
1899 struct mea **tmea = (struct mea **)mem_tgt;
1900 struct mea *mea = (struct mea *)disk_src;
1901 struct lmv_obd *lmv = &obd->u.lmv;
1905 mea_size = sizeof(struct lustre_id) *
1906 lmv->desc.ld_tgt_count + sizeof(struct mea);
1907 if (mem_tgt == NULL)
1910 if (*mem_tgt != NULL && disk_src == NULL) {
1911 OBD_FREE(*tmea, mea_size);
1915 LASSERT(mea_size == mdsize);
1917 OBD_ALLOC(*tmea, mea_size);
1924 (*tmea)->mea_magic = le32_to_cpu(mea->mea_magic);
1925 (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
1926 (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
1928 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1929 (*tmea)->mea_ids[i] = mea->mea_ids[i];
1930 id_le_to_cpu(&(*tmea)->mea_ids[i]);
1936 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1937 struct lov_stripe_md *ea, obd_count oa_bufs,
1938 struct brw_page *pgarr, struct obd_trans_info *oti)
1940 struct obd_device *obd = exp->exp_obd;
1941 struct lmv_obd *lmv = &obd->u.lmv;
1942 struct mea *mea = (struct mea *) ea;
1945 LASSERT(oa != NULL);
1946 LASSERT(ea != NULL);
1947 LASSERT(pgarr != NULL);
1948 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1950 oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1951 oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1952 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1954 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1955 oa, NULL, oa_bufs, pgarr, oti);
1959 struct obd_ops lmv_obd_ops = {
1960 .o_owner = THIS_MODULE,
1961 .o_attach = lmv_attach,
1962 .o_detach = lmv_detach,
1963 .o_setup = lmv_setup,
1964 .o_cleanup = lmv_cleanup,
1965 .o_connect = lmv_connect,
1966 .o_disconnect = lmv_disconnect,
1967 .o_statfs = lmv_statfs,
1968 .o_llog_init = lmv_llog_init,
1969 .o_llog_finish = lmv_llog_finish,
1970 .o_get_info = lmv_get_info,
1971 .o_set_info = lmv_set_info,
1972 .o_create = lmv_obd_create,
1973 .o_packmd = lmv_packmd,
1974 .o_unpackmd = lmv_unpackmd,
1976 .o_init_ea_size = lmv_init_ea_size,
1977 .o_notify = lmv_notify,
1978 .o_iocontrol = lmv_iocontrol,
1979 .o_getready = lmv_getready,
1982 struct md_ops lmv_md_ops = {
1983 .m_getstatus = lmv_getstatus,
1984 .m_getattr = lmv_getattr,
1985 .m_change_cbdata = lmv_change_cbdata,
1986 .m_change_cbdata_name = lmv_change_cbdata_name,
1987 .m_close = lmv_close,
1988 .m_create = lmv_create,
1989 .m_done_writing = lmv_done_writing,
1990 .m_enqueue = lmv_enqueue,
1991 .m_getattr_lock = lmv_getattr_lock,
1992 .m_intent_lock = lmv_intent_lock,
1994 .m_rename = lmv_rename,
1995 .m_setattr = lmv_setattr,
1997 .m_readpage = lmv_readpage,
1998 .m_unlink = lmv_unlink,
1999 .m_get_real_obd = lmv_get_real_obd,
2000 .m_valid_attrs = lmv_valid_attrs,
2001 .m_delete_inode = lmv_delete_inode,
2004 int __init lmv_init(void)
2006 struct lprocfs_static_vars lvars;
2009 obj_cache = kmem_cache_create("lmv_objects",
2010 sizeof(struct lmv_obj),
2013 CERROR("error allocating lmv objects cache\n");
2017 lprocfs_init_vars(lmv, &lvars);
2018 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2020 OBD_LMV_DEVICENAME);
2022 kmem_cache_destroy(obj_cache);
2028 static void lmv_exit(void)
2030 class_unregister_type(OBD_LMV_DEVICENAME);
2032 LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2033 "can't free lmv objects cache, %d object(s)"
2034 "still in use\n", atomic_read(&obj_cache_count));
2037 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2038 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2039 MODULE_LICENSE("GPL");
2041 module_init(lmv_init);
2042 module_exit(lmv_exit);