4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
39 #include <linux/slab.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
45 #include <asm/div64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
49 #include <liblustre.h>
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lprocfs_status.h>
58 #include <lustre_lite.h>
59 #include <lustre_fid.h>
60 #include "lmv_internal.h"
62 static void lmv_activate_target(struct lmv_obd *lmv,
63 struct lmv_tgt_desc *tgt,
66 if (tgt->ltd_active == activate)
69 tgt->ltd_active = activate;
70 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
76 * -EINVAL : UUID can't be found in the LMV's target list
77 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
78 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
80 static int lmv_set_mdc_active(struct lmv_obd *lmv,
81 const struct obd_uuid *uuid,
84 struct lmv_tgt_desc *tgt = NULL;
85 struct obd_device *obd;
90 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
91 lmv, uuid->uuid, activate);
93 spin_lock(&lmv->lmv_lock);
94 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
96 if (tgt == NULL || tgt->ltd_exp == NULL)
99 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
100 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
102 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
106 if (i == lmv->desc.ld_tgt_count)
107 GOTO(out_lmv_lock, rc = -EINVAL);
109 obd = class_exp2obd(tgt->ltd_exp);
111 GOTO(out_lmv_lock, rc = -ENOTCONN);
113 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
114 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
115 obd->obd_type->typ_name, i);
116 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
118 if (tgt->ltd_active == activate) {
119 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
120 activate ? "" : "in");
121 GOTO(out_lmv_lock, rc);
124 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
125 activate ? "" : "in");
126 lmv_activate_target(lmv, tgt, activate);
130 spin_unlock(&lmv->lmv_lock);
134 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
136 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
137 struct lmv_tgt_desc *tgt = lmv->tgts[0];
139 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
142 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
143 enum obd_notify_event ev, void *data)
145 struct obd_connect_data *conn_data;
146 struct lmv_obd *lmv = &obd->u.lmv;
147 struct obd_uuid *uuid;
151 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
152 CERROR("unexpected notification of %s %s!\n",
153 watched->obd_type->typ_name,
158 uuid = &watched->u.cli.cl_target_uuid;
159 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
161 * Set MDC as active before notifying the observer, so the
162 * observer can use the MDC normally.
164 rc = lmv_set_mdc_active(lmv, uuid,
165 ev == OBD_NOTIFY_ACTIVE);
167 CERROR("%sactivation of %s failed: %d\n",
168 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
172 } else if (ev == OBD_NOTIFY_OCD) {
173 conn_data = &watched->u.cli.cl_import->imp_connect_data;
175 * XXX: Make sure that ocd_connect_flags from all targets are
176 * the same. Otherwise one of MDTs runs wrong version or
177 * something like this. --umka
179 obd->obd_self_export->exp_connect_data = *conn_data;
182 else if (ev == OBD_NOTIFY_DISCON) {
184 * For disconnect event, flush fld cache for failout MDS case.
186 fld_client_flush(&lmv->lmv_fld);
190 * Pass the notification up the chain.
192 if (obd->obd_observer)
193 rc = obd_notify(obd->obd_observer, watched, ev, data);
199 * This is fake connect function. Its purpose is to initialize lmv and say
200 * caller that everything is okay. Real connection will be performed later.
202 static int lmv_connect(const struct lu_env *env,
203 struct obd_export **exp, struct obd_device *obd,
204 struct obd_uuid *cluuid, struct obd_connect_data *data,
208 struct proc_dir_entry *lmv_proc_dir;
210 struct lmv_obd *lmv = &obd->u.lmv;
211 struct lustre_handle conn = { 0 };
216 * We don't want to actually do the underlying connections more than
217 * once, so keep track.
220 if (lmv->refcount > 1) {
225 rc = class_connect(&conn, obd, cluuid);
227 CERROR("class_connection() returned %d\n", rc);
231 *exp = class_conn2export(&conn);
232 class_export_get(*exp);
236 lmv->cluuid = *cluuid;
239 lmv->conn_data = *data;
242 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
244 if (IS_ERR(lmv_proc_dir)) {
245 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
246 obd->obd_type->typ_name, obd->obd_name);
252 * All real clients should perform actual connection right away, because
253 * it is possible, that LMV will not have opportunity to connect targets
254 * and MDC stuff will be called directly, for instance while reading
255 * ../mdc/../kbytesfree procfs file, etc.
257 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
258 rc = lmv_check_connect(obd);
263 lprocfs_remove(&lmv_proc_dir);
270 static void lmv_set_timeouts(struct obd_device *obd)
276 if (lmv->server_timeout == 0)
279 if (lmv->connected == 0)
282 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
283 struct lmv_tgt_desc *tgt = lmv->tgts[i];
285 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
288 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
289 KEY_INTERMDS, 0, NULL, NULL);
293 static int lmv_init_ea_size(struct obd_export *exp, int easize,
294 int def_easize, int cookiesize)
296 struct obd_device *obd = exp->exp_obd;
297 struct lmv_obd *lmv = &obd->u.lmv;
303 if (lmv->max_easize < easize) {
304 lmv->max_easize = easize;
307 if (lmv->max_def_easize < def_easize) {
308 lmv->max_def_easize = def_easize;
311 if (lmv->max_cookiesize < cookiesize) {
312 lmv->max_cookiesize = cookiesize;
318 if (lmv->connected == 0)
321 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
322 struct lmv_tgt_desc *tgt = lmv->tgts[i];
324 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
325 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
329 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
332 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
333 " rc = %d.\n", obd->obd_name, i, rc);
340 #define MAX_STRING_SIZE 128
342 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
345 struct proc_dir_entry *lmv_proc_dir;
347 struct lmv_obd *lmv = &obd->u.lmv;
348 struct obd_uuid *cluuid = &lmv->cluuid;
349 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
350 struct obd_device *mdc_obd;
351 struct obd_export *mdc_exp;
352 struct lu_fld_target target;
356 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
359 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
363 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
364 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
365 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
368 if (!mdc_obd->obd_set_up) {
369 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
373 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
374 &lmv->conn_data, NULL);
376 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
381 * Init fid sequence client for this mdc and add new fld target.
383 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
387 target.ft_srv = NULL;
388 target.ft_exp = mdc_exp;
389 target.ft_idx = tgt->ltd_idx;
391 fld_client_add_target(&lmv->lmv_fld, &target);
393 rc = obd_register_observer(mdc_obd, obd);
395 obd_disconnect(mdc_exp);
396 CERROR("target %s register_observer error %d\n",
397 tgt->ltd_uuid.uuid, rc);
401 if (obd->obd_observer) {
403 * Tell the observer about the new target.
