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 "lmv_internal.h"
60 static void lmv_activate_target(struct lmv_obd *lmv,
61 struct lmv_tgt_desc *tgt,
64 if (tgt->ltd_active == activate)
67 tgt->ltd_active = activate;
68 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
74 * -EINVAL : UUID can't be found in the LMV's target list
75 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
76 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
78 static int lmv_set_mdc_active(struct lmv_obd *lmv,
79 const struct obd_uuid *uuid,
82 struct lmv_tgt_desc *tgt = NULL;
83 struct obd_device *obd;
88 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
89 lmv, uuid->uuid, activate);
91 spin_lock(&lmv->lmv_lock);
92 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
94 if (tgt == NULL || tgt->ltd_exp == NULL)
97 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
98 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
100 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
104 if (i == lmv->desc.ld_tgt_count)
105 GOTO(out_lmv_lock, rc = -EINVAL);
107 obd = class_exp2obd(tgt->ltd_exp);
109 GOTO(out_lmv_lock, rc = -ENOTCONN);
111 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
112 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
113 obd->obd_type->typ_name, i);
114 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
116 if (tgt->ltd_active == activate) {
117 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
118 activate ? "" : "in");
119 GOTO(out_lmv_lock, rc);
122 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
123 activate ? "" : "in");
124 lmv_activate_target(lmv, tgt, activate);
128 spin_unlock(&lmv->lmv_lock);
132 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
134 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
135 struct lmv_tgt_desc *tgt = lmv->tgts[0];
137 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
141 enum obd_notify_event ev, void *data)
143 struct obd_connect_data *conn_data;
144 struct lmv_obd *lmv = &obd->u.lmv;
145 struct obd_uuid *uuid;
149 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
150 CERROR("unexpected notification of %s %s!\n",
151 watched->obd_type->typ_name,
156 uuid = &watched->u.cli.cl_target_uuid;
157 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
159 * Set MDC as active before notifying the observer, so the
160 * observer can use the MDC normally.
162 rc = lmv_set_mdc_active(lmv, uuid,
163 ev == OBD_NOTIFY_ACTIVE);
165 CERROR("%sactivation of %s failed: %d\n",
166 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
170 } else if (ev == OBD_NOTIFY_OCD) {
171 conn_data = &watched->u.cli.cl_import->imp_connect_data;
173 * XXX: Make sure that ocd_connect_flags from all targets are
174 * the same. Otherwise one of MDTs runs wrong version or
175 * something like this. --umka
177 obd->obd_self_export->exp_connect_data = *conn_data;
180 else if (ev == OBD_NOTIFY_DISCON) {
182 * For disconnect event, flush fld cache for failout MDS case.
184 fld_client_flush(&lmv->lmv_fld);
188 * Pass the notification up the chain.
190 if (obd->obd_observer)
191 rc = obd_notify(obd->obd_observer, watched, ev, data);
197 * This is fake connect function. Its purpose is to initialize lmv and say
198 * caller that everything is okay. Real connection will be performed later.
200 static int lmv_connect(const struct lu_env *env,
201 struct obd_export **exp, struct obd_device *obd,
202 struct obd_uuid *cluuid, struct obd_connect_data *data,
205 struct lmv_obd *lmv = &obd->u.lmv;
206 struct lustre_handle conn = { 0 };
211 * We don't want to actually do the underlying connections more than
212 * once, so keep track.
215 if (lmv->refcount > 1) {
220 rc = class_connect(&conn, obd, cluuid);
222 CERROR("class_connection() returned %d\n", rc);
226 *exp = class_conn2export(&conn);
227 class_export_get(*exp);
231 lmv->cluuid = *cluuid;
234 lmv->conn_data = *data;
236 if (lmv->targets_proc_entry == NULL) {
237 lmv->targets_proc_entry = lprocfs_register("target_obds",
240 if (IS_ERR(lmv->targets_proc_entry)) {
241 CERROR("%s: cannot register "
242 "/proc/fs/lustre/%s/%s/target_obds\n",
243 obd->obd_name, obd->obd_type->typ_name,
245 lmv->targets_proc_entry = NULL;
250 * All real clients should perform actual connection right away, because
251 * it is possible, that LMV will not have opportunity to connect targets
252 * and MDC stuff will be called directly, for instance while reading
253 * ../mdc/../kbytesfree procfs file, etc.
255 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
256 rc = lmv_check_connect(obd);
258 if (rc && lmv->targets_proc_entry != NULL)
259 lprocfs_remove(&lmv->targets_proc_entry);
263 static int lmv_init_ea_size(struct obd_export *exp,
264 __u32 easize, __u32 def_easize,
265 __u32 cookiesize, __u32 def_cookiesize)
267 struct obd_device *obd = exp->exp_obd;
268 struct lmv_obd *lmv = &obd->u.lmv;
274 if (lmv->max_easize < easize) {
275 lmv->max_easize = easize;
278 if (lmv->max_def_easize < def_easize) {
279 lmv->max_def_easize = def_easize;
282 if (lmv->max_cookiesize < cookiesize) {
283 lmv->max_cookiesize = cookiesize;
286 if (lmv->max_def_cookiesize < def_cookiesize) {
287 lmv->max_def_cookiesize = def_cookiesize;
293 if (lmv->connected == 0)
296 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
297 struct lmv_tgt_desc *tgt = lmv->tgts[i];
299 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
300 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
304 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
305 cookiesize, def_cookiesize);
307 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
308 " rc = %d\n", obd->obd_name, i, rc);
315 #define MAX_STRING_SIZE 128
317 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
319 struct lmv_obd *lmv = &obd->u.lmv;
320 struct obd_uuid *cluuid = &lmv->cluuid;
321 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
322 struct obd_device *mdc_obd;
323 struct obd_export *mdc_exp;
324 struct lu_fld_target target;
328 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
331 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
335 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
336 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
337 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
340 if (!mdc_obd->obd_set_up) {
341 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
345 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
346 &lmv->conn_data, NULL);
348 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
353 * Init fid sequence client for this mdc and add new fld target.
355 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
359 target.ft_srv = NULL;
360 target.ft_exp = mdc_exp;
361 target.ft_idx = tgt->ltd_idx;
363 fld_client_add_target(&lmv->lmv_fld, &target);
365 rc = obd_register_observer(mdc_obd, obd);
367 obd_disconnect(mdc_exp);
368 CERROR("target %s register_observer error %d\n",
369 tgt->ltd_uuid.uuid, rc);
373 if (obd->obd_observer) {
375 * Tell the observer about the new target.
377 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
379 (void *)(tgt - lmv->tgts[0]));
381 obd_disconnect(mdc_exp);
387 tgt->ltd_exp = mdc_exp;
388 lmv->desc.ld_active_tgt_count++;
390 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
391 lmv->max_cookiesize, lmv->max_def_cookiesize);
393 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
394 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
395 atomic_read(&obd->obd_refcount));
397 if (lmv->targets_proc_entry != NULL) {
398 struct proc_dir_entry *mdc_symlink;
400 LASSERT(mdc_obd->obd_type != NULL);
401 LASSERT(mdc_obd->obd_type->typ_name != NULL);
402 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
403 lmv->targets_proc_entry,
405 mdc_obd->obd_type->typ_name,
407 if (mdc_symlink == NULL) {
408 CERROR("cannot register LMV target "
409 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
410 obd->obd_type->typ_name, obd->obd_name,
417 static void lmv_del_target(struct lmv_obd *lmv, int index)
419 if (lmv->tgts[index] == NULL)
422 OBD_FREE_PTR(lmv->tgts[index]);
423 lmv->tgts[index] = NULL;
427 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
428 __u32 index, int gen)
430 struct lmv_obd *lmv = &obd->u.lmv;
431 struct lmv_tgt_desc *tgt;
432 int orig_tgt_count = 0;
436 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
438 mutex_lock(&lmv->lmv_init_mutex);
440 if (lmv->desc.ld_tgt_count == 0) {
441 struct obd_device *mdc_obd;
443 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
446 mutex_unlock(&lmv->lmv_init_mutex);
447 CERROR("%s: Target %s not attached: rc = %d\n",
448 obd->obd_name, uuidp->uuid, -EINVAL);
453 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
454 tgt = lmv->tgts[index];
455 CERROR("%s: UUID %s already assigned at LOV target index %d:"
456 " rc = %d\n", obd->obd_name,
457 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
458 mutex_unlock(&lmv->lmv_init_mutex);
462 if (index >= lmv->tgts_size) {
463 /* We need to reallocate the lmv target array. */
