4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
39 #include <linux/slab.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
45 #include <linux/math64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
49 #include <liblustre.h>
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_lmv.h>
58 #include <lprocfs_status.h>
59 #include <cl_object.h>
61 #include <lustre_lite.h>
62 #include <lustre_fid.h>
63 #include <lustre_ioctl.h>
64 #include "lmv_internal.h"
66 /* This hash is only for testing purpose */
67 static inline unsigned int
68 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
71 const unsigned char *p = (const unsigned char *)name;
73 while (--namelen >= 0)
81 static inline unsigned int
82 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
86 hash = lustre_hash_fnv_1a_64(name, namelen);
93 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
94 const char *name, int namelen)
97 __u32 hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
100 if (stripe_count <= 1)
103 /* for migrating object, always start from 0 stripe */
104 if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
108 case LMV_HASH_TYPE_ALL_CHARS:
109 idx = lmv_hash_all_chars(stripe_count, name, namelen);
111 case LMV_HASH_TYPE_FNV_1A_64:
112 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
115 CERROR("Unknown hash type 0x%x\n", hash_type);
119 CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
125 static void lmv_activate_target(struct lmv_obd *lmv,
126 struct lmv_tgt_desc *tgt,
129 if (tgt->ltd_active == activate)
132 tgt->ltd_active = activate;
133 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
139 * -EINVAL : UUID can't be found in the LMV's target list
140 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
141 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
143 static int lmv_set_mdc_active(struct lmv_obd *lmv,
144 const struct obd_uuid *uuid,
147 struct lmv_tgt_desc *tgt = NULL;
148 struct obd_device *obd;
153 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
154 lmv, uuid->uuid, activate);
156 spin_lock(&lmv->lmv_lock);
157 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
159 if (tgt == NULL || tgt->ltd_exp == NULL)
162 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
163 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
165 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
169 if (i == lmv->desc.ld_tgt_count)
170 GOTO(out_lmv_lock, rc = -EINVAL);
172 obd = class_exp2obd(tgt->ltd_exp);
174 GOTO(out_lmv_lock, rc = -ENOTCONN);
176 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
177 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
178 obd->obd_type->typ_name, i);
179 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
181 if (tgt->ltd_active == activate) {
182 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
183 activate ? "" : "in");
184 GOTO(out_lmv_lock, rc);
187 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
188 activate ? "" : "in");
189 lmv_activate_target(lmv, tgt, activate);
193 spin_unlock(&lmv->lmv_lock);
197 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
199 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
200 struct lmv_tgt_desc *tgt = lmv->tgts[0];
202 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
205 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
206 enum obd_notify_event ev, void *data)
208 struct obd_connect_data *conn_data;
209 struct lmv_obd *lmv = &obd->u.lmv;
210 struct obd_uuid *uuid;
214 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
215 CERROR("unexpected notification of %s %s!\n",
216 watched->obd_type->typ_name,
221 uuid = &watched->u.cli.cl_target_uuid;
222 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
224 * Set MDC as active before notifying the observer, so the
225 * observer can use the MDC normally.
227 rc = lmv_set_mdc_active(lmv, uuid,
228 ev == OBD_NOTIFY_ACTIVE);
230 CERROR("%sactivation of %s failed: %d\n",
231 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
235 } else if (ev == OBD_NOTIFY_OCD) {
236 conn_data = &watched->u.cli.cl_import->imp_connect_data;
238 * XXX: Make sure that ocd_connect_flags from all targets are
239 * the same. Otherwise one of MDTs runs wrong version or
240 * something like this. --umka
242 obd->obd_self_export->exp_connect_data = *conn_data;
245 else if (ev == OBD_NOTIFY_DISCON) {
247 * For disconnect event, flush fld cache for failout MDS case.
249 fld_client_flush(&lmv->lmv_fld);
253 * Pass the notification up the chain.
255 if (obd->obd_observer)
256 rc = obd_notify(obd->obd_observer, watched, ev, data);
262 * This is fake connect function. Its purpose is to initialize lmv and say
263 * caller that everything is okay. Real connection will be performed later.
