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/user_namespace.h>
42 #ifdef HAVE_UIDGID_HEADER
43 # include <linux/uidgid.h>
45 #include <linux/slab.h>
46 #include <linux/pagemap.h>
48 #include <linux/math64.h>
49 #include <linux/seq_file.h>
50 #include <linux/namei.h>
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_lmv.h>
58 #include <lprocfs_status.h>
59 #include <cl_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_ioctl.h>
62 #include <lustre_kernelcomm.h>
63 #include "lmv_internal.h"
65 static void lmv_activate_target(struct lmv_obd *lmv,
66 struct lmv_tgt_desc *tgt,
69 if (tgt->ltd_active == activate)
72 tgt->ltd_active = activate;
73 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
79 * -EINVAL : UUID can't be found in the LMV's target list
80 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
81 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
83 static int lmv_set_mdc_active(struct lmv_obd *lmv,
84 const struct obd_uuid *uuid,
87 struct lmv_tgt_desc *tgt = NULL;
88 struct obd_device *obd;
93 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
94 lmv, uuid->uuid, activate);
96 spin_lock(&lmv->lmv_lock);
97 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
99 if (tgt == NULL || tgt->ltd_exp == NULL)
102 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
103 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
105 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
109 if (i == lmv->desc.ld_tgt_count)
110 GOTO(out_lmv_lock, rc = -EINVAL);
112 obd = class_exp2obd(tgt->ltd_exp);
114 GOTO(out_lmv_lock, rc = -ENOTCONN);
116 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
117 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
118 obd->obd_type->typ_name, i);
119 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
121 if (tgt->ltd_active == activate) {
122 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
123 activate ? "" : "in");
124 GOTO(out_lmv_lock, rc);
127 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
128 activate ? "" : "in");
129 lmv_activate_target(lmv, tgt, activate);
133 spin_unlock(&lmv->lmv_lock);
137 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
139 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
140 struct lmv_tgt_desc *tgt = lmv->tgts[0];
142 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
145 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
146 enum obd_notify_event ev, void *data)
148 struct obd_connect_data *conn_data;
149 struct lmv_obd *lmv = &obd->u.lmv;
150 struct obd_uuid *uuid;
154 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
155 CERROR("unexpected notification of %s %s!\n",
156 watched->obd_type->typ_name,
161 uuid = &watched->u.cli.cl_target_uuid;
162 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
164 * Set MDC as active before notifying the observer, so the
165 * observer can use the MDC normally.
167 rc = lmv_set_mdc_active(lmv, uuid,
168 ev == OBD_NOTIFY_ACTIVE);
170 CERROR("%sactivation of %s failed: %d\n",
171 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
175 } else if (ev == OBD_NOTIFY_OCD) {
176 conn_data = &watched->u.cli.cl_import->imp_connect_data;
178 * XXX: Make sure that ocd_connect_flags from all targets are
179 * the same. Otherwise one of MDTs runs wrong version or
180 * something like this. --umka
182 obd->obd_self_export->exp_connect_data = *conn_data;
185 else if (ev == OBD_NOTIFY_DISCON) {
187 * For disconnect event, flush fld cache for failout MDS case.
189 fld_client_flush(&lmv->lmv_fld);
193 * Pass the notification up the chain.
195 if (obd->obd_observer)
196 rc = obd_notify(obd->obd_observer, watched, ev, data);
202 * This is fake connect function. Its purpose is to initialize lmv and say
203 * caller that everything is okay. Real connection will be performed later.
205 static int lmv_connect(const struct lu_env *env,
206 struct obd_export **exp, struct obd_device *obd,
207 struct obd_uuid *cluuid, struct obd_connect_data *data,
210 struct lmv_obd *lmv = &obd->u.lmv;
211 struct lustre_handle conn = { 0 };
216 * We don't want to actually do the underlying connections more than
217 * once, so keep track.
220 if (lmv->refcount > 1) {
225 rc = class_connect(&conn, obd, cluuid);
227 CERROR("class_connection() returned %d\n", rc);
231 *exp = class_conn2export(&conn);
232 class_export_get(*exp);
236 lmv->cluuid = *cluuid;
239 lmv->conn_data = *data;
241 if (lmv->targets_proc_entry == NULL) {
242 lmv->targets_proc_entry = lprocfs_register("target_obds",
245 if (IS_ERR(lmv->targets_proc_entry)) {
246 CERROR("%s: cannot register "
247 "/proc/fs/lustre/%s/%s/target_obds\n",
248 obd->obd_name, obd->obd_type->typ_name,
250 lmv->targets_proc_entry = NULL;
255 * All real clients should perform actual connection right away, because
256 * it is possible, that LMV will not have opportunity to connect targets
257 * and MDC stuff will be called directly, for instance while reading
258 * ../mdc/../kbytesfree procfs file, etc.
260 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
261 rc = lmv_check_connect(obd);
263 if (rc && lmv->targets_proc_entry != NULL)
264 lprocfs_remove(&lmv->targets_proc_entry);
268 static int lmv_init_ea_size(struct obd_export *exp,
269 __u32 easize, __u32 def_easize,
270 __u32 cookiesize, __u32 def_cookiesize)
272 struct obd_device *obd = exp->exp_obd;
273 struct lmv_obd *lmv = &obd->u.lmv;
279 if (lmv->max_easize < easize) {
280 lmv->max_easize = easize;
283 if (lmv->max_def_easize < def_easize) {
284 lmv->max_def_easize = def_easize;
287 if (lmv->max_cookiesize < cookiesize) {
288 lmv->max_cookiesize = cookiesize;
291 if (lmv->max_def_cookiesize < def_cookiesize) {
292 lmv->max_def_cookiesize = def_cookiesize;
298 if (lmv->connected == 0)
301 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
302 struct lmv_tgt_desc *tgt = lmv->tgts[i];
304 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
305 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
309 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
310 cookiesize, def_cookiesize);
312 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
313 " rc = %d\n", obd->obd_name, i, rc);
320 #define MAX_STRING_SIZE 128
322 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
324 struct lmv_obd *lmv = &obd->u.lmv;
325 struct obd_uuid *cluuid = &lmv->cluuid;
326 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
327 struct obd_device *mdc_obd;
328 struct obd_export *mdc_exp;
329 struct lu_fld_target target;
333 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
336 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
340 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
341 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
342 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
345 if (!mdc_obd->obd_set_up) {
346 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
350 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
351 &lmv->conn_data, NULL);
353 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
358 * Init fid sequence client for this mdc and add new fld target.
360 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
364 target.ft_srv = NULL;
365 target.ft_exp = mdc_exp;
366 target.ft_idx = tgt->ltd_idx;
368 fld_client_add_target(&lmv->lmv_fld, &target);
370 rc = obd_register_observer(mdc_obd, obd);
372 obd_disconnect(mdc_exp);
373 CERROR("target %s register_observer error %d\n",
374 tgt->ltd_uuid.uuid, rc);
378 if (obd->obd_observer) {
380 * Tell the observer about the new target.