464 struct lmv_tgt_desc **newtgts, **old = NULL;
468 while (newsize < index + 1)
469 newsize = newsize << 1;
470 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
471 if (newtgts == NULL) {
472 mutex_unlock(&lmv->lmv_init_mutex);
476 if (lmv->tgts_size) {
477 memcpy(newtgts, lmv->tgts,
478 sizeof(*newtgts) * lmv->tgts_size);
480 oldsize = lmv->tgts_size;
484 lmv->tgts_size = newsize;
487 OBD_FREE(old, sizeof(*old) * oldsize);
489 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
495 mutex_unlock(&lmv->lmv_init_mutex);
499 mutex_init(&tgt->ltd_fid_mutex);
500 tgt->ltd_idx = index;
501 tgt->ltd_uuid = *uuidp;
503 lmv->tgts[index] = tgt;
504 if (index >= lmv->desc.ld_tgt_count) {
505 orig_tgt_count = lmv->desc.ld_tgt_count;
506 lmv->desc.ld_tgt_count = index + 1;
509 if (lmv->connected) {
510 rc = lmv_connect_mdc(obd, tgt);
512 spin_lock(&lmv->lmv_lock);
513 if (lmv->desc.ld_tgt_count == index + 1)
514 lmv->desc.ld_tgt_count = orig_tgt_count;
515 memset(tgt, 0, sizeof(*tgt));
516 spin_unlock(&lmv->lmv_lock);
518 int easize = sizeof(struct lmv_stripe_md) +
519 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
520 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
524 mutex_unlock(&lmv->lmv_init_mutex);
528 int lmv_check_connect(struct obd_device *obd)
530 struct lmv_obd *lmv = &obd->u.lmv;
531 struct lmv_tgt_desc *tgt;
540 mutex_lock(&lmv->lmv_init_mutex);
541 if (lmv->connected) {
542 mutex_unlock(&lmv->lmv_init_mutex);
546 if (lmv->desc.ld_tgt_count == 0) {
547 mutex_unlock(&lmv->lmv_init_mutex);
548 CERROR("%s: no targets configured.\n", obd->obd_name);
552 LASSERT(lmv->tgts != NULL);
554 if (lmv->tgts[0] == NULL) {
555 mutex_unlock(&lmv->lmv_init_mutex);
556 CERROR("%s: no target configured for index 0.\n",
561 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
562 lmv->cluuid.uuid, obd->obd_name);
564 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
568 rc = lmv_connect_mdc(obd, tgt);
573 class_export_put(lmv->exp);
575 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
576 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
577 mutex_unlock(&lmv->lmv_init_mutex);
588 --lmv->desc.ld_active_tgt_count;
589 rc2 = obd_disconnect(tgt->ltd_exp);
591 CERROR("LMV target %s disconnect on "
592 "MDC idx %d: error %d\n",
593 tgt->ltd_uuid.uuid, i, rc2);
597 class_disconnect(lmv->exp);
598 mutex_unlock(&lmv->lmv_init_mutex);
602 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
604 struct lmv_obd *lmv = &obd->u.lmv;
605 struct obd_device *mdc_obd;
609 LASSERT(tgt != NULL);
610 LASSERT(obd != NULL);
612 mdc_obd = class_exp2obd(tgt->ltd_exp);
615 mdc_obd->obd_force = obd->obd_force;
616 mdc_obd->obd_fail = obd->obd_fail;
617 mdc_obd->obd_no_recov = obd->obd_no_recov;
620 if (lmv->targets_proc_entry != NULL)
621 lprocfs_remove_proc_entry(mdc_obd->obd_name,
622 lmv->targets_proc_entry);
624 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
626 CERROR("Can't finanize fids factory\n");
628 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
629 tgt->ltd_exp->exp_obd->obd_name,
630 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
632 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
633 rc = obd_disconnect(tgt->ltd_exp);
635 if (tgt->ltd_active) {
636 CERROR("Target %s disconnect error %d\n",
637 tgt->ltd_uuid.uuid, rc);
641 lmv_activate_target(lmv, tgt, 0);
646 static int lmv_disconnect(struct obd_export *exp)
648 struct obd_device *obd = class_exp2obd(exp);
649 struct lmv_obd *lmv = &obd->u.lmv;
658 * Only disconnect the underlying layers on the final disconnect.
661 if (lmv->refcount != 0)
664 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
665 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
668 lmv_disconnect_mdc(obd, lmv->tgts[i]);
671 if (lmv->targets_proc_entry != NULL)
672 lprocfs_remove(&lmv->targets_proc_entry);
674 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
675 obd->obd_type->typ_name, obd->obd_name);
679 * This is the case when no real connection is established by
680 * lmv_check_connect().
683 class_export_put(exp);
684 rc = class_disconnect(exp);
685 if (lmv->refcount == 0)
690 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
692 struct obd_device *obddev = class_exp2obd(exp);
693 struct lmv_obd *lmv = &obddev->u.lmv;
694 struct getinfo_fid2path *gf;
695 struct lmv_tgt_desc *tgt;
696 struct getinfo_fid2path *remote_gf = NULL;
697 int remote_gf_size = 0;
700 gf = (struct getinfo_fid2path *)karg;
701 tgt = lmv_find_target(lmv, &gf->gf_fid);
703 RETURN(PTR_ERR(tgt));
706 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
707 if (rc != 0 && rc != -EREMOTE)
708 GOTO(out_fid2path, rc);
710 /* If remote_gf != NULL, it means just building the
711 * path on the remote MDT, copy this path segement to gf */
712 if (remote_gf != NULL) {
713 struct getinfo_fid2path *ori_gf;
716 ori_gf = (struct getinfo_fid2path *)karg;
717 if (strlen(ori_gf->gf_path) +
718 strlen(gf->gf_path) > ori_gf->gf_pathlen)
719 GOTO(out_fid2path, rc = -EOVERFLOW);
721 ptr = ori_gf->gf_path;
723 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
724 strlen(ori_gf->gf_path));
726 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
727 ptr += strlen(gf->gf_path);
731 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
732 tgt->ltd_exp->exp_obd->obd_name,
733 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
737 GOTO(out_fid2path, rc);
739 /* sigh, has to go to another MDT to do path building further */
740 if (remote_gf == NULL) {
741 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
742 OBD_ALLOC(remote_gf, remote_gf_size);
743 if (remote_gf == NULL)
744 GOTO(out_fid2path, rc = -ENOMEM);
745 remote_gf->gf_pathlen = PATH_MAX;
748 if (!fid_is_sane(&gf->gf_fid)) {
749 CERROR("%s: invalid FID "DFID": rc = %d\n",
750 tgt->ltd_exp->exp_obd->obd_name,
751 PFID(&gf->gf_fid), -EINVAL);
752 GOTO(out_fid2path, rc = -EINVAL);
755 tgt = lmv_find_target(lmv, &gf->gf_fid);
757 GOTO(out_fid2path, rc = -EINVAL);
759 remote_gf->gf_fid = gf->gf_fid;
760 remote_gf->gf_recno = -1;
761 remote_gf->gf_linkno = -1;
762 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
764 goto repeat_fid2path;
767 if (remote_gf != NULL)
768 OBD_FREE(remote_gf, remote_gf_size);
772 static int lmv_hsm_req_count(struct lmv_obd *lmv,
773 const struct hsm_user_request *hur,
774 const struct lmv_tgt_desc *tgt_mds)
778 struct lmv_tgt_desc *curr_tgt;
780 /* count how many requests must be sent to the given target */
781 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
782 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
783 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
789 static void lmv_hsm_req_build(struct lmv_obd *lmv,
790 struct hsm_user_request *hur_in,
791 const struct lmv_tgt_desc *tgt_mds,
792 struct hsm_user_request *hur_out)
795 struct lmv_tgt_desc *curr_tgt;
797 /* build the hsm_user_request for the given target */
798 hur_out->hur_request = hur_in->hur_request;
800 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
801 curr_tgt = lmv_find_target(lmv,
802 &hur_in->hur_user_item[i].hui_fid);
803 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
804 hur_out->hur_user_item[nr_out] =
805 hur_in->hur_user_item[i];
809 hur_out->hur_request.hr_itemcount = nr_out;
810 memcpy(hur_data(hur_out), hur_data(hur_in),
811 hur_in->hur_request.hr_data_len);
814 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
815 struct lustre_kernelcomm *lk, void *uarg)
819 struct kkuc_ct_data *kcd = NULL;
822 /* unregister request (call from llapi_hsm_copytool_fini) */
823 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
824 struct lmv_tgt_desc *tgt = lmv->tgts[i];
826 if (tgt == NULL || tgt->ltd_exp == NULL)
828 /* best effort: try to clean as much as possible
829 * (continue on error) */
830 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
833 /* Whatever the result, remove copytool from kuc groups.
834 * Unreached coordinators will get EPIPE on next requests
835 * and will unregister automatically.
837 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
844 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
845 struct lustre_kernelcomm *lk, void *uarg)
850 bool any_set = false;
851 struct kkuc_ct_data *kcd;
854 /* All or nothing: try to register to all MDS.
855 * In case of failure, unregister from previous MDS,
856 * except if it because of inactive target. */
857 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
858 struct lmv_tgt_desc *tgt = lmv->tgts[i];
860 if (tgt == NULL || tgt->ltd_exp == NULL)
862 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
864 if (tgt->ltd_active) {
865 /* permanent error */
866 CERROR("%s: iocontrol MDC %s on MDT"
867 " idx %d cmd %x: err = %d\n",
868 class_exp2obd(lmv->exp)->obd_name,
869 tgt->ltd_uuid.uuid, i, cmd, err);
871 lk->lk_flags |= LK_FLG_STOP;
872 /* unregister from previous MDS */
873 for (j = 0; j < i; j++) {
875 if (tgt == NULL || tgt->ltd_exp == NULL)
877 obd_iocontrol(cmd, tgt->ltd_exp, len,
882 /* else: transient error.
