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, 2014, 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>
56 #include <lustre_fid.h>
57 #include <lustre_ioctl.h>
58 #include <lustre_kernelcomm.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_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,
265 __u32 easize, __u32 def_easize,
266 __u32 cookiesize, __u32 def_cookiesize)
268 struct obd_device *obd = exp->exp_obd;
269 struct lmv_obd *lmv = &obd->u.lmv;
275 if (lmv->max_easize < easize) {
276 lmv->max_easize = easize;
279 if (lmv->max_def_easize < def_easize) {
280 lmv->max_def_easize = def_easize;
283 if (lmv->max_cookiesize < cookiesize) {
284 lmv->max_cookiesize = cookiesize;
287 if (lmv->max_def_cookiesize < def_cookiesize) {
288 lmv->max_def_cookiesize = def_cookiesize;
294 if (lmv->connected == 0)
297 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
298 struct lmv_tgt_desc *tgt = lmv->tgts[i];
300 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
301 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
305 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
306 cookiesize, def_cookiesize);
308 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
309 " rc = %d\n", obd->obd_name, i, rc);
316 #define MAX_STRING_SIZE 128
318 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
320 struct lmv_obd *lmv = &obd->u.lmv;
321 struct obd_uuid *cluuid = &lmv->cluuid;
322 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
323 struct obd_device *mdc_obd;
324 struct obd_export *mdc_exp;
325 struct lu_fld_target target;
329 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
332 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
336 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
337 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
338 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
341 if (!mdc_obd->obd_set_up) {
342 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
346 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
347 &lmv->conn_data, NULL);
349 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
354 * Init fid sequence client for this mdc and add new fld target.
356 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
360 target.ft_srv = NULL;
361 target.ft_exp = mdc_exp;
362 target.ft_idx = tgt->ltd_idx;
364 fld_client_add_target(&lmv->lmv_fld, &target);
366 rc = obd_register_observer(mdc_obd, obd);
368 obd_disconnect(mdc_exp);
369 CERROR("target %s register_observer error %d\n",
370 tgt->ltd_uuid.uuid, rc);
374 if (obd->obd_observer) {
376 * Tell the observer about the new target.
378 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
380 (void *)(tgt - lmv->tgts[0]));
382 obd_disconnect(mdc_exp);
388 tgt->ltd_exp = mdc_exp;
389 lmv->desc.ld_active_tgt_count++;
391 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
392 lmv->max_cookiesize, lmv->max_def_cookiesize);
394 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
395 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
396 atomic_read(&obd->obd_refcount));
398 if (lmv->targets_proc_entry != NULL) {
399 struct proc_dir_entry *mdc_symlink;
401 LASSERT(mdc_obd->obd_type != NULL);
402 LASSERT(mdc_obd->obd_type->typ_name != NULL);
403 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
404 lmv->targets_proc_entry,
406 mdc_obd->obd_type->typ_name,
408 if (mdc_symlink == NULL) {
409 CERROR("cannot register LMV target "
410 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
411 obd->obd_type->typ_name, obd->obd_name,
418 static void lmv_del_target(struct lmv_obd *lmv, int index)
420 if (lmv->tgts[index] == NULL)
423 OBD_FREE_PTR(lmv->tgts[index]);
424 lmv->tgts[index] = NULL;
428 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
429 __u32 index, int gen)
431 struct lmv_obd *lmv = &obd->u.lmv;
432 struct lmv_tgt_desc *tgt;
433 int orig_tgt_count = 0;
437 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
439 mutex_lock(&lmv->lmv_init_mutex);
441 if (lmv->desc.ld_tgt_count == 0) {
442 struct obd_device *mdc_obd;
444 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
447 mutex_unlock(&lmv->lmv_init_mutex);
448 CERROR("%s: Target %s not attached: rc = %d\n",
449 obd->obd_name, uuidp->uuid, -EINVAL);
454 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
455 tgt = lmv->tgts[index];
456 CERROR("%s: UUID %s already assigned at LOV target index %d:"
457 " rc = %d\n", obd->obd_name,
458 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
459 mutex_unlock(&lmv->lmv_init_mutex);
463 if (index >= lmv->tgts_size) {
464 /* We need to reallocate the lmv target array. */
465 struct lmv_tgt_desc **newtgts, **old = NULL;
469 while (newsize < index + 1)
470 newsize = newsize << 1;
471 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
472 if (newtgts == NULL) {
473 mutex_unlock(&lmv->lmv_init_mutex);
477 if (lmv->tgts_size) {
478 memcpy(newtgts, lmv->tgts,
479 sizeof(*newtgts) * lmv->tgts_size);
481 oldsize = lmv->tgts_size;
485 lmv->tgts_size = newsize;
488 OBD_FREE(old, sizeof(*old) * oldsize);
490 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
496 mutex_unlock(&lmv->lmv_init_mutex);
500 mutex_init(&tgt->ltd_fid_mutex);
501 tgt->ltd_idx = index;
502 tgt->ltd_uuid = *uuidp;
504 lmv->tgts[index] = tgt;
505 if (index >= lmv->desc.ld_tgt_count) {
506 orig_tgt_count = lmv->desc.ld_tgt_count;
507 lmv->desc.ld_tgt_count = index + 1;
510 if (lmv->connected) {
511 rc = lmv_connect_mdc(obd, tgt);
513 spin_lock(&lmv->lmv_lock);
514 if (lmv->desc.ld_tgt_count == index + 1)
515 lmv->desc.ld_tgt_count = orig_tgt_count;
516 memset(tgt, 0, sizeof(*tgt));
517 spin_unlock(&lmv->lmv_lock);
519 int easize = sizeof(struct lmv_stripe_md) +
520 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
521 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
525 mutex_unlock(&lmv->lmv_init_mutex);
529 int lmv_check_connect(struct obd_device *obd)
531 struct lmv_obd *lmv = &obd->u.lmv;
532 struct lmv_tgt_desc *tgt;
541 mutex_lock(&lmv->lmv_init_mutex);
542 if (lmv->connected) {
543 mutex_unlock(&lmv->lmv_init_mutex);
547 if (lmv->desc.ld_tgt_count == 0) {
548 mutex_unlock(&lmv->lmv_init_mutex);
549 CERROR("%s: no targets configured.\n", obd->obd_name);
553 LASSERT(lmv->tgts != NULL);
555 if (lmv->tgts[0] == NULL) {
556 mutex_unlock(&lmv->lmv_init_mutex);
557 CERROR("%s: no target configured for index 0.\n",
562 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
563 lmv->cluuid.uuid, obd->obd_name);
565 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
569 rc = lmv_connect_mdc(obd, tgt);
574 class_export_put(lmv->exp);
576 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
577 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
578 mutex_unlock(&lmv->lmv_init_mutex);
589 --lmv->desc.ld_active_tgt_count;
590 rc2 = obd_disconnect(tgt->ltd_exp);
592 CERROR("LMV target %s disconnect on "
593 "MDC idx %d: error %d\n",
594 tgt->ltd_uuid.uuid, i, rc2);
598 class_disconnect(lmv->exp);
599 mutex_unlock(&lmv->lmv_init_mutex);
603 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
605 struct lmv_obd *lmv = &obd->u.lmv;
606 struct obd_device *mdc_obd;
610 LASSERT(tgt != NULL);
611 LASSERT(obd != NULL);
613 mdc_obd = class_exp2obd(tgt->ltd_exp);
616 mdc_obd->obd_force = obd->obd_force;
617 mdc_obd->obd_fail = obd->obd_fail;
618 mdc_obd->obd_no_recov = obd->obd_no_recov;
621 if (lmv->targets_proc_entry != NULL)
622 lprocfs_remove_proc_entry(mdc_obd->obd_name,
623 lmv->targets_proc_entry);
625 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
627 CERROR("Can't finanize fids factory\n");
629 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
630 tgt->ltd_exp->exp_obd->obd_name,
631 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
633 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
634 rc = obd_disconnect(tgt->ltd_exp);
636 if (tgt->ltd_active) {
637 CERROR("Target %s disconnect error %d\n",
638 tgt->ltd_uuid.uuid, rc);
642 lmv_activate_target(lmv, tgt, 0);
647 static int lmv_disconnect(struct obd_export *exp)
649 struct obd_device *obd = class_exp2obd(exp);
650 struct lmv_obd *lmv = &obd->u.lmv;
659 * Only disconnect the underlying layers on the final disconnect.
662 if (lmv->refcount != 0)
665 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
666 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
669 lmv_disconnect_mdc(obd, lmv->tgts[i]);
672 if (lmv->targets_proc_entry != NULL)
673 lprocfs_remove(&lmv->targets_proc_entry);
675 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
676 obd->obd_type->typ_name, obd->obd_name);
680 * This is the case when no real connection is established by
681 * lmv_check_connect().