382 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
384 (void *)(tgt - lmv->tgts[0]));
386 obd_disconnect(mdc_exp);
392 tgt->ltd_exp = mdc_exp;
393 lmv->desc.ld_active_tgt_count++;
395 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
396 lmv->max_cookiesize, lmv->max_def_cookiesize);
398 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
399 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
400 atomic_read(&obd->obd_refcount));
402 if (lmv->targets_proc_entry != NULL) {
403 struct proc_dir_entry *mdc_symlink;
405 LASSERT(mdc_obd->obd_type != NULL);
406 LASSERT(mdc_obd->obd_type->typ_name != NULL);
407 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
408 lmv->targets_proc_entry,
410 mdc_obd->obd_type->typ_name,
412 if (mdc_symlink == NULL) {
413 CERROR("cannot register LMV target "
414 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
415 obd->obd_type->typ_name, obd->obd_name,
422 static void lmv_del_target(struct lmv_obd *lmv, int index)
424 if (lmv->tgts[index] == NULL)
427 OBD_FREE_PTR(lmv->tgts[index]);
428 lmv->tgts[index] = NULL;
432 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
433 __u32 index, int gen)
435 struct lmv_obd *lmv = &obd->u.lmv;
436 struct lmv_tgt_desc *tgt;
437 int orig_tgt_count = 0;
441 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
443 mutex_lock(&lmv->lmv_init_mutex);
445 if (lmv->desc.ld_tgt_count == 0) {
446 struct obd_device *mdc_obd;
448 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
451 mutex_unlock(&lmv->lmv_init_mutex);
452 CERROR("%s: Target %s not attached: rc = %d\n",
453 obd->obd_name, uuidp->uuid, -EINVAL);
458 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
459 tgt = lmv->tgts[index];
460 CERROR("%s: UUID %s already assigned at LOV target index %d:"
461 " rc = %d\n", obd->obd_name,
462 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
463 mutex_unlock(&lmv->lmv_init_mutex);
467 if (index >= lmv->tgts_size) {
468 /* We need to reallocate the lmv target array. */
469 struct lmv_tgt_desc **newtgts, **old = NULL;
473 while (newsize < index + 1)
474 newsize = newsize << 1;
475 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
476 if (newtgts == NULL) {
477 mutex_unlock(&lmv->lmv_init_mutex);
481 if (lmv->tgts_size) {
482 memcpy(newtgts, lmv->tgts,
483 sizeof(*newtgts) * lmv->tgts_size);
485 oldsize = lmv->tgts_size;
489 lmv->tgts_size = newsize;
492 OBD_FREE(old, sizeof(*old) * oldsize);
494 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
500 mutex_unlock(&lmv->lmv_init_mutex);
504 mutex_init(&tgt->ltd_fid_mutex);
505 tgt->ltd_idx = index;
506 tgt->ltd_uuid = *uuidp;
508 lmv->tgts[index] = tgt;
509 if (index >= lmv->desc.ld_tgt_count) {
510 orig_tgt_count = lmv->desc.ld_tgt_count;
511 lmv->desc.ld_tgt_count = index + 1;
514 if (lmv->connected) {
515 rc = lmv_connect_mdc(obd, tgt);
517 spin_lock(&lmv->lmv_lock);
518 if (lmv->desc.ld_tgt_count == index + 1)
519 lmv->desc.ld_tgt_count = orig_tgt_count;
520 memset(tgt, 0, sizeof(*tgt));
521 spin_unlock(&lmv->lmv_lock);
523 int easize = sizeof(struct lmv_stripe_md) +
524 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
525 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
529 mutex_unlock(&lmv->lmv_init_mutex);
533 int lmv_check_connect(struct obd_device *obd)
535 struct lmv_obd *lmv = &obd->u.lmv;
536 struct lmv_tgt_desc *tgt;
545 mutex_lock(&lmv->lmv_init_mutex);
546 if (lmv->connected) {
547 mutex_unlock(&lmv->lmv_init_mutex);
551 if (lmv->desc.ld_tgt_count == 0) {
552 mutex_unlock(&lmv->lmv_init_mutex);
553 CERROR("%s: no targets configured.\n", obd->obd_name);
557 LASSERT(lmv->tgts != NULL);
559 if (lmv->tgts[0] == NULL) {
560 mutex_unlock(&lmv->lmv_init_mutex);
561 CERROR("%s: no target configured for index 0.\n",
566 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
567 lmv->cluuid.uuid, obd->obd_name);
569 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
573 rc = lmv_connect_mdc(obd, tgt);
578 class_export_put(lmv->exp);
580 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
581 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
582 mutex_unlock(&lmv->lmv_init_mutex);
593 --lmv->desc.ld_active_tgt_count;
594 rc2 = obd_disconnect(tgt->ltd_exp);
596 CERROR("LMV target %s disconnect on "
597 "MDC idx %d: error %d\n",
598 tgt->ltd_uuid.uuid, i, rc2);
602 class_disconnect(lmv->exp);
603 mutex_unlock(&lmv->lmv_init_mutex);
607 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
609 struct lmv_obd *lmv = &obd->u.lmv;
610 struct obd_device *mdc_obd;
614 LASSERT(tgt != NULL);
615 LASSERT(obd != NULL);
617 mdc_obd = class_exp2obd(tgt->ltd_exp);
620 mdc_obd->obd_force = obd->obd_force;
621 mdc_obd->obd_fail = obd->obd_fail;
622 mdc_obd->obd_no_recov = obd->obd_no_recov;
625 if (lmv->targets_proc_entry != NULL)
626 lprocfs_remove_proc_entry(mdc_obd->obd_name,
627 lmv->targets_proc_entry);
629 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
631 CERROR("Can't finanize fids factory\n");
633 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
634 tgt->ltd_exp->exp_obd->obd_name,
635 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
637 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
638 rc = obd_disconnect(tgt->ltd_exp);
640 if (tgt->ltd_active) {
641 CERROR("Target %s disconnect error %d\n",
642 tgt->ltd_uuid.uuid, rc);
646 lmv_activate_target(lmv, tgt, 0);
651 static int lmv_disconnect(struct obd_export *exp)
653 struct obd_device *obd = class_exp2obd(exp);
654 struct lmv_obd *lmv = &obd->u.lmv;
663 * Only disconnect the underlying layers on the final disconnect.
666 if (lmv->refcount != 0)
669 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
670 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
673 lmv_disconnect_mdc(obd, lmv->tgts[i]);
676 if (lmv->targets_proc_entry != NULL)
677 lprocfs_remove(&lmv->targets_proc_entry);
679 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
680 obd->obd_type->typ_name, obd->obd_name);
684 * This is the case when no real connection is established by
685 * lmv_check_connect().
688 class_export_put(exp);
689 rc = class_disconnect(exp);
690 if (lmv->refcount == 0)
695 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
697 struct obd_device *obddev = class_exp2obd(exp);
698 struct lmv_obd *lmv = &obddev->u.lmv;
699 struct getinfo_fid2path *gf;
700 struct lmv_tgt_desc *tgt;
701 struct getinfo_fid2path *remote_gf = NULL;
702 int remote_gf_size = 0;
705 gf = (struct getinfo_fid2path *)karg;
706 tgt = lmv_find_target(lmv, &gf->gf_fid);
708 RETURN(PTR_ERR(tgt));
711 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
712 if (rc != 0 && rc != -EREMOTE)
713 GOTO(out_fid2path, rc);
715 /* If remote_gf != NULL, it means just building the
716 * path on the remote MDT, copy this path segement to gf */
717 if (remote_gf != NULL) {
718 struct getinfo_fid2path *ori_gf;
721 ori_gf = (struct getinfo_fid2path *)karg;
722 if (strlen(ori_gf->gf_path) +
723 strlen(gf->gf_path) > ori_gf->gf_pathlen)
724 GOTO(out_fid2path, rc = -EOVERFLOW);
726 ptr = ori_gf->gf_path;
728 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
729 strlen(ori_gf->gf_path));
731 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
732 ptr += strlen(gf->gf_path);
736 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
737 tgt->ltd_exp->exp_obd->obd_name,
738 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
742 GOTO(out_fid2path, rc);
744 /* sigh, has to go to another MDT to do path building further */
745 if (remote_gf == NULL) {
746 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
747 OBD_ALLOC(remote_gf, remote_gf_size);
748 if (remote_gf == NULL)
749 GOTO(out_fid2path, rc = -ENOMEM);
750 remote_gf->gf_pathlen = PATH_MAX;
753 if (!fid_is_sane(&gf->gf_fid)) {
754 CERROR("%s: invalid FID "DFID": rc = %d\n",
755 tgt->ltd_exp->exp_obd->obd_name,
756 PFID(&gf->gf_fid), -EINVAL);
757 GOTO(out_fid2path, rc = -EINVAL);
760 tgt = lmv_find_target(lmv, &gf->gf_fid);
762 GOTO(out_fid2path, rc = -EINVAL);
764 remote_gf->gf_fid = gf->gf_fid;
765 remote_gf->gf_recno = -1;
766 remote_gf->gf_linkno = -1;
767 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
769 goto repeat_fid2path;
772 if (remote_gf != NULL)
773 OBD_FREE(remote_gf, remote_gf_size);
777 static int lmv_hsm_req_count(struct lmv_obd *lmv,
778 const struct hsm_user_request *hur,
779 const struct lmv_tgt_desc *tgt_mds)
783 struct lmv_tgt_desc *curr_tgt;
785 /* count how many requests must be sent to the given target */
786 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
787 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
788 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
794 static void lmv_hsm_req_build(struct lmv_obd *lmv,
795 struct hsm_user_request *hur_in,
796 const struct lmv_tgt_desc *tgt_mds,
797 struct hsm_user_request *hur_out)
800 struct lmv_tgt_desc *curr_tgt;
802 /* build the hsm_user_request for the given target */
803 hur_out->hur_request = hur_in->hur_request;
805 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
806 curr_tgt = lmv_find_target(lmv,
807 &hur_in->hur_user_item[i].hui_fid);
808 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
809 hur_out->hur_user_item[nr_out] =
810 hur_in->hur_user_item[i];
814 hur_out->hur_request.hr_itemcount = nr_out;
815 memcpy(hur_data(hur_out), hur_data(hur_in),
816 hur_in->hur_request.hr_data_len);
819 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
820 struct lustre_kernelcomm *lk, void *uarg)
824 struct kkuc_ct_data *kcd = NULL;
827 /* unregister request (call from llapi_hsm_copytool_fini) */
828 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
829 struct lmv_tgt_desc *tgt = lmv->tgts[i];
831 if (tgt == NULL || tgt->ltd_exp == NULL)
833 /* best effort: try to clean as much as possible
834 * (continue on error) */
835 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
838 /* Whatever the result, remove copytool from kuc groups.
839 * Unreached coordinators will get EPIPE on next requests
840 * and will unregister automatically.
842 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
849 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
850 struct lustre_kernelcomm *lk, void *uarg)
855 bool any_set = false;
856 struct kkuc_ct_data *kcd;
859 /* All or nothing: try to register to all MDS.
860 * In case of failure, unregister from previous MDS,
861 * except if it because of inactive target. */
862 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
863 struct lmv_tgt_desc *tgt = lmv->tgts[i];
865 if (tgt == NULL || tgt->ltd_exp == NULL)
867 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
869 if (tgt->ltd_active) {
870 /* permanent error */
871 CERROR("%s: iocontrol MDC %s on MDT"
872 " idx %d cmd %x: err = %d\n",
873 class_exp2obd(lmv->exp)->obd_name,
874 tgt->ltd_uuid.uuid, i, cmd, err);
876 lk->lk_flags |= LK_FLG_STOP;
877 /* unregister from previous MDS */
878 for (j = 0; j < i; j++) {
880 if (tgt == NULL || tgt->ltd_exp == NULL)
882 obd_iocontrol(cmd, tgt->ltd_exp, len,
887 /* else: transient error.