883 * kuc will register to the missing MDT
891 /* no registration done: return error */
894 /* at least one registration done, with no failure */
895 filp = fget(lk->lk_wfd);
904 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
905 kcd->kcd_uuid = lmv->cluuid;
906 kcd->kcd_archive = lk->lk_data;
908 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
922 int len, void *karg, void *uarg)
924 struct obd_device *obddev = class_exp2obd(exp);
925 struct lmv_obd *lmv = &obddev->u.lmv;
926 struct lmv_tgt_desc *tgt = NULL;
930 __u32 count = lmv->desc.ld_tgt_count;
937 case IOC_OBD_STATFS: {
938 struct obd_ioctl_data *data = karg;
939 struct obd_device *mdc_obd;
940 struct obd_statfs stat_buf = {0};
943 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
944 if ((index >= count))
947 tgt = lmv->tgts[index];
948 if (tgt == NULL || !tgt->ltd_active)
951 mdc_obd = class_exp2obd(tgt->ltd_exp);
956 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
957 min((int) data->ioc_plen2,
958 (int) sizeof(struct obd_uuid))))
961 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
962 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
966 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
967 min((int) data->ioc_plen1,
968 (int) sizeof(stat_buf))))
972 case OBD_IOC_QUOTACTL: {
973 struct if_quotactl *qctl = karg;
974 struct obd_quotactl *oqctl;
976 if (qctl->qc_valid == QC_MDTIDX) {
977 if (count <= qctl->qc_idx)
980 tgt = lmv->tgts[qctl->qc_idx];
981 if (tgt == NULL || tgt->ltd_exp == NULL)
983 } else if (qctl->qc_valid == QC_UUID) {
984 for (i = 0; i < count; i++) {
988 if (!obd_uuid_equals(&tgt->ltd_uuid,
992 if (tgt->ltd_exp == NULL)
1004 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005 OBD_ALLOC_PTR(oqctl);
1009 QCTL_COPY(oqctl, qctl);
1010 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1012 QCTL_COPY(qctl, oqctl);
1013 qctl->qc_valid = QC_MDTIDX;
1014 qctl->obd_uuid = tgt->ltd_uuid;
1016 OBD_FREE_PTR(oqctl);
1019 case OBD_IOC_CHANGELOG_SEND:
1020 case OBD_IOC_CHANGELOG_CLEAR: {
1021 struct ioc_changelog *icc = karg;
1023 if (icc->icc_mdtindex >= count)
1026 tgt = lmv->tgts[icc->icc_mdtindex];
1027 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1029 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1032 case LL_IOC_GET_CONNECT_FLAGS: {
1034 if (tgt == NULL || tgt->ltd_exp == NULL)
1036 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1039 case LL_IOC_FID2MDTIDX: {
1040 struct lu_fid *fid = karg;
1043 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1047 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048 * point to user space memory for FID2MDTIDX. */
1049 *(__u32 *)uarg = mdt_index;
1052 case OBD_IOC_FID2PATH: {
1053 rc = lmv_fid2path(exp, len, karg, uarg);
1056 case LL_IOC_HSM_STATE_GET:
1057 case LL_IOC_HSM_STATE_SET:
1058 case LL_IOC_HSM_ACTION: {
1059 struct md_op_data *op_data = karg;
1061 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1063 RETURN(PTR_ERR(tgt));
1065 if (tgt->ltd_exp == NULL)
1068 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1071 case LL_IOC_HSM_PROGRESS: {
1072 const struct hsm_progress_kernel *hpk = karg;
1074 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1076 RETURN(PTR_ERR(tgt));
1077 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1080 case LL_IOC_HSM_REQUEST: {
1081 struct hsm_user_request *hur = karg;
1082 unsigned int reqcount = hur->hur_request.hr_itemcount;
1087 /* if the request is about a single fid
1088 * or if there is a single MDS, no need to split
1090 if (reqcount == 1 || count == 1) {
1091 tgt = lmv_find_target(lmv,
1092 &hur->hur_user_item[0].hui_fid);
1094 RETURN(PTR_ERR(tgt));
1095 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1097 /* split fid list to their respective MDS */
1098 for (i = 0; i < count; i++) {
1099 unsigned int nr, reqlen;
1101 struct hsm_user_request *req;
1104 if (tgt == NULL || tgt->ltd_exp == NULL)
1107 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108 if (nr == 0) /* nothing for this MDS */
1111 /* build a request with fids for this MDS */
1112 reqlen = offsetof(typeof(*hur),
1114 + hur->hur_request.hr_data_len;
1115 OBD_ALLOC_LARGE(req, reqlen);
1119 lmv_hsm_req_build(lmv, hur, tgt, req);
1121 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1123 if (rc1 != 0 && rc == 0)
1125 OBD_FREE_LARGE(req, reqlen);
1130 case LL_IOC_LOV_SWAP_LAYOUTS: {
1131 struct md_op_data *op_data = karg;
1132 struct lmv_tgt_desc *tgt1, *tgt2;
1134 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1136 RETURN(PTR_ERR(tgt1));
1138 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1140 RETURN(PTR_ERR(tgt2));
1142 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1145 /* only files on same MDT can have their layouts swapped */
1146 if (tgt1->ltd_idx != tgt2->ltd_idx)
1149 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1152 case LL_IOC_HSM_CT_START: {
1153 struct lustre_kernelcomm *lk = karg;
1154 if (lk->lk_flags & LK_FLG_STOP)
1155 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1157 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1161 for (i = 0; i < count; i++) {
1162 struct obd_device *mdc_obd;
1166 if (tgt == NULL || tgt->ltd_exp == NULL)
1168 /* ll_umount_begin() sets force flag but for lmv, not
1169 * mdc. Let's pass it through */
1170 mdc_obd = class_exp2obd(tgt->ltd_exp);
1171 mdc_obd->obd_force = obddev->obd_force;
1172 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1173 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1176 if (tgt->ltd_active) {
1177 CERROR("error: iocontrol MDC %s on MDT"
1178 " idx %d cmd %x: err = %d\n",
1179 tgt->ltd_uuid.uuid, i, cmd, err);
1193 static int lmv_all_chars_policy(int count, const char *name,
1204 static int lmv_nid_policy(struct lmv_obd *lmv)
1206 struct obd_import *imp;
1210 * XXX: To get nid we assume that underlying obd device is mdc.
1212 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1213 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1214 return id % lmv->desc.ld_tgt_count;
1217 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1218 placement_policy_t placement)
1220 switch (placement) {
1221 case PLACEMENT_CHAR_POLICY:
1222 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1224 op_data->op_namelen);
1225 case PLACEMENT_NID_POLICY:
1226 return lmv_nid_policy(lmv);
1232 CERROR("Unsupported placement policy %x\n", placement);
1238 * This is _inode_ placement policy function (not name).
1240 static int lmv_placement_policy(struct obd_device *obd,
1241 struct md_op_data *op_data, u32 *mds)
1243 struct lmv_obd *lmv = &obd->u.lmv;
1246 LASSERT(mds != NULL);
1248 if (lmv->desc.ld_tgt_count == 1) {
1253 if (op_data->op_default_stripe_offset != -1) {
1254 *mds = op_data->op_default_stripe_offset;
1259 * If stripe_offset is provided during setdirstripe
1260 * (setdirstripe -i xx), xx MDS will be choosen.
1262 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1263 struct lmv_user_md *lum;
1265 lum = op_data->op_data;
1267 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1268 *mds = le32_to_cpu(lum->lum_stripe_offset);
1270 /* -1 means default, which will be in the same MDT with
1272 *mds = op_data->op_mds;
1273 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1276 /* Allocate new fid on target according to operation type and
1277 * parent home mds. */
1278 *mds = op_data->op_mds;
1284 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1286 struct lmv_tgt_desc *tgt;
1290 tgt = lmv_get_target(lmv, mds, NULL);
1292 RETURN(PTR_ERR(tgt));
1295 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1296 * on server that seq in new allocated fid is not yet known.
1298 mutex_lock(&tgt->ltd_fid_mutex);
1300 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1301 GOTO(out, rc = -ENODEV);
1304 * Asking underlying tgt layer to allocate new fid.