684 class_export_put(exp);
685 rc = class_disconnect(exp);
686 if (lmv->refcount == 0)
691 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
693 struct obd_device *obddev = class_exp2obd(exp);
694 struct lmv_obd *lmv = &obddev->u.lmv;
695 struct getinfo_fid2path *gf;
696 struct lmv_tgt_desc *tgt;
697 struct getinfo_fid2path *remote_gf = NULL;
698 int remote_gf_size = 0;
701 gf = (struct getinfo_fid2path *)karg;
702 tgt = lmv_find_target(lmv, &gf->gf_fid);
704 RETURN(PTR_ERR(tgt));
707 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
708 if (rc != 0 && rc != -EREMOTE)
709 GOTO(out_fid2path, rc);
711 /* If remote_gf != NULL, it means just building the
712 * path on the remote MDT, copy this path segement to gf */
713 if (remote_gf != NULL) {
714 struct getinfo_fid2path *ori_gf;
717 ori_gf = (struct getinfo_fid2path *)karg;
718 if (strlen(ori_gf->gf_path) +
719 strlen(gf->gf_path) > ori_gf->gf_pathlen)
720 GOTO(out_fid2path, rc = -EOVERFLOW);
722 ptr = ori_gf->gf_path;
724 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
725 strlen(ori_gf->gf_path));
727 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
728 ptr += strlen(gf->gf_path);
732 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
733 tgt->ltd_exp->exp_obd->obd_name,
734 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
738 GOTO(out_fid2path, rc);
740 /* sigh, has to go to another MDT to do path building further */
741 if (remote_gf == NULL) {
742 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
743 OBD_ALLOC(remote_gf, remote_gf_size);
744 if (remote_gf == NULL)
745 GOTO(out_fid2path, rc = -ENOMEM);
746 remote_gf->gf_pathlen = PATH_MAX;
749 if (!fid_is_sane(&gf->gf_fid)) {
750 CERROR("%s: invalid FID "DFID": rc = %d\n",
751 tgt->ltd_exp->exp_obd->obd_name,
752 PFID(&gf->gf_fid), -EINVAL);
753 GOTO(out_fid2path, rc = -EINVAL);
756 tgt = lmv_find_target(lmv, &gf->gf_fid);
758 GOTO(out_fid2path, rc = -EINVAL);
760 remote_gf->gf_fid = gf->gf_fid;
761 remote_gf->gf_recno = -1;
762 remote_gf->gf_linkno = -1;
763 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
765 goto repeat_fid2path;
768 if (remote_gf != NULL)
769 OBD_FREE(remote_gf, remote_gf_size);
773 static int lmv_hsm_req_count(struct lmv_obd *lmv,
774 const struct hsm_user_request *hur,
775 const struct lmv_tgt_desc *tgt_mds)
779 struct lmv_tgt_desc *curr_tgt;
781 /* count how many requests must be sent to the given target */
782 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
783 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
784 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
790 static void lmv_hsm_req_build(struct lmv_obd *lmv,
791 struct hsm_user_request *hur_in,
792 const struct lmv_tgt_desc *tgt_mds,
793 struct hsm_user_request *hur_out)
796 struct lmv_tgt_desc *curr_tgt;
798 /* build the hsm_user_request for the given target */
799 hur_out->hur_request = hur_in->hur_request;
801 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
802 curr_tgt = lmv_find_target(lmv,
803 &hur_in->hur_user_item[i].hui_fid);
804 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
805 hur_out->hur_user_item[nr_out] =
806 hur_in->hur_user_item[i];
810 hur_out->hur_request.hr_itemcount = nr_out;
811 memcpy(hur_data(hur_out), hur_data(hur_in),
812 hur_in->hur_request.hr_data_len);
815 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
816 struct lustre_kernelcomm *lk, void *uarg)
820 struct kkuc_ct_data *kcd = NULL;
823 /* unregister request (call from llapi_hsm_copytool_fini) */
824 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
825 struct lmv_tgt_desc *tgt = lmv->tgts[i];
827 if (tgt == NULL || tgt->ltd_exp == NULL)
829 /* best effort: try to clean as much as possible
830 * (continue on error) */
831 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
834 /* Whatever the result, remove copytool from kuc groups.
835 * Unreached coordinators will get EPIPE on next requests
836 * and will unregister automatically.
838 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
845 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
846 struct lustre_kernelcomm *lk, void *uarg)
851 bool any_set = false;
852 struct kkuc_ct_data *kcd;
855 /* All or nothing: try to register to all MDS.
856 * In case of failure, unregister from previous MDS,
857 * except if it because of inactive target. */
858 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
859 struct lmv_tgt_desc *tgt = lmv->tgts[i];
861 if (tgt == NULL || tgt->ltd_exp == NULL)
863 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
865 if (tgt->ltd_active) {
866 /* permanent error */
867 CERROR("%s: iocontrol MDC %s on MDT"
868 " idx %d cmd %x: err = %d\n",
869 class_exp2obd(lmv->exp)->obd_name,
870 tgt->ltd_uuid.uuid, i, cmd, err);
872 lk->lk_flags |= LK_FLG_STOP;
873 /* unregister from previous MDS */
874 for (j = 0; j < i; j++) {
876 if (tgt == NULL || tgt->ltd_exp == NULL)
878 obd_iocontrol(cmd, tgt->ltd_exp, len,
883 /* else: transient error.
884 * kuc will register to the missing MDT
892 /* no registration done: return error */
895 /* at least one registration done, with no failure */
896 filp = fget(lk->lk_wfd);
905 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
906 kcd->kcd_uuid = lmv->cluuid;
907 kcd->kcd_archive = lk->lk_data;
909 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
922 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
923 int len, void *karg, void *uarg)
925 struct obd_device *obddev = class_exp2obd(exp);
926 struct lmv_obd *lmv = &obddev->u.lmv;
927 struct lmv_tgt_desc *tgt = NULL;
931 __u32 count = lmv->desc.ld_tgt_count;
938 case IOC_OBD_STATFS: {
939 struct obd_ioctl_data *data = karg;
940 struct obd_device *mdc_obd;
941 struct obd_statfs stat_buf = {0};
944 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
945 if ((index >= count))
948 tgt = lmv->tgts[index];
949 if (tgt == NULL || !tgt->ltd_active)
952 mdc_obd = class_exp2obd(tgt->ltd_exp);
957 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
958 min((int) data->ioc_plen2,
959 (int) sizeof(struct obd_uuid))))
962 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
963 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
967 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
968 min((int) data->ioc_plen1,
969 (int) sizeof(stat_buf))))
973 case OBD_IOC_QUOTACTL: {
974 struct if_quotactl *qctl = karg;
975 struct obd_quotactl *oqctl;
977 if (qctl->qc_valid == QC_MDTIDX) {
978 if (count <= qctl->qc_idx)
981 tgt = lmv->tgts[qctl->qc_idx];
982 if (tgt == NULL || tgt->ltd_exp == NULL)
984 } else if (qctl->qc_valid == QC_UUID) {
985 for (i = 0; i < count; i++) {
989 if (!obd_uuid_equals(&tgt->ltd_uuid,
993 if (tgt->ltd_exp == NULL)
1005 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1006 OBD_ALLOC_PTR(oqctl);
1010 QCTL_COPY(oqctl, qctl);
1011 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1013 QCTL_COPY(qctl, oqctl);
1014 qctl->qc_valid = QC_MDTIDX;
1015 qctl->obd_uuid = tgt->ltd_uuid;
1017 OBD_FREE_PTR(oqctl);
1020 case OBD_IOC_CHANGELOG_SEND:
1021 case OBD_IOC_CHANGELOG_CLEAR: {
1022 struct ioc_changelog *icc = karg;
1024 if (icc->icc_mdtindex >= count)
1027 tgt = lmv->tgts[icc->icc_mdtindex];
1028 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1030 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1033 case LL_IOC_GET_CONNECT_FLAGS: {
1035 if (tgt == NULL || tgt->ltd_exp == NULL)
1037 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1040 case LL_IOC_FID2MDTIDX: {
1041 struct lu_fid *fid = karg;
1044 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1048 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1049 * point to user space memory for FID2MDTIDX. */
1050 *(__u32 *)uarg = mdt_index;
1053 case OBD_IOC_FID2PATH: {
1054 rc = lmv_fid2path(exp, len, karg, uarg);
1057 case LL_IOC_HSM_STATE_GET:
1058 case LL_IOC_HSM_STATE_SET:
1059 case LL_IOC_HSM_ACTION: {
1060 struct md_op_data *op_data = karg;
1062 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1064 RETURN(PTR_ERR(tgt));
1066 if (tgt->ltd_exp == NULL)
1069 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1072 case LL_IOC_HSM_PROGRESS: {
1073 const struct hsm_progress_kernel *hpk = karg;
1075 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1077 RETURN(PTR_ERR(tgt));