888 * kuc will register to the missing MDT
896 /* no registration done: return error */
899 /* at least one registration done, with no failure */
900 filp = fget(lk->lk_wfd);
909 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
910 kcd->kcd_uuid = lmv->cluuid;
911 kcd->kcd_archive = lk->lk_data;
913 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
926 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
927 int len, void *karg, void *uarg)
929 struct obd_device *obddev = class_exp2obd(exp);
930 struct lmv_obd *lmv = &obddev->u.lmv;
931 struct lmv_tgt_desc *tgt = NULL;
935 __u32 count = lmv->desc.ld_tgt_count;
942 case IOC_OBD_STATFS: {
943 struct obd_ioctl_data *data = karg;
944 struct obd_device *mdc_obd;
945 struct obd_statfs stat_buf = {0};
948 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
949 if ((index >= count))
952 tgt = lmv->tgts[index];
953 if (tgt == NULL || !tgt->ltd_active)
956 mdc_obd = class_exp2obd(tgt->ltd_exp);
961 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
962 min((int) data->ioc_plen2,
963 (int) sizeof(struct obd_uuid))))
966 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
967 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
971 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
972 min((int) data->ioc_plen1,
973 (int) sizeof(stat_buf))))
977 case OBD_IOC_QUOTACTL: {
978 struct if_quotactl *qctl = karg;
979 struct obd_quotactl *oqctl;
981 if (qctl->qc_valid == QC_MDTIDX) {
982 if (count <= qctl->qc_idx)
985 tgt = lmv->tgts[qctl->qc_idx];
986 if (tgt == NULL || tgt->ltd_exp == NULL)
988 } else if (qctl->qc_valid == QC_UUID) {
989 for (i = 0; i < count; i++) {
993 if (!obd_uuid_equals(&tgt->ltd_uuid,
997 if (tgt->ltd_exp == NULL)
1009 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1010 OBD_ALLOC_PTR(oqctl);
1014 QCTL_COPY(oqctl, qctl);
1015 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1017 QCTL_COPY(qctl, oqctl);
1018 qctl->qc_valid = QC_MDTIDX;
1019 qctl->obd_uuid = tgt->ltd_uuid;
1021 OBD_FREE_PTR(oqctl);
1024 case OBD_IOC_CHANGELOG_SEND:
1025 case OBD_IOC_CHANGELOG_CLEAR: {
1026 struct ioc_changelog *icc = karg;
1028 if (icc->icc_mdtindex >= count)
1031 tgt = lmv->tgts[icc->icc_mdtindex];
1032 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1034 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1037 case LL_IOC_GET_CONNECT_FLAGS: {
1039 if (tgt == NULL || tgt->ltd_exp == NULL)
1041 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1044 case LL_IOC_FID2MDTIDX: {
1045 struct lu_fid *fid = karg;
1048 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1052 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1053 * point to user space memory for FID2MDTIDX. */
1054 *(__u32 *)uarg = mdt_index;
1057 case OBD_IOC_FID2PATH: {
1058 rc = lmv_fid2path(exp, len, karg, uarg);
1061 case LL_IOC_HSM_STATE_GET:
1062 case LL_IOC_HSM_STATE_SET:
1063 case LL_IOC_HSM_ACTION: {
1064 struct md_op_data *op_data = karg;
1066 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1068 RETURN(PTR_ERR(tgt));
1070 if (tgt->ltd_exp == NULL)
1073 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1076 case LL_IOC_HSM_PROGRESS: {
1077 const struct hsm_progress_kernel *hpk = karg;
1079 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1081 RETURN(PTR_ERR(tgt));
1082 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1085 case LL_IOC_HSM_REQUEST: {
1086 struct hsm_user_request *hur = karg;
1087 unsigned int reqcount = hur->hur_request.hr_itemcount;
1092 /* if the request is about a single fid
1093 * or if there is a single MDS, no need to split
1095 if (reqcount == 1 || count == 1) {
1096 tgt = lmv_find_target(lmv,
1097 &hur->hur_user_item[0].hui_fid);
1099 RETURN(PTR_ERR(tgt));
1100 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1102 /* split fid list to their respective MDS */
1103 for (i = 0; i < count; i++) {
1104 unsigned int nr, reqlen;
1106 struct hsm_user_request *req;
1109 if (tgt == NULL || tgt->ltd_exp == NULL)
1112 nr = lmv_hsm_req_count(lmv, hur, tgt);
1113 if (nr == 0) /* nothing for this MDS */
1116 /* build a request with fids for this MDS */
1117 reqlen = offsetof(typeof(*hur),
1119 + hur->hur_request.hr_data_len;
1120 OBD_ALLOC_LARGE(req, reqlen);
1124 lmv_hsm_req_build(lmv, hur, tgt, req);
1126 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1128 if (rc1 != 0 && rc == 0)
1130 OBD_FREE_LARGE(req, reqlen);
1135 case LL_IOC_LOV_SWAP_LAYOUTS: {
1136 struct md_op_data *op_data = karg;
1137 struct lmv_tgt_desc *tgt1, *tgt2;
1139 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1141 RETURN(PTR_ERR(tgt1));
1143 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1145 RETURN(PTR_ERR(tgt2));
1147 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1150 /* only files on same MDT can have their layouts swapped */
1151 if (tgt1->ltd_idx != tgt2->ltd_idx)
1154 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1157 case LL_IOC_HSM_CT_START: {
1158 struct lustre_kernelcomm *lk = karg;
1159 if (lk->lk_flags & LK_FLG_STOP)
1160 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1162 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1166 for (i = 0; i < count; i++) {
1167 struct obd_device *mdc_obd;
1171 if (tgt == NULL || tgt->ltd_exp == NULL)
1173 /* ll_umount_begin() sets force flag but for lmv, not
1174 * mdc. Let's pass it through */
1175 mdc_obd = class_exp2obd(tgt->ltd_exp);
1176 mdc_obd->obd_force = obddev->obd_force;
1177 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1178 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1181 if (tgt->ltd_active) {
1182 CERROR("error: iocontrol MDC %s on MDT"
1183 " idx %d cmd %x: err = %d\n",
1184 tgt->ltd_uuid.uuid, i, cmd, err);
1198 static int lmv_all_chars_policy(int count, const char *name,
1209 static int lmv_nid_policy(struct lmv_obd *lmv)
1211 struct obd_import *imp;
1215 * XXX: To get nid we assume that underlying obd device is mdc.
1217 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1218 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1219 return id % lmv->desc.ld_tgt_count;
1222 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1223 placement_policy_t placement)
1225 switch (placement) {
1226 case PLACEMENT_CHAR_POLICY:
1227 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1229 op_data->op_namelen);
1230 case PLACEMENT_NID_POLICY:
1231 return lmv_nid_policy(lmv);
1237 CERROR("Unsupported placement policy %x\n", placement);
1243 * This is _inode_ placement policy function (not name).
1245 static int lmv_placement_policy(struct obd_device *obd,
1246 struct md_op_data *op_data, u32 *mds)
1248 struct lmv_obd *lmv = &obd->u.lmv;
1251 LASSERT(mds != NULL);
1253 if (lmv->desc.ld_tgt_count == 1) {
1258 if (op_data->op_default_stripe_offset != -1) {
1259 *mds = op_data->op_default_stripe_offset;
1264 * If stripe_offset is provided during setdirstripe
1265 * (setdirstripe -i xx), xx MDS will be choosen.
1267 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1268 struct lmv_user_md *lum;
1270 lum = op_data->op_data;
1272 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1273 *mds = le32_to_cpu(lum->lum_stripe_offset);
1275 /* -1 means default, which will be in the same MDT with
1277 *mds = op_data->op_mds;
1278 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1281 /* Allocate new fid on target according to operation type and
1282 * parent home mds. */
1283 *mds = op_data->op_mds;
1289 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1291 struct lmv_tgt_desc *tgt;
1295 tgt = lmv_get_target(lmv, mds, NULL);
1297 RETURN(PTR_ERR(tgt));
1300 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1301 * on server that seq in new allocated fid is not yet known.
1303 mutex_lock(&tgt->ltd_fid_mutex);
1305 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1306 GOTO(out, rc = -ENODEV);
1309 * Asking underlying tgt layer to allocate new fid.