405 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
407 (void *)(tgt - lmv->tgts[0]));
409 obd_disconnect(mdc_exp);
415 tgt->ltd_exp = mdc_exp;
416 lmv->desc.ld_active_tgt_count++;
418 md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
419 lmv->max_def_easize, lmv->max_cookiesize);
421 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
422 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
423 cfs_atomic_read(&obd->obd_refcount));
426 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
428 struct proc_dir_entry *mdc_symlink;
430 LASSERT(mdc_obd->obd_type != NULL);
431 LASSERT(mdc_obd->obd_type->typ_name != NULL);
432 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
435 mdc_obd->obd_type->typ_name,
437 if (mdc_symlink == NULL) {
438 CERROR("Could not register LMV target "
439 "/proc/fs/lustre/%s/%s/target_obds/%s.",
440 obd->obd_type->typ_name, obd->obd_name,
442 lprocfs_remove(&lmv_proc_dir);
450 static void lmv_del_target(struct lmv_obd *lmv, int index)
452 if (lmv->tgts[index] == NULL)
455 OBD_FREE_PTR(lmv->tgts[index]);
456 lmv->tgts[index] = NULL;
460 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
461 __u32 index, int gen)
463 struct lmv_obd *lmv = &obd->u.lmv;
464 struct lmv_tgt_desc *tgt;
468 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
472 if (lmv->desc.ld_tgt_count == 0) {
473 struct obd_device *mdc_obd;
475 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
478 lmv_init_unlock(lmv);
479 CERROR("%s: Target %s not attached: rc = %d\n",
480 obd->obd_name, uuidp->uuid, -EINVAL);
485 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
486 tgt = lmv->tgts[index];
487 CERROR("%s: UUID %s already assigned at LOV target index %d:"
488 " rc = %d\n", obd->obd_name,
489 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
490 lmv_init_unlock(lmv);
494 if (index >= lmv->tgts_size) {
495 /* We need to reallocate the lmv target array. */
496 struct lmv_tgt_desc **newtgts, **old = NULL;
500 while (newsize < index + 1)
501 newsize = newsize << 1;
502 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
503 if (newtgts == NULL) {
504 lmv_init_unlock(lmv);
508 if (lmv->tgts_size) {
509 memcpy(newtgts, lmv->tgts,
510 sizeof(*newtgts) * lmv->tgts_size);
512 oldsize = lmv->tgts_size;
516 lmv->tgts_size = newsize;
519 OBD_FREE(old, sizeof(*old) * oldsize);
521 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
527 lmv_init_unlock(lmv);
531 mutex_init(&tgt->ltd_fid_mutex);
532 tgt->ltd_idx = index;
533 tgt->ltd_uuid = *uuidp;
535 lmv->tgts[index] = tgt;
536 if (index >= lmv->desc.ld_tgt_count)
537 lmv->desc.ld_tgt_count = index + 1;
539 if (lmv->connected) {
540 rc = lmv_connect_mdc(obd, tgt);
542 spin_lock(&lmv->lmv_lock);
543 lmv->desc.ld_tgt_count--;
544 memset(tgt, 0, sizeof(*tgt));
545 spin_unlock(&lmv->lmv_lock);
547 int easize = sizeof(struct lmv_stripe_md) +
548 lmv->desc.ld_tgt_count *
549 sizeof(struct lu_fid);
550 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
554 lmv_init_unlock(lmv);
558 int lmv_check_connect(struct obd_device *obd)
560 struct lmv_obd *lmv = &obd->u.lmv;
561 struct lmv_tgt_desc *tgt;
571 if (lmv->connected) {
572 lmv_init_unlock(lmv);
576 if (lmv->desc.ld_tgt_count == 0) {
577 lmv_init_unlock(lmv);
578 CERROR("%s: no targets configured.\n", obd->obd_name);
582 LASSERT(lmv->tgts != NULL);
584 if (lmv->tgts[0] == NULL) {
585 lmv_init_unlock(lmv);
586 CERROR("%s: no target configured for index 0.\n",
591 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
592 lmv->cluuid.uuid, obd->obd_name);
594 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
598 rc = lmv_connect_mdc(obd, tgt);
603 lmv_set_timeouts(obd);
604 class_export_put(lmv->exp);
606 easize = lmv_get_easize(lmv);
607 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
608 lmv_init_unlock(lmv);
619 --lmv->desc.ld_active_tgt_count;
620 rc2 = obd_disconnect(tgt->ltd_exp);
622 CERROR("LMV target %s disconnect on "
623 "MDC idx %d: error %d\n",
624 tgt->ltd_uuid.uuid, i, rc2);
628 class_disconnect(lmv->exp);
629 lmv_init_unlock(lmv);
633 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
636 struct proc_dir_entry *lmv_proc_dir;
638 struct lmv_obd *lmv = &obd->u.lmv;
639 struct obd_device *mdc_obd;
643 LASSERT(tgt != NULL);
644 LASSERT(obd != NULL);
646 mdc_obd = class_exp2obd(tgt->ltd_exp);
649 mdc_obd->obd_force = obd->obd_force;
650 mdc_obd->obd_fail = obd->obd_fail;
651 mdc_obd->obd_no_recov = obd->obd_no_recov;
655 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
657 struct proc_dir_entry *mdc_symlink;
659 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
661 lprocfs_remove(&mdc_symlink);
663 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
664 obd->obd_type->typ_name, obd->obd_name,
669 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
671 CERROR("Can't finanize fids factory\n");
673 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
674 tgt->ltd_exp->exp_obd->obd_name,
675 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
677 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
678 rc = obd_disconnect(tgt->ltd_exp);
680 if (tgt->ltd_active) {
681 CERROR("Target %s disconnect error %d\n",
682 tgt->ltd_uuid.uuid, rc);
686 lmv_activate_target(lmv, tgt, 0);
691 static int lmv_disconnect(struct obd_export *exp)
693 struct obd_device *obd = class_exp2obd(exp);
695 struct proc_dir_entry *lmv_proc_dir;
697 struct lmv_obd *lmv = &obd->u.lmv;
706 * Only disconnect the underlying layers on the final disconnect.
709 if (lmv->refcount != 0)
712 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
713 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
716 lmv_disconnect_mdc(obd, lmv->tgts[i]);
720 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
722 lprocfs_remove(&lmv_proc_dir);
724 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
725 obd->obd_type->typ_name, obd->obd_name);
731 * This is the case when no real connection is established by
732 * lmv_check_connect().
735 class_export_put(exp);
736 rc = class_disconnect(exp);
737 if (lmv->refcount == 0)
742 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
744 struct obd_device *obddev = class_exp2obd(exp);
745 struct lmv_obd *lmv = &obddev->u.lmv;
746 struct getinfo_fid2path *gf;
747 struct lmv_tgt_desc *tgt;
748 struct getinfo_fid2path *remote_gf = NULL;
749 int remote_gf_size = 0;
752 gf = (struct getinfo_fid2path *)karg;
753 tgt = lmv_find_target(lmv, &gf->gf_fid);
755 RETURN(PTR_ERR(tgt));
758 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
759 if (rc != 0 && rc != -EREMOTE)
760 GOTO(out_fid2path, rc);
762 /* If remote_gf != NULL, it means just building the
763 * path on the remote MDT, copy this path segement to gf */
764 if (remote_gf != NULL) {
765 struct getinfo_fid2path *ori_gf;
768 ori_gf = (struct getinfo_fid2path *)karg;
769 if (strlen(ori_gf->gf_path) +
770 strlen(gf->gf_path) > ori_gf->gf_pathlen)
771 GOTO(out_fid2path, rc = -EOVERFLOW);
773 ptr = ori_gf->gf_path;
775 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
776 strlen(ori_gf->gf_path));
778 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
779 ptr += strlen(gf->gf_path);
783 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
784 tgt->ltd_exp->exp_obd->obd_name,
785 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
789 GOTO(out_fid2path, rc);
791 /* sigh, has to go to another MDT to do path building further */
792 if (remote_gf == NULL) {
793 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
794 OBD_ALLOC(remote_gf, remote_gf_size);
795 if (remote_gf == NULL)
796 GOTO(out_fid2path, rc = -ENOMEM);
797 remote_gf->gf_pathlen = PATH_MAX;
800 if (!fid_is_sane(&gf->gf_fid)) {
801 CERROR("%s: invalid FID "DFID": rc = %d\n",
802 tgt->ltd_exp->exp_obd->obd_name,
803 PFID(&gf->gf_fid), -EINVAL);
804 GOTO(out_fid2path, rc = -EINVAL);
807 tgt = lmv_find_target(lmv, &gf->gf_fid);
809 GOTO(out_fid2path, rc = -EINVAL);
811 remote_gf->gf_fid = gf->gf_fid;
812 remote_gf->gf_recno = -1;
813 remote_gf->gf_linkno = -1;
814 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
816 goto repeat_fid2path;
819 if (remote_gf != NULL)
820 OBD_FREE(remote_gf, remote_gf_size);
824 static int lmv_hsm_req_count(struct lmv_obd *lmv,
825 const struct hsm_user_request *hur,
826 const struct lmv_tgt_desc *tgt_mds)
830 struct lmv_tgt_desc *curr_tgt;
832 /* count how many requests must be sent to the given target */
833 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
834 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
835 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
841 static void lmv_hsm_req_build(struct lmv_obd *lmv,
842 struct hsm_user_request *hur_in,
843 const struct lmv_tgt_desc *tgt_mds,
844 struct hsm_user_request *hur_out)
847 struct lmv_tgt_desc *curr_tgt;
849 /* build the hsm_user_request for the given target */
850 hur_out->hur_request = hur_in->hur_request;
852 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
853 curr_tgt = lmv_find_target(lmv,
854 &hur_in->hur_user_item[i].hui_fid);
855 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
856 hur_out->hur_user_item[nr_out] =
857 hur_in->hur_user_item[i];
861 hur_out->hur_request.hr_itemcount = nr_out;
862 memcpy(hur_data(hur_out), hur_data(hur_in),
863 hur_in->hur_request.hr_data_len);
866 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
867 struct lustre_kernelcomm *lk, void *uarg)
871 struct kkuc_ct_data *kcd = NULL;
874 /* unregister request (call from llapi_hsm_copytool_fini) */
875 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
876 struct lmv_tgt_desc *tgt = lmv->tgts[i];
878 if (tgt == NULL || tgt->ltd_exp == NULL)
880 /* best effort: try to clean as much as possible
881 * (continue on error) */
882 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
885 /* Whatever the result, remove copytool from kuc groups.