1306 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1308 LASSERT(fid_is_sane(fid));
1314 mutex_unlock(&tgt->ltd_fid_mutex);
1318 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1319 struct lu_fid *fid, struct md_op_data *op_data)
1321 struct obd_device *obd = class_exp2obd(exp);
1322 struct lmv_obd *lmv = &obd->u.lmv;
1327 LASSERT(op_data != NULL);
1328 LASSERT(fid != NULL);
1330 rc = lmv_placement_policy(obd, op_data, &mds);
1332 CERROR("Can't get target for allocating fid, "
1337 rc = __lmv_fid_alloc(lmv, fid, mds);
1339 CERROR("Can't alloc new fid, rc %d\n", rc);
1346 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1348 struct lmv_obd *lmv = &obd->u.lmv;
1349 struct lmv_desc *desc;
1353 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1354 CERROR("LMV setup requires a descriptor\n");
1358 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1359 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1360 CERROR("Lmv descriptor size wrong: %d > %d\n",
1361 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1365 lmv->tgts_size = 32U;
1366 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1367 if (lmv->tgts == NULL)
1370 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1371 lmv->desc.ld_tgt_count = 0;
1372 lmv->desc.ld_active_tgt_count = 0;
1373 lmv->max_cookiesize = 0;
1374 lmv->max_def_easize = 0;
1375 lmv->max_easize = 0;
1376 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1378 spin_lock_init(&lmv->lmv_lock);
1379 mutex_init(&lmv->lmv_init_mutex);
1381 #ifdef CONFIG_PROC_FS
1382 obd->obd_vars = lprocfs_lmv_obd_vars;
1383 lprocfs_obd_setup(obd);
1384 lprocfs_alloc_md_stats(obd, 0);
1385 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1386 0444, &lmv_proc_target_fops, obd);
1388 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1391 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1392 LUSTRE_CLI_FLD_HASH_DHT);
1394 CERROR("Can't init FLD, err %d\n", rc);
1404 static int lmv_cleanup(struct obd_device *obd)
1406 struct lmv_obd *lmv = &obd->u.lmv;
1409 fld_client_fini(&lmv->lmv_fld);
1410 if (lmv->tgts != NULL) {
1412 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1413 if (lmv->tgts[i] == NULL)
1415 lmv_del_target(lmv, i);
1417 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1423 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1425 struct lustre_cfg *lcfg = buf;
1426 struct obd_uuid obd_uuid;
1432 switch (lcfg->lcfg_command) {
1434 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1435 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1436 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1437 GOTO(out, rc = -EINVAL);
1439 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1441 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1442 GOTO(out, rc = -EINVAL);
1443 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1444 GOTO(out, rc = -EINVAL);
1445 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1448 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1449 GOTO(out, rc = -EINVAL);
1455 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1456 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1458 struct obd_device *obd = class_exp2obd(exp);
1459 struct lmv_obd *lmv = &obd->u.lmv;
1460 struct obd_statfs *temp;
1465 rc = lmv_check_connect(obd);
1469 OBD_ALLOC(temp, sizeof(*temp));
1473 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1474 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1477 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1480 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1481 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1483 GOTO(out_free_temp, rc);
1488 /* If the statfs is from mount, it will needs
1489 * retrieve necessary information from MDT0.
1490 * i.e. mount does not need the merged osfs
1492 * And also clients can be mounted as long as
1493 * MDT0 is in service*/
1494 if (flags & OBD_STATFS_FOR_MDT0)
1495 GOTO(out_free_temp, rc);
1497 osfs->os_bavail += temp->os_bavail;
1498 osfs->os_blocks += temp->os_blocks;
1499 osfs->os_ffree += temp->os_ffree;
1500 osfs->os_files += temp->os_files;
1506 OBD_FREE(temp, sizeof(*temp));
1510 static int lmv_getstatus(struct obd_export *exp,
1512 struct obd_capa **pc)
1514 struct obd_device *obd = exp->exp_obd;
1515 struct lmv_obd *lmv = &obd->u.lmv;
1519 rc = lmv_check_connect(obd);
1523 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1527 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1528 struct obd_capa *oc, u64 valid, const char *name,
1529 const char *input, int input_size, int output_size,
1530 int flags, struct ptlrpc_request **request)
1532 struct obd_device *obd = exp->exp_obd;
1533 struct lmv_obd *lmv = &obd->u.lmv;
1534 struct lmv_tgt_desc *tgt;
1538 rc = lmv_check_connect(obd);
1542 tgt = lmv_find_target(lmv, fid);
1544 RETURN(PTR_ERR(tgt));
1546 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1547 input_size, output_size, flags, request);
1552 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1553 struct obd_capa *oc, u64 valid, const char *name,
1554 const char *input, int input_size, int output_size,
1555 int flags, __u32 suppgid,
1556 struct ptlrpc_request **request)
1558 struct obd_device *obd = exp->exp_obd;
1559 struct lmv_obd *lmv = &obd->u.lmv;
1560 struct lmv_tgt_desc *tgt;
1564 rc = lmv_check_connect(obd);
1568 tgt = lmv_find_target(lmv, fid);
1570 RETURN(PTR_ERR(tgt));
1572 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1573 input_size, output_size, flags, suppgid,
1579 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1580 struct ptlrpc_request **request)
1582 struct obd_device *obd = exp->exp_obd;
1583 struct lmv_obd *lmv = &obd->u.lmv;
1584 struct lmv_tgt_desc *tgt;
1588 rc = lmv_check_connect(obd);
1592 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1594 RETURN(PTR_ERR(tgt));
1596 if (op_data->op_flags & MF_GET_MDT_IDX) {
1597 op_data->op_mds = tgt->ltd_idx;
1601 rc = md_getattr(tgt->ltd_exp, op_data, request);
1606 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1608 struct obd_device *obd = exp->exp_obd;
1609 struct lmv_obd *lmv = &obd->u.lmv;
1614 rc = lmv_check_connect(obd);
1618 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1621 * With DNE every object can have two locks in different namespaces:
1622 * lookup lock in space of MDT storing direntry and update/open lock in
1623 * space of MDT storing inode.
1625 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1626 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1628 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1634 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1635 ldlm_iterator_t it, void *data)
1637 struct obd_device *obd = exp->exp_obd;
1638 struct lmv_obd *lmv = &obd->u.lmv;
1644 rc = lmv_check_connect(obd);
1648 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1651 * With DNE every object can have two locks in different namespaces:
1652 * lookup lock in space of MDT storing direntry and update/open lock in
1653 * space of MDT storing inode. Try the MDT that the FID maps to first,
1654 * since this can be easily found, and only try others if that fails.
1656 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1657 i < lmv->desc.ld_tgt_count;
1658 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1660 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1661 obd->obd_name, PFID(fid), tgt);
1665 if (lmv->tgts[tgt] == NULL ||
1666 lmv->tgts[tgt]->ltd_exp == NULL)
1669 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1678 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1679 struct md_open_data *mod, struct ptlrpc_request **request)
1681 struct obd_device *obd = exp->exp_obd;
1682 struct lmv_obd *lmv = &obd->u.lmv;
1683 struct lmv_tgt_desc *tgt;
1687 rc = lmv_check_connect(obd);
1691 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1693 RETURN(PTR_ERR(tgt));
1695 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1696 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1701 * Choosing the MDT by name or FID in @op_data.
1702 * For non-striped directory, it will locate MDT by fid.
1703 * For striped-directory, it will locate MDT by name. And also
1704 * it will reset op_fid1 with the FID of the choosen stripe.
1706 struct lmv_tgt_desc *
1707 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1708 const char *name, int namelen, struct lu_fid *fid,
1711 struct lmv_tgt_desc *tgt;
1712 const struct lmv_oinfo *oinfo;
1714 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1715 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1716 RETURN(ERR_PTR(-EBADF));
1717 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1719 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1721 RETURN(ERR_CAST(oinfo));
1725 *fid = oinfo->lmo_fid;
1727 *mds = oinfo->lmo_mds;
1729 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1731 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1732 PFID(&oinfo->lmo_fid));
1737 * Locate mds by fid or name
1739 * For striped directory (lsm != NULL), it will locate the stripe
1740 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1741 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1742 * walk through all of stripes to locate the entry.
1744 * For normal direcotry, it will locate MDS by FID directly.
1745 * \param[in] lmv LMV device
1746 * \param[in] op_data client MD stack parameters, name, namelen
1748 * \param[in] fid object FID used to locate MDS.
1750 * retval pointer to the lmv_tgt_desc if succeed.
1751 * ERR_PTR(errno) if failed.
