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
38 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/slab.h>
42 #include <linux/pagemap.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
48 #include <lustre/lustre_idl.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
57 #include <lustre_fid.h>
58 #include <lustre_ioctl.h>
59 #include "lmv_internal.h"
61 static void lmv_activate_target(struct lmv_obd *lmv,
62 struct lmv_tgt_desc *tgt,
65 if (tgt->ltd_active == activate)
68 tgt->ltd_active = activate;
69 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
75 * -EINVAL : UUID can't be found in the LMV's target list
76 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
77 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
79 static int lmv_set_mdc_active(struct lmv_obd *lmv,
80 const struct obd_uuid *uuid,
83 struct lmv_tgt_desc *tgt = NULL;
84 struct obd_device *obd;
89 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
90 lmv, uuid->uuid, activate);
92 spin_lock(&lmv->lmv_lock);
93 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
95 if (tgt == NULL || tgt->ltd_exp == NULL)
98 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
99 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
101 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
105 if (i == lmv->desc.ld_tgt_count)
106 GOTO(out_lmv_lock, rc = -EINVAL);
108 obd = class_exp2obd(tgt->ltd_exp);
110 GOTO(out_lmv_lock, rc = -ENOTCONN);
112 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
113 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
114 obd->obd_type->typ_name, i);
115 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
117 if (tgt->ltd_active == activate) {
118 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
119 activate ? "" : "in");
120 GOTO(out_lmv_lock, rc);
123 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
124 activate ? "" : "in");
125 lmv_activate_target(lmv, tgt, activate);
129 spin_unlock(&lmv->lmv_lock);
133 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
135 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
136 struct lmv_tgt_desc *tgt = lmv->tgts[0];
138 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
142 enum obd_notify_event ev, void *data)
144 struct obd_connect_data *conn_data;
145 struct lmv_obd *lmv = &obd->u.lmv;
146 struct obd_uuid *uuid;
150 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
151 CERROR("unexpected notification of %s %s!\n",
152 watched->obd_type->typ_name,
157 uuid = &watched->u.cli.cl_target_uuid;
158 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
160 * Set MDC as active before notifying the observer, so the
161 * observer can use the MDC normally.
163 rc = lmv_set_mdc_active(lmv, uuid,
164 ev == OBD_NOTIFY_ACTIVE);
166 CERROR("%sactivation of %s failed: %d\n",
167 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
171 } else if (ev == OBD_NOTIFY_OCD) {
172 conn_data = &watched->u.cli.cl_import->imp_connect_data;
174 * XXX: Make sure that ocd_connect_flags from all targets are
175 * the same. Otherwise one of MDTs runs wrong version or
176 * something like this. --umka
178 obd->obd_self_export->exp_connect_data = *conn_data;
181 else if (ev == OBD_NOTIFY_DISCON) {
183 * For disconnect event, flush fld cache for failout MDS case.
185 fld_client_flush(&lmv->lmv_fld);
189 * Pass the notification up the chain.
191 if (obd->obd_observer)
192 rc = obd_notify(obd->obd_observer, watched, ev, data);
198 * This is fake connect function. Its purpose is to initialize lmv and say
199 * caller that everything is okay. Real connection will be performed later.
201 static int lmv_connect(const struct lu_env *env,
202 struct obd_export **exp, struct obd_device *obd,
203 struct obd_uuid *cluuid, struct obd_connect_data *data,
206 struct lmv_obd *lmv = &obd->u.lmv;
207 struct lustre_handle conn = { 0 };
212 * We don't want to actually do the underlying connections more than
213 * once, so keep track.
216 if (lmv->refcount > 1) {
221 rc = class_connect(&conn, obd, cluuid);
223 CERROR("class_connection() returned %d\n", rc);
227 *exp = class_conn2export(&conn);
228 class_export_get(*exp);
232 lmv->cluuid = *cluuid;
235 lmv->conn_data = *data;
237 if (lmv->targets_proc_entry == NULL) {
238 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
241 if (IS_ERR(lmv->targets_proc_entry)) {
242 CERROR("%s: cannot register "
243 "/proc/fs/lustre/%s/%s/target_obds\n",
244 obd->obd_name, obd->obd_type->typ_name,
246 lmv->targets_proc_entry = NULL;
251 * All real clients should perform actual connection right away, because
252 * it is possible, that LMV will not have opportunity to connect targets
253 * and MDC stuff will be called directly, for instance while reading
254 * ../mdc/../kbytesfree procfs file, etc.
256 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
257 rc = lmv_check_connect(obd);
259 if (rc && lmv->targets_proc_entry != NULL)
260 lprocfs_remove(&lmv->targets_proc_entry);
264 static int lmv_init_ea_size(struct obd_export *exp, int easize,
265 int def_easize, int cookiesize, int def_cookiesize)
267 struct obd_device *obd = exp->exp_obd;
268 struct lmv_obd *lmv = &obd->u.lmv;
274 if (lmv->max_easize < easize) {
275 lmv->max_easize = easize;
278 if (lmv->max_def_easize < def_easize) {
279 lmv->max_def_easize = def_easize;
282 if (lmv->max_cookiesize < cookiesize) {
283 lmv->max_cookiesize = cookiesize;
286 if (lmv->max_def_cookiesize < def_cookiesize) {
287 lmv->max_def_cookiesize = def_cookiesize;
293 if (lmv->connected == 0)
296 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
297 struct lmv_tgt_desc *tgt = lmv->tgts[i];
299 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
300 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
304 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
305 cookiesize, def_cookiesize);
307 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
308 " rc = %d\n", obd->obd_name, i, rc);
315 #define MAX_STRING_SIZE 128
317 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
319 struct lmv_obd *lmv = &obd->u.lmv;
320 struct obd_uuid *cluuid = &lmv->cluuid;
321 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
322 struct obd_device *mdc_obd;
323 struct obd_export *mdc_exp;
324 struct lu_fld_target target;
328 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
331 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
335 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
336 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
337 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
340 if (!mdc_obd->obd_set_up) {
341 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
345 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
346 &lmv->conn_data, NULL);
348 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
353 * Init fid sequence client for this mdc and add new fld target.
355 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
359 target.ft_srv = NULL;
360 target.ft_exp = mdc_exp;
361 target.ft_idx = tgt->ltd_idx;
363 fld_client_add_target(&lmv->lmv_fld, &target);
365 rc = obd_register_observer(mdc_obd, obd);
367 obd_disconnect(mdc_exp);
368 CERROR("target %s register_observer error %d\n",
369 tgt->ltd_uuid.uuid, rc);
373 if (obd->obd_observer) {
375 * Tell the observer about the new target.