1078 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1081 case LL_IOC_HSM_REQUEST: {
1082 struct hsm_user_request *hur = karg;
1083 unsigned int reqcount = hur->hur_request.hr_itemcount;
1088 /* if the request is about a single fid
1089 * or if there is a single MDS, no need to split
1091 if (reqcount == 1 || count == 1) {
1092 tgt = lmv_find_target(lmv,
1093 &hur->hur_user_item[0].hui_fid);
1095 RETURN(PTR_ERR(tgt));
1096 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1098 /* split fid list to their respective MDS */
1099 for (i = 0; i < count; i++) {
1100 unsigned int nr, reqlen;
1102 struct hsm_user_request *req;
1105 if (tgt == NULL || tgt->ltd_exp == NULL)
1108 nr = lmv_hsm_req_count(lmv, hur, tgt);
1109 if (nr == 0) /* nothing for this MDS */
1112 /* build a request with fids for this MDS */
1113 reqlen = offsetof(typeof(*hur),
1115 + hur->hur_request.hr_data_len;
1116 OBD_ALLOC_LARGE(req, reqlen);
1120 lmv_hsm_req_build(lmv, hur, tgt, req);
1122 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1124 if (rc1 != 0 && rc == 0)
1126 OBD_FREE_LARGE(req, reqlen);
1131 case LL_IOC_LOV_SWAP_LAYOUTS: {
1132 struct md_op_data *op_data = karg;
1133 struct lmv_tgt_desc *tgt1, *tgt2;
1135 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1137 RETURN(PTR_ERR(tgt1));
1139 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1141 RETURN(PTR_ERR(tgt2));
1143 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1146 /* only files on same MDT can have their layouts swapped */
1147 if (tgt1->ltd_idx != tgt2->ltd_idx)
1150 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1153 case LL_IOC_HSM_CT_START: {
1154 struct lustre_kernelcomm *lk = karg;
1155 if (lk->lk_flags & LK_FLG_STOP)
1156 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1158 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1162 for (i = 0; i < count; i++) {
1163 struct obd_device *mdc_obd;
1167 if (tgt == NULL || tgt->ltd_exp == NULL)
1169 /* ll_umount_begin() sets force flag but for lmv, not
1170 * mdc. Let's pass it through */
1171 mdc_obd = class_exp2obd(tgt->ltd_exp);
1172 mdc_obd->obd_force = obddev->obd_force;
1173 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1174 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1177 if (tgt->ltd_active) {
1178 CERROR("error: iocontrol MDC %s on MDT"
1179 " idx %d cmd %x: err = %d\n",
1180 tgt->ltd_uuid.uuid, i, cmd, err);
1194 static int lmv_all_chars_policy(int count, const char *name,
1205 static int lmv_nid_policy(struct lmv_obd *lmv)
1207 struct obd_import *imp;
1211 * XXX: To get nid we assume that underlying obd device is mdc.
1213 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1214 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1215 return id % lmv->desc.ld_tgt_count;
1218 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1219 placement_policy_t placement)
1221 switch (placement) {
1222 case PLACEMENT_CHAR_POLICY:
1223 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1225 op_data->op_namelen);
1226 case PLACEMENT_NID_POLICY:
1227 return lmv_nid_policy(lmv);
1233 CERROR("Unsupported placement policy %x\n", placement);
1239 * This is _inode_ placement policy function (not name).
1241 static int lmv_placement_policy(struct obd_device *obd,
1242 struct md_op_data *op_data, u32 *mds)
1244 struct lmv_obd *lmv = &obd->u.lmv;
1247 LASSERT(mds != NULL);
1249 if (lmv->desc.ld_tgt_count == 1) {
1254 if (op_data->op_default_stripe_offset != -1) {
1255 *mds = op_data->op_default_stripe_offset;
1260 * If stripe_offset is provided during setdirstripe
1261 * (setdirstripe -i xx), xx MDS will be choosen.
1263 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1264 struct lmv_user_md *lum;
1266 lum = op_data->op_data;
1268 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1269 *mds = le32_to_cpu(lum->lum_stripe_offset);
1271 /* -1 means default, which will be in the same MDT with
1273 *mds = op_data->op_mds;
1274 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1277 /* Allocate new fid on target according to operation type and
1278 * parent home mds. */
1279 *mds = op_data->op_mds;
1285 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1287 struct lmv_tgt_desc *tgt;
1291 tgt = lmv_get_target(lmv, mds, NULL);
1293 RETURN(PTR_ERR(tgt));
1296 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1297 * on server that seq in new allocated fid is not yet known.
1299 mutex_lock(&tgt->ltd_fid_mutex);
1301 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1302 GOTO(out, rc = -ENODEV);
1305 * Asking underlying tgt layer to allocate new fid.
1307 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1309 LASSERT(fid_is_sane(fid));
1315 mutex_unlock(&tgt->ltd_fid_mutex);
1319 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1320 struct lu_fid *fid, struct md_op_data *op_data)
1322 struct obd_device *obd = class_exp2obd(exp);
1323 struct lmv_obd *lmv = &obd->u.lmv;
1328 LASSERT(op_data != NULL);
1329 LASSERT(fid != NULL);
1331 rc = lmv_placement_policy(obd, op_data, &mds);
1333 CERROR("Can't get target for allocating fid, "
1338 rc = __lmv_fid_alloc(lmv, fid, mds);
1340 CERROR("Can't alloc new fid, rc %d\n", rc);
1347 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1349 struct lmv_obd *lmv = &obd->u.lmv;
1350 struct lmv_desc *desc;
1354 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1355 CERROR("LMV setup requires a descriptor\n");
1359 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1360 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1361 CERROR("Lmv descriptor size wrong: %d > %d\n",
1362 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1366 lmv->tgts_size = 32U;
1367 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1368 if (lmv->tgts == NULL)
1371 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1372 lmv->desc.ld_tgt_count = 0;
1373 lmv->desc.ld_active_tgt_count = 0;
1374 lmv->max_cookiesize = 0;
1375 lmv->max_def_easize = 0;
1376 lmv->max_easize = 0;
1377 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1379 spin_lock_init(&lmv->lmv_lock);
1380 mutex_init(&lmv->lmv_init_mutex);
1382 #ifdef CONFIG_PROC_FS
1383 obd->obd_vars = lprocfs_lmv_obd_vars;
1384 lprocfs_obd_setup(obd);
1385 lprocfs_alloc_md_stats(obd, 0);
1386 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1387 0444, &lmv_proc_target_fops, obd);
1389 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1392 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1393 LUSTRE_CLI_FLD_HASH_DHT);
1395 CERROR("Can't init FLD, err %d\n", rc);
1405 static int lmv_cleanup(struct obd_device *obd)
1407 struct lmv_obd *lmv = &obd->u.lmv;
1410 fld_client_fini(&lmv->lmv_fld);
1411 if (lmv->tgts != NULL) {
1413 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1414 if (lmv->tgts[i] == NULL)
1416 lmv_del_target(lmv, i);
1418 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1424 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1426 struct lustre_cfg *lcfg = buf;
1427 struct obd_uuid obd_uuid;
1433 switch (lcfg->lcfg_command) {
1435 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1436 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1437 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1438 GOTO(out, rc = -EINVAL);
1440 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1442 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1443 GOTO(out, rc = -EINVAL);
1444 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1445 GOTO(out, rc = -EINVAL);
1446 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1449 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1450 GOTO(out, rc = -EINVAL);
1456 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1457 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1459 struct obd_device *obd = class_exp2obd(exp);
1460 struct lmv_obd *lmv = &obd->u.lmv;
1461 struct obd_statfs *temp;
1466 rc = lmv_check_connect(obd);
1470 OBD_ALLOC(temp, sizeof(*temp));
1474 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1475 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1478 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1481 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1482 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1484 GOTO(out_free_temp, rc);
1489 /* If the statfs is from mount, it will needs
1490 * retrieve necessary information from MDT0.