265 static int lmv_connect(const struct lu_env *env,
266 struct obd_export **exp, struct obd_device *obd,
267 struct obd_uuid *cluuid, struct obd_connect_data *data,
270 struct lmv_obd *lmv = &obd->u.lmv;
271 struct lustre_handle conn = { 0 };
276 * We don't want to actually do the underlying connections more than
277 * once, so keep track.
280 if (lmv->refcount > 1) {
285 rc = class_connect(&conn, obd, cluuid);
287 CERROR("class_connection() returned %d\n", rc);
291 *exp = class_conn2export(&conn);
292 class_export_get(*exp);
296 lmv->cluuid = *cluuid;
299 lmv->conn_data = *data;
301 if (obd->obd_type->typ_procsym == NULL) {
302 obd->obd_type->typ_procsym = lprocfs_seq_register("target_obds",
305 if (IS_ERR(obd->obd_type->typ_procsym)) {
306 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
307 obd->obd_type->typ_name, obd->obd_name);
308 obd->obd_type->typ_procsym = NULL;
313 * All real clients should perform actual connection right away, because
314 * it is possible, that LMV will not have opportunity to connect targets
315 * and MDC stuff will be called directly, for instance while reading
316 * ../mdc/../kbytesfree procfs file, etc.
318 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
319 rc = lmv_check_connect(obd);
321 if (rc && obd->obd_type->typ_procsym != NULL)
322 lprocfs_remove(&obd->obd_type->typ_procsym);
326 static void lmv_set_timeouts(struct obd_device *obd)
332 if (lmv->server_timeout == 0)
335 if (lmv->connected == 0)
338 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
339 struct lmv_tgt_desc *tgt = lmv->tgts[i];
341 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
344 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
345 KEY_INTERMDS, 0, NULL, NULL);
349 static int lmv_init_ea_size(struct obd_export *exp, int easize,
350 int def_easize, int cookiesize, int def_cookiesize)
352 struct obd_device *obd = exp->exp_obd;
353 struct lmv_obd *lmv = &obd->u.lmv;
359 if (lmv->max_easize < easize) {
360 lmv->max_easize = easize;
363 if (lmv->max_def_easize < def_easize) {
364 lmv->max_def_easize = def_easize;
367 if (lmv->max_cookiesize < cookiesize) {
368 lmv->max_cookiesize = cookiesize;
371 if (lmv->max_def_cookiesize < def_cookiesize) {
372 lmv->max_def_cookiesize = def_cookiesize;
378 if (lmv->connected == 0)
381 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
382 struct lmv_tgt_desc *tgt = lmv->tgts[i];
384 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
385 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
389 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
390 cookiesize, def_cookiesize);
392 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
393 " rc = %d.\n", obd->obd_name, i, rc);
400 #define MAX_STRING_SIZE 128
402 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
404 struct lmv_obd *lmv = &obd->u.lmv;
405 struct obd_uuid *cluuid = &lmv->cluuid;
406 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
407 struct obd_device *mdc_obd;
408 struct obd_export *mdc_exp;
409 struct lu_fld_target target;
413 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
416 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
420 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
421 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
422 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
425 if (!mdc_obd->obd_set_up) {
426 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
430 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
431 &lmv->conn_data, NULL);
433 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
438 * Init fid sequence client for this mdc and add new fld target.
440 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
444 target.ft_srv = NULL;
445 target.ft_exp = mdc_exp;
446 target.ft_idx = tgt->ltd_idx;
448 fld_client_add_target(&lmv->lmv_fld, &target);
450 rc = obd_register_observer(mdc_obd, obd);
452 obd_disconnect(mdc_exp);
453 CERROR("target %s register_observer error %d\n",
454 tgt->ltd_uuid.uuid, rc);
458 if (obd->obd_observer) {
460 * Tell the observer about the new target.