1753 struct lmv_tgt_desc*
1754 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1757 struct lmv_stripe_md *lsm = op_data->op_mea1;
1758 struct lmv_tgt_desc *tgt;
1760 /* During creating VOLATILE file, it should honor the mdt
1761 * index if the file under striped dir is being restored, see
1763 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1764 (int)op_data->op_mds != -1 && lsm != NULL) {
1766 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1770 /* refill the right parent fid */
1771 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1772 struct lmv_oinfo *oinfo;
1774 oinfo = &lsm->lsm_md_oinfo[i];
1775 if (oinfo->lmo_mds == op_data->op_mds) {
1776 *fid = oinfo->lmo_fid;
1781 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1782 if (i == lsm->lsm_md_stripe_count)
1783 tgt = ERR_PTR(-EINVAL);
1788 if (lsm == NULL || op_data->op_namelen == 0) {
1789 tgt = lmv_find_target(lmv, fid);
1793 op_data->op_mds = tgt->ltd_idx;
1797 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1798 op_data->op_namelen, fid,
1802 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1803 const void *data, size_t datalen, umode_t mode, uid_t uid,
1804 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1805 struct ptlrpc_request **request)
1807 struct obd_device *obd = exp->exp_obd;
1808 struct lmv_obd *lmv = &obd->u.lmv;
1809 struct lmv_tgt_desc *tgt;
1813 rc = lmv_check_connect(obd);
1817 if (!lmv->desc.ld_active_tgt_count)
1820 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1822 RETURN(PTR_ERR(tgt));
1824 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1825 (int)op_data->op_namelen, op_data->op_name,
1826 PFID(&op_data->op_fid1), op_data->op_mds);
1828 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1831 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1832 /* Send the create request to the MDT where the object
1833 * will be located */
1834 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1836 RETURN(PTR_ERR(tgt));
1838 op_data->op_mds = tgt->ltd_idx;
1840 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1843 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1844 PFID(&op_data->op_fid2), op_data->op_mds);
1846 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1847 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1848 cap_effective, rdev, request);
1850 if (*request == NULL)
1852 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1858 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1859 const union ldlm_policy_data *policy,
1860 struct lookup_intent *it, struct md_op_data *op_data,
1861 struct lustre_handle *lockh, __u64 extra_lock_flags)
1863 struct obd_device *obd = exp->exp_obd;
1864 struct lmv_obd *lmv = &obd->u.lmv;
1865 struct lmv_tgt_desc *tgt;
1869 rc = lmv_check_connect(obd);
1873 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1874 LL_IT2STR(it), PFID(&op_data->op_fid1));
1876 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1878 RETURN(PTR_ERR(tgt));
1880 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1881 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1883 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1890 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1891 struct ptlrpc_request **preq)
1893 struct ptlrpc_request *req = NULL;
1894 struct obd_device *obd = exp->exp_obd;
1895 struct lmv_obd *lmv = &obd->u.lmv;
1896 struct lmv_tgt_desc *tgt;
1897 struct mdt_body *body;
1901 rc = lmv_check_connect(obd);
1905 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1907 RETURN(PTR_ERR(tgt));
1909 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1910 (int)op_data->op_namelen, op_data->op_name,
1911 PFID(&op_data->op_fid1), tgt->ltd_idx);
1913 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1917 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1918 LASSERT(body != NULL);
1920 if (body->mbo_valid & OBD_MD_MDS) {
1921 struct lu_fid rid = body->mbo_fid1;
1922 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1925 tgt = lmv_find_target(lmv, &rid);
1927 ptlrpc_req_finished(*preq);
1929 RETURN(PTR_ERR(tgt));
1932 op_data->op_fid1 = rid;
1933 op_data->op_valid |= OBD_MD_FLCROSSREF;
1934 op_data->op_namelen = 0;
1935 op_data->op_name = NULL;
1936 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1937 ptlrpc_req_finished(*preq);
1944 #define md_op_data_fid(op_data, fl) \
1945 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1946 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1947 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1948 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1951 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1952 struct md_op_data *op_data,
1953 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1955 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1956 struct obd_device *obd = exp->exp_obd;
1957 struct lmv_obd *lmv = &obd->u.lmv;
1958 ldlm_policy_data_t policy = {{ 0 }};
1962 if (!fid_is_sane(fid))
1966 tgt = lmv_find_target(lmv, fid);
1968 RETURN(PTR_ERR(tgt));
1971 if (tgt->ltd_idx != op_tgt) {
1972 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1973 policy.l_inodebits.bits = bits;
1974 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1975 mode, LCF_ASYNC, NULL);
1978 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1980 op_data->op_flags |= flag;
1988 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1991 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1992 struct ptlrpc_request **request)
1994 struct obd_device *obd = exp->exp_obd;
1995 struct lmv_obd *lmv = &obd->u.lmv;
1996 struct lmv_tgt_desc *tgt;
2000 rc = lmv_check_connect(obd);
2004 LASSERT(op_data->op_namelen != 0);
2006 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2007 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2008 op_data->op_name, PFID(&op_data->op_fid1));
2010 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2011 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2012 op_data->op_cap = cfs_curproc_cap_pack();
2013 if (op_data->op_mea2 != NULL) {
2014 struct lmv_stripe_md *lsm = op_data->op_mea2;
2015 const struct lmv_oinfo *oinfo;
2017 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2018 op_data->op_namelen);
2020 RETURN(PTR_ERR(oinfo));
2022 op_data->op_fid2 = oinfo->lmo_fid;
2025 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2027 RETURN(PTR_ERR(tgt));
2030 * Cancel UPDATE lock on child (fid1).
2032 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2033 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2034 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2038 rc = md_link(tgt->ltd_exp, op_data, request);
2043 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2044 const char *old, size_t oldlen,
2045 const char *new, size_t newlen,
2046 struct ptlrpc_request **request)
2048 struct obd_device *obd = exp->exp_obd;
2049 struct lmv_obd *lmv = &obd->u.lmv;
2050 struct lmv_tgt_desc *src_tgt;
2054 LASSERT(oldlen != 0);
2056 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2057 (int)oldlen, old, PFID(&op_data->op_fid1),
2058 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2059 (int)newlen, new, PFID(&op_data->op_fid2),
2060 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2062 rc = lmv_check_connect(obd);
2066 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2067 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2068 op_data->op_cap = cfs_curproc_cap_pack();
2069 if (op_data->op_cli_flags & CLI_MIGRATE) {
2070 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2071 PFID(&op_data->op_fid3));
2073 if (op_data->op_mea1 != NULL) {
2074 struct lmv_stripe_md *lsm = op_data->op_mea1;
2075 struct lmv_tgt_desc *tmp;
2077 /* Fix the parent fid for striped dir */
2078 tmp = lmv_locate_target_for_name(lmv, lsm, old,
2083 RETURN(PTR_ERR(tmp));
2086 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2090 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2092 if (op_data->op_mea1 != NULL) {
2093 struct lmv_stripe_md *lsm = op_data->op_mea1;
2095 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2099 if (IS_ERR(src_tgt))
2100 RETURN(PTR_ERR(src_tgt));
2102 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2103 if (IS_ERR(src_tgt))
2104 RETURN(PTR_ERR(src_tgt));
2106 op_data->op_mds = src_tgt->ltd_idx;
2109 if (op_data->op_mea2) {
2110 struct lmv_stripe_md *lsm = op_data->op_mea2;
2111 const struct lmv_oinfo *oinfo;
2113 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2115 RETURN(PTR_ERR(oinfo));
2117 op_data->op_fid2 = oinfo->lmo_fid;
2120 if (IS_ERR(src_tgt))
2121 RETURN(PTR_ERR(src_tgt));
2124 * LOOKUP lock on src child (fid3) should also be cancelled for
2125 * src_tgt in mdc_rename.
2127 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2130 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2133 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2134 LCK_EX, MDS_INODELOCK_UPDATE,
2135 MF_MDC_CANCEL_FID2);
2140 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2142 if (fid_is_sane(&op_data->op_fid3)) {
2143 struct lmv_tgt_desc *tgt;
2145 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2147 RETURN(PTR_ERR(tgt));
2149 /* Cancel LOOKUP lock on its parent */
2150 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2151 LCK_EX, MDS_INODELOCK_LOOKUP,
2152 MF_MDC_CANCEL_FID3);
2156 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2157 LCK_EX, MDS_INODELOCK_FULL,
2158 MF_MDC_CANCEL_FID3);
2164 * Cancel all the locks on tgt child (fid4).
2166 if (fid_is_sane(&op_data->op_fid4))
2167 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2168 LCK_EX, MDS_INODELOCK_FULL,
2169 MF_MDC_CANCEL_FID4);
2171 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2172 op_data->op_mds, PFID(&op_data->op_fid2));
2174 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2180 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2181 void *ea, size_t ealen, struct ptlrpc_request **request)
2183 struct obd_device *obd = exp->exp_obd;
2184 struct lmv_obd *lmv = &obd->u.lmv;
2185 struct lmv_tgt_desc *tgt;
2189 rc = lmv_check_connect(obd);
2193 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2194 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2196 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2197 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2199 RETURN(PTR_ERR(tgt));
2201 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2206 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2207 struct obd_capa *oc, struct ptlrpc_request **request)
2209 struct obd_device *obd = exp->exp_obd;
2210 struct lmv_obd *lmv = &obd->u.lmv;
2211 struct lmv_tgt_desc *tgt;
2215 rc = lmv_check_connect(obd);
2219 tgt = lmv_find_target(lmv, fid);
2221 RETURN(PTR_ERR(tgt));
2223 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2228 * Get current minimum entry from striped directory
2230 * This function will search the dir entry, whose hash value is the
2231 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2232 * only being called for striped directory.
2234 * \param[in] exp export of LMV
2235 * \param[in] op_data parameters transferred beween client MD stack
2236 * stripe_information will be included in this
2238 * \param[in] cb_op ldlm callback being used in enqueue in
2240 * \param[in] hash_offset the hash value, which is used to locate
2241 * minum(closet) dir entry
2242 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2243 * index of last entry, so to avoid hash conflict
2244 * between stripes. It will also be used to
2245 * return the stripe index of current dir entry.
2246 * \param[in|out] entp the minum entry and it also is being used
2247 * to input the last dir entry to resolve the
2250 * \param[out] ppage the page which holds the minum entry
2252 * \retval = 0 get the entry successfully
2253 * negative errno (< 0) does not get the entry
2255 static int lmv_get_min_striped_entry(struct obd_export *exp,
2256 struct md_op_data *op_data,
2257 struct md_callback *cb_op,
2258 __u64 hash_offset, int *stripe_offset,
2259 struct lu_dirent **entp,
2260 struct page **ppage)
2262 struct obd_device *obd = exp->exp_obd;
2263 struct lmv_obd *lmv = &obd->u.lmv;
2264 struct lmv_stripe_md *lsm = op_data->op_mea1;
2265 struct lmv_tgt_desc *tgt;
2267 struct lu_dirent *min_ent = NULL;
2268 struct page *min_page = NULL;
2274 stripe_count = lsm->lsm_md_stripe_count;
2275 for (i = 0; i < stripe_count; i++) {
2276 struct lu_dirent *ent = NULL;
2277 struct page *page = NULL;
2278 struct lu_dirpage *dp;
2279 __u64 stripe_hash = hash_offset;
2281 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2283 GOTO(out, rc = PTR_ERR(tgt));
2285 /* op_data will be shared by each stripe, so we need
2286 * reset these value for each stripe */
2287 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2288 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2289 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2291 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2296 dp = page_address(page);
2297 for (ent = lu_dirent_start(dp); ent != NULL;
2298 ent = lu_dirent_next(ent)) {
2299 /* Skip dummy entry */
2300 if (le16_to_cpu(ent->lde_namelen) == 0)
2303 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2306 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2307 (*entp == ent || i < *stripe_offset))
2310 /* skip . and .. for other stripes */
2312 (strncmp(ent->lde_name, ".",
2313 le16_to_cpu(ent->lde_namelen)) == 0 ||
2314 strncmp(ent->lde_name, "..",
2315 le16_to_cpu(ent->lde_namelen)) == 0))
2321 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2324 page_cache_release(page);
2327 /* reach the end of current stripe, go to next stripe */
2328 if (stripe_hash == MDS_DIR_END_OFF)
2334 if (min_ent != NULL) {
2335 if (le64_to_cpu(min_ent->lde_hash) >
2336 le64_to_cpu(ent->lde_hash)) {
2339 page_cache_release(min_page);
2344 page_cache_release(page);
2355 if (*ppage != NULL) {
2357 page_cache_release(*ppage);
2359 *stripe_offset = min_idx;
2366 * Build dir entry page from a striped directory
2368 * This function gets one entry by @offset from a striped directory. It will
2369 * read entries from all of stripes, and choose one closest to the required
2370 * offset(&offset). A few notes
2371 * 1. skip . and .. for non-zero stripes, because there can only have one .