1311 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1313 LASSERT(fid_is_sane(fid));
1319 mutex_unlock(&tgt->ltd_fid_mutex);
1323 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1324 struct lu_fid *fid, struct md_op_data *op_data)
1326 struct obd_device *obd = class_exp2obd(exp);
1327 struct lmv_obd *lmv = &obd->u.lmv;
1332 LASSERT(op_data != NULL);
1333 LASSERT(fid != NULL);
1335 rc = lmv_placement_policy(obd, op_data, &mds);
1337 CERROR("Can't get target for allocating fid, "
1342 rc = __lmv_fid_alloc(lmv, fid, mds);
1344 CERROR("Can't alloc new fid, rc %d\n", rc);
1351 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1353 struct lmv_obd *lmv = &obd->u.lmv;
1354 struct lmv_desc *desc;
1358 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1359 CERROR("LMV setup requires a descriptor\n");
1363 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1364 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1365 CERROR("Lmv descriptor size wrong: %d > %d\n",
1366 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1370 lmv->tgts_size = 32U;
1371 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1372 if (lmv->tgts == NULL)
1375 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1376 lmv->desc.ld_tgt_count = 0;
1377 lmv->desc.ld_active_tgt_count = 0;
1378 lmv->max_cookiesize = 0;
1379 lmv->max_def_easize = 0;
1380 lmv->max_easize = 0;
1381 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1383 spin_lock_init(&lmv->lmv_lock);
1384 mutex_init(&lmv->lmv_init_mutex);
1386 #ifdef CONFIG_PROC_FS
1387 obd->obd_vars = lprocfs_lmv_obd_vars;
1388 lprocfs_obd_setup(obd);
1389 lprocfs_alloc_md_stats(obd, 0);
1390 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1391 0444, &lmv_proc_target_fops, obd);
1393 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1396 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1397 LUSTRE_CLI_FLD_HASH_DHT);
1399 CERROR("Can't init FLD, err %d\n", rc);
1409 static int lmv_cleanup(struct obd_device *obd)
1411 struct lmv_obd *lmv = &obd->u.lmv;
1414 fld_client_fini(&lmv->lmv_fld);
1415 if (lmv->tgts != NULL) {
1417 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1418 if (lmv->tgts[i] == NULL)
1420 lmv_del_target(lmv, i);
1422 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1428 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1430 struct lustre_cfg *lcfg = buf;
1431 struct obd_uuid obd_uuid;
1437 switch (lcfg->lcfg_command) {
1439 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1440 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1441 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1442 GOTO(out, rc = -EINVAL);
1444 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1446 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1447 GOTO(out, rc = -EINVAL);
1448 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1449 GOTO(out, rc = -EINVAL);
1450 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1453 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1454 GOTO(out, rc = -EINVAL);
1460 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1461 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1463 struct obd_device *obd = class_exp2obd(exp);
1464 struct lmv_obd *lmv = &obd->u.lmv;
1465 struct obd_statfs *temp;
1470 rc = lmv_check_connect(obd);
1474 OBD_ALLOC(temp, sizeof(*temp));
1478 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1479 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1482 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1485 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1486 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1488 GOTO(out_free_temp, rc);
1493 /* If the statfs is from mount, it will needs
1494 * retrieve necessary information from MDT0.
1495 * i.e. mount does not need the merged osfs
1497 * And also clients can be mounted as long as
1498 * MDT0 is in service*/
1499 if (flags & OBD_STATFS_FOR_MDT0)
1500 GOTO(out_free_temp, rc);
1502 osfs->os_bavail += temp->os_bavail;
1503 osfs->os_blocks += temp->os_blocks;
1504 osfs->os_ffree += temp->os_ffree;
1505 osfs->os_files += temp->os_files;
1511 OBD_FREE(temp, sizeof(*temp));
1515 static int lmv_getstatus(struct obd_export *exp,
1517 struct obd_capa **pc)
1519 struct obd_device *obd = exp->exp_obd;
1520 struct lmv_obd *lmv = &obd->u.lmv;
1524 rc = lmv_check_connect(obd);
1528 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1532 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1533 struct obd_capa *oc, u64 valid, const char *name,
1534 const char *input, int input_size, int output_size,
1535 int flags, struct ptlrpc_request **request)
1537 struct obd_device *obd = exp->exp_obd;
1538 struct lmv_obd *lmv = &obd->u.lmv;
1539 struct lmv_tgt_desc *tgt;
1543 rc = lmv_check_connect(obd);
1547 tgt = lmv_find_target(lmv, fid);
1549 RETURN(PTR_ERR(tgt));
1551 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1552 input_size, output_size, flags, request);
1557 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1558 struct obd_capa *oc, u64 valid, const char *name,
1559 const char *input, int input_size, int output_size,
1560 int flags, __u32 suppgid,
1561 struct ptlrpc_request **request)
1563 struct obd_device *obd = exp->exp_obd;
1564 struct lmv_obd *lmv = &obd->u.lmv;
1565 struct lmv_tgt_desc *tgt;
1569 rc = lmv_check_connect(obd);
1573 tgt = lmv_find_target(lmv, fid);
1575 RETURN(PTR_ERR(tgt));
1577 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1578 input_size, output_size, flags, suppgid,
1584 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1585 struct ptlrpc_request **request)
1587 struct obd_device *obd = exp->exp_obd;
1588 struct lmv_obd *lmv = &obd->u.lmv;
1589 struct lmv_tgt_desc *tgt;
1593 rc = lmv_check_connect(obd);
1597 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1599 RETURN(PTR_ERR(tgt));
1601 if (op_data->op_flags & MF_GET_MDT_IDX) {
1602 op_data->op_mds = tgt->ltd_idx;
1606 rc = md_getattr(tgt->ltd_exp, op_data, request);
1611 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1613 struct obd_device *obd = exp->exp_obd;
1614 struct lmv_obd *lmv = &obd->u.lmv;
1619 rc = lmv_check_connect(obd);
1623 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1626 * With DNE every object can have two locks in different namespaces:
1627 * lookup lock in space of MDT storing direntry and update/open lock in
1628 * space of MDT storing inode.
1630 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1631 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1633 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1639 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1640 ldlm_iterator_t it, void *data)
1642 struct obd_device *obd = exp->exp_obd;
1643 struct lmv_obd *lmv = &obd->u.lmv;
1649 rc = lmv_check_connect(obd);
1653 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1656 * With DNE every object can have two locks in different namespaces:
1657 * lookup lock in space of MDT storing direntry and update/open lock in
1658 * space of MDT storing inode. Try the MDT that the FID maps to first,
1659 * since this can be easily found, and only try others if that fails.
1661 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1662 i < lmv->desc.ld_tgt_count;
1663 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1665 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1666 obd->obd_name, PFID(fid), tgt);
1670 if (lmv->tgts[tgt] == NULL ||
1671 lmv->tgts[tgt]->ltd_exp == NULL)
1674 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1683 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1684 struct md_open_data *mod, struct ptlrpc_request **request)
1686 struct obd_device *obd = exp->exp_obd;
1687 struct lmv_obd *lmv = &obd->u.lmv;
1688 struct lmv_tgt_desc *tgt;
1692 rc = lmv_check_connect(obd);
1696 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1698 RETURN(PTR_ERR(tgt));
1700 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1701 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1706 * Choosing the MDT by name or FID in @op_data.
1707 * For non-striped directory, it will locate MDT by fid.
1708 * For striped-directory, it will locate MDT by name. And also
1709 * it will reset op_fid1 with the FID of the choosen stripe.
1711 struct lmv_tgt_desc *
1712 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1713 const char *name, int namelen, struct lu_fid *fid,
1716 struct lmv_tgt_desc *tgt;
1717 const struct lmv_oinfo *oinfo;
1719 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1720 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1721 RETURN(ERR_PTR(-EBADF));
1722 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1724 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1726 RETURN(ERR_CAST(oinfo));
1730 *fid = oinfo->lmo_fid;
1732 *mds = oinfo->lmo_mds;
1734 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1736 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1737 PFID(&oinfo->lmo_fid));
1742 * Locate mds by fid or name
1744 * For striped directory (lsm != NULL), it will locate the stripe
1745 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1746 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1747 * walk through all of stripes to locate the entry.
1749 * For normal direcotry, it will locate MDS by FID directly.
1750 * \param[in] lmv LMV device
1751 * \param[in] op_data client MD stack parameters, name, namelen
1753 * \param[in] fid object FID used to locate MDS.
1755 * retval pointer to the lmv_tgt_desc if succeed.
1756 * ERR_PTR(errno) if failed.