886 * Unreached coordinators will get EPIPE on next requests
887 * and will unregister automatically.
889 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
896 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
897 struct lustre_kernelcomm *lk, void *uarg)
902 bool any_set = false;
903 struct kkuc_ct_data *kcd;
906 /* All or nothing: try to register to all MDS.
907 * In case of failure, unregister from previous MDS,
908 * except if it because of inactive target. */
909 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
910 struct lmv_tgt_desc *tgt = lmv->tgts[i];
912 if (tgt == NULL || tgt->ltd_exp == NULL)
914 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
916 if (tgt->ltd_active) {
917 /* permanent error */
918 CERROR("%s: iocontrol MDC %s on MDT"
919 " idx %d cmd %x: err = %d\n",
920 class_exp2obd(lmv->exp)->obd_name,
921 tgt->ltd_uuid.uuid, i, cmd, err);
923 lk->lk_flags |= LK_FLG_STOP;
924 /* unregister from previous MDS */
925 for (j = 0; j < i; j++) {
927 if (tgt == NULL || tgt->ltd_exp == NULL)
929 obd_iocontrol(cmd, tgt->ltd_exp, len,
934 /* else: transient error.
935 * kuc will register to the missing MDT
943 /* no registration done: return error */
946 /* at least one registration done, with no failure */
947 filp = fget(lk->lk_wfd);
956 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
957 kcd->kcd_uuid = lmv->cluuid;
958 kcd->kcd_archive = lk->lk_data;
960 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
973 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
974 int len, void *karg, void *uarg)
976 struct obd_device *obddev = class_exp2obd(exp);
977 struct lmv_obd *lmv = &obddev->u.lmv;
978 struct lmv_tgt_desc *tgt = NULL;
982 __u32 count = lmv->desc.ld_tgt_count;
989 case IOC_OBD_STATFS: {
990 struct obd_ioctl_data *data = karg;
991 struct obd_device *mdc_obd;
992 struct obd_statfs stat_buf = {0};
995 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
996 if ((index >= count))
999 tgt = lmv->tgts[index];
1000 if (tgt == NULL || !tgt->ltd_active)
1003 mdc_obd = class_exp2obd(tgt->ltd_exp);
1008 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1009 min((int) data->ioc_plen2,
1010 (int) sizeof(struct obd_uuid))))
1013 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1014 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1018 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1019 min((int) data->ioc_plen1,
1020 (int) sizeof(stat_buf))))
1024 case OBD_IOC_QUOTACTL: {
1025 struct if_quotactl *qctl = karg;
1026 struct obd_quotactl *oqctl;
1028 if (qctl->qc_valid == QC_MDTIDX) {
1029 if (count <= qctl->qc_idx)
1032 tgt = lmv->tgts[qctl->qc_idx];
1033 if (tgt == NULL || tgt->ltd_exp == NULL)
1035 } else if (qctl->qc_valid == QC_UUID) {
1036 for (i = 0; i < count; i++) {
1040 if (!obd_uuid_equals(&tgt->ltd_uuid,
1044 if (tgt->ltd_exp == NULL)
1056 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1057 OBD_ALLOC_PTR(oqctl);
1061 QCTL_COPY(oqctl, qctl);
1062 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1064 QCTL_COPY(qctl, oqctl);
1065 qctl->qc_valid = QC_MDTIDX;
1066 qctl->obd_uuid = tgt->ltd_uuid;
1068 OBD_FREE_PTR(oqctl);
1071 case OBD_IOC_CHANGELOG_SEND:
1072 case OBD_IOC_CHANGELOG_CLEAR: {
1073 struct ioc_changelog *icc = karg;
1075 if (icc->icc_mdtindex >= count)
1078 tgt = lmv->tgts[icc->icc_mdtindex];
1079 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1081 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1084 case LL_IOC_GET_CONNECT_FLAGS: {
1086 if (tgt == NULL || tgt->ltd_exp == NULL)
1088 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1091 case OBD_IOC_FID2PATH: {
1092 rc = lmv_fid2path(exp, len, karg, uarg);
1095 case LL_IOC_HSM_STATE_GET:
1096 case LL_IOC_HSM_STATE_SET:
1097 case LL_IOC_HSM_ACTION: {
1098 struct md_op_data *op_data = karg;
1100 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1102 RETURN(PTR_ERR(tgt));
1104 if (tgt->ltd_exp == NULL)
1107 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1110 case LL_IOC_HSM_PROGRESS: {
1111 const struct hsm_progress_kernel *hpk = karg;
1113 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1115 RETURN(PTR_ERR(tgt));
1116 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1119 case LL_IOC_HSM_REQUEST: {
1120 struct hsm_user_request *hur = karg;
1121 unsigned int reqcount = hur->hur_request.hr_itemcount;
1126 /* if the request is about a single fid
1127 * or if there is a single MDS, no need to split
1129 if (reqcount == 1 || count == 1) {
1130 tgt = lmv_find_target(lmv,
1131 &hur->hur_user_item[0].hui_fid);
1133 RETURN(PTR_ERR(tgt));
1134 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1136 /* split fid list to their respective MDS */
1137 for (i = 0; i < count; i++) {
1138 unsigned int nr, reqlen;
1140 struct hsm_user_request *req;
1143 if (tgt == NULL || tgt->ltd_exp == NULL)
1146 nr = lmv_hsm_req_count(lmv, hur, tgt);
1147 if (nr == 0) /* nothing for this MDS */
1150 /* build a request with fids for this MDS */
1151 reqlen = offsetof(typeof(*hur),
1153 + hur->hur_request.hr_data_len;
1154 OBD_ALLOC_LARGE(req, reqlen);
1158 lmv_hsm_req_build(lmv, hur, tgt, req);
1160 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1162 if (rc1 != 0 && rc == 0)
1164 OBD_FREE_LARGE(req, reqlen);
1169 case LL_IOC_LOV_SWAP_LAYOUTS: {
1170 struct md_op_data *op_data = karg;
1171 struct lmv_tgt_desc *tgt1, *tgt2;
1173 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1175 RETURN(PTR_ERR(tgt1));
1177 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1179 RETURN(PTR_ERR(tgt2));
1181 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1184 /* only files on same MDT can have their layouts swapped */
1185 if (tgt1->ltd_idx != tgt2->ltd_idx)
1188 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1191 case LL_IOC_HSM_CT_START: {
1192 struct lustre_kernelcomm *lk = karg;
1193 if (lk->lk_flags & LK_FLG_STOP)
1194 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1196 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1200 for (i = 0; i < count; i++) {
1201 struct obd_device *mdc_obd;
1205 if (tgt == NULL || tgt->ltd_exp == NULL)
1207 /* ll_umount_begin() sets force flag but for lmv, not
1208 * mdc. Let's pass it through */
1209 mdc_obd = class_exp2obd(tgt->ltd_exp);
1210 mdc_obd->obd_force = obddev->obd_force;
1211 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1212 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1215 if (tgt->ltd_active) {
1216 CERROR("error: iocontrol MDC %s on MDT"
1217 " idx %d cmd %x: err = %d\n",
1218 tgt->ltd_uuid.uuid, i, cmd, err);
1232 static int lmv_all_chars_policy(int count, const char *name,
1243 static int lmv_nid_policy(struct lmv_obd *lmv)
1245 struct obd_import *imp;
1249 * XXX: To get nid we assume that underlying obd device is mdc.