377 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
379 (void *)(tgt - lmv->tgts[0]));
381 obd_disconnect(mdc_exp);
387 tgt->ltd_exp = mdc_exp;
388 lmv->desc.ld_active_tgt_count++;
390 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
391 lmv->max_cookiesize, lmv->max_def_cookiesize);
393 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
394 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
395 atomic_read(&obd->obd_refcount));
397 if (lmv->targets_proc_entry != NULL) {
398 struct proc_dir_entry *mdc_symlink;
400 LASSERT(mdc_obd->obd_type != NULL);
401 LASSERT(mdc_obd->obd_type->typ_name != NULL);
402 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
403 lmv->targets_proc_entry,
405 mdc_obd->obd_type->typ_name,
407 if (mdc_symlink == NULL) {
408 CERROR("cannot register LMV target "
409 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
410 obd->obd_type->typ_name, obd->obd_name,
417 static void lmv_del_target(struct lmv_obd *lmv, int index)
419 if (lmv->tgts[index] == NULL)
422 OBD_FREE_PTR(lmv->tgts[index]);
423 lmv->tgts[index] = NULL;
427 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
428 __u32 index, int gen)
430 struct lmv_obd *lmv = &obd->u.lmv;
431 struct lmv_tgt_desc *tgt;
432 int orig_tgt_count = 0;
436 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
440 if (lmv->desc.ld_tgt_count == 0) {
441 struct obd_device *mdc_obd;
443 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
446 lmv_init_unlock(lmv);
447 CERROR("%s: Target %s not attached: rc = %d\n",
448 obd->obd_name, uuidp->uuid, -EINVAL);
453 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
454 tgt = lmv->tgts[index];
455 CERROR("%s: UUID %s already assigned at LOV target index %d:"
456 " rc = %d\n", obd->obd_name,
457 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
458 lmv_init_unlock(lmv);
462 if (index >= lmv->tgts_size) {
463 /* We need to reallocate the lmv target array. */
464 struct lmv_tgt_desc **newtgts, **old = NULL;
468 while (newsize < index + 1)
469 newsize = newsize << 1;
470 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
471 if (newtgts == NULL) {
472 lmv_init_unlock(lmv);
476 if (lmv->tgts_size) {
477 memcpy(newtgts, lmv->tgts,
478 sizeof(*newtgts) * lmv->tgts_size);
480 oldsize = lmv->tgts_size;
484 lmv->tgts_size = newsize;
487 OBD_FREE(old, sizeof(*old) * oldsize);
489 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
495 lmv_init_unlock(lmv);
499 mutex_init(&tgt->ltd_fid_mutex);
500 tgt->ltd_idx = index;
501 tgt->ltd_uuid = *uuidp;
503 lmv->tgts[index] = tgt;
504 if (index >= lmv->desc.ld_tgt_count) {
505 orig_tgt_count = lmv->desc.ld_tgt_count;
506 lmv->desc.ld_tgt_count = index + 1;
509 if (lmv->connected) {
510 rc = lmv_connect_mdc(obd, tgt);
512 spin_lock(&lmv->lmv_lock);
513 if (lmv->desc.ld_tgt_count == index + 1)
514 lmv->desc.ld_tgt_count = orig_tgt_count;
515 memset(tgt, 0, sizeof(*tgt));
516 spin_unlock(&lmv->lmv_lock);
518 int easize = sizeof(struct lmv_stripe_md) +
519 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
520 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
524 lmv_init_unlock(lmv);
528 int lmv_check_connect(struct obd_device *obd)
530 struct lmv_obd *lmv = &obd->u.lmv;
531 struct lmv_tgt_desc *tgt;
541 if (lmv->connected) {
542 lmv_init_unlock(lmv);
546 if (lmv->desc.ld_tgt_count == 0) {
547 lmv_init_unlock(lmv);
548 CERROR("%s: no targets configured.\n", obd->obd_name);
552 LASSERT(lmv->tgts != NULL);
554 if (lmv->tgts[0] == NULL) {
555 lmv_init_unlock(lmv);
556 CERROR("%s: no target configured for index 0.\n",
561 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
562 lmv->cluuid.uuid, obd->obd_name);
564 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
568 rc = lmv_connect_mdc(obd, tgt);
573 class_export_put(lmv->exp);
575 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
576 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
577 lmv_init_unlock(lmv);
588 --lmv->desc.ld_active_tgt_count;
589 rc2 = obd_disconnect(tgt->ltd_exp);
591 CERROR("LMV target %s disconnect on "
592 "MDC idx %d: error %d\n",
593 tgt->ltd_uuid.uuid, i, rc2);
597 class_disconnect(lmv->exp);
598 lmv_init_unlock(lmv);
602 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
604 struct lmv_obd *lmv = &obd->u.lmv;
605 struct obd_device *mdc_obd;
609 LASSERT(tgt != NULL);
610 LASSERT(obd != NULL);
612 mdc_obd = class_exp2obd(tgt->ltd_exp);
615 mdc_obd->obd_force = obd->obd_force;
616 mdc_obd->obd_fail = obd->obd_fail;
617 mdc_obd->obd_no_recov = obd->obd_no_recov;
620 if (lmv->targets_proc_entry != NULL)
621 lprocfs_remove_proc_entry(mdc_obd->obd_name,
622 lmv->targets_proc_entry);
624 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
626 CERROR("Can't finanize fids factory\n");
628 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
629 tgt->ltd_exp->exp_obd->obd_name,
630 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
632 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
633 rc = obd_disconnect(tgt->ltd_exp);
635 if (tgt->ltd_active) {
636 CERROR("Target %s disconnect error %d\n",
637 tgt->ltd_uuid.uuid, rc);
641 lmv_activate_target(lmv, tgt, 0);
646 static int lmv_disconnect(struct obd_export *exp)
648 struct obd_device *obd = class_exp2obd(exp);
649 struct lmv_obd *lmv = &obd->u.lmv;
658 * Only disconnect the underlying layers on the final disconnect.
661 if (lmv->refcount != 0)
664 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
665 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
668 lmv_disconnect_mdc(obd, lmv->tgts[i]);
671 if (lmv->targets_proc_entry != NULL)
672 lprocfs_remove(&lmv->targets_proc_entry);
674 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
675 obd->obd_type->typ_name, obd->obd_name);
679 * This is the case when no real connection is established by
680 * lmv_check_connect().
683 class_export_put(exp);
684 rc = class_disconnect(exp);
685 if (lmv->refcount == 0)
690 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
692 struct obd_device *obddev = class_exp2obd(exp);
693 struct lmv_obd *lmv = &obddev->u.lmv;
694 struct getinfo_fid2path *gf;
695 struct lmv_tgt_desc *tgt;
696 struct getinfo_fid2path *remote_gf = NULL;
697 int remote_gf_size = 0;
700 gf = (struct getinfo_fid2path *)karg;
701 tgt = lmv_find_target(lmv, &gf->gf_fid);
703 RETURN(PTR_ERR(tgt));
706 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
707 if (rc != 0 && rc != -EREMOTE)
708 GOTO(out_fid2path, rc);
710 /* If remote_gf != NULL, it means just building the
711 * path on the remote MDT, copy this path segement to gf */
712 if (remote_gf != NULL) {
713 struct getinfo_fid2path *ori_gf;
716 ori_gf = (struct getinfo_fid2path *)karg;
717 if (strlen(ori_gf->gf_path) +
718 strlen(gf->gf_path) > ori_gf->gf_pathlen)
719 GOTO(out_fid2path, rc = -EOVERFLOW);
721 ptr = ori_gf->gf_path;
723 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
724 strlen(ori_gf->gf_path));
726 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
727 ptr += strlen(gf->gf_path);
731 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
732 tgt->ltd_exp->exp_obd->obd_name,
733 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
737 GOTO(out_fid2path, rc);
739 /* sigh, has to go to another MDT to do path building further */
740 if (remote_gf == NULL) {
741 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
742 OBD_ALLOC(remote_gf, remote_gf_size);
743 if (remote_gf == NULL)
744 GOTO(out_fid2path, rc = -ENOMEM);
745 remote_gf->gf_pathlen = PATH_MAX;
748 if (!fid_is_sane(&gf->gf_fid)) {
749 CERROR("%s: invalid FID "DFID": rc = %d\n",
750 tgt->ltd_exp->exp_obd->obd_name,
751 PFID(&gf->gf_fid), -EINVAL);
752 GOTO(out_fid2path, rc = -EINVAL);
755 tgt = lmv_find_target(lmv, &gf->gf_fid);
757 GOTO(out_fid2path, rc = -EINVAL);
759 remote_gf->gf_fid = gf->gf_fid;
760 remote_gf->gf_recno = -1;
761 remote_gf->gf_linkno = -1;
762 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
764 goto repeat_fid2path;
767 if (remote_gf != NULL)
768 OBD_FREE(remote_gf, remote_gf_size);
772 static int lmv_hsm_req_count(struct lmv_obd *lmv,
773 const struct hsm_user_request *hur,
774 const struct lmv_tgt_desc *tgt_mds)
778 struct lmv_tgt_desc *curr_tgt;
780 /* count how many requests must be sent to the given target */
781 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
782 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
783 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
789 static void lmv_hsm_req_build(struct lmv_obd *lmv,
790 struct hsm_user_request *hur_in,
791 const struct lmv_tgt_desc *tgt_mds,
792 struct hsm_user_request *hur_out)
795 struct lmv_tgt_desc *curr_tgt;
797 /* build the hsm_user_request for the given target */
798 hur_out->hur_request = hur_in->hur_request;
800 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
801 curr_tgt = lmv_find_target(lmv,
802 &hur_in->hur_user_item[i].hui_fid);
803 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
804 hur_out->hur_user_item[nr_out] =
805 hur_in->hur_user_item[i];
809 hur_out->hur_request.hr_itemcount = nr_out;
810 memcpy(hur_data(hur_out), hur_data(hur_in),
811 hur_in->hur_request.hr_data_len);
814 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
815 struct lustre_kernelcomm *lk, void *uarg)
819 struct kkuc_ct_data *kcd = NULL;
822 /* unregister request (call from llapi_hsm_copytool_fini) */
823 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
824 struct lmv_tgt_desc *tgt = lmv->tgts[i];
826 if (tgt == NULL || tgt->ltd_exp == NULL)
828 /* best effort: try to clean as much as possible
829 * (continue on error) */
830 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
833 /* Whatever the result, remove copytool from kuc groups.
834 * Unreached coordinators will get EPIPE on next requests
835 * and will unregister automatically.
837 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
844 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
845 struct lustre_kernelcomm *lk, void *uarg)
850 bool any_set = false;
851 struct kkuc_ct_data *kcd;
854 /* All or nothing: try to register to all MDS.
855 * In case of failure, unregister from previous MDS,
856 * except if it because of inactive target. */
857 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
858 struct lmv_tgt_desc *tgt = lmv->tgts[i];
860 if (tgt == NULL || tgt->ltd_exp == NULL)
862 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
864 if (tgt->ltd_active) {
865 /* permanent error */
866 CERROR("%s: iocontrol MDC %s on MDT"
867 " idx %d cmd %x: err = %d\n",
868 class_exp2obd(lmv->exp)->obd_name,
869 tgt->ltd_uuid.uuid, i, cmd, err);
871 lk->lk_flags |= LK_FLG_STOP;
872 /* unregister from previous MDS */
873 for (j = 0; j < i; j++) {
875 if (tgt == NULL || tgt->ltd_exp == NULL)
877 obd_iocontrol(cmd, tgt->ltd_exp, len,
882 /* else: transient error.