1758 struct lmv_tgt_desc*
1759 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1762 struct lmv_stripe_md *lsm = op_data->op_mea1;
1763 struct lmv_tgt_desc *tgt;
1765 /* During creating VOLATILE file, it should honor the mdt
1766 * index if the file under striped dir is being restored, see
1768 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1769 (int)op_data->op_mds != -1 && lsm != NULL) {
1771 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1775 /* refill the right parent fid */
1776 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1777 struct lmv_oinfo *oinfo;
1779 oinfo = &lsm->lsm_md_oinfo[i];
1780 if (oinfo->lmo_mds == op_data->op_mds) {
1781 *fid = oinfo->lmo_fid;
1786 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1787 if (i == lsm->lsm_md_stripe_count)
1788 tgt = ERR_PTR(-EINVAL);
1793 if (lsm == NULL || op_data->op_namelen == 0) {
1794 tgt = lmv_find_target(lmv, fid);
1798 op_data->op_mds = tgt->ltd_idx;
1802 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1803 op_data->op_namelen, fid,
1807 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1808 const void *data, size_t datalen, umode_t mode, uid_t uid,
1809 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1810 struct ptlrpc_request **request)
1812 struct obd_device *obd = exp->exp_obd;
1813 struct lmv_obd *lmv = &obd->u.lmv;
1814 struct lmv_tgt_desc *tgt;
1818 rc = lmv_check_connect(obd);
1822 if (!lmv->desc.ld_active_tgt_count)
1825 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1827 RETURN(PTR_ERR(tgt));
1829 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1830 (int)op_data->op_namelen, op_data->op_name,
1831 PFID(&op_data->op_fid1), op_data->op_mds);
1833 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1836 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1837 /* Send the create request to the MDT where the object
1838 * will be located */
1839 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1841 RETURN(PTR_ERR(tgt));
1843 op_data->op_mds = tgt->ltd_idx;
1845 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1848 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1849 PFID(&op_data->op_fid2), op_data->op_mds);
1851 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1852 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1853 cap_effective, rdev, request);
1855 if (*request == NULL)
1857 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1863 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1864 const union ldlm_policy_data *policy,
1865 struct lookup_intent *it, struct md_op_data *op_data,
1866 struct lustre_handle *lockh, __u64 extra_lock_flags)
1868 struct obd_device *obd = exp->exp_obd;
1869 struct lmv_obd *lmv = &obd->u.lmv;
1870 struct lmv_tgt_desc *tgt;
1874 rc = lmv_check_connect(obd);
1878 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1879 LL_IT2STR(it), PFID(&op_data->op_fid1));
1881 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1883 RETURN(PTR_ERR(tgt));
1885 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1886 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1888 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1895 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1896 struct ptlrpc_request **preq)
1898 struct ptlrpc_request *req = NULL;
1899 struct obd_device *obd = exp->exp_obd;
1900 struct lmv_obd *lmv = &obd->u.lmv;
1901 struct lmv_tgt_desc *tgt;
1902 struct mdt_body *body;
1906 rc = lmv_check_connect(obd);
1910 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1912 RETURN(PTR_ERR(tgt));
1914 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1915 (int)op_data->op_namelen, op_data->op_name,
1916 PFID(&op_data->op_fid1), tgt->ltd_idx);
1918 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1922 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1923 LASSERT(body != NULL);
1925 if (body->mbo_valid & OBD_MD_MDS) {
1926 struct lu_fid rid = body->mbo_fid1;
1927 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1930 tgt = lmv_find_target(lmv, &rid);
1932 ptlrpc_req_finished(*preq);
1934 RETURN(PTR_ERR(tgt));
1937 op_data->op_fid1 = rid;
1938 op_data->op_valid |= OBD_MD_FLCROSSREF;
1939 op_data->op_namelen = 0;
1940 op_data->op_name = NULL;
1941 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1942 ptlrpc_req_finished(*preq);
1949 #define md_op_data_fid(op_data, fl) \
1950 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1951 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1952 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1953 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1956 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1957 struct md_op_data *op_data,
1958 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1960 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1961 struct obd_device *obd = exp->exp_obd;
1962 struct lmv_obd *lmv = &obd->u.lmv;
1963 ldlm_policy_data_t policy = {{ 0 }};
1967 if (!fid_is_sane(fid))
1971 tgt = lmv_find_target(lmv, fid);
1973 RETURN(PTR_ERR(tgt));
1976 if (tgt->ltd_idx != op_tgt) {
1977 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1978 policy.l_inodebits.bits = bits;
1979 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1980 mode, LCF_ASYNC, NULL);
1983 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1985 op_data->op_flags |= flag;
1993 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1996 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1997 struct ptlrpc_request **request)
1999 struct obd_device *obd = exp->exp_obd;
2000 struct lmv_obd *lmv = &obd->u.lmv;
2001 struct lmv_tgt_desc *tgt;
2005 rc = lmv_check_connect(obd);
2009 LASSERT(op_data->op_namelen != 0);
2011 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2012 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2013 op_data->op_name, PFID(&op_data->op_fid1));
2015 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2016 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2017 op_data->op_cap = cfs_curproc_cap_pack();
2018 if (op_data->op_mea2 != NULL) {
2019 struct lmv_stripe_md *lsm = op_data->op_mea2;
2020 const struct lmv_oinfo *oinfo;
2022 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2023 op_data->op_namelen);
2025 RETURN(PTR_ERR(oinfo));
2027 op_data->op_fid2 = oinfo->lmo_fid;
2030 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2032 RETURN(PTR_ERR(tgt));
2035 * Cancel UPDATE lock on child (fid1).
2037 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2038 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2039 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2043 rc = md_link(tgt->ltd_exp, op_data, request);
2048 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2049 const char *old, size_t oldlen,
2050 const char *new, size_t newlen,
2051 struct ptlrpc_request **request)
2053 struct obd_device *obd = exp->exp_obd;
2054 struct lmv_obd *lmv = &obd->u.lmv;
2055 struct lmv_tgt_desc *src_tgt;
2059 LASSERT(oldlen != 0);
2061 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2062 (int)oldlen, old, PFID(&op_data->op_fid1),
2063 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2064 (int)newlen, new, PFID(&op_data->op_fid2),
2065 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2067 rc = lmv_check_connect(obd);
2071 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2072 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2073 op_data->op_cap = cfs_curproc_cap_pack();
2074 if (op_data->op_cli_flags & CLI_MIGRATE) {
2075 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2076 PFID(&op_data->op_fid3));
2078 if (op_data->op_mea1 != NULL) {
2079 struct lmv_stripe_md *lsm = op_data->op_mea1;
2080 struct lmv_tgt_desc *tmp;
2082 /* Fix the parent fid for striped dir */
2083 tmp = lmv_locate_target_for_name(lmv, lsm, old,
2088 RETURN(PTR_ERR(tmp));
2091 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2095 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2097 if (op_data->op_mea1 != NULL) {
2098 struct lmv_stripe_md *lsm = op_data->op_mea1;
2100 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2104 if (IS_ERR(src_tgt))
2105 RETURN(PTR_ERR(src_tgt));
2107 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2108 if (IS_ERR(src_tgt))
2109 RETURN(PTR_ERR(src_tgt));
2111 op_data->op_mds = src_tgt->ltd_idx;
2114 if (op_data->op_mea2) {
2115 struct lmv_stripe_md *lsm = op_data->op_mea2;
2116 const struct lmv_oinfo *oinfo;
2118 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2120 RETURN(PTR_ERR(oinfo));
2122 op_data->op_fid2 = oinfo->lmo_fid;
2125 if (IS_ERR(src_tgt))
2126 RETURN(PTR_ERR(src_tgt));
2129 * LOOKUP lock on src child (fid3) should also be cancelled for
2130 * src_tgt in mdc_rename.
2132 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2135 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2138 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2139 LCK_EX, MDS_INODELOCK_UPDATE,
2140 MF_MDC_CANCEL_FID2);
2145 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2147 if (fid_is_sane(&op_data->op_fid3)) {
2148 struct lmv_tgt_desc *tgt;
2150 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2152 RETURN(PTR_ERR(tgt));
2154 /* Cancel LOOKUP lock on its parent */
2155 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2156 LCK_EX, MDS_INODELOCK_LOOKUP,
2157 MF_MDC_CANCEL_FID3);
2161 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2162 LCK_EX, MDS_INODELOCK_FULL,
2163 MF_MDC_CANCEL_FID3);
2169 * Cancel all the locks on tgt child (fid4).
2171 if (fid_is_sane(&op_data->op_fid4))
2172 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2173 LCK_EX, MDS_INODELOCK_FULL,
2174 MF_MDC_CANCEL_FID4);
2176 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2177 op_data->op_mds, PFID(&op_data->op_fid2));
2179 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2185 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2186 void *ea, size_t ealen, struct ptlrpc_request **request)
2188 struct obd_device *obd = exp->exp_obd;
2189 struct lmv_obd *lmv = &obd->u.lmv;
2190 struct lmv_tgt_desc *tgt;
2194 rc = lmv_check_connect(obd);
2198 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2199 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2201 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2202 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2204 RETURN(PTR_ERR(tgt));
2206 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2211 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2212 struct obd_capa *oc, struct ptlrpc_request **request)
2214 struct obd_device *obd = exp->exp_obd;
2215 struct lmv_obd *lmv = &obd->u.lmv;
2216 struct lmv_tgt_desc *tgt;
2220 rc = lmv_check_connect(obd);
2224 tgt = lmv_find_target(lmv, fid);
2226 RETURN(PTR_ERR(tgt));
2228 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2233 * Get current minimum entry from striped directory
2235 * This function will search the dir entry, whose hash value is the
2236 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2237 * only being called for striped directory.
2239 * \param[in] exp export of LMV
2240 * \param[in] op_data parameters transferred beween client MD stack
2241 * stripe_information will be included in this
2243 * \param[in] cb_op ldlm callback being used in enqueue in
2245 * \param[in] hash_offset the hash value, which is used to locate
2246 * minum(closet) dir entry
2247 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2248 * index of last entry, so to avoid hash conflict
2249 * between stripes. It will also be used to
2250 * return the stripe index of current dir entry.