1491 * i.e. mount does not need the merged osfs
1493 * And also clients can be mounted as long as
1494 * MDT0 is in service*/
1495 if (flags & OBD_STATFS_FOR_MDT0)
1496 GOTO(out_free_temp, rc);
1498 osfs->os_bavail += temp->os_bavail;
1499 osfs->os_blocks += temp->os_blocks;
1500 osfs->os_ffree += temp->os_ffree;
1501 osfs->os_files += temp->os_files;
1507 OBD_FREE(temp, sizeof(*temp));
1511 static int lmv_getstatus(struct obd_export *exp,
1513 struct obd_capa **pc)
1515 struct obd_device *obd = exp->exp_obd;
1516 struct lmv_obd *lmv = &obd->u.lmv;
1520 rc = lmv_check_connect(obd);
1524 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1528 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1529 struct obd_capa *oc, u64 valid, const char *name,
1530 const char *input, int input_size, int output_size,
1531 int flags, struct ptlrpc_request **request)
1533 struct obd_device *obd = exp->exp_obd;
1534 struct lmv_obd *lmv = &obd->u.lmv;
1535 struct lmv_tgt_desc *tgt;
1539 rc = lmv_check_connect(obd);
1543 tgt = lmv_find_target(lmv, fid);
1545 RETURN(PTR_ERR(tgt));
1547 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1548 input_size, output_size, flags, request);
1553 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1554 struct obd_capa *oc, u64 valid, const char *name,
1555 const char *input, int input_size, int output_size,
1556 int flags, __u32 suppgid,
1557 struct ptlrpc_request **request)
1559 struct obd_device *obd = exp->exp_obd;
1560 struct lmv_obd *lmv = &obd->u.lmv;
1561 struct lmv_tgt_desc *tgt;
1565 rc = lmv_check_connect(obd);
1569 tgt = lmv_find_target(lmv, fid);
1571 RETURN(PTR_ERR(tgt));
1573 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1574 input_size, output_size, flags, suppgid,
1580 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1581 struct ptlrpc_request **request)
1583 struct obd_device *obd = exp->exp_obd;
1584 struct lmv_obd *lmv = &obd->u.lmv;
1585 struct lmv_tgt_desc *tgt;
1589 rc = lmv_check_connect(obd);
1593 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1595 RETURN(PTR_ERR(tgt));
1597 if (op_data->op_flags & MF_GET_MDT_IDX) {
1598 op_data->op_mds = tgt->ltd_idx;
1602 rc = md_getattr(tgt->ltd_exp, op_data, request);
1607 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1609 struct obd_device *obd = exp->exp_obd;
1610 struct lmv_obd *lmv = &obd->u.lmv;
1615 rc = lmv_check_connect(obd);
1619 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1622 * With DNE every object can have two locks in different namespaces:
1623 * lookup lock in space of MDT storing direntry and update/open lock in
1624 * space of MDT storing inode.
1626 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1627 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1629 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1635 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1636 ldlm_iterator_t it, void *data)
1638 struct obd_device *obd = exp->exp_obd;
1639 struct lmv_obd *lmv = &obd->u.lmv;
1645 rc = lmv_check_connect(obd);
1649 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1652 * With DNE every object can have two locks in different namespaces:
1653 * lookup lock in space of MDT storing direntry and update/open lock in
1654 * space of MDT storing inode. Try the MDT that the FID maps to first,
1655 * since this can be easily found, and only try others if that fails.
1657 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1658 i < lmv->desc.ld_tgt_count;
1659 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1661 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1662 obd->obd_name, PFID(fid), tgt);
1666 if (lmv->tgts[tgt] == NULL ||
1667 lmv->tgts[tgt]->ltd_exp == NULL)
1670 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1679 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1680 struct md_open_data *mod, struct ptlrpc_request **request)
1682 struct obd_device *obd = exp->exp_obd;
1683 struct lmv_obd *lmv = &obd->u.lmv;
1684 struct lmv_tgt_desc *tgt;
1688 rc = lmv_check_connect(obd);
1692 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1694 RETURN(PTR_ERR(tgt));
1696 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1697 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1702 * Choosing the MDT by name or FID in @op_data.
1703 * For non-striped directory, it will locate MDT by fid.
1704 * For striped-directory, it will locate MDT by name. And also
1705 * it will reset op_fid1 with the FID of the choosen stripe.
1707 struct lmv_tgt_desc *
1708 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1709 const char *name, int namelen, struct lu_fid *fid,
1712 struct lmv_tgt_desc *tgt;
1713 const struct lmv_oinfo *oinfo;
1715 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1716 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1717 RETURN(ERR_PTR(-EBADF));
1718 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1720 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1722 RETURN(ERR_CAST(oinfo));
1726 *fid = oinfo->lmo_fid;
1728 *mds = oinfo->lmo_mds;
1730 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1732 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1733 PFID(&oinfo->lmo_fid));
1738 * Locate mds by fid or name
1740 * For striped directory (lsm != NULL), it will locate the stripe
1741 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1742 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1743 * walk through all of stripes to locate the entry.
1745 * For normal direcotry, it will locate MDS by FID directly.
1746 * \param[in] lmv LMV device
1747 * \param[in] op_data client MD stack parameters, name, namelen
1749 * \param[in] fid object FID used to locate MDS.
1751 * retval pointer to the lmv_tgt_desc if succeed.
1752 * ERR_PTR(errno) if failed.
1754 struct lmv_tgt_desc*
1755 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1758 struct lmv_stripe_md *lsm = op_data->op_mea1;
1759 struct lmv_tgt_desc *tgt;
1761 /* During creating VOLATILE file, it should honor the mdt
1762 * index if the file under striped dir is being restored, see
1764 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1765 (int)op_data->op_mds != -1 && lsm != NULL) {
1767 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1771 /* refill the right parent fid */
1772 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1773 struct lmv_oinfo *oinfo;
1775 oinfo = &lsm->lsm_md_oinfo[i];
1776 if (oinfo->lmo_mds == op_data->op_mds) {
1777 *fid = oinfo->lmo_fid;
1782 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1783 if (i == lsm->lsm_md_stripe_count)
1784 tgt = ERR_PTR(-EINVAL);
1789 if (lsm == NULL || op_data->op_namelen == 0) {
1790 tgt = lmv_find_target(lmv, fid);
1794 op_data->op_mds = tgt->ltd_idx;
1798 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1799 op_data->op_namelen, fid,
1803 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1804 const void *data, size_t datalen, umode_t mode, uid_t uid,
1805 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1806 struct ptlrpc_request **request)
1808 struct obd_device *obd = exp->exp_obd;
1809 struct lmv_obd *lmv = &obd->u.lmv;
1810 struct lmv_tgt_desc *tgt;
1814 rc = lmv_check_connect(obd);
1818 if (!lmv->desc.ld_active_tgt_count)
1821 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1823 RETURN(PTR_ERR(tgt));
1825 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1826 (int)op_data->op_namelen, op_data->op_name,
1827 PFID(&op_data->op_fid1), op_data->op_mds);
1829 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1832 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1833 /* Send the create request to the MDT where the object
1834 * will be located */
1835 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1837 RETURN(PTR_ERR(tgt));
1839 op_data->op_mds = tgt->ltd_idx;
1841 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1844 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1845 PFID(&op_data->op_fid2), op_data->op_mds);
1847 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1848 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1849 cap_effective, rdev, request);
1851 if (*request == NULL)
1853 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1859 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1860 const union ldlm_policy_data *policy,
1861 struct lookup_intent *it, struct md_op_data *op_data,
1862 struct lustre_handle *lockh, __u64 extra_lock_flags)
1864 struct obd_device *obd = exp->exp_obd;
1865 struct lmv_obd *lmv = &obd->u.lmv;
1866 struct lmv_tgt_desc *tgt;
1870 rc = lmv_check_connect(obd);
1874 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1875 LL_IT2STR(it), PFID(&op_data->op_fid1));
1877 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1879 RETURN(PTR_ERR(tgt));
1881 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1882 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1884 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1891 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1892 struct ptlrpc_request **preq)
1894 struct ptlrpc_request *req = NULL;
1895 struct obd_device *obd = exp->exp_obd;
1896 struct lmv_obd *lmv = &obd->u.lmv;
1897 struct lmv_tgt_desc *tgt;
1898 struct mdt_body *body;
1902 rc = lmv_check_connect(obd);
1906 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1908 RETURN(PTR_ERR(tgt));
1910 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1911 (int)op_data->op_namelen, op_data->op_name,
1912 PFID(&op_data->op_fid1), tgt->ltd_idx);
1914 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1918 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1919 LASSERT(body != NULL);
1921 if (body->mbo_valid & OBD_MD_MDS) {
1922 struct lu_fid rid = body->mbo_fid1;
1923 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1926 tgt = lmv_find_target(lmv, &rid);
1928 ptlrpc_req_finished(*preq);
1930 RETURN(PTR_ERR(tgt));
1933 op_data->op_fid1 = rid;
1934 op_data->op_valid |= OBD_MD_FLCROSSREF;
1935 op_data->op_namelen = 0;
1936 op_data->op_name = NULL;
1937 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1938 ptlrpc_req_finished(*preq);
1945 #define md_op_data_fid(op_data, fl) \
1946 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1947 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1948 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1949 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1952 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1953 struct md_op_data *op_data,
1954 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1956 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1957 struct obd_device *obd = exp->exp_obd;
1958 struct lmv_obd *lmv = &obd->u.lmv;
1959 ldlm_policy_data_t policy = {{ 0 }};
1963 if (!fid_is_sane(fid))
1967 tgt = lmv_find_target(lmv, fid);
1969 RETURN(PTR_ERR(tgt));
1972 if (tgt->ltd_idx != op_tgt) {
1973 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1974 policy.l_inodebits.bits = bits;
1975 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1976 mode, LCF_ASYNC, NULL);
1979 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1981 op_data->op_flags |= flag;
1989 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1992 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1993 struct ptlrpc_request **request)
1995 struct obd_device *obd = exp->exp_obd;
1996 struct lmv_obd *lmv = &obd->u.lmv;
1997 struct lmv_tgt_desc *tgt;
2001 rc = lmv_check_connect(obd);
2005 LASSERT(op_data->op_namelen != 0);
2007 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2008 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2009 op_data->op_name, PFID(&op_data->op_fid1));
2011 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2012 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2013 op_data->op_cap = cfs_curproc_cap_pack();
2014 if (op_data->op_mea2 != NULL) {
2015 struct lmv_stripe_md *lsm = op_data->op_mea2;
2016 const struct lmv_oinfo *oinfo;
2018 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2019 op_data->op_namelen);
2021 RETURN(PTR_ERR(oinfo));
2023 op_data->op_fid2 = oinfo->lmo_fid;
2026 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2028 RETURN(PTR_ERR(tgt));
2031 * Cancel UPDATE lock on child (fid1).