1251 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1252 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1253 return id % lmv->desc.ld_tgt_count;
1256 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1257 placement_policy_t placement)
1259 switch (placement) {
1260 case PLACEMENT_CHAR_POLICY:
1261 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1263 op_data->op_namelen);
1264 case PLACEMENT_NID_POLICY:
1265 return lmv_nid_policy(lmv);
1271 CERROR("Unsupported placement policy %x\n", placement);
1277 * This is _inode_ placement policy function (not name).
1279 static int lmv_placement_policy(struct obd_device *obd,
1280 struct md_op_data *op_data,
1283 struct lmv_obd *lmv = &obd->u.lmv;
1286 LASSERT(mds != NULL);
1288 if (lmv->desc.ld_tgt_count == 1) {
1294 * If stripe_offset is provided during setdirstripe
1295 * (setdirstripe -i xx), xx MDS will be choosen.
1297 if (op_data->op_cli_flags & CLI_SET_MEA) {
1298 struct lmv_user_md *lum;
1300 lum = (struct lmv_user_md *)op_data->op_data;
1301 if (lum->lum_type == LMV_STRIPE_TYPE &&
1302 lum->lum_stripe_offset != -1) {
1303 if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1304 CERROR("%s: Stripe_offset %d > MDT count %d:"
1305 " rc = %d\n", obd->obd_name,
1306 lum->lum_stripe_offset,
1307 lmv->desc.ld_tgt_count, -ERANGE);
1310 *mds = lum->lum_stripe_offset;
1315 /* Allocate new fid on target according to operation type and parent
1317 *mds = op_data->op_mds;
1321 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1324 struct lmv_tgt_desc *tgt;
1328 tgt = lmv_get_target(lmv, mds);
1330 RETURN(PTR_ERR(tgt));
1333 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1334 * on server that seq in new allocated fid is not yet known.
1336 mutex_lock(&tgt->ltd_fid_mutex);
1338 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1339 GOTO(out, rc = -ENODEV);
1342 * Asking underlaying tgt layer to allocate new fid.
1344 rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1346 LASSERT(fid_is_sane(fid));
1352 mutex_unlock(&tgt->ltd_fid_mutex);
1356 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1357 struct md_op_data *op_data)
1359 struct obd_device *obd = class_exp2obd(exp);
1360 struct lmv_obd *lmv = &obd->u.lmv;
1365 LASSERT(op_data != NULL);
1366 LASSERT(fid != NULL);
1368 rc = lmv_placement_policy(obd, op_data, &mds);
1370 CERROR("Can't get target for allocating fid, "
1375 rc = __lmv_fid_alloc(lmv, fid, mds);
1377 CERROR("Can't alloc new fid, rc %d\n", rc);
1384 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1386 struct lmv_obd *lmv = &obd->u.lmv;
1387 struct lprocfs_static_vars lvars;
1388 struct lmv_desc *desc;
1392 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1393 CERROR("LMV setup requires a descriptor\n");
1397 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1398 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1399 CERROR("Lmv descriptor size wrong: %d > %d\n",
1400 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1404 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1405 if (lmv->tgts == NULL)
1407 lmv->tgts_size = 32;
1409 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1410 lmv->desc.ld_tgt_count = 0;
1411 lmv->desc.ld_active_tgt_count = 0;
1412 lmv->max_cookiesize = 0;
1413 lmv->max_def_easize = 0;
1414 lmv->max_easize = 0;
1415 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1417 spin_lock_init(&lmv->lmv_lock);
1418 mutex_init(&lmv->init_mutex);
1420 lprocfs_lmv_init_vars(&lvars);
1422 lprocfs_obd_setup(obd, lvars.obd_vars);
1423 lprocfs_alloc_md_stats(obd, 0);
1426 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1427 0444, &lmv_proc_target_fops, obd);
1429 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1433 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1434 LUSTRE_CLI_FLD_HASH_DHT);
1436 CERROR("Can't init FLD, err %d\n", rc);
1446 static int lmv_cleanup(struct obd_device *obd)
1448 struct lmv_obd *lmv = &obd->u.lmv;
1451 fld_client_fini(&lmv->lmv_fld);
1452 if (lmv->tgts != NULL) {
1454 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1455 if (lmv->tgts[i] == NULL)
1457 lmv_del_target(lmv, i);
1459 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1465 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1467 struct lustre_cfg *lcfg = buf;
1468 struct obd_uuid obd_uuid;
1474 switch (lcfg->lcfg_command) {
1476 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1477 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1478 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1479 GOTO(out, rc = -EINVAL);
1481 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1483 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1484 GOTO(out, rc = -EINVAL);
1485 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1486 GOTO(out, rc = -EINVAL);
1487 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1490 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1491 GOTO(out, rc = -EINVAL);
1497 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1498 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1500 struct obd_device *obd = class_exp2obd(exp);
1501 struct lmv_obd *lmv = &obd->u.lmv;
1502 struct obd_statfs *temp;
1507 rc = lmv_check_connect(obd);
1511 OBD_ALLOC(temp, sizeof(*temp));
1515 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1516 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1519 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1522 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1523 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1525 GOTO(out_free_temp, rc);
1530 /* If the statfs is from mount, it will needs
1531 * retrieve necessary information from MDT0.