2251 * \param[in|out] entp the minum entry and it also is being used
2252 * to input the last dir entry to resolve the
2255 * \param[out] ppage the page which holds the minum entry
2257 * \retval = 0 get the entry successfully
2258 * negative errno (< 0) does not get the entry
2260 static int lmv_get_min_striped_entry(struct obd_export *exp,
2261 struct md_op_data *op_data,
2262 struct md_callback *cb_op,
2263 __u64 hash_offset, int *stripe_offset,
2264 struct lu_dirent **entp,
2265 struct page **ppage)
2267 struct obd_device *obd = exp->exp_obd;
2268 struct lmv_obd *lmv = &obd->u.lmv;
2269 struct lmv_stripe_md *lsm = op_data->op_mea1;
2270 struct lmv_tgt_desc *tgt;
2272 struct lu_dirent *min_ent = NULL;
2273 struct page *min_page = NULL;
2279 stripe_count = lsm->lsm_md_stripe_count;
2280 for (i = 0; i < stripe_count; i++) {
2281 struct lu_dirent *ent = NULL;
2282 struct page *page = NULL;
2283 struct lu_dirpage *dp;
2284 __u64 stripe_hash = hash_offset;
2286 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2288 GOTO(out, rc = PTR_ERR(tgt));
2290 /* op_data will be shared by each stripe, so we need
2291 * reset these value for each stripe */
2292 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2293 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2294 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2296 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2301 dp = page_address(page);
2302 for (ent = lu_dirent_start(dp); ent != NULL;
2303 ent = lu_dirent_next(ent)) {
2304 /* Skip dummy entry */
2305 if (le16_to_cpu(ent->lde_namelen) == 0)
2308 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2311 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2312 (*entp == ent || i < *stripe_offset))
2315 /* skip . and .. for other stripes */
2317 (strncmp(ent->lde_name, ".",
2318 le16_to_cpu(ent->lde_namelen)) == 0 ||
2319 strncmp(ent->lde_name, "..",
2320 le16_to_cpu(ent->lde_namelen)) == 0))
2326 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2329 page_cache_release(page);
2332 /* reach the end of current stripe, go to next stripe */
2333 if (stripe_hash == MDS_DIR_END_OFF)
2339 if (min_ent != NULL) {
2340 if (le64_to_cpu(min_ent->lde_hash) >
2341 le64_to_cpu(ent->lde_hash)) {
2344 page_cache_release(min_page);
2349 page_cache_release(page);
2360 if (*ppage != NULL) {
2362 page_cache_release(*ppage);
2364 *stripe_offset = min_idx;
2371 * Build dir entry page from a striped directory
2373 * This function gets one entry by @offset from a striped directory. It will
2374 * read entries from all of stripes, and choose one closest to the required
2375 * offset(&offset). A few notes
2376 * 1. skip . and .. for non-zero stripes, because there can only have one .
2377 * and .. in a directory.
2378 * 2. op_data will be shared by all of stripes, instead of allocating new
2379 * one, so need to restore before reusing.
2380 * 3. release the entry page if that is not being chosen.
2382 * \param[in] exp obd export refer to LMV
2383 * \param[in] op_data hold those MD parameters of read_entry
2384 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2385 * \param[out] ldp the entry being read
2386 * \param[out] ppage the page holding the entry. Note: because the entry
2387 * will be accessed in upper layer, so we need hold the
2388 * page until the usages of entry is finished, see
2389 * ll_dir_entry_next.
2391 * retval =0 if get entry successfully
2392 * <0 cannot get entry
2394 static int lmv_read_striped_page(struct obd_export *exp,
2395 struct md_op_data *op_data,
2396 struct md_callback *cb_op,
2397 __u64 offset, struct page **ppage)
2399 struct obd_device *obd = exp->exp_obd;
2400 struct lu_fid master_fid = op_data->op_fid1;
2401 struct inode *master_inode = op_data->op_data;
2402 __u64 hash_offset = offset;
2403 struct lu_dirpage *dp;
2404 struct page *min_ent_page = NULL;
2405 struct page *ent_page = NULL;
2406 struct lu_dirent *ent;
2409 struct lu_dirent *min_ent = NULL;
2410 struct lu_dirent *last_ent;
2415 rc = lmv_check_connect(obd);
2419 /* Allocate a page and read entries from all of stripes and fill
2420 * the page by hash order */
2421 ent_page = alloc_page(GFP_KERNEL);
2422 if (ent_page == NULL)
2425 /* Initialize the entry page */
2426 dp = kmap(ent_page);
2427 memset(dp, 0, sizeof(*dp));
2428 dp->ldp_hash_start = cpu_to_le64(offset);
2429 dp->ldp_flags |= LDF_COLLIDE;
2432 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2438 /* Find the minum entry from all sub-stripes */
2439 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2445 /* If it can not get minum entry, it means it already reaches
2446 * the end of this directory */
2447 if (min_ent == NULL) {
2448 last_ent->lde_reclen = 0;
2449 hash_offset = MDS_DIR_END_OFF;
2453 ent_size = le16_to_cpu(min_ent->lde_reclen);
2455 /* the last entry lde_reclen is 0, but it might not
2456 * the end of this entry of this temporay entry */
2458 ent_size = lu_dirent_calc_size(
2459 le16_to_cpu(min_ent->lde_namelen),
2460 le32_to_cpu(min_ent->lde_attrs));
2461 if (ent_size > left_bytes) {
2462 last_ent->lde_reclen = cpu_to_le16(0);
2463 hash_offset = le64_to_cpu(min_ent->lde_hash);
2467 memcpy(ent, min_ent, ent_size);
2469 /* Replace . with master FID and Replace .. with the parent FID
2470 * of master object */
2471 if (strncmp(ent->lde_name, ".",
2472 le16_to_cpu(ent->lde_namelen)) == 0 &&
2473 le16_to_cpu(ent->lde_namelen) == 1)
2474 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2475 else if (strncmp(ent->lde_name, "..",
2476 le16_to_cpu(ent->lde_namelen)) == 0 &&
2477 le16_to_cpu(ent->lde_namelen) == 2)
2478 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2480 left_bytes -= ent_size;
2481 ent->lde_reclen = cpu_to_le16(ent_size);
2483 ent = (void *)ent + ent_size;
2484 hash_offset = le64_to_cpu(min_ent->lde_hash);
2485 if (hash_offset == MDS_DIR_END_OFF) {
2486 last_ent->lde_reclen = 0;
2491 if (min_ent_page != NULL) {
2492 kunmap(min_ent_page);
2493 page_cache_release(min_ent_page);
2496 if (unlikely(rc != 0)) {
2497 __free_page(ent_page);
2501 dp->ldp_flags |= LDF_EMPTY;
2502 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2503 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2506 /* We do not want to allocate md_op_data during each
2507 * dir entry reading, so op_data will be shared by every stripe,
2508 * then we need to restore it back to original value before
2509 * return to the upper layer */
2510 op_data->op_fid1 = master_fid;
2511 op_data->op_fid2 = master_fid;
2512 op_data->op_data = master_inode;
2519 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2520 struct md_callback *cb_op, __u64 offset,
2521 struct page **ppage)
2523 struct obd_device *obd = exp->exp_obd;
2524 struct lmv_obd *lmv = &obd->u.lmv;
2525 struct lmv_stripe_md *lsm = op_data->op_mea1;
2526 struct lmv_tgt_desc *tgt;
2530 rc = lmv_check_connect(obd);
2534 if (unlikely(lsm != NULL)) {
2535 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2539 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2541 RETURN(PTR_ERR(tgt));
2543 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2549 * Unlink a file/directory
2551 * Unlink a file or directory under the parent dir. The unlink request
2552 * usually will be sent to the MDT where the child is located, but if
2553 * the client does not have the child FID then request will be sent to the
2554 * MDT where the parent is located.
2556 * If the parent is a striped directory then it also needs to locate which
2557 * stripe the name of the child is located, and replace the parent FID
2558 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2559 * it will walk through all of sub-stripes until the child is being
2562 * \param[in] exp export refer to LMV
2563 * \param[in] op_data different parameters transferred beween client
2564 * MD stacks, name, namelen, FIDs etc.
2565 * op_fid1 is the parent FID, op_fid2 is the child
2567 * \param[out] request point to the request of unlink.
2569 * retval 0 if succeed
2570 * negative errno if failed.
2572 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2573 struct ptlrpc_request **request)
2575 struct obd_device *obd = exp->exp_obd;
2576 struct lmv_obd *lmv = &obd->u.lmv;
2577 struct lmv_tgt_desc *tgt = NULL;
2578 struct lmv_tgt_desc *parent_tgt = NULL;
2579 struct mdt_body *body;
2581 int stripe_index = 0;
2582 struct lmv_stripe_md *lsm = op_data->op_mea1;
2585 rc = lmv_check_connect(obd);
2589 /* For striped dir, we need to locate the parent as well */
2591 struct lmv_tgt_desc *tmp;
2593 LASSERT(op_data->op_name != NULL &&
2594 op_data->op_namelen != 0);
2596 tmp = lmv_locate_target_for_name(lmv, lsm,
2598 op_data->op_namelen,
2602 /* return -EBADFD means unknown hash type, might
2603 * need try all sub-stripe here */
2604 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2605 RETURN(PTR_ERR(tmp));
2607 /* Note: both migrating dir and unknown hash dir need to
2608 * try all of sub-stripes, so we need start search the
2609 * name from stripe 0, but migrating dir is already handled
2610 * inside lmv_locate_target_for_name(), so we only check
2611 * unknown hash type directory here */
2612 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2613 struct lmv_oinfo *oinfo;
2615 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2617 op_data->op_fid1 = oinfo->lmo_fid;
2618 op_data->op_mds = oinfo->lmo_mds;
2623 /* Send unlink requests to the MDT where the child is located */
2624 if (likely(!fid_is_zero(&op_data->op_fid2)))
2625 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2626 else if (lsm != NULL)
2627 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2629 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2632 RETURN(PTR_ERR(tgt));
2634 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2635 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2636 op_data->op_cap = cfs_curproc_cap_pack();
2639 * If child's fid is given, cancel unused locks for it if it is from
2640 * another export than parent.
2642 * LOOKUP lock for child (fid3) should also be cancelled on parent
2643 * tgt_tgt in mdc_unlink().
2645 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2648 * Cancel FULL locks on child (fid3).