2033 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2034 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2035 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2039 rc = md_link(tgt->ltd_exp, op_data, request);
2044 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2045 const char *old, size_t oldlen,
2046 const char *new, size_t 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;
2055 LASSERT(oldlen != 0);
2057 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2058 (int)oldlen, old, PFID(&op_data->op_fid1),
2059 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2060 (int)newlen, new, PFID(&op_data->op_fid2),
2061 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2063 rc = lmv_check_connect(obd);
2067 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2068 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2069 op_data->op_cap = cfs_curproc_cap_pack();
2070 if (op_data->op_cli_flags & CLI_MIGRATE) {
2071 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2072 PFID(&op_data->op_fid3));
2074 if (op_data->op_mea1 != NULL) {
2075 struct lmv_stripe_md *lsm = op_data->op_mea1;
2076 struct lmv_tgt_desc *tmp;
2078 /* Fix the parent fid for striped dir */
2079 tmp = lmv_locate_target_for_name(lmv, lsm, old,
2084 RETURN(PTR_ERR(tmp));
2087 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2091 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2093 if (op_data->op_mea1 != NULL) {
2094 struct lmv_stripe_md *lsm = op_data->op_mea1;
2096 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2100 if (IS_ERR(src_tgt))
2101 RETURN(PTR_ERR(src_tgt));
2103 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2104 if (IS_ERR(src_tgt))
2105 RETURN(PTR_ERR(src_tgt));
2107 op_data->op_mds = src_tgt->ltd_idx;
2110 if (op_data->op_mea2) {
2111 struct lmv_stripe_md *lsm = op_data->op_mea2;
2112 const struct lmv_oinfo *oinfo;
2114 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2116 RETURN(PTR_ERR(oinfo));
2118 op_data->op_fid2 = oinfo->lmo_fid;
2121 if (IS_ERR(src_tgt))
2122 RETURN(PTR_ERR(src_tgt));
2125 * LOOKUP lock on src child (fid3) should also be cancelled for
2126 * src_tgt in mdc_rename.
2128 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2131 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2134 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2135 LCK_EX, MDS_INODELOCK_UPDATE,
2136 MF_MDC_CANCEL_FID2);
2141 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2143 if (fid_is_sane(&op_data->op_fid3)) {
2144 struct lmv_tgt_desc *tgt;
2146 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2148 RETURN(PTR_ERR(tgt));
2150 /* Cancel LOOKUP lock on its parent */
2151 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2152 LCK_EX, MDS_INODELOCK_LOOKUP,
2153 MF_MDC_CANCEL_FID3);
2157 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2158 LCK_EX, MDS_INODELOCK_FULL,
2159 MF_MDC_CANCEL_FID3);
2165 * Cancel all the locks on tgt child (fid4).
2167 if (fid_is_sane(&op_data->op_fid4))
2168 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2169 LCK_EX, MDS_INODELOCK_FULL,
2170 MF_MDC_CANCEL_FID4);
2172 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2173 op_data->op_mds, PFID(&op_data->op_fid2));
2175 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2181 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2182 void *ea, size_t ealen, struct ptlrpc_request **request)
2184 struct obd_device *obd = exp->exp_obd;
2185 struct lmv_obd *lmv = &obd->u.lmv;
2186 struct lmv_tgt_desc *tgt;
2190 rc = lmv_check_connect(obd);
2194 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2195 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2197 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2198 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2200 RETURN(PTR_ERR(tgt));
2202 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2207 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2208 struct obd_capa *oc, struct ptlrpc_request **request)
2210 struct obd_device *obd = exp->exp_obd;
2211 struct lmv_obd *lmv = &obd->u.lmv;
2212 struct lmv_tgt_desc *tgt;
2216 rc = lmv_check_connect(obd);
2220 tgt = lmv_find_target(lmv, fid);
2222 RETURN(PTR_ERR(tgt));
2224 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2229 * Get current minimum entry from striped directory
2231 * This function will search the dir entry, whose hash value is the
2232 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2233 * only being called for striped directory.
2235 * \param[in] exp export of LMV
2236 * \param[in] op_data parameters transferred beween client MD stack
2237 * stripe_information will be included in this
2239 * \param[in] cb_op ldlm callback being used in enqueue in
2241 * \param[in] hash_offset the hash value, which is used to locate
2242 * minum(closet) dir entry
2243 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2244 * index of last entry, so to avoid hash conflict
2245 * between stripes. It will also be used to
2246 * return the stripe index of current dir entry.
2247 * \param[in|out] entp the minum entry and it also is being used
2248 * to input the last dir entry to resolve the
2251 * \param[out] ppage the page which holds the minum entry
2253 * \retval = 0 get the entry successfully
2254 * negative errno (< 0) does not get the entry
2256 static int lmv_get_min_striped_entry(struct obd_export *exp,
2257 struct md_op_data *op_data,
2258 struct md_callback *cb_op,
2259 __u64 hash_offset, int *stripe_offset,
2260 struct lu_dirent **entp,
2261 struct page **ppage)
2263 struct obd_device *obd = exp->exp_obd;
2264 struct lmv_obd *lmv = &obd->u.lmv;
2265 struct lmv_stripe_md *lsm = op_data->op_mea1;
2266 struct lmv_tgt_desc *tgt;
2268 struct lu_dirent *min_ent = NULL;
2269 struct page *min_page = NULL;
2275 stripe_count = lsm->lsm_md_stripe_count;
2276 for (i = 0; i < stripe_count; i++) {
2277 struct lu_dirent *ent = NULL;
2278 struct page *page = NULL;
2279 struct lu_dirpage *dp;
2280 __u64 stripe_hash = hash_offset;
2282 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2284 GOTO(out, rc = PTR_ERR(tgt));
2286 /* op_data will be shared by each stripe, so we need
2287 * reset these value for each stripe */
2288 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2289 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2290 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2292 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2297 dp = page_address(page);
2298 for (ent = lu_dirent_start(dp); ent != NULL;
2299 ent = lu_dirent_next(ent)) {
2300 /* Skip dummy entry */
2301 if (le16_to_cpu(ent->lde_namelen) == 0)
2304 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2307 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2308 (*entp == ent || i < *stripe_offset))
2311 /* skip . and .. for other stripes */
2313 (strncmp(ent->lde_name, ".",
2314 le16_to_cpu(ent->lde_namelen)) == 0 ||
2315 strncmp(ent->lde_name, "..",
2316 le16_to_cpu(ent->lde_namelen)) == 0))
2322 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2325 page_cache_release(page);
2328 /* reach the end of current stripe, go to next stripe */
2329 if (stripe_hash == MDS_DIR_END_OFF)
2335 if (min_ent != NULL) {
2336 if (le64_to_cpu(min_ent->lde_hash) >
2337 le64_to_cpu(ent->lde_hash)) {
2340 page_cache_release(min_page);
2345 page_cache_release(page);
2356 if (*ppage != NULL) {
2358 page_cache_release(*ppage);
2360 *stripe_offset = min_idx;
2367 * Build dir entry page from a striped directory
2369 * This function gets one entry by @offset from a striped directory. It will
2370 * read entries from all of stripes, and choose one closest to the required
2371 * offset(&offset). A few notes
2372 * 1. skip . and .. for non-zero stripes, because there can only have one .
2373 * and .. in a directory.
2374 * 2. op_data will be shared by all of stripes, instead of allocating new
2375 * one, so need to restore before reusing.
2376 * 3. release the entry page if that is not being chosen.