883 * kuc will register to the missing MDT
891 /* no registration done: return error */
894 /* at least one registration done, with no failure */
895 filp = fget(lk->lk_wfd);
904 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
905 kcd->kcd_uuid = lmv->cluuid;
906 kcd->kcd_archive = lk->lk_data;
908 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
922 int len, void *karg, void *uarg)
924 struct obd_device *obddev = class_exp2obd(exp);
925 struct lmv_obd *lmv = &obddev->u.lmv;
926 struct lmv_tgt_desc *tgt = NULL;
930 __u32 count = lmv->desc.ld_tgt_count;
937 case IOC_OBD_STATFS: {
938 struct obd_ioctl_data *data = karg;
939 struct obd_device *mdc_obd;
940 struct obd_statfs stat_buf = {0};
943 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
944 if ((index >= count))
947 tgt = lmv->tgts[index];
948 if (tgt == NULL || !tgt->ltd_active)
951 mdc_obd = class_exp2obd(tgt->ltd_exp);
956 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
957 min((int) data->ioc_plen2,
958 (int) sizeof(struct obd_uuid))))
961 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
962 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
966 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
967 min((int) data->ioc_plen1,
968 (int) sizeof(stat_buf))))
972 case OBD_IOC_QUOTACTL: {
973 struct if_quotactl *qctl = karg;
974 struct obd_quotactl *oqctl;
976 if (qctl->qc_valid == QC_MDTIDX) {
977 if (count <= qctl->qc_idx)
980 tgt = lmv->tgts[qctl->qc_idx];
981 if (tgt == NULL || tgt->ltd_exp == NULL)
983 } else if (qctl->qc_valid == QC_UUID) {
984 for (i = 0; i < count; i++) {
988 if (!obd_uuid_equals(&tgt->ltd_uuid,
992 if (tgt->ltd_exp == NULL)
1004 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005 OBD_ALLOC_PTR(oqctl);
1009 QCTL_COPY(oqctl, qctl);
1010 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1012 QCTL_COPY(qctl, oqctl);
1013 qctl->qc_valid = QC_MDTIDX;
1014 qctl->obd_uuid = tgt->ltd_uuid;
1016 OBD_FREE_PTR(oqctl);
1019 case OBD_IOC_CHANGELOG_SEND:
1020 case OBD_IOC_CHANGELOG_CLEAR: {
1021 struct ioc_changelog *icc = karg;
1023 if (icc->icc_mdtindex >= count)
1026 tgt = lmv->tgts[icc->icc_mdtindex];
1027 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1029 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1032 case LL_IOC_GET_CONNECT_FLAGS: {
1034 if (tgt == NULL || tgt->ltd_exp == NULL)
1036 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1039 case LL_IOC_FID2MDTIDX: {
1040 struct lu_fid *fid = karg;
1043 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1047 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048 * point to user space memory for FID2MDTIDX. */
1049 *(__u32 *)uarg = mdt_index;
1052 case OBD_IOC_FID2PATH: {
1053 rc = lmv_fid2path(exp, len, karg, uarg);
1056 case LL_IOC_HSM_STATE_GET:
1057 case LL_IOC_HSM_STATE_SET:
1058 case LL_IOC_HSM_ACTION: {
1059 struct md_op_data *op_data = karg;
1061 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1063 RETURN(PTR_ERR(tgt));
1065 if (tgt->ltd_exp == NULL)
1068 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1071 case LL_IOC_HSM_PROGRESS: {
1072 const struct hsm_progress_kernel *hpk = karg;
1074 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1076 RETURN(PTR_ERR(tgt));
1077 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1080 case LL_IOC_HSM_REQUEST: {
1081 struct hsm_user_request *hur = karg;
1082 unsigned int reqcount = hur->hur_request.hr_itemcount;
1087 /* if the request is about a single fid
1088 * or if there is a single MDS, no need to split
1090 if (reqcount == 1 || count == 1) {
1091 tgt = lmv_find_target(lmv,
1092 &hur->hur_user_item[0].hui_fid);
1094 RETURN(PTR_ERR(tgt));
1095 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1097 /* split fid list to their respective MDS */
1098 for (i = 0; i < count; i++) {
1099 unsigned int nr, reqlen;
1101 struct hsm_user_request *req;
1104 if (tgt == NULL || tgt->ltd_exp == NULL)
1107 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108 if (nr == 0) /* nothing for this MDS */
1111 /* build a request with fids for this MDS */
1112 reqlen = offsetof(typeof(*hur),
1114 + hur->hur_request.hr_data_len;
1115 OBD_ALLOC_LARGE(req, reqlen);
1119 lmv_hsm_req_build(lmv, hur, tgt, req);
1121 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1123 if (rc1 != 0 && rc == 0)
1125 OBD_FREE_LARGE(req, reqlen);
1130 case LL_IOC_LOV_SWAP_LAYOUTS: {
1131 struct md_op_data *op_data = karg;
1132 struct lmv_tgt_desc *tgt1, *tgt2;
1134 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1136 RETURN(PTR_ERR(tgt1));
1138 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1140 RETURN(PTR_ERR(tgt2));
1142 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1145 /* only files on same MDT can have their layouts swapped */
1146 if (tgt1->ltd_idx != tgt2->ltd_idx)
1149 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1152 case LL_IOC_HSM_CT_START: {
1153 struct lustre_kernelcomm *lk = karg;
1154 if (lk->lk_flags & LK_FLG_STOP)
1155 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1157 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1161 for (i = 0; i < count; i++) {
1162 struct obd_device *mdc_obd;
1166 if (tgt == NULL || tgt->ltd_exp == NULL)
1168 /* ll_umount_begin() sets force flag but for lmv, not
1169 * mdc. Let's pass it through */
1170 mdc_obd = class_exp2obd(tgt->ltd_exp);
1171 mdc_obd->obd_force = obddev->obd_force;
1172 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1173 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1176 if (tgt->ltd_active) {
1177 CERROR("error: iocontrol MDC %s on MDT"
1178 " idx %d cmd %x: err = %d\n",
1179 tgt->ltd_uuid.uuid, i, cmd, err);
1193 static int lmv_all_chars_policy(int count, const char *name,
1204 static int lmv_nid_policy(struct lmv_obd *lmv)
1206 struct obd_import *imp;
1210 * XXX: To get nid we assume that underlying obd device is mdc.
1212 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1213 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1214 return id % lmv->desc.ld_tgt_count;
1217 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1218 placement_policy_t placement)
1220 switch (placement) {
1221 case PLACEMENT_CHAR_POLICY:
1222 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1224 op_data->op_namelen);
1225 case PLACEMENT_NID_POLICY:
1226 return lmv_nid_policy(lmv);
1232 CERROR("Unsupported placement policy %x\n", placement);
1238 * This is _inode_ placement policy function (not name).
1240 static int lmv_placement_policy(struct obd_device *obd,
1241 struct md_op_data *op_data, u32 *mds)
1243 struct lmv_obd *lmv = &obd->u.lmv;
1246 LASSERT(mds != NULL);
1248 if (lmv->desc.ld_tgt_count == 1) {
1254 * If stripe_offset is provided during setdirstripe
1255 * (setdirstripe -i xx), xx MDS will be choosen.
1257 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1258 struct lmv_user_md *lum;
1260 lum = op_data->op_data;
1262 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1263 *mds = le32_to_cpu(lum->lum_stripe_offset);
1265 /* -1 means default, which will be in the same MDT with
1267 *mds = op_data->op_mds;
1268 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1271 /* Allocate new fid on target according to operation type and
1272 * parent home mds. */
1273 *mds = op_data->op_mds;
1279 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1281 struct lmv_tgt_desc *tgt;
1285 tgt = lmv_get_target(lmv, mds, NULL);
1287 RETURN(PTR_ERR(tgt));
1290 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1291 * on server that seq in new allocated fid is not yet known.
1293 mutex_lock(&tgt->ltd_fid_mutex);
1295 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1296 GOTO(out, rc = -ENODEV);
1299 * Asking underlying tgt layer to allocate new fid.