1532 * i.e. mount does not need the merged osfs
1534 * And also clients can be mounted as long as
1535 * MDT0 is in service*/
1536 if (flags & OBD_STATFS_FOR_MDT0)
1537 GOTO(out_free_temp, rc);
1539 osfs->os_bavail += temp->os_bavail;
1540 osfs->os_blocks += temp->os_blocks;
1541 osfs->os_ffree += temp->os_ffree;
1542 osfs->os_files += temp->os_files;
1548 OBD_FREE(temp, sizeof(*temp));
1552 static int lmv_getstatus(struct obd_export *exp,
1554 struct obd_capa **pc)
1556 struct obd_device *obd = exp->exp_obd;
1557 struct lmv_obd *lmv = &obd->u.lmv;
1561 rc = lmv_check_connect(obd);
1565 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1569 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1570 struct obd_capa *oc, obd_valid valid, const char *name,
1571 const char *input, int input_size, int output_size,
1572 int flags, struct ptlrpc_request **request)
1574 struct obd_device *obd = exp->exp_obd;
1575 struct lmv_obd *lmv = &obd->u.lmv;
1576 struct lmv_tgt_desc *tgt;
1580 rc = lmv_check_connect(obd);
1584 tgt = lmv_find_target(lmv, fid);
1586 RETURN(PTR_ERR(tgt));
1588 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1589 input_size, output_size, flags, request);
1594 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1595 struct obd_capa *oc, obd_valid valid, const char *name,
1596 const char *input, int input_size, int output_size,
1597 int flags, __u32 suppgid,
1598 struct ptlrpc_request **request)
1600 struct obd_device *obd = exp->exp_obd;
1601 struct lmv_obd *lmv = &obd->u.lmv;
1602 struct lmv_tgt_desc *tgt;
1606 rc = lmv_check_connect(obd);
1610 tgt = lmv_find_target(lmv, fid);
1612 RETURN(PTR_ERR(tgt));
1614 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1615 input_size, output_size, flags, suppgid,
1621 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1622 struct ptlrpc_request **request)
1624 struct obd_device *obd = exp->exp_obd;
1625 struct lmv_obd *lmv = &obd->u.lmv;
1626 struct lmv_tgt_desc *tgt;
1630 rc = lmv_check_connect(obd);
1634 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1636 RETURN(PTR_ERR(tgt));
1638 if (op_data->op_flags & MF_GET_MDT_IDX) {
1639 op_data->op_mds = tgt->ltd_idx;
1643 rc = md_getattr(tgt->ltd_exp, op_data, request);
1648 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1650 struct obd_device *obd = exp->exp_obd;
1651 struct lmv_obd *lmv = &obd->u.lmv;
1656 rc = lmv_check_connect(obd);
1660 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1663 * With DNE every object can have two locks in different namespaces:
1664 * lookup lock in space of MDT storing direntry and update/open lock in
1665 * space of MDT storing inode.
1667 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1668 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1670 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1676 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1677 ldlm_iterator_t it, void *data)
1679 struct obd_device *obd = exp->exp_obd;
1680 struct lmv_obd *lmv = &obd->u.lmv;
1685 rc = lmv_check_connect(obd);
1689 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1692 * With DNE every object can have two locks in different namespaces:
1693 * lookup lock in space of MDT storing direntry and update/open lock in
1694 * space of MDT storing inode.
1696 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1697 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1699 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1708 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1709 struct md_open_data *mod, struct ptlrpc_request **request)
1711 struct obd_device *obd = exp->exp_obd;
1712 struct lmv_obd *lmv = &obd->u.lmv;
1713 struct lmv_tgt_desc *tgt;
1717 rc = lmv_check_connect(obd);
1721 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1723 RETURN(PTR_ERR(tgt));
1725 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1726 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1731 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1734 struct lmv_tgt_desc *tgt;
1736 tgt = lmv_find_target(lmv, fid);
1740 op_data->op_mds = tgt->ltd_idx;
1745 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1746 const void *data, int datalen, int mode, __u32 uid,
1747 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1748 struct ptlrpc_request **request)
1750 struct obd_device *obd = exp->exp_obd;
1751 struct lmv_obd *lmv = &obd->u.lmv;
1752 struct lmv_tgt_desc *tgt;
1756 rc = lmv_check_connect(obd);
1760 if (!lmv->desc.ld_active_tgt_count)
1763 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1765 RETURN(PTR_ERR(tgt));
1767 rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1771 CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1772 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1775 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1776 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1777 cap_effective, rdev, request);
1780 if (*request == NULL)
1782 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1787 static int lmv_done_writing(struct obd_export *exp,
1788 struct md_op_data *op_data,
1789 struct md_open_data *mod)
1791 struct obd_device *obd = exp->exp_obd;
1792 struct lmv_obd *lmv = &obd->u.lmv;
1793 struct lmv_tgt_desc *tgt;
1797 rc = lmv_check_connect(obd);
1801 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1803 RETURN(PTR_ERR(tgt));
1805 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1810 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1811 struct lookup_intent *it, struct md_op_data *op_data,
1812 struct lustre_handle *lockh, void *lmm, int lmmsize,
1813 __u64 extra_lock_flags)
1815 struct ptlrpc_request *req = it->d.lustre.it_data;
1816 struct obd_device *obd = exp->exp_obd;
1817 struct lmv_obd *lmv = &obd->u.lmv;
1818 struct lustre_handle plock;
1819 struct lmv_tgt_desc *tgt;
1820 struct md_op_data *rdata;
1822 struct mdt_body *body;
1827 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1828 LASSERT(body != NULL);
1830 if (!(body->valid & OBD_MD_MDS))
1833 CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1834 LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1837 * We got LOOKUP lock, but we really need attrs.
1839 pmode = it->d.lustre.it_lock_mode;
1840 LASSERT(pmode != 0);
1841 memcpy(&plock, lockh, sizeof(plock));
1842 it->d.lustre.it_lock_mode = 0;
1843 it->d.lustre.it_data = NULL;
1846 it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1847 ptlrpc_req_finished(req);
1849 tgt = lmv_find_target(lmv, &fid1);
1851 GOTO(out, rc = PTR_ERR(tgt));
1853 OBD_ALLOC_PTR(rdata);
1855 GOTO(out, rc = -ENOMEM);
1857 rdata->op_fid1 = fid1;
1858 rdata->op_bias = MDS_CROSS_REF;
1860 rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1861 lmm, lmmsize, NULL, extra_lock_flags);
1862 OBD_FREE_PTR(rdata);
1865 ldlm_lock_decref(&plock, pmode);
1870 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1871 struct lookup_intent *it, struct md_op_data *op_data,
1872 struct lustre_handle *lockh, void *lmm, int lmmsize,
1873 struct ptlrpc_request **req, __u64 extra_lock_flags)
1875 struct obd_device *obd = exp->exp_obd;
1876 struct lmv_obd *lmv = &obd->u.lmv;
1877 struct lmv_tgt_desc *tgt;
1881 rc = lmv_check_connect(obd);
1885 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1886 LL_IT2STR(it), PFID(&op_data->op_fid1));
1888 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1890 RETURN(PTR_ERR(tgt));
1892 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1893 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1895 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1896 lmm, lmmsize, req, extra_lock_flags);
1898 if (rc == 0 && it && it->it_op == IT_OPEN) {
1899 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1900 lmm, lmmsize, extra_lock_flags);
1906 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1907 struct ptlrpc_request **request)
1909 struct ptlrpc_request *req = NULL;
1910 struct obd_device *obd = exp->exp_obd;
1911 struct lmv_obd *lmv = &obd->u.lmv;
1912 struct lmv_tgt_desc *tgt;
1913 struct mdt_body *body;
1917 rc = lmv_check_connect(obd);
1921 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1923 RETURN(PTR_ERR(tgt));
1925 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1926 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1929 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1933 body = req_capsule_server_get(&(*request)->rq_pill,
1935 LASSERT(body != NULL);
1937 if (body->valid & OBD_MD_MDS) {
1938 struct lu_fid rid = body->fid1;
1939 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1942 tgt = lmv_find_target(lmv, &rid);
1944 ptlrpc_req_finished(*request);
1945 RETURN(PTR_ERR(tgt));
1948 op_data->op_fid1 = rid;
1949 op_data->op_valid |= OBD_MD_FLCROSSREF;
1950 op_data->op_namelen = 0;
1951 op_data->op_name = NULL;
1952 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1953 ptlrpc_req_finished(*request);
1960 #define md_op_data_fid(op_data, fl) \
1961 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1962 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1963 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1964 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1967 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1968 int op_tgt, ldlm_mode_t mode, int bits, int flag)
1970 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1971 struct obd_device *obd = exp->exp_obd;
1972 struct lmv_obd *lmv = &obd->u.lmv;
1973 struct lmv_tgt_desc *tgt;
1974 ldlm_policy_data_t policy = {{0}};
1978 if (!fid_is_sane(fid))
1981 tgt = lmv_find_target(lmv, fid);
1983 RETURN(PTR_ERR(tgt));
1985 if (tgt->ltd_idx != op_tgt) {
1986 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1987 policy.l_inodebits.bits = bits;
1988 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1989 mode, LCF_ASYNC, NULL);
1992 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1994 op_data->op_flags |= flag;
2002 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2005 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2006 struct ptlrpc_request **request)
2008 struct obd_device *obd = exp->exp_obd;
2009 struct lmv_obd *lmv = &obd->u.lmv;
2010 struct lmv_tgt_desc *tgt;
2014 rc = lmv_check_connect(obd);
2018 LASSERT(op_data->op_namelen != 0);
2020 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2021 PFID(&op_data->op_fid2), op_data->op_namelen,
2022 op_data->op_name, PFID(&op_data->op_fid1));
2024 op_data->op_fsuid = current_fsuid();
2025 op_data->op_fsgid = current_fsgid();
2026 op_data->op_cap = cfs_curproc_cap_pack();
2027 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2029 RETURN(PTR_ERR(tgt));
2032 * Cancel UPDATE lock on child (fid1).