2378 * \param[in] exp obd export refer to LMV
2379 * \param[in] op_data hold those MD parameters of read_entry
2380 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2381 * \param[out] ldp the entry being read
2382 * \param[out] ppage the page holding the entry. Note: because the entry
2383 * will be accessed in upper layer, so we need hold the
2384 * page until the usages of entry is finished, see
2385 * ll_dir_entry_next.
2387 * retval =0 if get entry successfully
2388 * <0 cannot get entry
2390 static int lmv_read_striped_page(struct obd_export *exp,
2391 struct md_op_data *op_data,
2392 struct md_callback *cb_op,
2393 __u64 offset, struct page **ppage)
2395 struct obd_device *obd = exp->exp_obd;
2396 struct lu_fid master_fid = op_data->op_fid1;
2397 struct inode *master_inode = op_data->op_data;
2398 __u64 hash_offset = offset;
2399 struct lu_dirpage *dp;
2400 struct page *min_ent_page = NULL;
2401 struct page *ent_page = NULL;
2402 struct lu_dirent *ent;
2405 struct lu_dirent *min_ent = NULL;
2406 struct lu_dirent *last_ent;
2411 rc = lmv_check_connect(obd);
2415 /* Allocate a page and read entries from all of stripes and fill
2416 * the page by hash order */
2417 ent_page = alloc_page(GFP_KERNEL);
2418 if (ent_page == NULL)
2421 /* Initialize the entry page */
2422 dp = kmap(ent_page);
2423 memset(dp, 0, sizeof(*dp));
2424 dp->ldp_hash_start = cpu_to_le64(offset);
2425 dp->ldp_flags |= LDF_COLLIDE;
2428 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2434 /* Find the minum entry from all sub-stripes */
2435 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2441 /* If it can not get minum entry, it means it already reaches
2442 * the end of this directory */
2443 if (min_ent == NULL) {
2444 last_ent->lde_reclen = 0;
2445 hash_offset = MDS_DIR_END_OFF;
2449 ent_size = le16_to_cpu(min_ent->lde_reclen);
2451 /* the last entry lde_reclen is 0, but it might not
2452 * the end of this entry of this temporay entry */
2454 ent_size = lu_dirent_calc_size(
2455 le16_to_cpu(min_ent->lde_namelen),
2456 le32_to_cpu(min_ent->lde_attrs));
2457 if (ent_size > left_bytes) {
2458 last_ent->lde_reclen = cpu_to_le16(0);
2459 hash_offset = le64_to_cpu(min_ent->lde_hash);
2463 memcpy(ent, min_ent, ent_size);
2465 /* Replace . with master FID and Replace .. with the parent FID
2466 * of master object */
2467 if (strncmp(ent->lde_name, ".",
2468 le16_to_cpu(ent->lde_namelen)) == 0 &&
2469 le16_to_cpu(ent->lde_namelen) == 1)
2470 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2471 else if (strncmp(ent->lde_name, "..",
2472 le16_to_cpu(ent->lde_namelen)) == 0 &&
2473 le16_to_cpu(ent->lde_namelen) == 2)
2474 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2476 left_bytes -= ent_size;
2477 ent->lde_reclen = cpu_to_le16(ent_size);
2479 ent = (void *)ent + ent_size;
2480 hash_offset = le64_to_cpu(min_ent->lde_hash);
2481 if (hash_offset == MDS_DIR_END_OFF) {
2482 last_ent->lde_reclen = 0;
2487 if (min_ent_page != NULL) {
2488 kunmap(min_ent_page);
2489 page_cache_release(min_ent_page);
2492 if (unlikely(rc != 0)) {
2493 __free_page(ent_page);
2497 dp->ldp_flags |= LDF_EMPTY;
2498 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2499 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2502 /* We do not want to allocate md_op_data during each
2503 * dir entry reading, so op_data will be shared by every stripe,
2504 * then we need to restore it back to original value before
2505 * return to the upper layer */
2506 op_data->op_fid1 = master_fid;
2507 op_data->op_fid2 = master_fid;
2508 op_data->op_data = master_inode;
2515 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2516 struct md_callback *cb_op, __u64 offset,
2517 struct page **ppage)
2519 struct obd_device *obd = exp->exp_obd;
2520 struct lmv_obd *lmv = &obd->u.lmv;
2521 struct lmv_stripe_md *lsm = op_data->op_mea1;
2522 struct lmv_tgt_desc *tgt;
2526 rc = lmv_check_connect(obd);
2530 if (unlikely(lsm != NULL)) {
2531 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2535 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2537 RETURN(PTR_ERR(tgt));
2539 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2545 * Unlink a file/directory
2547 * Unlink a file or directory under the parent dir. The unlink request
2548 * usually will be sent to the MDT where the child is located, but if
2549 * the client does not have the child FID then request will be sent to the
2550 * MDT where the parent is located.
2552 * If the parent is a striped directory then it also needs to locate which
2553 * stripe the name of the child is located, and replace the parent FID
2554 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2555 * it will walk through all of sub-stripes until the child is being
2558 * \param[in] exp export refer to LMV
2559 * \param[in] op_data different parameters transferred beween client
2560 * MD stacks, name, namelen, FIDs etc.
2561 * op_fid1 is the parent FID, op_fid2 is the child
2563 * \param[out] request point to the request of unlink.
2565 * retval 0 if succeed
2566 * negative errno if failed.
2568 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2569 struct ptlrpc_request **request)
2571 struct obd_device *obd = exp->exp_obd;
2572 struct lmv_obd *lmv = &obd->u.lmv;
2573 struct lmv_tgt_desc *tgt = NULL;
2574 struct lmv_tgt_desc *parent_tgt = NULL;
2575 struct mdt_body *body;
2577 int stripe_index = 0;
2578 struct lmv_stripe_md *lsm = op_data->op_mea1;
2581 rc = lmv_check_connect(obd);
2585 /* For striped dir, we need to locate the parent as well */
2587 struct lmv_tgt_desc *tmp;
2589 LASSERT(op_data->op_name != NULL &&
2590 op_data->op_namelen != 0);
2592 tmp = lmv_locate_target_for_name(lmv, lsm,
2594 op_data->op_namelen,
2598 /* return -EBADFD means unknown hash type, might
2599 * need try all sub-stripe here */
2600 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2601 RETURN(PTR_ERR(tmp));
2603 /* Note: both migrating dir and unknown hash dir need to
2604 * try all of sub-stripes, so we need start search the
2605 * name from stripe 0, but migrating dir is already handled
2606 * inside lmv_locate_target_for_name(), so we only check
2607 * unknown hash type directory here */
2608 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2609 struct lmv_oinfo *oinfo;
2611 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2613 op_data->op_fid1 = oinfo->lmo_fid;
2614 op_data->op_mds = oinfo->lmo_mds;
2619 /* Send unlink requests to the MDT where the child is located */
2620 if (likely(!fid_is_zero(&op_data->op_fid2)))
2621 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2622 else if (lsm != NULL)
2623 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2625 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2628 RETURN(PTR_ERR(tgt));
2630 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2631 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2632 op_data->op_cap = cfs_curproc_cap_pack();
2635 * If child's fid is given, cancel unused locks for it if it is from
2636 * another export than parent.
2638 * LOOKUP lock for child (fid3) should also be cancelled on parent
2639 * tgt_tgt in mdc_unlink().
2641 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2644 * Cancel FULL locks on child (fid3).
2646 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2647 if (IS_ERR(parent_tgt))
2648 RETURN(PTR_ERR(parent_tgt));
2650 if (parent_tgt != tgt) {
2651 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2652 LCK_EX, MDS_INODELOCK_LOOKUP,
2653 MF_MDC_CANCEL_FID3);
2656 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2657 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2661 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2662 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2664 rc = md_unlink(tgt->ltd_exp, op_data, request);
2665 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2668 /* Try next stripe if it is needed. */
2669 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2670 struct lmv_oinfo *oinfo;
2673 if (stripe_index >= lsm->lsm_md_stripe_count)
2676 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2678 op_data->op_fid1 = oinfo->lmo_fid;
2679 op_data->op_mds = oinfo->lmo_mds;
2681 ptlrpc_req_finished(*request);
2684 goto try_next_stripe;
2687 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2691 /* Not cross-ref case, just get out of here. */
2692 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2695 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2696 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2698 /* This is a remote object, try remote MDT, Note: it may
2699 * try more than 1 time here, Considering following case
2700 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2701 * 1. Initially A does not know where remote1 is, it send
2702 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2703 * resend unlink RPC to MDT1 (retry 1st time).
2705 * 2. During the unlink RPC in flight,
2706 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2707 * and create new remote1, but on MDT0
2709 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2710 * /mnt/lustre, then lookup get fid of remote1, and find
2711 * it is remote dir again, and replay -EREMOTE again.