462 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
464 (void *)(tgt - lmv->tgts[0]));
466 obd_disconnect(mdc_exp);
472 tgt->ltd_exp = mdc_exp;
473 lmv->desc.ld_active_tgt_count++;
475 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
476 lmv->max_cookiesize, lmv->max_def_cookiesize);
478 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
479 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
480 atomic_read(&obd->obd_refcount));
482 if (obd->obd_type->typ_procsym != NULL) {
483 struct proc_dir_entry *mdc_symlink;
485 LASSERT(mdc_obd->obd_type != NULL);
486 LASSERT(mdc_obd->obd_type->typ_name != NULL);
487 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
488 obd->obd_type->typ_procsym,
490 mdc_obd->obd_type->typ_name,
492 if (mdc_symlink == NULL) {
493 CERROR("Could not register LMV target "
494 "/proc/fs/lustre/%s/%s/target_obds/%s.",
495 obd->obd_type->typ_name, obd->obd_name,
497 lprocfs_remove(&obd->obd_type->typ_procsym);
503 static void lmv_del_target(struct lmv_obd *lmv, int index)
505 if (lmv->tgts[index] == NULL)
508 OBD_FREE_PTR(lmv->tgts[index]);
509 lmv->tgts[index] = NULL;
513 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
514 __u32 index, int gen)
516 struct lmv_obd *lmv = &obd->u.lmv;
517 struct lmv_tgt_desc *tgt;
521 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
525 if (lmv->desc.ld_tgt_count == 0) {
526 struct obd_device *mdc_obd;
528 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
531 lmv_init_unlock(lmv);
532 CERROR("%s: Target %s not attached: rc = %d\n",
533 obd->obd_name, uuidp->uuid, -EINVAL);
538 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
539 tgt = lmv->tgts[index];
540 CERROR("%s: UUID %s already assigned at LOV target index %d:"
541 " rc = %d\n", obd->obd_name,
542 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
543 lmv_init_unlock(lmv);
547 if (index >= lmv->tgts_size) {
548 /* We need to reallocate the lmv target array. */
549 struct lmv_tgt_desc **newtgts, **old = NULL;
553 while (newsize < index + 1)
554 newsize = newsize << 1;
555 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
556 if (newtgts == NULL) {
557 lmv_init_unlock(lmv);
561 if (lmv->tgts_size) {
562 memcpy(newtgts, lmv->tgts,
563 sizeof(*newtgts) * lmv->tgts_size);
565 oldsize = lmv->tgts_size;
569 lmv->tgts_size = newsize;
572 OBD_FREE(old, sizeof(*old) * oldsize);
574 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
580 lmv_init_unlock(lmv);
584 mutex_init(&tgt->ltd_fid_mutex);
585 tgt->ltd_idx = index;
586 tgt->ltd_uuid = *uuidp;
588 lmv->tgts[index] = tgt;
589 if (index >= lmv->desc.ld_tgt_count)
590 lmv->desc.ld_tgt_count = index + 1;
592 if (lmv->connected) {
593 rc = lmv_connect_mdc(obd, tgt);
595 spin_lock(&lmv->lmv_lock);
596 lmv->desc.ld_tgt_count--;
597 memset(tgt, 0, sizeof(*tgt));
598 spin_unlock(&lmv->lmv_lock);
600 int easize = sizeof(struct lmv_stripe_md) +
601 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
602 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
606 lmv_init_unlock(lmv);
610 int lmv_check_connect(struct obd_device *obd)
612 struct lmv_obd *lmv = &obd->u.lmv;
613 struct lmv_tgt_desc *tgt;
623 if (lmv->connected) {
624 lmv_init_unlock(lmv);
628 if (lmv->desc.ld_tgt_count == 0) {
629 lmv_init_unlock(lmv);
630 CERROR("%s: no targets configured.\n", obd->obd_name);
634 LASSERT(lmv->tgts != NULL);
636 if (lmv->tgts[0] == NULL) {
637 lmv_init_unlock(lmv);
638 CERROR("%s: no target configured for index 0.\n",
643 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
644 lmv->cluuid.uuid, obd->obd_name);