2372 * and .. in a directory.
2373 * 2. op_data will be shared by all of stripes, instead of allocating new
2374 * one, so need to restore before reusing.
2375 * 3. release the entry page if that is not being chosen.
2377 * \param[in] exp obd export refer to LMV
2378 * \param[in] op_data hold those MD parameters of read_entry
2379 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2380 * \param[out] ldp the entry being read
2381 * \param[out] ppage the page holding the entry. Note: because the entry
2382 * will be accessed in upper layer, so we need hold the
2383 * page until the usages of entry is finished, see
2384 * ll_dir_entry_next.
2386 * retval =0 if get entry successfully
2387 * <0 cannot get entry
2389 static int lmv_read_striped_page(struct obd_export *exp,
2390 struct md_op_data *op_data,
2391 struct md_callback *cb_op,
2392 __u64 offset, struct page **ppage)
2394 struct obd_device *obd = exp->exp_obd;
2395 struct lu_fid master_fid = op_data->op_fid1;
2396 struct inode *master_inode = op_data->op_data;
2397 __u64 hash_offset = offset;
2398 struct lu_dirpage *dp;
2399 struct page *min_ent_page = NULL;
2400 struct page *ent_page = NULL;
2401 struct lu_dirent *ent;
2404 struct lu_dirent *min_ent = NULL;
2405 struct lu_dirent *last_ent;
2410 rc = lmv_check_connect(obd);
2414 /* Allocate a page and read entries from all of stripes and fill
2415 * the page by hash order */
2416 ent_page = alloc_page(GFP_KERNEL);
2417 if (ent_page == NULL)
2420 /* Initialize the entry page */
2421 dp = kmap(ent_page);
2422 memset(dp, 0, sizeof(*dp));
2423 dp->ldp_hash_start = cpu_to_le64(offset);
2424 dp->ldp_flags |= LDF_COLLIDE;
2427 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2433 /* Find the minum entry from all sub-stripes */
2434 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2440 /* If it can not get minum entry, it means it already reaches
2441 * the end of this directory */
2442 if (min_ent == NULL) {
2443 last_ent->lde_reclen = 0;
2444 hash_offset = MDS_DIR_END_OFF;
2448 ent_size = le16_to_cpu(min_ent->lde_reclen);
2450 /* the last entry lde_reclen is 0, but it might not
2451 * the end of this entry of this temporay entry */
2453 ent_size = lu_dirent_calc_size(
2454 le16_to_cpu(min_ent->lde_namelen),
2455 le32_to_cpu(min_ent->lde_attrs));
2456 if (ent_size > left_bytes) {
2457 last_ent->lde_reclen = cpu_to_le16(0);
2458 hash_offset = le64_to_cpu(min_ent->lde_hash);
2462 memcpy(ent, min_ent, ent_size);
2464 /* Replace . with master FID and Replace .. with the parent FID
2465 * of master object */
2466 if (strncmp(ent->lde_name, ".",
2467 le16_to_cpu(ent->lde_namelen)) == 0 &&
2468 le16_to_cpu(ent->lde_namelen) == 1)
2469 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2470 else if (strncmp(ent->lde_name, "..",
2471 le16_to_cpu(ent->lde_namelen)) == 0 &&
2472 le16_to_cpu(ent->lde_namelen) == 2)
2473 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2475 left_bytes -= ent_size;
2476 ent->lde_reclen = cpu_to_le16(ent_size);
2478 ent = (void *)ent + ent_size;
2479 hash_offset = le64_to_cpu(min_ent->lde_hash);
2480 if (hash_offset == MDS_DIR_END_OFF) {
2481 last_ent->lde_reclen = 0;
2486 if (min_ent_page != NULL) {
2487 kunmap(min_ent_page);
2488 page_cache_release(min_ent_page);
2491 if (unlikely(rc != 0)) {
2492 __free_page(ent_page);
2496 dp->ldp_flags |= LDF_EMPTY;
2497 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2498 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2501 /* We do not want to allocate md_op_data during each
2502 * dir entry reading, so op_data will be shared by every stripe,
2503 * then we need to restore it back to original value before
2504 * return to the upper layer */
2505 op_data->op_fid1 = master_fid;
2506 op_data->op_fid2 = master_fid;
2507 op_data->op_data = master_inode;
2514 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2515 struct md_callback *cb_op, __u64 offset,
2516 struct page **ppage)
2518 struct obd_device *obd = exp->exp_obd;
2519 struct lmv_obd *lmv = &obd->u.lmv;
2520 struct lmv_stripe_md *lsm = op_data->op_mea1;
2521 struct lmv_tgt_desc *tgt;
2525 rc = lmv_check_connect(obd);
2529 if (unlikely(lsm != NULL)) {
2530 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2534 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2536 RETURN(PTR_ERR(tgt));
2538 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2544 * Unlink a file/directory
2546 * Unlink a file or directory under the parent dir. The unlink request
2547 * usually will be sent to the MDT where the child is located, but if
2548 * the client does not have the child FID then request will be sent to the
2549 * MDT where the parent is located.
2551 * If the parent is a striped directory then it also needs to locate which
2552 * stripe the name of the child is located, and replace the parent FID
2553 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2554 * it will walk through all of sub-stripes until the child is being
2557 * \param[in] exp export refer to LMV
2558 * \param[in] op_data different parameters transferred beween client
2559 * MD stacks, name, namelen, FIDs etc.
2560 * op_fid1 is the parent FID, op_fid2 is the child
2562 * \param[out] request point to the request of unlink.
2564 * retval 0 if succeed
2565 * negative errno if failed.
2567 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2568 struct ptlrpc_request **request)
2570 struct obd_device *obd = exp->exp_obd;
2571 struct lmv_obd *lmv = &obd->u.lmv;
2572 struct lmv_tgt_desc *tgt = NULL;
2573 struct lmv_tgt_desc *parent_tgt = NULL;
2574 struct mdt_body *body;
2576 int stripe_index = 0;
2577 struct lmv_stripe_md *lsm = op_data->op_mea1;
2580 rc = lmv_check_connect(obd);
2584 /* For striped dir, we need to locate the parent as well */
2586 struct lmv_tgt_desc *tmp;
2588 LASSERT(op_data->op_name != NULL &&
2589 op_data->op_namelen != 0);
2591 tmp = lmv_locate_target_for_name(lmv, lsm,
2593 op_data->op_namelen,
2597 /* return -EBADFD means unknown hash type, might
2598 * need try all sub-stripe here */
2599 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2600 RETURN(PTR_ERR(tmp));
2602 /* Note: both migrating dir and unknown hash dir need to
2603 * try all of sub-stripes, so we need start search the
2604 * name from stripe 0, but migrating dir is already handled
2605 * inside lmv_locate_target_for_name(), so we only check
2606 * unknown hash type directory here */
2607 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2608 struct lmv_oinfo *oinfo;
2610 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2612 op_data->op_fid1 = oinfo->lmo_fid;
2613 op_data->op_mds = oinfo->lmo_mds;
2618 /* Send unlink requests to the MDT where the child is located */
2619 if (likely(!fid_is_zero(&op_data->op_fid2)))
2620 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2621 else if (lsm != NULL)
2622 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2624 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2627 RETURN(PTR_ERR(tgt));
2629 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2630 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2631 op_data->op_cap = cfs_curproc_cap_pack();
2634 * If child's fid is given, cancel unused locks for it if it is from
2635 * another export than parent.
2637 * LOOKUP lock for child (fid3) should also be cancelled on parent
2638 * tgt_tgt in mdc_unlink().
2640 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2643 * Cancel FULL locks on child (fid3).