1301 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1303 LASSERT(fid_is_sane(fid));
1309 mutex_unlock(&tgt->ltd_fid_mutex);
1313 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1314 struct lu_fid *fid, struct md_op_data *op_data)
1316 struct obd_device *obd = class_exp2obd(exp);
1317 struct lmv_obd *lmv = &obd->u.lmv;
1322 LASSERT(op_data != NULL);
1323 LASSERT(fid != NULL);
1325 rc = lmv_placement_policy(obd, op_data, &mds);
1327 CERROR("Can't get target for allocating fid, "
1332 rc = __lmv_fid_alloc(lmv, fid, mds);
1334 CERROR("Can't alloc new fid, rc %d\n", rc);
1341 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1343 struct lmv_obd *lmv = &obd->u.lmv;
1344 struct lmv_desc *desc;
1348 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1349 CERROR("LMV setup requires a descriptor\n");
1353 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1354 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1355 CERROR("Lmv descriptor size wrong: %d > %d\n",
1356 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1360 lmv->tgts_size = 32U;
1361 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1362 if (lmv->tgts == NULL)
1365 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1366 lmv->desc.ld_tgt_count = 0;
1367 lmv->desc.ld_active_tgt_count = 0;
1368 lmv->max_cookiesize = 0;
1369 lmv->max_def_easize = 0;
1370 lmv->max_easize = 0;
1371 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1373 spin_lock_init(&lmv->lmv_lock);
1374 mutex_init(&lmv->init_mutex);
1377 obd->obd_vars = lprocfs_lmv_obd_vars;
1378 lprocfs_obd_setup(obd);
1379 lprocfs_alloc_md_stats(obd, 0);
1380 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1381 0444, &lmv_proc_target_fops, obd);
1383 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1386 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1387 LUSTRE_CLI_FLD_HASH_DHT);
1389 CERROR("Can't init FLD, err %d\n", rc);
1399 static int lmv_cleanup(struct obd_device *obd)
1401 struct lmv_obd *lmv = &obd->u.lmv;
1404 fld_client_fini(&lmv->lmv_fld);
1405 if (lmv->tgts != NULL) {
1407 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1408 if (lmv->tgts[i] == NULL)
1410 lmv_del_target(lmv, i);
1412 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1418 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1420 struct lustre_cfg *lcfg = buf;
1421 struct obd_uuid obd_uuid;
1427 switch (lcfg->lcfg_command) {
1429 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1430 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1431 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1432 GOTO(out, rc = -EINVAL);
1434 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1436 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1437 GOTO(out, rc = -EINVAL);
1438 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1439 GOTO(out, rc = -EINVAL);
1440 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1443 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1444 GOTO(out, rc = -EINVAL);
1450 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1451 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1453 struct obd_device *obd = class_exp2obd(exp);
1454 struct lmv_obd *lmv = &obd->u.lmv;
1455 struct obd_statfs *temp;
1460 rc = lmv_check_connect(obd);
1464 OBD_ALLOC(temp, sizeof(*temp));
1468 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1469 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1472 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1475 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1476 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1478 GOTO(out_free_temp, rc);
1483 /* If the statfs is from mount, it will needs
1484 * retrieve necessary information from MDT0.
1485 * i.e. mount does not need the merged osfs
1487 * And also clients can be mounted as long as
1488 * MDT0 is in service*/
1489 if (flags & OBD_STATFS_FOR_MDT0)
1490 GOTO(out_free_temp, rc);
1492 osfs->os_bavail += temp->os_bavail;
1493 osfs->os_blocks += temp->os_blocks;
1494 osfs->os_ffree += temp->os_ffree;
1495 osfs->os_files += temp->os_files;
1501 OBD_FREE(temp, sizeof(*temp));
1505 static int lmv_getstatus(struct obd_export *exp,
1507 struct obd_capa **pc)
1509 struct obd_device *obd = exp->exp_obd;
1510 struct lmv_obd *lmv = &obd->u.lmv;
1514 rc = lmv_check_connect(obd);
1518 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1522 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1523 struct obd_capa *oc, obd_valid valid, const char *name,
1524 const char *input, int input_size, int output_size,
1525 int flags, struct ptlrpc_request **request)
1527 struct obd_device *obd = exp->exp_obd;
1528 struct lmv_obd *lmv = &obd->u.lmv;
1529 struct lmv_tgt_desc *tgt;
1533 rc = lmv_check_connect(obd);
1537 tgt = lmv_find_target(lmv, fid);
1539 RETURN(PTR_ERR(tgt));
1541 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1542 input_size, output_size, flags, request);
1547 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1548 struct obd_capa *oc, obd_valid valid, const char *name,
1549 const char *input, int input_size, int output_size,
1550 int flags, __u32 suppgid,
1551 struct ptlrpc_request **request)
1553 struct obd_device *obd = exp->exp_obd;
1554 struct lmv_obd *lmv = &obd->u.lmv;
1555 struct lmv_tgt_desc *tgt;
1559 rc = lmv_check_connect(obd);
1563 tgt = lmv_find_target(lmv, fid);
1565 RETURN(PTR_ERR(tgt));
1567 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1568 input_size, output_size, flags, suppgid,
1574 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1575 struct ptlrpc_request **request)
1577 struct obd_device *obd = exp->exp_obd;
1578 struct lmv_obd *lmv = &obd->u.lmv;
1579 struct lmv_tgt_desc *tgt;
1583 rc = lmv_check_connect(obd);
1587 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1589 RETURN(PTR_ERR(tgt));
1591 if (op_data->op_flags & MF_GET_MDT_IDX) {
1592 op_data->op_mds = tgt->ltd_idx;
1596 rc = md_getattr(tgt->ltd_exp, op_data, request);
1601 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1603 struct obd_device *obd = exp->exp_obd;
1604 struct lmv_obd *lmv = &obd->u.lmv;
1609 rc = lmv_check_connect(obd);
1613 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1616 * With DNE every object can have two locks in different namespaces:
1617 * lookup lock in space of MDT storing direntry and update/open lock in
1618 * space of MDT storing inode.
1620 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1621 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1623 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1629 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1630 ldlm_iterator_t it, void *data)
1632 struct obd_device *obd = exp->exp_obd;
1633 struct lmv_obd *lmv = &obd->u.lmv;
1639 rc = lmv_check_connect(obd);
1643 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1646 * With DNE every object can have two locks in different namespaces:
1647 * lookup lock in space of MDT storing direntry and update/open lock in
1648 * space of MDT storing inode. Try the MDT that the FID maps to first,
1649 * since this can be easily found, and only try others if that fails.
1651 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1652 i < lmv->desc.ld_tgt_count;
1653 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1655 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1656 obd->obd_name, PFID(fid), tgt);
1660 if (lmv->tgts[tgt] == NULL ||
1661 lmv->tgts[tgt]->ltd_exp == NULL)
1664 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1673 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1674 struct md_open_data *mod, struct ptlrpc_request **request)
1676 struct obd_device *obd = exp->exp_obd;
1677 struct lmv_obd *lmv = &obd->u.lmv;
1678 struct lmv_tgt_desc *tgt;
1682 rc = lmv_check_connect(obd);
1686 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1688 RETURN(PTR_ERR(tgt));
1690 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1691 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1696 * Choosing the MDT by name or FID in @op_data.
1697 * For non-striped directory, it will locate MDT by fid.
1698 * For striped-directory, it will locate MDT by name. And also
1699 * it will reset op_fid1 with the FID of the choosen stripe.
1701 struct lmv_tgt_desc *
1702 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1703 const char *name, int namelen, struct lu_fid *fid,
1706 struct lmv_tgt_desc *tgt;
1707 const struct lmv_oinfo *oinfo;
1709 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1711 RETURN(ERR_CAST(oinfo));
1712 *fid = oinfo->lmo_fid;
1713 *mds = oinfo->lmo_mds;
1714 tgt = lmv_get_target(lmv, *mds, NULL);
1716 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1721 * Locate mds by fid or name
1723 * For striped directory (lsm != NULL), it will locate the stripe
1724 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1725 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1726 * walk through all of stripes to locate the entry.
1728 * For normal direcotry, it will locate MDS by FID directly.
1729 * \param[in] lmv LMV device
1730 * \param[in] op_data client MD stack parameters, name, namelen
1732 * \param[in] fid object FID used to locate MDS.
1734 * retval pointer to the lmv_tgt_desc if succeed.
1735 * ERR_PTR(errno) if failed.