2713 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2715 * In theory, it might try unlimited time here, but it should
2716 * be very rare case. */
2717 op_data->op_fid2 = body->mbo_fid1;
2718 ptlrpc_req_finished(*request);
2724 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2726 struct lmv_obd *lmv = &obd->u.lmv;
2730 case OBD_CLEANUP_EARLY:
2731 /* XXX: here should be calling obd_precleanup() down to
2734 case OBD_CLEANUP_EXPORTS:
2735 fld_client_proc_fini(&lmv->lmv_fld);
2736 lprocfs_obd_cleanup(obd);
2737 lprocfs_free_md_stats(obd);
2746 * Get by key a value associated with a LMV device.
2748 * Dispatch request to lower-layer devices as needed.
2750 * \param[in] env execution environment for this thread
2751 * \param[in] exp export for the LMV device
2752 * \param[in] keylen length of key identifier
2753 * \param[in] key identifier of key to get value for
2754 * \param[in] vallen size of \a val
2755 * \param[out] val pointer to storage location for value
2756 * \param[in] lsm optional striping metadata of object
2758 * \retval 0 on success
2759 * \retval negative negated errno on failure
2761 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2762 __u32 keylen, void *key, __u32 *vallen, void *val,
2763 struct lov_stripe_md *lsm)
2765 struct obd_device *obd;
2766 struct lmv_obd *lmv;
2770 obd = class_exp2obd(exp);
2772 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2773 exp->exp_handle.h_cookie);
2778 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2781 rc = lmv_check_connect(obd);
2785 LASSERT(*vallen == sizeof(__u32));
2786 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2787 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2789 * All tgts should be connected when this gets called.
2791 if (tgt == NULL || tgt->ltd_exp == NULL)
2794 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2799 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2800 KEY_IS(KEY_DEFAULT_EASIZE) ||
2801 KEY_IS(KEY_MAX_COOKIESIZE) ||
2802 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2803 KEY_IS(KEY_CONN_DATA)) {
2804 rc = lmv_check_connect(obd);
2809 * Forwarding this request to first MDS, it should know LOV
2812 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2814 if (!rc && KEY_IS(KEY_CONN_DATA))
2815 exp->exp_connect_data = *(struct obd_connect_data *)val;
2817 } else if (KEY_IS(KEY_TGT_COUNT)) {
2818 *((int *)val) = lmv->desc.ld_tgt_count;
2822 CDEBUG(D_IOCTL, "Invalid key\n");
2827 * Asynchronously set by key a value associated with a LMV device.
2829 * Dispatch request to lower-layer devices as needed.
2831 * \param[in] env execution environment for this thread
2832 * \param[in] exp export for the LMV device
2833 * \param[in] keylen length of key identifier
2834 * \param[in] key identifier of key to store value for
2835 * \param[in] vallen size of value to store
2836 * \param[in] val pointer to data to be stored
2837 * \param[in] set optional list of related ptlrpc requests
2839 * \retval 0 on success
2840 * \retval negative negated errno on failure
2842 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2843 __u32 keylen, void *key, __u32 vallen, void *val,
2844 struct ptlrpc_request_set *set)
2846 struct lmv_tgt_desc *tgt = NULL;
2847 struct obd_device *obd;
2848 struct lmv_obd *lmv;
2852 obd = class_exp2obd(exp);
2854 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2855 exp->exp_handle.h_cookie);
2860 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2861 KEY_IS(KEY_DEFAULT_EASIZE)) {
2864 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2867 if (tgt == NULL || tgt->ltd_exp == NULL)
2870 err = obd_set_info_async(env, tgt->ltd_exp,
2871 keylen, key, vallen, val, set);
2882 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2883 struct lmv_mds_md_v1 *lmm1)
2888 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2889 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2890 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2891 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2892 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2893 sizeof(lmm1->lmv_pool_name));
2894 if (cplen >= sizeof(lmm1->lmv_pool_name))
2897 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2898 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2899 &lsm->lsm_md_oinfo[i].lmo_fid);
2903 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2907 bool allocated = false;
2911 LASSERT(lmmp != NULL);
2913 if (*lmmp != NULL && lsm == NULL) {
2916 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2917 lmm_size = lmv_mds_md_size(stripe_count,
2918 le32_to_cpu((*lmmp)->lmv_magic));
2921 OBD_FREE(*lmmp, lmm_size);
2927 if (*lmmp == NULL && lsm == NULL) {
2928 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2929 LASSERT(lmm_size > 0);
2930 OBD_ALLOC(*lmmp, lmm_size);
2933 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2934 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2939 LASSERT(lsm != NULL);
2940 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2941 if (*lmmp == NULL) {
2942 OBD_ALLOC(*lmmp, lmm_size);
2948 switch (lsm->lsm_md_magic) {
2950 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2957 if (rc != 0 && allocated) {
2958 OBD_FREE(*lmmp, lmm_size);
2965 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2966 const struct lmv_mds_md_v1 *lmm1)
2968 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2975 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2976 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2977 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2978 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2979 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2981 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2982 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2983 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2984 sizeof(lsm->lsm_md_pool_name));
2986 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2989 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2990 "layout_version %d\n", lsm->lsm_md_stripe_count,
2991 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2992 lsm->lsm_md_layout_version);
2994 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2995 for (i = 0; i < stripe_count; i++) {
2996 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2997 &lmm1->lmv_stripe_fids[i]);
2998 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2999 &lsm->lsm_md_oinfo[i].lmo_mds);
3002 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3003 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3009 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3010 const union lmv_mds_md *lmm, int stripe_count)
3012 struct lmv_stripe_md *lsm;
3015 bool allocated = false;
3018 LASSERT(lsmp != NULL);
3022 if (lsm != NULL && lmm == NULL) {
3024 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3025 /* For migrating inode, the master stripe and master
3026 * object will be the same, so do not need iput, see
3027 * ll_update_lsm_md */
3028 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3029 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3030 iput(lsm->lsm_md_oinfo[i].lmo_root);
3032 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3033 OBD_FREE(lsm, lsm_size);
3039 if (lsm == NULL && lmm == NULL) {
3040 lsm_size = lmv_stripe_md_size(stripe_count);
3041 OBD_ALLOC(lsm, lsm_size);
3044 lsm->lsm_md_stripe_count = stripe_count;
3049 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3053 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3054 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3055 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3056 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3061 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3062 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3065 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3066 * stripecount should be 0 then.
3068 lsm_size = lmv_stripe_md_size(0);
3070 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3072 OBD_ALLOC(lsm, lsm_size);
3079 switch (le32_to_cpu(lmm->lmv_magic)) {
3081 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3084 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3085 le32_to_cpu(lmm->lmv_magic));
3090 if (rc != 0 && allocated) {
3091 OBD_FREE(lsm, lsm_size);
3098 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3100 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3103 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3105 lmv_unpack_md(NULL, &lsm, NULL, 0);
3107 EXPORT_SYMBOL(lmv_free_memmd);
3109 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3110 struct lov_mds_md *lmm, int disk_len)
3112 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3113 (union lmv_mds_md *)lmm, disk_len);
3116 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3117 struct lov_stripe_md *lsm)
3119 struct obd_device *obd = exp->exp_obd;
3120 struct lmv_obd *lmv_obd = &obd->u.lmv;
3121 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3126 stripe_count = lmv->lsm_md_stripe_count;
3128 stripe_count = lmv_obd->desc.ld_tgt_count;
3130 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3133 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3136 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3137 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3138 ldlm_cancel_flags_t flags, void *opaque)
3140 struct obd_device *obd = exp->exp_obd;
3141 struct lmv_obd *lmv = &obd->u.lmv;
3147 LASSERT(fid != NULL);
3149 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3150 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3152 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3155 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3163 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3166 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3167 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3171 if (tgt == NULL || tgt->ltd_exp == NULL)
3173 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3177 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3178 const struct lu_fid *fid, ldlm_type_t type,
3179 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3180 struct lustre_handle *lockh)
3182 struct obd_device *obd = exp->exp_obd;
3183 struct lmv_obd *lmv = &obd->u.lmv;
3189 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3192 * With DNE every object can have two locks in different namespaces:
3193 * lookup lock in space of MDT storing direntry and update/open lock in
3194 * space of MDT storing inode. Try the MDT that the FID maps to first,
3195 * since this can be easily found, and only try others if that fails.