646 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
650 rc = lmv_connect_mdc(obd, tgt);
655 lmv_set_timeouts(obd);
656 class_export_put(lmv->exp);
658 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
659 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
660 lmv_init_unlock(lmv);
671 --lmv->desc.ld_active_tgt_count;
672 rc2 = obd_disconnect(tgt->ltd_exp);
674 CERROR("LMV target %s disconnect on "
675 "MDC idx %d: error %d\n",
676 tgt->ltd_uuid.uuid, i, rc2);
680 class_disconnect(lmv->exp);
681 lmv_init_unlock(lmv);
685 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
687 struct lmv_obd *lmv = &obd->u.lmv;
688 struct obd_device *mdc_obd;
692 LASSERT(tgt != NULL);
693 LASSERT(obd != NULL);
695 mdc_obd = class_exp2obd(tgt->ltd_exp);
698 mdc_obd->obd_force = obd->obd_force;
699 mdc_obd->obd_fail = obd->obd_fail;
700 mdc_obd->obd_no_recov = obd->obd_no_recov;
703 if (obd->obd_type->typ_procsym != NULL)
704 lprocfs_remove_proc_entry(mdc_obd->obd_name,
705 obd->obd_type->typ_procsym);
707 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
709 CERROR("Can't finanize fids factory\n");
711 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
712 tgt->ltd_exp->exp_obd->obd_name,
713 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
715 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
716 rc = obd_disconnect(tgt->ltd_exp);
718 if (tgt->ltd_active) {
719 CERROR("Target %s disconnect error %d\n",
720 tgt->ltd_uuid.uuid, rc);
724 lmv_activate_target(lmv, tgt, 0);
729 static int lmv_disconnect(struct obd_export *exp)
731 struct obd_device *obd = class_exp2obd(exp);
732 struct lmv_obd *lmv = &obd->u.lmv;
741 * Only disconnect the underlying layers on the final disconnect.
744 if (lmv->refcount != 0)
747 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
748 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
751 lmv_disconnect_mdc(obd, lmv->tgts[i]);
754 if (obd->obd_type->typ_procsym != NULL)
755 lprocfs_remove(&obd->obd_type->typ_procsym);
757 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
758 obd->obd_type->typ_name, obd->obd_name);
762 * This is the case when no real connection is established by
763 * lmv_check_connect().
766 class_export_put(exp);
767 rc = class_disconnect(exp);
768 if (lmv->refcount == 0)
773 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
775 struct obd_device *obddev = class_exp2obd(exp);
776 struct lmv_obd *lmv = &obddev->u.lmv;
777 struct getinfo_fid2path *gf;
778 struct lmv_tgt_desc *tgt;
779 struct getinfo_fid2path *remote_gf = NULL;
780 int remote_gf_size = 0;
783 gf = (struct getinfo_fid2path *)karg;
784 tgt = lmv_find_target(lmv, &gf->gf_fid);
786 RETURN(PTR_ERR(tgt));
789 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
790 if (rc != 0 && rc != -EREMOTE)
791 GOTO(out_fid2path, rc);
793 /* If remote_gf != NULL, it means just building the
794 * path on the remote MDT, copy this path segement to gf */
795 if (remote_gf != NULL) {
796 struct getinfo_fid2path *ori_gf;
799 ori_gf = (struct getinfo_fid2path *)karg;
800 if (strlen(ori_gf->gf_path) +
801 strlen(gf->gf_path) > ori_gf->gf_pathlen)
802 GOTO(out_fid2path, rc = -EOVERFLOW);
804 ptr = ori_gf->gf_path;
806 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
807 strlen(ori_gf->gf_path));
809 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
810 ptr += strlen(gf->gf_path);
814 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
815 tgt->ltd_exp->exp_obd->obd_name,
816 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
820 GOTO(out_fid2path, rc);