1737 struct lmv_tgt_desc*
1738 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1741 struct lmv_stripe_md *lsm = op_data->op_mea1;
1742 struct lmv_tgt_desc *tgt;
1744 /* During creating VOLATILE file, it should honor the mdt
1745 * index if the file under striped dir is being restored, see
1747 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1748 (int)op_data->op_mds != -1 && lsm != NULL) {
1750 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1754 /* refill the right parent fid */
1755 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1756 struct lmv_oinfo *oinfo;
1758 oinfo = &lsm->lsm_md_oinfo[i];
1759 if (oinfo->lmo_mds == op_data->op_mds) {
1760 *fid = oinfo->lmo_fid;
1765 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1766 if (i == lsm->lsm_md_stripe_count)
1767 tgt = ERR_PTR(-EINVAL);
1772 if (lsm == NULL || op_data->op_namelen == 0) {
1773 tgt = lmv_find_target(lmv, fid);
1777 op_data->op_mds = tgt->ltd_idx;
1781 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1782 op_data->op_namelen, fid,
1786 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1787 const void *data, int datalen, int mode, __u32 uid,
1788 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1789 struct ptlrpc_request **request)
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 if (!lmv->desc.ld_active_tgt_count)
1804 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1806 RETURN(PTR_ERR(tgt));
1808 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1809 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1812 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1816 /* Send the create request to the MDT where the object
1817 * will be located */
1818 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1820 RETURN(PTR_ERR(tgt));
1822 op_data->op_mds = tgt->ltd_idx;
1824 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1825 PFID(&op_data->op_fid2), op_data->op_mds);
1827 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1828 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1829 cap_effective, rdev, request);
1831 if (*request == NULL)
1833 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1838 static int lmv_done_writing(struct obd_export *exp,
1839 struct md_op_data *op_data,
1840 struct md_open_data *mod)
1842 struct obd_device *obd = exp->exp_obd;
1843 struct lmv_obd *lmv = &obd->u.lmv;
1844 struct lmv_tgt_desc *tgt;
1848 rc = lmv_check_connect(obd);
1852 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1854 RETURN(PTR_ERR(tgt));
1856 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1861 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1862 const union ldlm_policy_data *policy,
1863 struct lookup_intent *it, struct md_op_data *op_data,
1864 struct lustre_handle *lockh, __u64 extra_lock_flags)
1866 struct obd_device *obd = exp->exp_obd;
1867 struct lmv_obd *lmv = &obd->u.lmv;
1868 struct lmv_tgt_desc *tgt;
1872 rc = lmv_check_connect(obd);
1876 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1877 LL_IT2STR(it), PFID(&op_data->op_fid1));
1879 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1881 RETURN(PTR_ERR(tgt));
1883 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1884 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1886 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1893 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1894 struct ptlrpc_request **preq)
1896 struct ptlrpc_request *req = NULL;
1897 struct obd_device *obd = exp->exp_obd;
1898 struct lmv_obd *lmv = &obd->u.lmv;
1899 struct lmv_tgt_desc *tgt;
1900 struct mdt_body *body;
1904 rc = lmv_check_connect(obd);
1908 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1910 RETURN(PTR_ERR(tgt));
1912 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%u\n",
1913 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1916 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1920 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1921 LASSERT(body != NULL);
1923 if (body->mbo_valid & OBD_MD_MDS) {
1924 struct lu_fid rid = body->mbo_fid1;
1925 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1928 tgt = lmv_find_target(lmv, &rid);
1930 ptlrpc_req_finished(*preq);
1932 RETURN(PTR_ERR(tgt));
1935 op_data->op_fid1 = rid;
1936 op_data->op_valid |= OBD_MD_FLCROSSREF;
1937 op_data->op_namelen = 0;
1938 op_data->op_name = NULL;
1939 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1940 ptlrpc_req_finished(*preq);
1947 #define md_op_data_fid(op_data, fl) \
1948 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1949 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1950 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1951 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1954 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1955 struct md_op_data *op_data,
1956 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1958 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1959 struct obd_device *obd = exp->exp_obd;
1960 struct lmv_obd *lmv = &obd->u.lmv;
1961 ldlm_policy_data_t policy = {{ 0 }};
1965 if (!fid_is_sane(fid))
1969 tgt = lmv_find_target(lmv, fid);
1971 RETURN(PTR_ERR(tgt));
1974 if (tgt->ltd_idx != op_tgt) {
1975 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1976 policy.l_inodebits.bits = bits;
1977 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1978 mode, LCF_ASYNC, NULL);
1981 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1983 op_data->op_flags |= flag;
1991 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1994 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1995 struct ptlrpc_request **request)
1997 struct obd_device *obd = exp->exp_obd;
1998 struct lmv_obd *lmv = &obd->u.lmv;
1999 struct lmv_tgt_desc *tgt;
2003 rc = lmv_check_connect(obd);
2007 LASSERT(op_data->op_namelen != 0);
2009 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2010 PFID(&op_data->op_fid2), op_data->op_namelen,
2011 op_data->op_name, PFID(&op_data->op_fid1));
2013 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2014 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2015 op_data->op_cap = cfs_curproc_cap_pack();
2016 if (op_data->op_mea2 != NULL) {
2017 struct lmv_stripe_md *lsm = op_data->op_mea2;
2018 const struct lmv_oinfo *oinfo;
2020 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2021 op_data->op_namelen);
2023 RETURN(PTR_ERR(oinfo));
2025 op_data->op_fid2 = oinfo->lmo_fid;
2028 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2030 RETURN(PTR_ERR(tgt));
2033 * Cancel UPDATE lock on child (fid1).
2035 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2036 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2037 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2041 rc = md_link(tgt->ltd_exp, op_data, request);
2046 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2047 const char *old, int oldlen, const char *new, int newlen,
2048 struct ptlrpc_request **request)
2050 struct obd_device *obd = exp->exp_obd;
2051 struct lmv_obd *lmv = &obd->u.lmv;
2052 struct lmv_tgt_desc *src_tgt;
2056 LASSERT(oldlen != 0);
2058 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2059 oldlen, old, PFID(&op_data->op_fid1),
2060 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2061 newlen, new, PFID(&op_data->op_fid2),
2062 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2064 rc = lmv_check_connect(obd);
2068 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2069 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2070 op_data->op_cap = cfs_curproc_cap_pack();
2071 if (op_data->op_cli_flags & CLI_MIGRATE) {
2072 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2073 PFID(&op_data->op_fid3));
2074 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2077 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2079 if (op_data->op_mea1 != NULL) {
2080 struct lmv_stripe_md *lsm = op_data->op_mea1;
2082 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2086 if (IS_ERR(src_tgt))
2087 RETURN(PTR_ERR(src_tgt));
2089 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2090 if (IS_ERR(src_tgt))
2091 RETURN(PTR_ERR(src_tgt));
2093 op_data->op_mds = src_tgt->ltd_idx;
2096 if (op_data->op_mea2) {
2097 struct lmv_stripe_md *lsm = op_data->op_mea2;
2098 const struct lmv_oinfo *oinfo;
2100 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2102 RETURN(PTR_ERR(oinfo));
2104 op_data->op_fid2 = oinfo->lmo_fid;
2107 if (IS_ERR(src_tgt))
2108 RETURN(PTR_ERR(src_tgt));
2111 * LOOKUP lock on src child (fid3) should also be cancelled for
2112 * src_tgt in mdc_rename.
2114 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2117 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2120 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2121 LCK_EX, MDS_INODELOCK_UPDATE,
2122 MF_MDC_CANCEL_FID2);
2127 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2129 if (fid_is_sane(&op_data->op_fid3)) {
2130 struct lmv_tgt_desc *tgt;
2132 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2134 RETURN(PTR_ERR(tgt));
2136 /* Cancel LOOKUP lock on its parent */
2137 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2138 LCK_EX, MDS_INODELOCK_LOOKUP,
2139 MF_MDC_CANCEL_FID3);
2143 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2144 LCK_EX, MDS_INODELOCK_FULL,
2145 MF_MDC_CANCEL_FID3);
2151 * Cancel all the locks on tgt child (fid4).
2153 if (fid_is_sane(&op_data->op_fid4))
2154 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2155 LCK_EX, MDS_INODELOCK_FULL,
2156 MF_MDC_CANCEL_FID4);
2158 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2159 op_data->op_mds, PFID(&op_data->op_fid2));
2161 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2167 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2168 void *ea, int ealen, void *ea2, int ea2len,
2169 struct ptlrpc_request **request,
2170 struct md_open_data **mod)
2172 struct obd_device *obd = exp->exp_obd;
2173 struct lmv_obd *lmv = &obd->u.lmv;
2174 struct lmv_tgt_desc *tgt;
2178 rc = lmv_check_connect(obd);
2182 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2183 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2185 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2186 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2188 RETURN(PTR_ERR(tgt));
2190 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2191 ea2len, request, mod);
2196 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2197 struct obd_capa *oc, struct ptlrpc_request **request)
2199 struct obd_device *obd = exp->exp_obd;
2200 struct lmv_obd *lmv = &obd->u.lmv;
2201 struct lmv_tgt_desc *tgt;
2205 rc = lmv_check_connect(obd);
2209 tgt = lmv_find_target(lmv, fid);
2211 RETURN(PTR_ERR(tgt));
2213 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2218 * Get current minimum entry from striped directory
2220 * This function will search the dir entry, whose hash value is the
2221 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2222 * only being called for striped directory.
2224 * \param[in] exp export of LMV
2225 * \param[in] op_data parameters transferred beween client MD stack
2226 * stripe_information will be included in this
2228 * \param[in] cb_op ldlm callback being used in enqueue in
2230 * \param[in] hash_offset the hash value, which is used to locate
2231 * minum(closet) dir entry
2232 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2233 * index of last entry, so to avoid hash conflict
2234 * between stripes. It will also be used to
2235 * return the stripe index of current dir entry.