2034 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2035 rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2036 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2040 rc = md_link(tgt->ltd_exp, op_data, request);
2045 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2046 const char *old, int oldlen, const char *new, int newlen,
2047 struct ptlrpc_request **request)
2049 struct obd_device *obd = exp->exp_obd;
2050 struct lmv_obd *lmv = &obd->u.lmv;
2051 struct lmv_tgt_desc *src_tgt;
2052 struct lmv_tgt_desc *tgt_tgt;
2056 LASSERT(oldlen != 0);
2058 CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2059 oldlen, old, PFID(&op_data->op_fid1),
2060 newlen, new, PFID(&op_data->op_fid2));
2062 rc = lmv_check_connect(obd);
2066 op_data->op_fsuid = current_fsuid();
2067 op_data->op_fsgid = current_fsgid();
2068 op_data->op_cap = cfs_curproc_cap_pack();
2069 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2070 if (IS_ERR(src_tgt))
2071 RETURN(PTR_ERR(src_tgt));
2073 tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2074 if (IS_ERR(tgt_tgt))
2075 RETURN(PTR_ERR(tgt_tgt));
2077 * LOOKUP lock on src child (fid3) should also be cancelled for
2078 * src_tgt in mdc_rename.
2080 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2083 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2086 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2087 LCK_EX, MDS_INODELOCK_UPDATE,
2088 MF_MDC_CANCEL_FID2);
2091 * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2094 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2095 LCK_EX, MDS_INODELOCK_LOOKUP,
2096 MF_MDC_CANCEL_FID4);
2100 * Cancel all the locks on tgt child (fid4).
2103 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2104 LCK_EX, MDS_INODELOCK_FULL,
2105 MF_MDC_CANCEL_FID4);
2108 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2109 new, newlen, request);
2113 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2114 void *ea, int ealen, void *ea2, int ea2len,
2115 struct ptlrpc_request **request,
2116 struct md_open_data **mod)
2118 struct obd_device *obd = exp->exp_obd;
2119 struct lmv_obd *lmv = &obd->u.lmv;
2120 struct lmv_tgt_desc *tgt;
2124 rc = lmv_check_connect(obd);
2128 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2129 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2131 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2132 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2134 RETURN(PTR_ERR(tgt));
2136 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2137 ea2len, request, mod);
2142 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2143 struct obd_capa *oc, struct ptlrpc_request **request)
2145 struct obd_device *obd = exp->exp_obd;
2146 struct lmv_obd *lmv = &obd->u.lmv;
2147 struct lmv_tgt_desc *tgt;
2151 rc = lmv_check_connect(obd);
2155 tgt = lmv_find_target(lmv, fid);
2157 RETURN(PTR_ERR(tgt));
2159 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2164 * Adjust a set of pages, each page containing an array of lu_dirpages,
2165 * so that each page can be used as a single logical lu_dirpage.
2167 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2168 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2169 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
2170 * value is used as a cookie to request the next lu_dirpage in a
2171 * directory listing that spans multiple pages (two in this example):
2174 * .|--------v------- -----.
2175 * |s|e|f|p|ent|ent| ... |ent|
2176 * '--|-------------- -----' Each CFS_PAGE contains a single
2177 * '------. lu_dirpage.
2178 * .---------v------- -----.
2179 * |s|e|f|p|ent| 0 | ... | 0 |
2180 * '----------------- -----'
2182 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2183 * larger than LU_PAGE_SIZE, a single host page may contain multiple
2184 * lu_dirpages. After reading the lu_dirpages from the MDS, the
2185 * ldp_hash_end of the first lu_dirpage refers to the one immediately
2186 * after it in the same CFS_PAGE (arrows simplified for brevity, but
2187 * in general e0==s1, e1==s2, etc.):
2189 * .-------------------- -----.
2190 * |s0|e0|f0|p|ent|ent| ... |ent|
2191 * |---v---------------- -----|
2192 * |s1|e1|f1|p|ent|ent| ... |ent|
2193 * |---v---------------- -----| Here, each CFS_PAGE contains
2194 * ... multiple lu_dirpages.
2195 * |---v---------------- -----|
2196 * |s'|e'|f'|p|ent|ent| ... |ent|
2197 * '---|---------------- -----'
2199 * .----------------------------.
2202 * This structure is transformed into a single logical lu_dirpage as follows:
2204 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2205 * labeled 'next CFS_PAGE'.
2207 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2208 * a hash collision with the next page exists.
2210 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2211 * to the first entry of the next lu_dirpage.
2213 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2214 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2218 for (i = 0; i < ncfspgs; i++) {
2219 struct lu_dirpage *dp = kmap(pages[i]);
2220 struct lu_dirpage *first = dp;
2221 struct lu_dirent *end_dirent = NULL;
2222 struct lu_dirent *ent;
2223 __u64 hash_end = dp->ldp_hash_end;
2224 __u32 flags = dp->ldp_flags;
2226 while (--nlupgs > 0) {
2227 ent = lu_dirent_start(dp);
2228 for (end_dirent = ent; ent != NULL;
2229 end_dirent = ent, ent = lu_dirent_next(ent));
2231 /* Advance dp to next lu_dirpage. */
2232 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2234 /* Check if we've reached the end of the CFS_PAGE. */
2235 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2238 /* Save the hash and flags of this lu_dirpage. */
2239 hash_end = dp->ldp_hash_end;
2240 flags = dp->ldp_flags;
2242 /* Check if lu_dirpage contains no entries. */
2246 /* Enlarge the end entry lde_reclen from 0 to
2247 * first entry of next lu_dirpage. */
2248 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2249 end_dirent->lde_reclen =
2250 cpu_to_le16((char *)(dp->ldp_entries) -
2251 (char *)end_dirent);
2254 first->ldp_hash_end = hash_end;
2255 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2256 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2260 LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2263 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2264 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2266 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
2267 struct page **pages, struct ptlrpc_request **request)
2269 struct obd_device *obd = exp->exp_obd;
2270 struct lmv_obd *lmv = &obd->u.lmv;
2271 __u64 offset = op_data->op_offset;
2273 int ncfspgs; /* pages read in PAGE_CACHE_SIZE */
2274 int nlupgs; /* pages read in LU_PAGE_SIZE */
2275 struct lmv_tgt_desc *tgt;
2278 rc = lmv_check_connect(obd);
2282 CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
2283 offset, PFID(&op_data->op_fid1));
2285 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2287 RETURN(PTR_ERR(tgt));
2289 rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2293 ncfspgs = ((*request)->rq_bulk->bd_nob_transferred +
2294 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2295 nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2296 LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2297 LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages);
2299 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs,
2300 op_data->op_npages);
2302 lmv_adjust_dirpages(pages, ncfspgs, nlupgs);
2307 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2308 struct ptlrpc_request **request)
2310 struct obd_device *obd = exp->exp_obd;
2311 struct lmv_obd *lmv = &obd->u.lmv;
2312 struct lmv_tgt_desc *tgt = NULL;
2313 struct mdt_body *body;
2317 rc = lmv_check_connect(obd);
2321 /* Send unlink requests to the MDT where the child is located */
2322 if (likely(!fid_is_zero(&op_data->op_fid2)))
2323 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2325 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2327 RETURN(PTR_ERR(tgt));
2329 op_data->op_fsuid = current_fsuid();
2330 op_data->op_fsgid = current_fsgid();
2331 op_data->op_cap = cfs_curproc_cap_pack();
2334 * If child's fid is given, cancel unused locks for it if it is from
2335 * another export than parent.