822 /* sigh, has to go to another MDT to do path building further */
823 if (remote_gf == NULL) {
824 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
825 OBD_ALLOC(remote_gf, remote_gf_size);
826 if (remote_gf == NULL)
827 GOTO(out_fid2path, rc = -ENOMEM);
828 remote_gf->gf_pathlen = PATH_MAX;
831 if (!fid_is_sane(&gf->gf_fid)) {
832 CERROR("%s: invalid FID "DFID": rc = %d\n",
833 tgt->ltd_exp->exp_obd->obd_name,
834 PFID(&gf->gf_fid), -EINVAL);
835 GOTO(out_fid2path, rc = -EINVAL);
838 tgt = lmv_find_target(lmv, &gf->gf_fid);
840 GOTO(out_fid2path, rc = -EINVAL);
842 remote_gf->gf_fid = gf->gf_fid;
843 remote_gf->gf_recno = -1;
844 remote_gf->gf_linkno = -1;
845 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
847 goto repeat_fid2path;
850 if (remote_gf != NULL)
851 OBD_FREE(remote_gf, remote_gf_size);
855 static int lmv_hsm_req_count(struct lmv_obd *lmv,
856 const struct hsm_user_request *hur,
857 const struct lmv_tgt_desc *tgt_mds)
861 struct lmv_tgt_desc *curr_tgt;
863 /* count how many requests must be sent to the given target */
864 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
865 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
866 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
872 static void lmv_hsm_req_build(struct lmv_obd *lmv,
873 struct hsm_user_request *hur_in,
874 const struct lmv_tgt_desc *tgt_mds,
875 struct hsm_user_request *hur_out)
878 struct lmv_tgt_desc *curr_tgt;
880 /* build the hsm_user_request for the given target */
881 hur_out->hur_request = hur_in->hur_request;
883 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
884 curr_tgt = lmv_find_target(lmv,
885 &hur_in->hur_user_item[i].hui_fid);
886 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
887 hur_out->hur_user_item[nr_out] =
888 hur_in->hur_user_item[i];
892 hur_out->hur_request.hr_itemcount = nr_out;
893 memcpy(hur_data(hur_out), hur_data(hur_in),
894 hur_in->hur_request.hr_data_len);
897 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
898 struct lustre_kernelcomm *lk, void *uarg)
902 struct kkuc_ct_data *kcd = NULL;
905 /* unregister request (call from llapi_hsm_copytool_fini) */
906 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
907 struct lmv_tgt_desc *tgt = lmv->tgts[i];
909 if (tgt == NULL || tgt->ltd_exp == NULL)
911 /* best effort: try to clean as much as possible
912 * (continue on error) */
913 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
916 /* Whatever the result, remove copytool from kuc groups.
917 * Unreached coordinators will get EPIPE on next requests
918 * and will unregister automatically.
920 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
927 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
928 struct lustre_kernelcomm *lk, void *uarg)
933 bool any_set = false;
934 struct kkuc_ct_data *kcd;
937 /* All or nothing: try to register to all MDS.
938 * In case of failure, unregister from previous MDS,
939 * except if it because of inactive target. */
940 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
941 struct lmv_tgt_desc *tgt = lmv->tgts[i];
943 if (tgt == NULL || tgt->ltd_exp == NULL)
945 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
947 if (tgt->ltd_active) {
948 /* permanent error */
949 CERROR("%s: iocontrol MDC %s on MDT"
950 " idx %d cmd %x: err = %d\n",
951 class_exp2obd(lmv->exp)->obd_name,
952 tgt->ltd_uuid.uuid, i, cmd, err);
954 lk->lk_flags |= LK_FLG_STOP;
955 /* unregister from previous MDS */
956 for (j = 0; j < i; j++) {
958 if (tgt == NULL || tgt->ltd_exp == NULL)
960 obd_iocontrol(cmd, tgt->ltd_exp, len,
965 /* else: transient error.