2236 * \param[in|out] entp the minum entry and it also is being used
2237 * to input the last dir entry to resolve the
2240 * \param[out] ppage the page which holds the minum entry
2242 * \retval = 0 get the entry successfully
2243 * negative errno (< 0) does not get the entry
2245 static int lmv_get_min_striped_entry(struct obd_export *exp,
2246 struct md_op_data *op_data,
2247 struct md_callback *cb_op,
2248 __u64 hash_offset, int *stripe_offset,
2249 struct lu_dirent **entp,
2250 struct page **ppage)
2252 struct obd_device *obd = exp->exp_obd;
2253 struct lmv_obd *lmv = &obd->u.lmv;
2254 struct lmv_stripe_md *lsm = op_data->op_mea1;
2255 struct lmv_tgt_desc *tgt;
2257 struct lu_dirent *min_ent = NULL;
2258 struct page *min_page = NULL;
2264 stripe_count = lsm->lsm_md_stripe_count;
2265 for (i = 0; i < stripe_count; i++) {
2266 struct lu_dirent *ent = NULL;
2267 struct page *page = NULL;
2268 struct lu_dirpage *dp;
2269 __u64 stripe_hash = hash_offset;
2271 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2273 GOTO(out, rc = PTR_ERR(tgt));
2275 /* op_data will be shared by each stripe, so we need
2276 * reset these value for each stripe */
2277 op_data->op_stripe_offset = i;
2278 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2279 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2280 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2282 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2287 dp = page_address(page);
2288 for (ent = lu_dirent_start(dp); ent != NULL;
2289 ent = lu_dirent_next(ent)) {
2290 /* Skip dummy entry */
2291 if (le16_to_cpu(ent->lde_namelen) == 0)
2294 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2297 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2298 (*entp == ent || i < *stripe_offset))
2301 /* skip . and .. for other stripes */
2303 (strncmp(ent->lde_name, ".",
2304 le16_to_cpu(ent->lde_namelen)) == 0 ||
2305 strncmp(ent->lde_name, "..",
2306 le16_to_cpu(ent->lde_namelen)) == 0))
2312 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2315 page_cache_release(page);
2318 /* reach the end of current stripe, go to next stripe */
2319 if (stripe_hash == MDS_DIR_END_OFF)
2325 if (min_ent != NULL) {
2326 if (le64_to_cpu(min_ent->lde_hash) >
2327 le64_to_cpu(ent->lde_hash)) {
2330 page_cache_release(min_page);
2335 page_cache_release(page);
2346 if (*ppage != NULL) {
2348 page_cache_release(*ppage);
2350 *stripe_offset = min_idx;
2357 * Build dir entry page from a striped directory
2359 * This function gets one entry by @offset from a striped directory. It will
2360 * read entries from all of stripes, and choose one closest to the required
2361 * offset(&offset). A few notes
2362 * 1. skip . and .. for non-zero stripes, because there can only have one .
2363 * and .. in a directory.
2364 * 2. op_data will be shared by all of stripes, instead of allocating new
2365 * one, so need to restore before reusing.
2366 * 3. release the entry page if that is not being chosen.
2368 * \param[in] exp obd export refer to LMV
2369 * \param[in] op_data hold those MD parameters of read_entry
2370 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2371 * \param[out] ldp the entry being read
2372 * \param[out] ppage the page holding the entry. Note: because the entry
2373 * will be accessed in upper layer, so we need hold the
2374 * page until the usages of entry is finished, see
2375 * ll_dir_entry_next.
2377 * retval =0 if get entry successfully
2378 * <0 cannot get entry
2380 static int lmv_read_striped_page(struct obd_export *exp,
2381 struct md_op_data *op_data,
2382 struct md_callback *cb_op,
2383 __u64 offset, struct page **ppage)
2385 struct obd_device *obd = exp->exp_obd;
2386 struct lu_fid master_fid = op_data->op_fid1;
2387 struct inode *master_inode = op_data->op_data;
2388 __u64 hash_offset = offset;
2389 struct lu_dirpage *dp;
2390 struct page *min_ent_page = NULL;
2391 struct page *ent_page = NULL;
2392 struct lu_dirent *ent;
2395 struct lu_dirent *min_ent = NULL;
2396 struct lu_dirent *last_ent;
2401 rc = lmv_check_connect(obd);
2405 /* Allocate a page and read entries from all of stripes and fill
2406 * the page by hash order */
2407 ent_page = alloc_page(GFP_KERNEL);
2408 if (ent_page == NULL)
2411 /* Initialize the entry page */
2412 dp = kmap(ent_page);
2413 memset(dp, 0, sizeof(*dp));
2414 dp->ldp_hash_start = cpu_to_le64(offset);
2415 dp->ldp_flags |= LDF_COLLIDE;
2418 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2424 /* Find the minum entry from all sub-stripes */
2425 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2431 /* If it can not get minum entry, it means it already reaches
2432 * the end of this directory */
2433 if (min_ent == NULL) {
2434 last_ent->lde_reclen = 0;
2435 hash_offset = MDS_DIR_END_OFF;
2439 ent_size = le16_to_cpu(min_ent->lde_reclen);
2441 /* the last entry lde_reclen is 0, but it might not
2442 * the end of this entry of this temporay entry */
2444 ent_size = lu_dirent_calc_size(
2445 le16_to_cpu(min_ent->lde_namelen),
2446 le32_to_cpu(min_ent->lde_attrs));
2447 if (ent_size > left_bytes) {
2448 last_ent->lde_reclen = cpu_to_le16(0);
2449 hash_offset = le64_to_cpu(min_ent->lde_hash);
2453 memcpy(ent, min_ent, ent_size);
2455 /* Replace . with master FID and Replace .. with the parent FID
2456 * of master object */
2457 if (strncmp(ent->lde_name, ".",
2458 le16_to_cpu(ent->lde_namelen)) == 0 &&
2459 le16_to_cpu(ent->lde_namelen) == 1)
2460 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2461 else if (strncmp(ent->lde_name, "..",
2462 le16_to_cpu(ent->lde_namelen)) == 0 &&
2463 le16_to_cpu(ent->lde_namelen) == 2)
2464 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2466 left_bytes -= ent_size;
2467 ent->lde_reclen = cpu_to_le16(ent_size);
2469 ent = (void *)ent + ent_size;
2470 hash_offset = le64_to_cpu(min_ent->lde_hash);
2471 if (hash_offset == MDS_DIR_END_OFF) {
2472 last_ent->lde_reclen = 0;
2477 if (min_ent_page != NULL) {
2478 kunmap(min_ent_page);
2479 page_cache_release(min_ent_page);
2482 if (unlikely(rc != 0)) {
2483 __free_page(ent_page);
2487 dp->ldp_flags |= LDF_EMPTY;
2488 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2489 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2492 /* We do not want to allocate md_op_data during each
2493 * dir entry reading, so op_data will be shared by every stripe,
2494 * then we need to restore it back to original value before
2495 * return to the upper layer */
2496 op_data->op_fid1 = master_fid;
2497 op_data->op_fid2 = master_fid;
2498 op_data->op_data = master_inode;
2505 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2506 struct md_callback *cb_op, __u64 offset,
2507 struct page **ppage)
2509 struct obd_device *obd = exp->exp_obd;
2510 struct lmv_obd *lmv = &obd->u.lmv;
2511 struct lmv_stripe_md *lsm = op_data->op_mea1;
2512 struct lmv_tgt_desc *tgt;
2516 rc = lmv_check_connect(obd);
2520 if (unlikely(lsm != NULL)) {
2521 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2525 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2527 RETURN(PTR_ERR(tgt));
2529 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2535 * Unlink a file/directory
2537 * Unlink a file or directory under the parent dir. The unlink request
2538 * usually will be sent to the MDT where the child is located, but if
2539 * the client does not have the child FID then request will be sent to the
2540 * MDT where the parent is located.
2542 * If the parent is a striped directory then it also needs to locate which
2543 * stripe the name of the child is located, and replace the parent FID
2544 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2545 * it will walk through all of sub-stripes until the child is being
2548 * \param[in] exp export refer to LMV
2549 * \param[in] op_data different parameters transferred beween client
2550 * MD stacks, name, namelen, FIDs etc.
2551 * op_fid1 is the parent FID, op_fid2 is the child
2553 * \param[out] request point to the request of unlink.
2555 * retval 0 if succeed
2556 * negative errno if failed.
2558 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2559 struct ptlrpc_request **request)
2561 struct obd_device *obd = exp->exp_obd;
2562 struct lmv_obd *lmv = &obd->u.lmv;
2563 struct lmv_tgt_desc *tgt = NULL;
2564 struct lmv_tgt_desc *parent_tgt = NULL;
2565 struct mdt_body *body;
2567 int stripe_index = 0;
2568 struct lmv_stripe_md *lsm = op_data->op_mea1;
2571 rc = lmv_check_connect(obd);
2575 /* For striped dir, we need to locate the parent as well */
2577 struct lmv_tgt_desc *tmp;
2579 LASSERT(op_data->op_name != NULL &&
2580 op_data->op_namelen != 0);
2582 tmp = lmv_locate_target_for_name(lmv, lsm,
2584 op_data->op_namelen,
2588 /* return -EBADFD means unknown hash type, might
2589 * need try all sub-stripe here */
2590 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2591 RETURN(PTR_ERR(tmp));
2593 /* Note: both migrating dir and unknown hash dir need to
2594 * try all of sub-stripes, so we need start search the
2595 * name from stripe 0, but migrating dir is already handled
2596 * inside lmv_locate_target_for_name(), so we only check
2597 * unknown hash type directory here */
2598 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2599 struct lmv_oinfo *oinfo;
2601 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2603 op_data->op_fid1 = oinfo->lmo_fid;
2604 op_data->op_mds = oinfo->lmo_mds;
2609 /* Send unlink requests to the MDT where the child is located */
2610 if (likely(!fid_is_zero(&op_data->op_fid2)))
2611 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2612 else if (lsm != NULL)
2613 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2615 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2618 RETURN(PTR_ERR(tgt));
2620 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2621 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2622 op_data->op_cap = cfs_curproc_cap_pack();
2625 * If child's fid is given, cancel unused locks for it if it is from
2626 * another export than parent.