2337 * LOOKUP lock for child (fid3) should also be cancelled on parent
2338 * tgt_tgt in mdc_unlink().
2340 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2343 * Cancel FULL locks on child (fid3).
2345 rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2346 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2351 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2352 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2354 rc = md_unlink(tgt->ltd_exp, op_data, request);
2355 if (rc != 0 && rc != -EREMOTE)
2358 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2362 /* Not cross-ref case, just get out of here. */
2363 if (likely(!(body->valid & OBD_MD_MDS)))
2366 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2367 exp->exp_obd->obd_name, PFID(&body->fid1));
2369 /* This is a remote object, try remote MDT, Note: it may
2370 * try more than 1 time here, Considering following case
2371 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2372 * 1. Initially A does not know where remote1 is, it send
2373 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2374 * resend unlink RPC to MDT1 (retry 1st time).
2376 * 2. During the unlink RPC in flight,
2377 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2378 * and create new remote1, but on MDT0
2380 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2381 * /mnt/lustre, then lookup get fid of remote1, and find
2382 * it is remote dir again, and replay -EREMOTE again.
2384 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2386 * In theory, it might try unlimited time here, but it should
2387 * be very rare case. */
2388 op_data->op_fid2 = body->fid1;
2389 ptlrpc_req_finished(*request);
2395 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2397 struct lmv_obd *lmv = &obd->u.lmv;
2401 case OBD_CLEANUP_EARLY:
2402 /* XXX: here should be calling obd_precleanup() down to
2405 case OBD_CLEANUP_EXPORTS:
2406 fld_client_proc_fini(&lmv->lmv_fld);
2407 lprocfs_obd_cleanup(obd);
2408 lprocfs_free_md_stats(obd);
2416 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2417 __u32 keylen, void *key, __u32 *vallen, void *val,
2418 struct lov_stripe_md *lsm)
2420 struct obd_device *obd;
2421 struct lmv_obd *lmv;
2425 obd = class_exp2obd(exp);
2427 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2428 exp->exp_handle.h_cookie);
2433 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2436 rc = lmv_check_connect(obd);
2440 LASSERT(*vallen == sizeof(__u32));
2441 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2442 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2444 * All tgts should be connected when this gets called.
2446 if (tgt == NULL || tgt->ltd_exp == NULL)
2449 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2454 } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2455 rc = lmv_check_connect(obd);
2460 * Forwarding this request to first MDS, it should know LOV
2463 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2465 if (!rc && KEY_IS(KEY_CONN_DATA))
2466 exp->exp_connect_data = *(struct obd_connect_data *)val;
2468 } else if (KEY_IS(KEY_TGT_COUNT)) {
2469 *((int *)val) = lmv->desc.ld_tgt_count;
2473 CDEBUG(D_IOCTL, "Invalid key\n");
2477 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2478 obd_count keylen, void *key, obd_count vallen,
2479 void *val, struct ptlrpc_request_set *set)
2481 struct lmv_tgt_desc *tgt;
2482 struct obd_device *obd;
2483 struct lmv_obd *lmv;
2487 obd = class_exp2obd(exp);
2489 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2490 exp->exp_handle.h_cookie);
2495 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2498 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2501 if (tgt == NULL || tgt->ltd_exp == NULL)
2504 err = obd_set_info_async(env, tgt->ltd_exp,
2505 keylen, key, vallen, val, set);
2516 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2517 struct lov_stripe_md *lsm)
2519 struct obd_device *obd = class_exp2obd(exp);
2520 struct lmv_obd *lmv = &obd->u.lmv;
2521 struct lmv_stripe_md *meap;
2522 struct lmv_stripe_md *lsmp;
2527 mea_size = lmv_get_easize(lmv);
2531 if (*lmmp && !lsm) {
2532 OBD_FREE_LARGE(*lmmp, mea_size);
2537 if (*lmmp == NULL) {
2538 OBD_ALLOC_LARGE(*lmmp, mea_size);
2546 lsmp = (struct lmv_stripe_md *)lsm;
2547 meap = (struct lmv_stripe_md *)*lmmp;
2549 if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2550 lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2553 meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2554 meap->mea_count = cpu_to_le32(lsmp->mea_count);
2555 meap->mea_master = cpu_to_le32(lsmp->mea_master);
2557 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2558 meap->mea_ids[i] = lsmp->mea_ids[i];
2559 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2565 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2566 struct lov_mds_md *lmm, int lmm_size)
2568 struct obd_device *obd = class_exp2obd(exp);
2569 struct lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp;
2570 struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm;
2571 struct lmv_obd *lmv = &obd->u.lmv;
2577 mea_size = lmv_get_easize(lmv);
2581 if (*lsmp != NULL && lmm == NULL) {
2582 OBD_FREE_LARGE(*tmea, mea_size);
2587 LASSERT(mea_size == lmm_size);
2589 OBD_ALLOC_LARGE(*tmea, mea_size);
2596 if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2597 mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2598 mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2600 magic = le32_to_cpu(mea->mea_magic);
2603 * Old mea is not handled here.
2605 CERROR("Old not supportable EA is found\n");
2609 (*tmea)->mea_magic = magic;
2610 (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2611 (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2613 for (i = 0; i < (*tmea)->mea_count; i++) {
2614 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2615 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2620 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2621 ldlm_policy_data_t *policy, ldlm_mode_t mode,
2622 ldlm_cancel_flags_t flags, void *opaque)
2624 struct obd_device *obd = exp->exp_obd;
2625 struct lmv_obd *lmv = &obd->u.lmv;
2631 LASSERT(fid != NULL);
2633 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2634 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2636 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2639 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2647 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2650 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2651 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2655 if (tgt == NULL || tgt->ltd_exp == NULL)
2657 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2661 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2662 const struct lu_fid *fid, ldlm_type_t type,
2663 ldlm_policy_data_t *policy, ldlm_mode_t mode,
2664 struct lustre_handle *lockh)
2666 struct obd_device *obd = exp->exp_obd;
2667 struct lmv_obd *lmv = &obd->u.lmv;
2672 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2675 * With CMD every object can have two locks in different namespaces:
2676 * lookup lock in space of mds storing direntry and update/open lock in
2677 * space of mds storing inode. Thus we check all targets, not only that
2678 * one fid was created in.