3197 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3198 i < lmv->desc.ld_tgt_count;
3199 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3201 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3202 obd->obd_name, PFID(fid), tgt);
3206 if (lmv->tgts[tgt] == NULL ||
3207 lmv->tgts[tgt]->ltd_exp == NULL ||
3208 lmv->tgts[tgt]->ltd_active == 0)
3211 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3212 type, policy, mode, lockh);
3220 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3221 struct obd_export *dt_exp, struct obd_export *md_exp,
3222 struct lustre_md *md)
3224 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3225 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3227 if (tgt == NULL || tgt->ltd_exp == NULL)
3230 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3233 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3235 struct obd_device *obd = exp->exp_obd;
3236 struct lmv_obd *lmv = &obd->u.lmv;
3237 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3240 if (md->lmv != NULL) {
3241 lmv_free_memmd(md->lmv);
3244 if (tgt == NULL || tgt->ltd_exp == NULL)
3246 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3249 int lmv_set_open_replay_data(struct obd_export *exp,
3250 struct obd_client_handle *och,
3251 struct lookup_intent *it)
3253 struct obd_device *obd = exp->exp_obd;
3254 struct lmv_obd *lmv = &obd->u.lmv;
3255 struct lmv_tgt_desc *tgt;
3258 tgt = lmv_find_target(lmv, &och->och_fid);
3260 RETURN(PTR_ERR(tgt));
3262 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3265 int lmv_clear_open_replay_data(struct obd_export *exp,
3266 struct obd_client_handle *och)
3268 struct obd_device *obd = exp->exp_obd;
3269 struct lmv_obd *lmv = &obd->u.lmv;
3270 struct lmv_tgt_desc *tgt;
3273 tgt = lmv_find_target(lmv, &och->och_fid);
3275 RETURN(PTR_ERR(tgt));
3277 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3280 static int lmv_get_remote_perm(struct obd_export *exp,
3281 const struct lu_fid *fid,
3282 struct obd_capa *oc, __u32 suppgid,
3283 struct ptlrpc_request **request)
3285 struct obd_device *obd = exp->exp_obd;
3286 struct lmv_obd *lmv = &obd->u.lmv;
3287 struct lmv_tgt_desc *tgt;
3291 rc = lmv_check_connect(obd);
3295 tgt = lmv_find_target(lmv, fid);
3297 RETURN(PTR_ERR(tgt));
3299 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3303 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3306 struct obd_device *obd = exp->exp_obd;
3307 struct lmv_obd *lmv = &obd->u.lmv;
3308 struct lmv_tgt_desc *tgt;
3312 rc = lmv_check_connect(obd);
3316 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3318 RETURN(PTR_ERR(tgt));
3320 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3324 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3325 const struct req_msg_field *field, struct obd_capa **oc)
3327 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3328 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3330 if (tgt == NULL || tgt->ltd_exp == NULL)
3332 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3335 int lmv_intent_getattr_async(struct obd_export *exp,
3336 struct md_enqueue_info *minfo,
3337 struct ldlm_enqueue_info *einfo)
3339 struct md_op_data *op_data = &minfo->mi_data;
3340 struct obd_device *obd = exp->exp_obd;
3341 struct lmv_obd *lmv = &obd->u.lmv;
3342 struct lmv_tgt_desc *tgt = NULL;
3346 rc = lmv_check_connect(obd);
3350 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3352 RETURN(PTR_ERR(tgt));
3354 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3358 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3359 struct lu_fid *fid, __u64 *bits)
3361 struct obd_device *obd = exp->exp_obd;
3362 struct lmv_obd *lmv = &obd->u.lmv;
3363 struct lmv_tgt_desc *tgt;
3367 rc = lmv_check_connect(obd);
3371 tgt = lmv_find_target(lmv, fid);
3373 RETURN(PTR_ERR(tgt));
3375 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3379 int lmv_get_fid_from_lsm(struct obd_export *exp,
3380 const struct lmv_stripe_md *lsm,
3381 const char *name, int namelen, struct lu_fid *fid)
3383 const struct lmv_oinfo *oinfo;
3385 LASSERT(lsm != NULL);
3386 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3388 return PTR_ERR(oinfo);
3390 *fid = oinfo->lmo_fid;
3396 * For lmv, only need to send request to master MDT, and the master MDT will
3397 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3398 * we directly fetch data from the slave MDTs.
3400 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3401 struct obd_quotactl *oqctl)
3403 struct obd_device *obd = class_exp2obd(exp);
3404 struct lmv_obd *lmv = &obd->u.lmv;
3405 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3408 __u64 curspace, curinodes;
3412 tgt->ltd_exp == NULL ||
3414 lmv->desc.ld_tgt_count == 0) {
3415 CERROR("master lmv inactive\n");
3419 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3420 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3424 curspace = curinodes = 0;
3425 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3429 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3432 err = obd_quotactl(tgt->ltd_exp, oqctl);
3434 CERROR("getquota on mdt %d failed. %d\n", i, err);
3438 curspace += oqctl->qc_dqblk.dqb_curspace;
3439 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3442 oqctl->qc_dqblk.dqb_curspace = curspace;
3443 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3448 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3449 struct obd_quotactl *oqctl)
3451 struct obd_device *obd = class_exp2obd(exp);
3452 struct lmv_obd *lmv = &obd->u.lmv;
3453 struct lmv_tgt_desc *tgt;
3458 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3461 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3462 CERROR("lmv idx %d inactive\n", i);
3466 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3474 static int lmv_merge_attr(struct obd_export *exp,
3475 const struct lmv_stripe_md *lsm,
3476 struct cl_attr *attr,
3477 ldlm_blocking_callback cb_blocking)
3482 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3486 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3487 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3489 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3490 " atime %lu ctime %lu, mtime %lu.\n",
3491 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3492 i_size_read(inode), (unsigned long long)inode->i_blocks,
3493 inode->i_nlink, LTIME_S(inode->i_atime),
3494 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3496 /* for slave stripe, it needs to subtract nlink for . and .. */
3498 attr->cat_nlink += inode->i_nlink - 2;
3500 attr->cat_nlink = inode->i_nlink;
3502 attr->cat_size += i_size_read(inode);
3503 attr->cat_blocks += inode->i_blocks;
3505 if (attr->cat_atime < LTIME_S(inode->i_atime))
3506 attr->cat_atime = LTIME_S(inode->i_atime);
3508 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3509 attr->cat_ctime = LTIME_S(inode->i_ctime);
3511 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3512 attr->cat_mtime = LTIME_S(inode->i_mtime);
3517 struct obd_ops lmv_obd_ops = {
3518 .o_owner = THIS_MODULE,
3519 .o_setup = lmv_setup,
3520 .o_cleanup = lmv_cleanup,
3521 .o_precleanup = lmv_precleanup,
3522 .o_process_config = lmv_process_config,
3523 .o_connect = lmv_connect,
3524 .o_disconnect = lmv_disconnect,
3525 .o_statfs = lmv_statfs,
3526 .o_get_info = lmv_get_info,
3527 .o_set_info_async = lmv_set_info_async,
3528 .o_packmd = lmv_packmd,
3529 .o_unpackmd = lmv_unpackmd,
3530 .o_notify = lmv_notify,
3531 .o_get_uuid = lmv_get_uuid,
3532 .o_iocontrol = lmv_iocontrol,
3533 .o_quotacheck = lmv_quotacheck,
3534 .o_quotactl = lmv_quotactl
3537 struct md_ops lmv_md_ops = {
3538 .m_getstatus = lmv_getstatus,
3539 .m_null_inode = lmv_null_inode,
3540 .m_find_cbdata = lmv_find_cbdata,
3541 .m_close = lmv_close,
3542 .m_create = lmv_create,
3543 .m_enqueue = lmv_enqueue,
3544 .m_getattr = lmv_getattr,
3545 .m_getxattr = lmv_getxattr,
3546 .m_getattr_name = lmv_getattr_name,
3547 .m_intent_lock = lmv_intent_lock,
3549 .m_rename = lmv_rename,
3550 .m_setattr = lmv_setattr,
3551 .m_setxattr = lmv_setxattr,
3552 .m_fsync = lmv_fsync,
3553 .m_read_page = lmv_read_page,
3554 .m_unlink = lmv_unlink,
3555 .m_init_ea_size = lmv_init_ea_size,
3556 .m_cancel_unused = lmv_cancel_unused,
3557 .m_set_lock_data = lmv_set_lock_data,
3558 .m_lock_match = lmv_lock_match,
3559 .m_get_lustre_md = lmv_get_lustre_md,
3560 .m_free_lustre_md = lmv_free_lustre_md,
3561 .m_merge_attr = lmv_merge_attr,
3562 .m_set_open_replay_data = lmv_set_open_replay_data,
3563 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3564 .m_renew_capa = lmv_renew_capa,
3565 .m_unpack_capa = lmv_unpack_capa,
3566 .m_get_remote_perm = lmv_get_remote_perm,
3567 .m_intent_getattr_async = lmv_intent_getattr_async,
3568 .m_revalidate_lock = lmv_revalidate_lock,
3569 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3572 int __init lmv_init(void)
3574 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3575 LUSTRE_LMV_NAME, NULL);
3578 static void lmv_exit(void)
3580 class_unregister_type(LUSTRE_LMV_NAME);
3583 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3584 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3585 MODULE_LICENSE("GPL");
3587 module_init(lmv_init);
3588 module_exit(lmv_exit);