2628 * LOOKUP lock for child (fid3) should also be cancelled on parent
2629 * tgt_tgt in mdc_unlink().
2631 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2634 * Cancel FULL locks on child (fid3).
2636 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2637 if (IS_ERR(parent_tgt))
2638 RETURN(PTR_ERR(parent_tgt));
2640 if (parent_tgt != tgt) {
2641 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2642 LCK_EX, MDS_INODELOCK_LOOKUP,
2643 MF_MDC_CANCEL_FID3);
2646 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2647 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2651 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2652 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2654 rc = md_unlink(tgt->ltd_exp, op_data, request);
2655 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2658 /* Try next stripe if it is needed. */
2659 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2660 struct lmv_oinfo *oinfo;
2663 if (stripe_index >= lsm->lsm_md_stripe_count)
2666 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2668 op_data->op_fid1 = oinfo->lmo_fid;
2669 op_data->op_mds = oinfo->lmo_mds;
2671 ptlrpc_req_finished(*request);
2674 goto try_next_stripe;
2677 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2681 /* Not cross-ref case, just get out of here. */
2682 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2685 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2686 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2688 /* This is a remote object, try remote MDT, Note: it may
2689 * try more than 1 time here, Considering following case
2690 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2691 * 1. Initially A does not know where remote1 is, it send
2692 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2693 * resend unlink RPC to MDT1 (retry 1st time).
2695 * 2. During the unlink RPC in flight,
2696 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2697 * and create new remote1, but on MDT0
2699 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2700 * /mnt/lustre, then lookup get fid of remote1, and find
2701 * it is remote dir again, and replay -EREMOTE again.
2703 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2705 * In theory, it might try unlimited time here, but it should
2706 * be very rare case. */
2707 op_data->op_fid2 = body->mbo_fid1;
2708 ptlrpc_req_finished(*request);
2714 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2716 struct lmv_obd *lmv = &obd->u.lmv;
2720 case OBD_CLEANUP_EARLY:
2721 /* XXX: here should be calling obd_precleanup() down to
2724 case OBD_CLEANUP_EXPORTS:
2725 fld_client_proc_fini(&lmv->lmv_fld);
2726 lprocfs_obd_cleanup(obd);
2727 lprocfs_free_md_stats(obd);
2735 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2736 __u32 keylen, void *key, __u32 *vallen, void *val,
2737 struct lov_stripe_md *lsm)
2739 struct obd_device *obd;
2740 struct lmv_obd *lmv;
2744 obd = class_exp2obd(exp);
2746 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2747 exp->exp_handle.h_cookie);
2752 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2755 rc = lmv_check_connect(obd);
2759 LASSERT(*vallen == sizeof(__u32));
2760 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2761 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2763 * All tgts should be connected when this gets called.
2765 if (tgt == NULL || tgt->ltd_exp == NULL)
2768 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2773 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2774 KEY_IS(KEY_DEFAULT_EASIZE) ||
2775 KEY_IS(KEY_MAX_COOKIESIZE) ||
2776 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2777 KEY_IS(KEY_CONN_DATA)) {
2778 rc = lmv_check_connect(obd);
2783 * Forwarding this request to first MDS, it should know LOV
2786 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2788 if (!rc && KEY_IS(KEY_CONN_DATA))
2789 exp->exp_connect_data = *(struct obd_connect_data *)val;
2791 } else if (KEY_IS(KEY_TGT_COUNT)) {
2792 *((int *)val) = lmv->desc.ld_tgt_count;
2796 CDEBUG(D_IOCTL, "Invalid key\n");
2800 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2801 obd_count keylen, void *key, obd_count vallen,
2802 void *val, struct ptlrpc_request_set *set)
2804 struct lmv_tgt_desc *tgt = NULL;
2805 struct obd_device *obd;
2806 struct lmv_obd *lmv;
2810 obd = class_exp2obd(exp);
2812 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2813 exp->exp_handle.h_cookie);
2818 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2821 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2824 if (tgt == NULL || tgt->ltd_exp == NULL)
2827 err = obd_set_info_async(env, tgt->ltd_exp,
2828 keylen, key, vallen, val, set);
2839 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2840 struct lmv_mds_md_v1 *lmm1)
2845 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2846 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2847 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2848 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2849 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2850 sizeof(lmm1->lmv_pool_name));
2851 if (cplen >= sizeof(lmm1->lmv_pool_name))
2854 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2855 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2856 &lsm->lsm_md_oinfo[i].lmo_fid);
2860 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2864 bool allocated = false;
2868 LASSERT(lmmp != NULL);
2870 if (*lmmp != NULL && lsm == NULL) {
2873 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2874 lmm_size = lmv_mds_md_size(stripe_count,
2875 le32_to_cpu((*lmmp)->lmv_magic));
2878 OBD_FREE(*lmmp, lmm_size);
2884 if (*lmmp == NULL && lsm == NULL) {
2885 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2886 LASSERT(lmm_size > 0);
2887 OBD_ALLOC(*lmmp, lmm_size);
2890 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2891 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2896 LASSERT(lsm != NULL);
2897 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2898 if (*lmmp == NULL) {
2899 OBD_ALLOC(*lmmp, lmm_size);
2905 switch (lsm->lsm_md_magic) {
2907 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2914 if (rc != 0 && allocated) {
2915 OBD_FREE(*lmmp, lmm_size);
2921 EXPORT_SYMBOL(lmv_pack_md);
2923 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2924 const struct lmv_mds_md_v1 *lmm1)
2926 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2933 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2934 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2935 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2936 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2937 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2939 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2940 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2941 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2942 sizeof(lsm->lsm_md_pool_name));
2944 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2947 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2948 "layout_version %d\n", lsm->lsm_md_stripe_count,
2949 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2950 lsm->lsm_md_layout_version);
2952 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2953 for (i = 0; i < stripe_count; i++) {
2954 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2955 &lmm1->lmv_stripe_fids[i]);
2956 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2957 &lsm->lsm_md_oinfo[i].lmo_mds);
2960 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2961 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2967 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2968 const union lmv_mds_md *lmm, int stripe_count)
2970 struct lmv_stripe_md *lsm;
2973 bool allocated = false;
2976 LASSERT(lsmp != NULL);
2980 if (lsm != NULL && lmm == NULL) {
2982 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2983 /* For migrating inode, the master stripe and master
2984 * object will be the same, so do not need iput, see
2985 * ll_update_lsm_md */
2986 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2987 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2988 iput(lsm->lsm_md_oinfo[i].lmo_root);
2990 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2991 OBD_FREE(lsm, lsm_size);
2997 if (lsm == NULL && lmm == NULL) {
2998 lsm_size = lmv_stripe_md_size(stripe_count);
2999 OBD_ALLOC(lsm, lsm_size);
3002 lsm->lsm_md_stripe_count = stripe_count;
3007 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3011 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3012 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3013 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3014 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3019 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3020 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3023 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3024 * stripecount should be 0 then.
3026 lsm_size = lmv_stripe_md_size(0);
3028 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3030 OBD_ALLOC(lsm, lsm_size);
3037 switch (le32_to_cpu(lmm->lmv_magic)) {
3039 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3042 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3043 le32_to_cpu(lmm->lmv_magic));
3048 if (rc != 0 && allocated) {
3049 OBD_FREE(lsm, lsm_size);
3056 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3058 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3060 EXPORT_SYMBOL(lmv_alloc_memmd);
3062 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3064 lmv_unpack_md(NULL, &lsm, NULL, 0);
3066 EXPORT_SYMBOL(lmv_free_memmd);
3068 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3069 struct lov_mds_md *lmm, int disk_len)
3071 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3072 (union lmv_mds_md *)lmm, disk_len);
3075 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3076 struct lov_stripe_md *lsm)
3078 struct obd_device *obd = exp->exp_obd;
3079 struct lmv_obd *lmv_obd = &obd->u.lmv;
3080 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3085 stripe_count = lmv->lsm_md_stripe_count;
3087 stripe_count = lmv_obd->desc.ld_tgt_count;
3089 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3092 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3095 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3096 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3097 ldlm_cancel_flags_t flags, void *opaque)
3099 struct obd_device *obd = exp->exp_obd;
3100 struct lmv_obd *lmv = &obd->u.lmv;
3106 LASSERT(fid != NULL);
3108 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3109 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3111 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3114 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3122 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3125 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3126 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3130 if (tgt == NULL || tgt->ltd_exp == NULL)
3132 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3136 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3137 const struct lu_fid *fid, ldlm_type_t type,
3138 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3139 struct lustre_handle *lockh)
3141 struct obd_device *obd = exp->exp_obd;
3142 struct lmv_obd *lmv = &obd->u.lmv;
3148 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3151 * With DNE every object can have two locks in different namespaces:
3152 * lookup lock in space of MDT storing direntry and update/open lock in
3153 * space of MDT storing inode. Try the MDT that the FID maps to first,
3154 * since this can be easily found, and only try others if that fails.