2680 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2681 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2683 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2686 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
2695 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2696 struct obd_export *dt_exp, struct obd_export *md_exp,
2697 struct lustre_md *md)
2699 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2700 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2702 if (tgt == NULL || tgt->ltd_exp == NULL)
2704 return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md);
2707 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2709 struct obd_device *obd = exp->exp_obd;
2710 struct lmv_obd *lmv = &obd->u.lmv;
2711 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2715 obd_free_memmd(exp, (void *)&md->mea);
2716 if (tgt == NULL || tgt->ltd_exp == NULL)
2718 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
2721 int lmv_set_open_replay_data(struct obd_export *exp,
2722 struct obd_client_handle *och,
2723 struct lookup_intent *it)
2725 struct obd_device *obd = exp->exp_obd;
2726 struct lmv_obd *lmv = &obd->u.lmv;
2727 struct lmv_tgt_desc *tgt;
2730 tgt = lmv_find_target(lmv, &och->och_fid);
2732 RETURN(PTR_ERR(tgt));
2734 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
2737 int lmv_clear_open_replay_data(struct obd_export *exp,
2738 struct obd_client_handle *och)
2740 struct obd_device *obd = exp->exp_obd;
2741 struct lmv_obd *lmv = &obd->u.lmv;
2742 struct lmv_tgt_desc *tgt;
2745 tgt = lmv_find_target(lmv, &och->och_fid);
2747 RETURN(PTR_ERR(tgt));
2749 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2752 static int lmv_get_remote_perm(struct obd_export *exp,
2753 const struct lu_fid *fid,
2754 struct obd_capa *oc, __u32 suppgid,
2755 struct ptlrpc_request **request)
2757 struct obd_device *obd = exp->exp_obd;
2758 struct lmv_obd *lmv = &obd->u.lmv;
2759 struct lmv_tgt_desc *tgt;
2763 rc = lmv_check_connect(obd);
2767 tgt = lmv_find_target(lmv, fid);
2769 RETURN(PTR_ERR(tgt));
2771 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2775 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2778 struct obd_device *obd = exp->exp_obd;
2779 struct lmv_obd *lmv = &obd->u.lmv;
2780 struct lmv_tgt_desc *tgt;
2784 rc = lmv_check_connect(obd);
2788 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2790 RETURN(PTR_ERR(tgt));
2792 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2796 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2797 const struct req_msg_field *field, struct obd_capa **oc)
2799 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2800 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2802 if (tgt == NULL || tgt->ltd_exp == NULL)
2804 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
2807 int lmv_intent_getattr_async(struct obd_export *exp,
2808 struct md_enqueue_info *minfo,
2809 struct ldlm_enqueue_info *einfo)
2811 struct md_op_data *op_data = &minfo->mi_data;
2812 struct obd_device *obd = exp->exp_obd;
2813 struct lmv_obd *lmv = &obd->u.lmv;
2814 struct lmv_tgt_desc *tgt = NULL;
2818 rc = lmv_check_connect(obd);
2822 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2824 RETURN(PTR_ERR(tgt));
2826 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2830 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2831 struct lu_fid *fid, __u64 *bits)
2833 struct obd_device *obd = exp->exp_obd;
2834 struct lmv_obd *lmv = &obd->u.lmv;
2835 struct lmv_tgt_desc *tgt;
2839 rc = lmv_check_connect(obd);
2843 tgt = lmv_find_target(lmv, fid);
2845 RETURN(PTR_ERR(tgt));
2847 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2852 * For lmv, only need to send request to master MDT, and the master MDT will
2853 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2854 * we directly fetch data from the slave MDTs.
2856 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2857 struct obd_quotactl *oqctl)
2859 struct obd_device *obd = class_exp2obd(exp);
2860 struct lmv_obd *lmv = &obd->u.lmv;
2861 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2864 __u64 curspace, curinodes;
2868 tgt->ltd_exp == NULL ||
2870 lmv->desc.ld_tgt_count == 0) {
2871 CERROR("master lmv inactive\n");
2875 if (oqctl->qc_cmd != Q_GETOQUOTA) {
2876 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2880 curspace = curinodes = 0;
2881 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2885 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2888 err = obd_quotactl(tgt->ltd_exp, oqctl);
2890 CERROR("getquota on mdt %d failed. %d\n", i, err);
2894 curspace += oqctl->qc_dqblk.dqb_curspace;
2895 curinodes += oqctl->qc_dqblk.dqb_curinodes;
2898 oqctl->qc_dqblk.dqb_curspace = curspace;
2899 oqctl->qc_dqblk.dqb_curinodes = curinodes;
2904 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2905 struct obd_quotactl *oqctl)
2907 struct obd_device *obd = class_exp2obd(exp);
2908 struct lmv_obd *lmv = &obd->u.lmv;
2909 struct lmv_tgt_desc *tgt;
2914 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2917 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2918 CERROR("lmv idx %d inactive\n", i);
2922 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2930 struct obd_ops lmv_obd_ops = {
2931 .o_owner = THIS_MODULE,
2932 .o_setup = lmv_setup,
2933 .o_cleanup = lmv_cleanup,
2934 .o_precleanup = lmv_precleanup,
2935 .o_process_config = lmv_process_config,
2936 .o_connect = lmv_connect,
2937 .o_disconnect = lmv_disconnect,
2938 .o_statfs = lmv_statfs,
2939 .o_get_info = lmv_get_info,
2940 .o_set_info_async = lmv_set_info_async,
2941 .o_packmd = lmv_packmd,
2942 .o_unpackmd = lmv_unpackmd,
2943 .o_notify = lmv_notify,
2944 .o_get_uuid = lmv_get_uuid,
2945 .o_iocontrol = lmv_iocontrol,
2946 .o_quotacheck = lmv_quotacheck,
2947 .o_quotactl = lmv_quotactl
2950 struct md_ops lmv_md_ops = {
2951 .m_getstatus = lmv_getstatus,
2952 .m_null_inode = lmv_null_inode,
2953 .m_find_cbdata = lmv_find_cbdata,
2954 .m_close = lmv_close,
2955 .m_create = lmv_create,
2956 .m_done_writing = lmv_done_writing,
2957 .m_enqueue = lmv_enqueue,
2958 .m_getattr = lmv_getattr,
2959 .m_getxattr = lmv_getxattr,
2960 .m_getattr_name = lmv_getattr_name,
2961 .m_intent_lock = lmv_intent_lock,
2963 .m_rename = lmv_rename,
2964 .m_setattr = lmv_setattr,
2965 .m_setxattr = lmv_setxattr,
2966 .m_fsync = lmv_fsync,
2967 .m_readpage = lmv_readpage,
2968 .m_unlink = lmv_unlink,
2969 .m_init_ea_size = lmv_init_ea_size,
2970 .m_cancel_unused = lmv_cancel_unused,
2971 .m_set_lock_data = lmv_set_lock_data,
2972 .m_lock_match = lmv_lock_match,
2973 .m_get_lustre_md = lmv_get_lustre_md,
2974 .m_free_lustre_md = lmv_free_lustre_md,
2975 .m_set_open_replay_data = lmv_set_open_replay_data,
2976 .m_clear_open_replay_data = lmv_clear_open_replay_data,
2977 .m_renew_capa = lmv_renew_capa,
2978 .m_unpack_capa = lmv_unpack_capa,
2979 .m_get_remote_perm = lmv_get_remote_perm,
2980 .m_intent_getattr_async = lmv_intent_getattr_async,
2981 .m_revalidate_lock = lmv_revalidate_lock
2984 int __init lmv_init(void)
2986 struct lprocfs_static_vars lvars;
2989 lprocfs_lmv_init_vars(&lvars);
2991 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2992 lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2997 static void lmv_exit(void)
2999 class_unregister_type(LUSTRE_LMV_NAME);
3002 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3003 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3004 MODULE_LICENSE("GPL");
3006 module_init(lmv_init);
3007 module_exit(lmv_exit);