2645 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2646 if (IS_ERR(parent_tgt))
2647 RETURN(PTR_ERR(parent_tgt));
2649 if (parent_tgt != tgt) {
2650 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2651 LCK_EX, MDS_INODELOCK_LOOKUP,
2652 MF_MDC_CANCEL_FID3);
2655 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2656 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2660 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2661 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2663 rc = md_unlink(tgt->ltd_exp, op_data, request);
2664 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2667 /* Try next stripe if it is needed. */
2668 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2669 struct lmv_oinfo *oinfo;
2672 if (stripe_index >= lsm->lsm_md_stripe_count)
2675 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2677 op_data->op_fid1 = oinfo->lmo_fid;
2678 op_data->op_mds = oinfo->lmo_mds;
2680 ptlrpc_req_finished(*request);
2683 goto try_next_stripe;
2686 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2690 /* Not cross-ref case, just get out of here. */
2691 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2694 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2695 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2697 /* This is a remote object, try remote MDT, Note: it may
2698 * try more than 1 time here, Considering following case
2699 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2700 * 1. Initially A does not know where remote1 is, it send
2701 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2702 * resend unlink RPC to MDT1 (retry 1st time).
2704 * 2. During the unlink RPC in flight,
2705 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2706 * and create new remote1, but on MDT0
2708 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2709 * /mnt/lustre, then lookup get fid of remote1, and find
2710 * it is remote dir again, and replay -EREMOTE again.
2712 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2714 * In theory, it might try unlimited time here, but it should
2715 * be very rare case. */
2716 op_data->op_fid2 = body->mbo_fid1;
2717 ptlrpc_req_finished(*request);
2723 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2725 struct lmv_obd *lmv = &obd->u.lmv;
2729 case OBD_CLEANUP_EARLY:
2730 /* XXX: here should be calling obd_precleanup() down to
2733 case OBD_CLEANUP_EXPORTS:
2734 fld_client_proc_fini(&lmv->lmv_fld);
2735 lprocfs_obd_cleanup(obd);
2736 lprocfs_free_md_stats(obd);
2745 * Get by key a value associated with a LMV device.
2747 * Dispatch request to lower-layer devices as needed.
2749 * \param[in] env execution environment for this thread
2750 * \param[in] exp export for the LMV device
2751 * \param[in] keylen length of key identifier
2752 * \param[in] key identifier of key to get value for
2753 * \param[in] vallen size of \a val
2754 * \param[out] val pointer to storage location for value
2755 * \param[in] lsm optional striping metadata of object
2757 * \retval 0 on success
2758 * \retval negative negated errno on failure
2760 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2761 __u32 keylen, void *key, __u32 *vallen, void *val,
2762 struct lov_stripe_md *lsm)
2764 struct obd_device *obd;
2765 struct lmv_obd *lmv;
2769 obd = class_exp2obd(exp);
2771 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2772 exp->exp_handle.h_cookie);
2777 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2780 rc = lmv_check_connect(obd);
2784 LASSERT(*vallen == sizeof(__u32));
2785 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2786 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2788 * All tgts should be connected when this gets called.
2790 if (tgt == NULL || tgt->ltd_exp == NULL)
2793 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2798 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2799 KEY_IS(KEY_DEFAULT_EASIZE) ||
2800 KEY_IS(KEY_MAX_COOKIESIZE) ||
2801 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2802 KEY_IS(KEY_CONN_DATA)) {
2803 rc = lmv_check_connect(obd);
2808 * Forwarding this request to first MDS, it should know LOV
2811 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2813 if (!rc && KEY_IS(KEY_CONN_DATA))
2814 exp->exp_connect_data = *(struct obd_connect_data *)val;
2816 } else if (KEY_IS(KEY_TGT_COUNT)) {
2817 *((int *)val) = lmv->desc.ld_tgt_count;
2821 CDEBUG(D_IOCTL, "Invalid key\n");
2826 * Asynchronously set by key a value associated with a LMV device.
2828 * Dispatch request to lower-layer devices as needed.
2830 * \param[in] env execution environment for this thread
2831 * \param[in] exp export for the LMV device
2832 * \param[in] keylen length of key identifier
2833 * \param[in] key identifier of key to store value for
2834 * \param[in] vallen size of value to store
2835 * \param[in] val pointer to data to be stored
2836 * \param[in] set optional list of related ptlrpc requests
2838 * \retval 0 on success
2839 * \retval negative negated errno on failure
2841 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2842 __u32 keylen, void *key, __u32 vallen, void *val,
2843 struct ptlrpc_request_set *set)
2845 struct lmv_tgt_desc *tgt = NULL;
2846 struct obd_device *obd;
2847 struct lmv_obd *lmv;
2851 obd = class_exp2obd(exp);
2853 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2854 exp->exp_handle.h_cookie);
2859 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2860 KEY_IS(KEY_DEFAULT_EASIZE)) {
2863 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2866 if (tgt == NULL || tgt->ltd_exp == NULL)
2869 err = obd_set_info_async(env, tgt->ltd_exp,
2870 keylen, key, vallen, val, set);
2881 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2882 struct lmv_mds_md_v1 *lmm1)
2887 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2888 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2889 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2890 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2891 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2892 sizeof(lmm1->lmv_pool_name));
2893 if (cplen >= sizeof(lmm1->lmv_pool_name))
2896 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2897 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2898 &lsm->lsm_md_oinfo[i].lmo_fid);
2902 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2906 bool allocated = false;
2910 LASSERT(lmmp != NULL);
2912 if (*lmmp != NULL && lsm == NULL) {
2915 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2916 lmm_size = lmv_mds_md_size(stripe_count,
2917 le32_to_cpu((*lmmp)->lmv_magic));
2920 OBD_FREE(*lmmp, lmm_size);
2926 if (*lmmp == NULL && lsm == NULL) {
2927 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2928 LASSERT(lmm_size > 0);
2929 OBD_ALLOC(*lmmp, lmm_size);
2932 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2933 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2938 LASSERT(lsm != NULL);
2939 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2940 if (*lmmp == NULL) {
2941 OBD_ALLOC(*lmmp, lmm_size);
2947 switch (lsm->lsm_md_magic) {
2949 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2956 if (rc != 0 && allocated) {
2957 OBD_FREE(*lmmp, lmm_size);
2964 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2965 const struct lmv_mds_md_v1 *lmm1)
2967 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2974 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2975 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2976 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2977 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2978 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2980 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2981 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2982 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2983 sizeof(lsm->lsm_md_pool_name));
2985 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2988 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2989 "layout_version %d\n", lsm->lsm_md_stripe_count,
2990 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2991 lsm->lsm_md_layout_version);
2993 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2994 for (i = 0; i < stripe_count; i++) {
2995 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2996 &lmm1->lmv_stripe_fids[i]);
2997 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2998 &lsm->lsm_md_oinfo[i].lmo_mds);
3001 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3002 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3008 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3009 const union lmv_mds_md *lmm, int stripe_count)
3011 struct lmv_stripe_md *lsm;
3014 bool allocated = false;
3017 LASSERT(lsmp != NULL);
3021 if (lsm != NULL && lmm == NULL) {
3023 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3024 /* For migrating inode, the master stripe and master
3025 * object will be the same, so do not need iput, see
3026 * ll_update_lsm_md */
3027 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3028 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3029 iput(lsm->lsm_md_oinfo[i].lmo_root);
3031 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3032 OBD_FREE(lsm, lsm_size);
3038 if (lsm == NULL && lmm == NULL) {
3039 lsm_size = lmv_stripe_md_size(stripe_count);
3040 OBD_ALLOC(lsm, lsm_size);
3043 lsm->lsm_md_stripe_count = stripe_count;
3048 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3052 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3053 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3054 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3055 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3060 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3061 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3064 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3065 * stripecount should be 0 then.
3067 lsm_size = lmv_stripe_md_size(0);
3069 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3071 OBD_ALLOC(lsm, lsm_size);
3078 switch (le32_to_cpu(lmm->lmv_magic)) {
3080 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3083 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3084 le32_to_cpu(lmm->lmv_magic));
3089 if (rc != 0 && allocated) {
3090 OBD_FREE(lsm, lsm_size);
3097 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3099 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3102 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3104 lmv_unpack_md(NULL, &lsm, NULL, 0);
3106 EXPORT_SYMBOL(lmv_free_memmd);
3108 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3109 struct lov_mds_md *lmm, int disk_len)
3111 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3112 (union lmv_mds_md *)lmm, disk_len);
3115 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3116 struct lov_stripe_md *lsm)
3118 struct obd_device *obd = exp->exp_obd;
3119 struct lmv_obd *lmv_obd = &obd->u.lmv;
3120 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3125 stripe_count = lmv->lsm_md_stripe_count;
3127 stripe_count = lmv_obd->desc.ld_tgt_count;
3129 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3132 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3135 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3136 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3137 ldlm_cancel_flags_t flags, void *opaque)
3139 struct obd_device *obd = exp->exp_obd;
3140 struct lmv_obd *lmv = &obd->u.lmv;
3146 LASSERT(fid != NULL);
3148 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3149 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3151 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3154 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3162 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3165 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3166 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3170 if (tgt == NULL || tgt->ltd_exp == NULL)
3172 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3176 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3177 const struct lu_fid *fid, ldlm_type_t type,
3178 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3179 struct lustre_handle *lockh)
3181 struct obd_device *obd = exp->exp_obd;
3182 struct lmv_obd *lmv = &obd->u.lmv;
3188 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3191 * With DNE every object can have two locks in different namespaces:
3192 * lookup lock in space of MDT storing direntry and update/open lock in
3193 * space of MDT storing inode. Try the MDT that the FID maps to first,
3194 * since this can be easily found, and only try others if that fails.