2650 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2651 if (IS_ERR(parent_tgt))
2652 RETURN(PTR_ERR(parent_tgt));
2654 if (parent_tgt != tgt) {
2655 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2656 LCK_EX, MDS_INODELOCK_LOOKUP,
2657 MF_MDC_CANCEL_FID3);
2660 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2661 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2665 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2666 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2668 rc = md_unlink(tgt->ltd_exp, op_data, request);
2669 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2672 /* Try next stripe if it is needed. */
2673 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2674 struct lmv_oinfo *oinfo;
2677 if (stripe_index >= lsm->lsm_md_stripe_count)
2680 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2682 op_data->op_fid1 = oinfo->lmo_fid;
2683 op_data->op_mds = oinfo->lmo_mds;
2685 ptlrpc_req_finished(*request);
2688 goto try_next_stripe;
2691 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2695 /* Not cross-ref case, just get out of here. */
2696 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2699 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2700 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2702 /* This is a remote object, try remote MDT, Note: it may
2703 * try more than 1 time here, Considering following case
2704 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2705 * 1. Initially A does not know where remote1 is, it send
2706 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2707 * resend unlink RPC to MDT1 (retry 1st time).
2709 * 2. During the unlink RPC in flight,
2710 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2711 * and create new remote1, but on MDT0
2713 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2714 * /mnt/lustre, then lookup get fid of remote1, and find
2715 * it is remote dir again, and replay -EREMOTE again.
2717 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2719 * In theory, it might try unlimited time here, but it should
2720 * be very rare case. */
2721 op_data->op_fid2 = body->mbo_fid1;
2722 ptlrpc_req_finished(*request);
2728 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2730 struct lmv_obd *lmv = &obd->u.lmv;
2734 case OBD_CLEANUP_EARLY:
2735 /* XXX: here should be calling obd_precleanup() down to
2738 case OBD_CLEANUP_EXPORTS:
2739 fld_client_proc_fini(&lmv->lmv_fld);
2740 lprocfs_obd_cleanup(obd);
2741 lprocfs_free_md_stats(obd);
2750 * Get by key a value associated with a LMV device.
2752 * Dispatch request to lower-layer devices as needed.
2754 * \param[in] env execution environment for this thread
2755 * \param[in] exp export for the LMV device
2756 * \param[in] keylen length of key identifier
2757 * \param[in] key identifier of key to get value for
2758 * \param[in] vallen size of \a val
2759 * \param[out] val pointer to storage location for value
2760 * \param[in] lsm optional striping metadata of object
2762 * \retval 0 on success
2763 * \retval negative negated errno on failure
2765 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2766 __u32 keylen, void *key, __u32 *vallen, void *val,
2767 struct lov_stripe_md *lsm)
2769 struct obd_device *obd;
2770 struct lmv_obd *lmv;
2774 obd = class_exp2obd(exp);
2776 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2777 exp->exp_handle.h_cookie);
2782 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2785 rc = lmv_check_connect(obd);
2789 LASSERT(*vallen == sizeof(__u32));
2790 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2791 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2793 * All tgts should be connected when this gets called.
2795 if (tgt == NULL || tgt->ltd_exp == NULL)
2798 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2803 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2804 KEY_IS(KEY_DEFAULT_EASIZE) ||
2805 KEY_IS(KEY_MAX_COOKIESIZE) ||
2806 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2807 KEY_IS(KEY_CONN_DATA)) {
2808 rc = lmv_check_connect(obd);
2813 * Forwarding this request to first MDS, it should know LOV
2816 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2818 if (!rc && KEY_IS(KEY_CONN_DATA))
2819 exp->exp_connect_data = *(struct obd_connect_data *)val;
2821 } else if (KEY_IS(KEY_TGT_COUNT)) {
2822 *((int *)val) = lmv->desc.ld_tgt_count;
2826 CDEBUG(D_IOCTL, "Invalid key\n");
2831 * Asynchronously set by key a value associated with a LMV device.
2833 * Dispatch request to lower-layer devices as needed.
2835 * \param[in] env execution environment for this thread
2836 * \param[in] exp export for the LMV device
2837 * \param[in] keylen length of key identifier
2838 * \param[in] key identifier of key to store value for
2839 * \param[in] vallen size of value to store
2840 * \param[in] val pointer to data to be stored
2841 * \param[in] set optional list of related ptlrpc requests
2843 * \retval 0 on success
2844 * \retval negative negated errno on failure
2846 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2847 __u32 keylen, void *key, __u32 vallen, void *val,
2848 struct ptlrpc_request_set *set)
2850 struct lmv_tgt_desc *tgt = NULL;
2851 struct obd_device *obd;
2852 struct lmv_obd *lmv;
2856 obd = class_exp2obd(exp);
2858 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2859 exp->exp_handle.h_cookie);
2864 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2865 KEY_IS(KEY_DEFAULT_EASIZE)) {
2868 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2871 if (tgt == NULL || tgt->ltd_exp == NULL)
2874 err = obd_set_info_async(env, tgt->ltd_exp,
2875 keylen, key, vallen, val, set);
2886 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2887 struct lmv_mds_md_v1 *lmm1)
2892 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2893 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2894 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2895 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2896 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2897 sizeof(lmm1->lmv_pool_name));
2898 if (cplen >= sizeof(lmm1->lmv_pool_name))
2901 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2902 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2903 &lsm->lsm_md_oinfo[i].lmo_fid);
2907 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2911 bool allocated = false;
2915 LASSERT(lmmp != NULL);
2917 if (*lmmp != NULL && lsm == NULL) {
2920 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2921 lmm_size = lmv_mds_md_size(stripe_count,
2922 le32_to_cpu((*lmmp)->lmv_magic));
2925 OBD_FREE(*lmmp, lmm_size);
2931 if (*lmmp == NULL && lsm == NULL) {
2932 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2933 LASSERT(lmm_size > 0);
2934 OBD_ALLOC(*lmmp, lmm_size);
2937 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2938 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2943 LASSERT(lsm != NULL);
2944 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2945 if (*lmmp == NULL) {
2946 OBD_ALLOC(*lmmp, lmm_size);
2952 switch (lsm->lsm_md_magic) {
2954 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2961 if (rc != 0 && allocated) {
2962 OBD_FREE(*lmmp, lmm_size);
2969 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2970 const struct lmv_mds_md_v1 *lmm1)
2972 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2979 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2980 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2981 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2982 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2983 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2985 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2986 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2987 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2988 sizeof(lsm->lsm_md_pool_name));
2990 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2993 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2994 "layout_version %d\n", lsm->lsm_md_stripe_count,
2995 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2996 lsm->lsm_md_layout_version);
2998 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2999 for (i = 0; i < stripe_count; i++) {
3000 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3001 &lmm1->lmv_stripe_fids[i]);
3002 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3003 &lsm->lsm_md_oinfo[i].lmo_mds);
3006 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3007 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3013 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3014 const union lmv_mds_md *lmm, int stripe_count)
3016 struct lmv_stripe_md *lsm;
3019 bool allocated = false;
3022 LASSERT(lsmp != NULL);
3026 if (lsm != NULL && lmm == NULL) {
3028 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3029 /* For migrating inode, the master stripe and master
3030 * object will be the same, so do not need iput, see
3031 * ll_update_lsm_md */
3032 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3033 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3034 iput(lsm->lsm_md_oinfo[i].lmo_root);
3036 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3037 OBD_FREE(lsm, lsm_size);
3043 if (lsm == NULL && lmm == NULL) {
3044 lsm_size = lmv_stripe_md_size(stripe_count);
3045 OBD_ALLOC(lsm, lsm_size);
3048 lsm->lsm_md_stripe_count = stripe_count;
3053 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3057 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3058 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3059 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3060 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3065 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3066 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3069 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3070 * stripecount should be 0 then.
3072 lsm_size = lmv_stripe_md_size(0);
3074 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3076 OBD_ALLOC(lsm, lsm_size);
3083 switch (le32_to_cpu(lmm->lmv_magic)) {
3085 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3088 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3089 le32_to_cpu(lmm->lmv_magic));
3094 if (rc != 0 && allocated) {
3095 OBD_FREE(lsm, lsm_size);
3102 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3104 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3107 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3109 lmv_unpack_md(NULL, &lsm, NULL, 0);
3111 EXPORT_SYMBOL(lmv_free_memmd);
3113 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3114 struct lov_mds_md *lmm, int disk_len)
3116 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3117 (union lmv_mds_md *)lmm, disk_len);
3120 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3121 struct lov_stripe_md *lsm)
3123 struct obd_device *obd = exp->exp_obd;
3124 struct lmv_obd *lmv_obd = &obd->u.lmv;
3125 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3130 stripe_count = lmv->lsm_md_stripe_count;
3132 stripe_count = lmv_obd->desc.ld_tgt_count;
3134 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3137 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3140 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3141 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3142 ldlm_cancel_flags_t flags, void *opaque)
3144 struct obd_device *obd = exp->exp_obd;
3145 struct lmv_obd *lmv = &obd->u.lmv;
3151 LASSERT(fid != NULL);
3153 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3154 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3156 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3159 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3167 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3170 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3171 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3175 if (tgt == NULL || tgt->ltd_exp == NULL)
3177 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3181 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3182 const struct lu_fid *fid, ldlm_type_t type,
3183 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3184 struct lustre_handle *lockh)
3186 struct obd_device *obd = exp->exp_obd;
3187 struct lmv_obd *lmv = &obd->u.lmv;
3193 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3196 * With DNE every object can have two locks in different namespaces:
3197 * lookup lock in space of MDT storing direntry and update/open lock in
3198 * space of MDT storing inode. Try the MDT that the FID maps to first,
3199 * since this can be easily found, and only try others if that fails.