3156 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3157 i < lmv->desc.ld_tgt_count;
3158 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3160 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3161 obd->obd_name, PFID(fid), tgt);
3165 if (lmv->tgts[tgt] == NULL ||
3166 lmv->tgts[tgt]->ltd_exp == NULL ||
3167 lmv->tgts[tgt]->ltd_active == 0)
3170 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3171 type, policy, mode, lockh);
3179 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3180 struct obd_export *dt_exp, struct obd_export *md_exp,
3181 struct lustre_md *md)
3183 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3184 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3186 if (tgt == NULL || tgt->ltd_exp == NULL)
3189 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3192 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3194 struct obd_device *obd = exp->exp_obd;
3195 struct lmv_obd *lmv = &obd->u.lmv;
3196 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3199 if (md->lmv != NULL) {
3200 lmv_free_memmd(md->lmv);
3203 if (tgt == NULL || tgt->ltd_exp == NULL)
3205 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3208 int lmv_set_open_replay_data(struct obd_export *exp,
3209 struct obd_client_handle *och,
3210 struct lookup_intent *it)
3212 struct obd_device *obd = exp->exp_obd;
3213 struct lmv_obd *lmv = &obd->u.lmv;
3214 struct lmv_tgt_desc *tgt;
3217 tgt = lmv_find_target(lmv, &och->och_fid);
3219 RETURN(PTR_ERR(tgt));
3221 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3224 int lmv_clear_open_replay_data(struct obd_export *exp,
3225 struct obd_client_handle *och)
3227 struct obd_device *obd = exp->exp_obd;
3228 struct lmv_obd *lmv = &obd->u.lmv;
3229 struct lmv_tgt_desc *tgt;
3232 tgt = lmv_find_target(lmv, &och->och_fid);
3234 RETURN(PTR_ERR(tgt));
3236 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3239 static int lmv_get_remote_perm(struct obd_export *exp,
3240 const struct lu_fid *fid,
3241 struct obd_capa *oc, __u32 suppgid,
3242 struct ptlrpc_request **request)
3244 struct obd_device *obd = exp->exp_obd;
3245 struct lmv_obd *lmv = &obd->u.lmv;
3246 struct lmv_tgt_desc *tgt;
3250 rc = lmv_check_connect(obd);
3254 tgt = lmv_find_target(lmv, fid);
3256 RETURN(PTR_ERR(tgt));
3258 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3262 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3265 struct obd_device *obd = exp->exp_obd;
3266 struct lmv_obd *lmv = &obd->u.lmv;
3267 struct lmv_tgt_desc *tgt;
3271 rc = lmv_check_connect(obd);
3275 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3277 RETURN(PTR_ERR(tgt));
3279 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3283 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3284 const struct req_msg_field *field, struct obd_capa **oc)
3286 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3287 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3289 if (tgt == NULL || tgt->ltd_exp == NULL)
3291 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3294 int lmv_intent_getattr_async(struct obd_export *exp,
3295 struct md_enqueue_info *minfo,
3296 struct ldlm_enqueue_info *einfo)
3298 struct md_op_data *op_data = &minfo->mi_data;
3299 struct obd_device *obd = exp->exp_obd;
3300 struct lmv_obd *lmv = &obd->u.lmv;
3301 struct lmv_tgt_desc *tgt = NULL;
3305 rc = lmv_check_connect(obd);
3309 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3311 RETURN(PTR_ERR(tgt));
3313 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3317 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3318 struct lu_fid *fid, __u64 *bits)
3320 struct obd_device *obd = exp->exp_obd;
3321 struct lmv_obd *lmv = &obd->u.lmv;
3322 struct lmv_tgt_desc *tgt;
3326 rc = lmv_check_connect(obd);
3330 tgt = lmv_find_target(lmv, fid);
3332 RETURN(PTR_ERR(tgt));
3334 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3338 int lmv_get_fid_from_lsm(struct obd_export *exp,
3339 const struct lmv_stripe_md *lsm,
3340 const char *name, int namelen, struct lu_fid *fid)
3342 const struct lmv_oinfo *oinfo;
3344 LASSERT(lsm != NULL);
3345 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3347 return PTR_ERR(oinfo);
3349 *fid = oinfo->lmo_fid;
3355 * For lmv, only need to send request to master MDT, and the master MDT will
3356 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3357 * we directly fetch data from the slave MDTs.
3359 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3360 struct obd_quotactl *oqctl)
3362 struct obd_device *obd = class_exp2obd(exp);
3363 struct lmv_obd *lmv = &obd->u.lmv;
3364 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3367 __u64 curspace, curinodes;
3371 tgt->ltd_exp == NULL ||
3373 lmv->desc.ld_tgt_count == 0) {
3374 CERROR("master lmv inactive\n");
3378 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3379 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3383 curspace = curinodes = 0;
3384 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3388 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3391 err = obd_quotactl(tgt->ltd_exp, oqctl);
3393 CERROR("getquota on mdt %d failed. %d\n", i, err);
3397 curspace += oqctl->qc_dqblk.dqb_curspace;
3398 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3401 oqctl->qc_dqblk.dqb_curspace = curspace;
3402 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3407 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3408 struct obd_quotactl *oqctl)
3410 struct obd_device *obd = class_exp2obd(exp);
3411 struct lmv_obd *lmv = &obd->u.lmv;
3412 struct lmv_tgt_desc *tgt;
3417 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3420 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3421 CERROR("lmv idx %d inactive\n", i);
3425 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3433 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3434 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3436 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3439 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3440 struct cl_attr *attr)
3444 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3445 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3447 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3448 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3449 i_size_read(inode), inode->i_nlink,
3450 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3451 LTIME_S(inode->i_mtime));
3453 /* for slave stripe, it needs to subtract nlink for . and .. */
3455 attr->cat_nlink += inode->i_nlink - 2;
3457 attr->cat_nlink = inode->i_nlink;
3459 attr->cat_size += i_size_read(inode);
3461 if (attr->cat_atime < LTIME_S(inode->i_atime))
3462 attr->cat_atime = LTIME_S(inode->i_atime);
3464 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3465 attr->cat_ctime = LTIME_S(inode->i_ctime);
3467 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3468 attr->cat_mtime = LTIME_S(inode->i_mtime);
3473 struct obd_ops lmv_obd_ops = {
3474 .o_owner = THIS_MODULE,
3475 .o_setup = lmv_setup,
3476 .o_cleanup = lmv_cleanup,
3477 .o_precleanup = lmv_precleanup,
3478 .o_process_config = lmv_process_config,
3479 .o_connect = lmv_connect,
3480 .o_disconnect = lmv_disconnect,
3481 .o_statfs = lmv_statfs,
3482 .o_get_info = lmv_get_info,
3483 .o_set_info_async = lmv_set_info_async,
3484 .o_packmd = lmv_packmd,
3485 .o_unpackmd = lmv_unpackmd,
3486 .o_notify = lmv_notify,
3487 .o_get_uuid = lmv_get_uuid,
3488 .o_iocontrol = lmv_iocontrol,
3489 .o_quotacheck = lmv_quotacheck,
3490 .o_quotactl = lmv_quotactl
3493 struct md_ops lmv_md_ops = {
3494 .m_getstatus = lmv_getstatus,
3495 .m_null_inode = lmv_null_inode,
3496 .m_find_cbdata = lmv_find_cbdata,
3497 .m_close = lmv_close,
3498 .m_create = lmv_create,
3499 .m_done_writing = lmv_done_writing,
3500 .m_enqueue = lmv_enqueue,
3501 .m_getattr = lmv_getattr,
3502 .m_getxattr = lmv_getxattr,
3503 .m_getattr_name = lmv_getattr_name,
3504 .m_intent_lock = lmv_intent_lock,
3506 .m_rename = lmv_rename,
3507 .m_setattr = lmv_setattr,
3508 .m_setxattr = lmv_setxattr,
3509 .m_fsync = lmv_fsync,
3510 .m_read_page = lmv_read_page,
3511 .m_unlink = lmv_unlink,
3512 .m_init_ea_size = lmv_init_ea_size,
3513 .m_cancel_unused = lmv_cancel_unused,
3514 .m_set_lock_data = lmv_set_lock_data,
3515 .m_lock_match = lmv_lock_match,
3516 .m_get_lustre_md = lmv_get_lustre_md,
3517 .m_free_lustre_md = lmv_free_lustre_md,
3518 .m_update_lsm_md = lmv_update_lsm_md,
3519 .m_merge_attr = lmv_merge_attr,
3520 .m_set_open_replay_data = lmv_set_open_replay_data,
3521 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3522 .m_renew_capa = lmv_renew_capa,
3523 .m_unpack_capa = lmv_unpack_capa,
3524 .m_get_remote_perm = lmv_get_remote_perm,
3525 .m_intent_getattr_async = lmv_intent_getattr_async,
3526 .m_revalidate_lock = lmv_revalidate_lock,
3527 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3530 int __init lmv_init(void)
3532 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3533 LUSTRE_LMV_NAME, NULL);
3536 static void lmv_exit(void)
3538 class_unregister_type(LUSTRE_LMV_NAME);
3541 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3542 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3543 MODULE_LICENSE("GPL");
3545 module_init(lmv_init);
3546 module_exit(lmv_exit);