3196 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3197 i < lmv->desc.ld_tgt_count;
3198 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3200 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3201 obd->obd_name, PFID(fid), tgt);
3205 if (lmv->tgts[tgt] == NULL ||
3206 lmv->tgts[tgt]->ltd_exp == NULL ||
3207 lmv->tgts[tgt]->ltd_active == 0)
3210 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3211 type, policy, mode, lockh);
3219 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3220 struct obd_export *dt_exp, struct obd_export *md_exp,
3221 struct lustre_md *md)
3223 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3224 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3226 if (tgt == NULL || tgt->ltd_exp == NULL)
3229 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3232 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3234 struct obd_device *obd = exp->exp_obd;
3235 struct lmv_obd *lmv = &obd->u.lmv;
3236 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3239 if (md->lmv != NULL) {
3240 lmv_free_memmd(md->lmv);
3243 if (tgt == NULL || tgt->ltd_exp == NULL)
3245 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3248 int lmv_set_open_replay_data(struct obd_export *exp,
3249 struct obd_client_handle *och,
3250 struct lookup_intent *it)
3252 struct obd_device *obd = exp->exp_obd;
3253 struct lmv_obd *lmv = &obd->u.lmv;
3254 struct lmv_tgt_desc *tgt;
3257 tgt = lmv_find_target(lmv, &och->och_fid);
3259 RETURN(PTR_ERR(tgt));
3261 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3264 int lmv_clear_open_replay_data(struct obd_export *exp,
3265 struct obd_client_handle *och)
3267 struct obd_device *obd = exp->exp_obd;
3268 struct lmv_obd *lmv = &obd->u.lmv;
3269 struct lmv_tgt_desc *tgt;
3272 tgt = lmv_find_target(lmv, &och->och_fid);
3274 RETURN(PTR_ERR(tgt));
3276 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3279 static int lmv_get_remote_perm(struct obd_export *exp,
3280 const struct lu_fid *fid,
3281 struct obd_capa *oc, __u32 suppgid,
3282 struct ptlrpc_request **request)
3284 struct obd_device *obd = exp->exp_obd;
3285 struct lmv_obd *lmv = &obd->u.lmv;
3286 struct lmv_tgt_desc *tgt;
3290 rc = lmv_check_connect(obd);
3294 tgt = lmv_find_target(lmv, fid);
3296 RETURN(PTR_ERR(tgt));
3298 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3302 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3305 struct obd_device *obd = exp->exp_obd;
3306 struct lmv_obd *lmv = &obd->u.lmv;
3307 struct lmv_tgt_desc *tgt;
3311 rc = lmv_check_connect(obd);
3315 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3317 RETURN(PTR_ERR(tgt));
3319 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3323 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3324 const struct req_msg_field *field, struct obd_capa **oc)
3326 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3327 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3329 if (tgt == NULL || tgt->ltd_exp == NULL)
3331 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3334 int lmv_intent_getattr_async(struct obd_export *exp,
3335 struct md_enqueue_info *minfo,
3336 struct ldlm_enqueue_info *einfo)
3338 struct md_op_data *op_data = &minfo->mi_data;
3339 struct obd_device *obd = exp->exp_obd;
3340 struct lmv_obd *lmv = &obd->u.lmv;
3341 struct lmv_tgt_desc *tgt = NULL;
3345 rc = lmv_check_connect(obd);
3349 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3351 RETURN(PTR_ERR(tgt));
3353 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3357 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3358 struct lu_fid *fid, __u64 *bits)
3360 struct obd_device *obd = exp->exp_obd;
3361 struct lmv_obd *lmv = &obd->u.lmv;
3362 struct lmv_tgt_desc *tgt;
3366 rc = lmv_check_connect(obd);
3370 tgt = lmv_find_target(lmv, fid);
3372 RETURN(PTR_ERR(tgt));
3374 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3378 int lmv_get_fid_from_lsm(struct obd_export *exp,
3379 const struct lmv_stripe_md *lsm,
3380 const char *name, int namelen, struct lu_fid *fid)
3382 const struct lmv_oinfo *oinfo;
3384 LASSERT(lsm != NULL);
3385 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3387 return PTR_ERR(oinfo);
3389 *fid = oinfo->lmo_fid;
3395 * For lmv, only need to send request to master MDT, and the master MDT will
3396 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3397 * we directly fetch data from the slave MDTs.
3399 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3400 struct obd_quotactl *oqctl)
3402 struct obd_device *obd = class_exp2obd(exp);
3403 struct lmv_obd *lmv = &obd->u.lmv;
3404 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3407 __u64 curspace, curinodes;
3411 tgt->ltd_exp == NULL ||
3413 lmv->desc.ld_tgt_count == 0) {
3414 CERROR("master lmv inactive\n");
3418 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3419 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3423 curspace = curinodes = 0;
3424 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3428 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3431 err = obd_quotactl(tgt->ltd_exp, oqctl);
3433 CERROR("getquota on mdt %d failed. %d\n", i, err);
3437 curspace += oqctl->qc_dqblk.dqb_curspace;
3438 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3441 oqctl->qc_dqblk.dqb_curspace = curspace;
3442 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3447 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3448 struct obd_quotactl *oqctl)
3450 struct obd_device *obd = class_exp2obd(exp);
3451 struct lmv_obd *lmv = &obd->u.lmv;
3452 struct lmv_tgt_desc *tgt;
3457 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3460 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3461 CERROR("lmv idx %d inactive\n", i);
3465 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3473 static int lmv_merge_attr(struct obd_export *exp,
3474 const struct lmv_stripe_md *lsm,
3475 struct cl_attr *attr,
3476 ldlm_blocking_callback cb_blocking)
3481 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3485 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3486 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3488 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3489 " atime %lu ctime %lu, mtime %lu.\n",
3490 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3491 i_size_read(inode), (unsigned long long)inode->i_blocks,
3492 inode->i_nlink, LTIME_S(inode->i_atime),
3493 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3495 /* for slave stripe, it needs to subtract nlink for . and .. */
3497 attr->cat_nlink += inode->i_nlink - 2;
3499 attr->cat_nlink = inode->i_nlink;
3501 attr->cat_size += i_size_read(inode);
3502 attr->cat_blocks += inode->i_blocks;
3504 if (attr->cat_atime < LTIME_S(inode->i_atime))
3505 attr->cat_atime = LTIME_S(inode->i_atime);
3507 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3508 attr->cat_ctime = LTIME_S(inode->i_ctime);
3510 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3511 attr->cat_mtime = LTIME_S(inode->i_mtime);
3516 struct obd_ops lmv_obd_ops = {
3517 .o_owner = THIS_MODULE,
3518 .o_setup = lmv_setup,
3519 .o_cleanup = lmv_cleanup,
3520 .o_precleanup = lmv_precleanup,
3521 .o_process_config = lmv_process_config,
3522 .o_connect = lmv_connect,
3523 .o_disconnect = lmv_disconnect,
3524 .o_statfs = lmv_statfs,
3525 .o_get_info = lmv_get_info,
3526 .o_set_info_async = lmv_set_info_async,
3527 .o_packmd = lmv_packmd,
3528 .o_unpackmd = lmv_unpackmd,
3529 .o_notify = lmv_notify,
3530 .o_get_uuid = lmv_get_uuid,
3531 .o_iocontrol = lmv_iocontrol,
3532 .o_quotacheck = lmv_quotacheck,
3533 .o_quotactl = lmv_quotactl
3536 struct md_ops lmv_md_ops = {
3537 .m_getstatus = lmv_getstatus,
3538 .m_null_inode = lmv_null_inode,
3539 .m_find_cbdata = lmv_find_cbdata,
3540 .m_close = lmv_close,
3541 .m_create = lmv_create,
3542 .m_enqueue = lmv_enqueue,
3543 .m_getattr = lmv_getattr,
3544 .m_getxattr = lmv_getxattr,
3545 .m_getattr_name = lmv_getattr_name,
3546 .m_intent_lock = lmv_intent_lock,
3548 .m_rename = lmv_rename,
3549 .m_setattr = lmv_setattr,
3550 .m_setxattr = lmv_setxattr,
3551 .m_fsync = lmv_fsync,
3552 .m_read_page = lmv_read_page,
3553 .m_unlink = lmv_unlink,
3554 .m_init_ea_size = lmv_init_ea_size,
3555 .m_cancel_unused = lmv_cancel_unused,
3556 .m_set_lock_data = lmv_set_lock_data,
3557 .m_lock_match = lmv_lock_match,
3558 .m_get_lustre_md = lmv_get_lustre_md,
3559 .m_free_lustre_md = lmv_free_lustre_md,
3560 .m_merge_attr = lmv_merge_attr,
3561 .m_set_open_replay_data = lmv_set_open_replay_data,
3562 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3563 .m_renew_capa = lmv_renew_capa,
3564 .m_unpack_capa = lmv_unpack_capa,
3565 .m_get_remote_perm = lmv_get_remote_perm,
3566 .m_intent_getattr_async = lmv_intent_getattr_async,
3567 .m_revalidate_lock = lmv_revalidate_lock,
3568 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3571 int __init lmv_init(void)
3573 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3574 LUSTRE_LMV_NAME, NULL);
3577 static void lmv_exit(void)
3579 class_unregister_type(LUSTRE_LMV_NAME);
3582 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3583 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3584 MODULE_LICENSE("GPL");
3586 module_init(lmv_init);
3587 module_exit(lmv_exit);