3201 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3202 i < lmv->desc.ld_tgt_count;
3203 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3205 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3206 obd->obd_name, PFID(fid), tgt);
3210 if (lmv->tgts[tgt] == NULL ||
3211 lmv->tgts[tgt]->ltd_exp == NULL ||
3212 lmv->tgts[tgt]->ltd_active == 0)
3215 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3216 type, policy, mode, lockh);
3224 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3225 struct obd_export *dt_exp, struct obd_export *md_exp,
3226 struct lustre_md *md)
3228 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3229 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3231 if (tgt == NULL || tgt->ltd_exp == NULL)
3234 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3237 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3239 struct obd_device *obd = exp->exp_obd;
3240 struct lmv_obd *lmv = &obd->u.lmv;
3241 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3244 if (md->lmv != NULL) {
3245 lmv_free_memmd(md->lmv);
3248 if (tgt == NULL || tgt->ltd_exp == NULL)
3250 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3253 int lmv_set_open_replay_data(struct obd_export *exp,
3254 struct obd_client_handle *och,
3255 struct lookup_intent *it)
3257 struct obd_device *obd = exp->exp_obd;
3258 struct lmv_obd *lmv = &obd->u.lmv;
3259 struct lmv_tgt_desc *tgt;
3262 tgt = lmv_find_target(lmv, &och->och_fid);
3264 RETURN(PTR_ERR(tgt));
3266 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3269 int lmv_clear_open_replay_data(struct obd_export *exp,
3270 struct obd_client_handle *och)
3272 struct obd_device *obd = exp->exp_obd;
3273 struct lmv_obd *lmv = &obd->u.lmv;
3274 struct lmv_tgt_desc *tgt;
3277 tgt = lmv_find_target(lmv, &och->och_fid);
3279 RETURN(PTR_ERR(tgt));
3281 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3284 static int lmv_get_remote_perm(struct obd_export *exp,
3285 const struct lu_fid *fid,
3286 struct obd_capa *oc, __u32 suppgid,
3287 struct ptlrpc_request **request)
3289 struct obd_device *obd = exp->exp_obd;
3290 struct lmv_obd *lmv = &obd->u.lmv;
3291 struct lmv_tgt_desc *tgt;
3295 rc = lmv_check_connect(obd);
3299 tgt = lmv_find_target(lmv, fid);
3301 RETURN(PTR_ERR(tgt));
3303 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3307 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3310 struct obd_device *obd = exp->exp_obd;
3311 struct lmv_obd *lmv = &obd->u.lmv;
3312 struct lmv_tgt_desc *tgt;
3316 rc = lmv_check_connect(obd);
3320 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3322 RETURN(PTR_ERR(tgt));
3324 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3328 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3329 const struct req_msg_field *field, struct obd_capa **oc)
3331 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3332 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3334 if (tgt == NULL || tgt->ltd_exp == NULL)
3336 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3339 int lmv_intent_getattr_async(struct obd_export *exp,
3340 struct md_enqueue_info *minfo,
3341 struct ldlm_enqueue_info *einfo)
3343 struct md_op_data *op_data = &minfo->mi_data;
3344 struct obd_device *obd = exp->exp_obd;
3345 struct lmv_obd *lmv = &obd->u.lmv;
3346 struct lmv_tgt_desc *tgt = NULL;
3350 rc = lmv_check_connect(obd);
3354 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3356 RETURN(PTR_ERR(tgt));
3358 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3362 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3363 struct lu_fid *fid, __u64 *bits)
3365 struct obd_device *obd = exp->exp_obd;
3366 struct lmv_obd *lmv = &obd->u.lmv;
3367 struct lmv_tgt_desc *tgt;
3371 rc = lmv_check_connect(obd);
3375 tgt = lmv_find_target(lmv, fid);
3377 RETURN(PTR_ERR(tgt));
3379 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3383 int lmv_get_fid_from_lsm(struct obd_export *exp,
3384 const struct lmv_stripe_md *lsm,
3385 const char *name, int namelen, struct lu_fid *fid)
3387 const struct lmv_oinfo *oinfo;
3389 LASSERT(lsm != NULL);
3390 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3392 return PTR_ERR(oinfo);
3394 *fid = oinfo->lmo_fid;
3400 * For lmv, only need to send request to master MDT, and the master MDT will
3401 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3402 * we directly fetch data from the slave MDTs.
3404 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3405 struct obd_quotactl *oqctl)
3407 struct obd_device *obd = class_exp2obd(exp);
3408 struct lmv_obd *lmv = &obd->u.lmv;
3409 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3412 __u64 curspace, curinodes;
3416 tgt->ltd_exp == NULL ||
3418 lmv->desc.ld_tgt_count == 0) {
3419 CERROR("master lmv inactive\n");
3423 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3424 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3428 curspace = curinodes = 0;
3429 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3433 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3436 err = obd_quotactl(tgt->ltd_exp, oqctl);
3438 CERROR("getquota on mdt %d failed. %d\n", i, err);
3442 curspace += oqctl->qc_dqblk.dqb_curspace;
3443 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3446 oqctl->qc_dqblk.dqb_curspace = curspace;
3447 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3452 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3453 struct obd_quotactl *oqctl)
3455 struct obd_device *obd = class_exp2obd(exp);
3456 struct lmv_obd *lmv = &obd->u.lmv;
3457 struct lmv_tgt_desc *tgt;
3462 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3465 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3466 CERROR("lmv idx %d inactive\n", i);
3470 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3478 static int lmv_merge_attr(struct obd_export *exp,
3479 const struct lmv_stripe_md *lsm,
3480 struct cl_attr *attr,
3481 ldlm_blocking_callback cb_blocking)
3486 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3490 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3491 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3493 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3494 " atime %lu ctime %lu, mtime %lu.\n",
3495 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3496 i_size_read(inode), (unsigned long long)inode->i_blocks,
3497 inode->i_nlink, LTIME_S(inode->i_atime),
3498 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3500 /* for slave stripe, it needs to subtract nlink for . and .. */
3502 attr->cat_nlink += inode->i_nlink - 2;
3504 attr->cat_nlink = inode->i_nlink;
3506 attr->cat_size += i_size_read(inode);
3507 attr->cat_blocks += inode->i_blocks;
3509 if (attr->cat_atime < LTIME_S(inode->i_atime))
3510 attr->cat_atime = LTIME_S(inode->i_atime);
3512 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3513 attr->cat_ctime = LTIME_S(inode->i_ctime);
3515 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3516 attr->cat_mtime = LTIME_S(inode->i_mtime);
3521 struct obd_ops lmv_obd_ops = {
3522 .o_owner = THIS_MODULE,
3523 .o_setup = lmv_setup,
3524 .o_cleanup = lmv_cleanup,
3525 .o_precleanup = lmv_precleanup,
3526 .o_process_config = lmv_process_config,
3527 .o_connect = lmv_connect,
3528 .o_disconnect = lmv_disconnect,
3529 .o_statfs = lmv_statfs,
3530 .o_get_info = lmv_get_info,
3531 .o_set_info_async = lmv_set_info_async,
3532 .o_packmd = lmv_packmd,
3533 .o_unpackmd = lmv_unpackmd,
3534 .o_notify = lmv_notify,
3535 .o_get_uuid = lmv_get_uuid,
3536 .o_iocontrol = lmv_iocontrol,
3537 .o_quotacheck = lmv_quotacheck,
3538 .o_quotactl = lmv_quotactl
3541 struct md_ops lmv_md_ops = {
3542 .m_getstatus = lmv_getstatus,
3543 .m_null_inode = lmv_null_inode,
3544 .m_find_cbdata = lmv_find_cbdata,
3545 .m_close = lmv_close,
3546 .m_create = lmv_create,
3547 .m_enqueue = lmv_enqueue,
3548 .m_getattr = lmv_getattr,
3549 .m_getxattr = lmv_getxattr,
3550 .m_getattr_name = lmv_getattr_name,
3551 .m_intent_lock = lmv_intent_lock,
3553 .m_rename = lmv_rename,
3554 .m_setattr = lmv_setattr,
3555 .m_setxattr = lmv_setxattr,
3556 .m_fsync = lmv_fsync,
3557 .m_read_page = lmv_read_page,
3558 .m_unlink = lmv_unlink,
3559 .m_init_ea_size = lmv_init_ea_size,
3560 .m_cancel_unused = lmv_cancel_unused,
3561 .m_set_lock_data = lmv_set_lock_data,
3562 .m_lock_match = lmv_lock_match,
3563 .m_get_lustre_md = lmv_get_lustre_md,
3564 .m_free_lustre_md = lmv_free_lustre_md,
3565 .m_merge_attr = lmv_merge_attr,
3566 .m_set_open_replay_data = lmv_set_open_replay_data,
3567 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3568 .m_renew_capa = lmv_renew_capa,
3569 .m_unpack_capa = lmv_unpack_capa,
3570 .m_get_remote_perm = lmv_get_remote_perm,
3571 .m_intent_getattr_async = lmv_intent_getattr_async,
3572 .m_revalidate_lock = lmv_revalidate_lock,
3573 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3576 int __init lmv_init(void)
3578 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3579 LUSTRE_LMV_NAME, NULL);
3582 static void lmv_exit(void)
3584 class_unregister_type(LUSTRE_LMV_NAME);
3587 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3588 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3589 MODULE_LICENSE("GPL");
3591 module_init(lmv_init);
3592 module_exit(lmv_exit);