966 * kuc will register to the missing MDT
974 /* no registration done: return error */
977 /* at least one registration done, with no failure */
978 filp = fget(lk->lk_wfd);
987 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
988 kcd->kcd_uuid = lmv->cluuid;
989 kcd->kcd_archive = lk->lk_data;
991 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
1004 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1005 int len, void *karg, void *uarg)
1007 struct obd_device *obddev = class_exp2obd(exp);
1008 struct lmv_obd *lmv = &obddev->u.lmv;
1009 struct lmv_tgt_desc *tgt = NULL;
1013 __u32 count = lmv->desc.ld_tgt_count;
1020 case IOC_OBD_STATFS: {
1021 struct obd_ioctl_data *data = karg;
1022 struct obd_device *mdc_obd;
1023 struct obd_statfs stat_buf = {0};
1026 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1027 if ((index >= count))
1030 tgt = lmv->tgts[index];
1031 if (tgt == NULL || !tgt->ltd_active)
1034 mdc_obd = class_exp2obd(tgt->ltd_exp);
1039 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1040 min((int) data->ioc_plen2,
1041 (int) sizeof(struct obd_uuid))))
1044 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1045 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1049 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1050 min((int) data->ioc_plen1,
1051 (int) sizeof(stat_buf))))
1055 case OBD_IOC_QUOTACTL: {
1056 struct if_quotactl *qctl = karg;
1057 struct obd_quotactl *oqctl;
1059 if (qctl->qc_valid == QC_MDTIDX) {
1060 if (count <= qctl->qc_idx)
1063 tgt = lmv->tgts[qctl->qc_idx];
1064 if (tgt == NULL || tgt->ltd_exp == NULL)
1066 } else if (qctl->qc_valid == QC_UUID) {
1067 for (i = 0; i < count; i++) {
1071 if (!obd_uuid_equals(&tgt->ltd_uuid,
1075 if (tgt->ltd_exp == NULL)
1087 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1088 OBD_ALLOC_PTR(oqctl);
1092 QCTL_COPY(oqctl, qctl);
1093 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1095 QCTL_COPY(qctl, oqctl);
1096 qctl->qc_valid = QC_MDTIDX;
1097 qctl->obd_uuid = tgt->ltd_uuid;
1099 OBD_FREE_PTR(oqctl);
1102 case OBD_IOC_CHANGELOG_SEND:
1103 case OBD_IOC_CHANGELOG_CLEAR: {
1104 struct ioc_changelog *icc = karg;
1106 if (icc->icc_mdtindex >= count)
1109 tgt = lmv->tgts[icc->icc_mdtindex];
1110 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1112 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1115 case LL_IOC_GET_CONNECT_FLAGS: {
1117 if (tgt == NULL || tgt->ltd_exp == NULL)
1119 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1122 case OBD_IOC_FID2PATH: {
1123 rc = lmv_fid2path(exp, len, karg, uarg);
1126 case LL_IOC_HSM_STATE_GET:
1127 case LL_IOC_HSM_STATE_SET:
1128 case LL_IOC_HSM_ACTION: {
1129 struct md_op_data *op_data = karg;
1131 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1133 RETURN(PTR_ERR(tgt));
1135 if (tgt->ltd_exp == NULL)
1138 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1141 case LL_IOC_HSM_PROGRESS: {
1142 const struct hsm_progress_kernel *hpk = karg;
1144 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1146 RETURN(PTR_ERR(tgt));
1147 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1150 case LL_IOC_HSM_REQUEST: {
1151 struct hsm_user_request *hur = karg;
1152 unsigned int reqcount = hur->hur_request.hr_itemcount;
1157 /* if the request is about a single fid
1158 * or if there is a single MDS, no need to split
1160 if (reqcount == 1 || count == 1) {
1161 tgt = lmv_find_target(lmv,
1162 &hur->hur_user_item[0].hui_fid);
1164 RETURN(PTR_ERR(tgt));
1165 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1167 /* split fid list to their respective MDS */
1168 for (i = 0; i < count; i++) {
1169 unsigned int nr, reqlen;
1171 struct hsm_user_request *req;
1174 if (tgt == NULL || tgt->ltd_exp == NULL)
1177 nr = lmv_hsm_req_count(lmv, hur, tgt);
1178 if (nr == 0) /* nothing for this MDS */
1181 /* build a request with fids for this MDS */
1182 reqlen = offsetof(typeof(*hur),
1184 + hur->hur_request.hr_data_len;
1185 OBD_ALLOC_LARGE(req, reqlen);
1189 lmv_hsm_req_build(lmv, hur, tgt, req);
1191 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1193 if (rc1 != 0 && rc == 0)
1195 OBD_FREE_LARGE(req, reqlen);
1200 case LL_IOC_LOV_SWAP_LAYOUTS: {
1201 struct md_op_data *op_data = karg;
1202 struct lmv_tgt_desc *tgt1, *tgt2;
1204 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1206 RETURN(PTR_ERR(tgt1));
1208 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1210 RETURN(PTR_ERR(tgt2));
1212 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1215 /* only files on same MDT can have their layouts swapped */
1216 if (tgt1->ltd_idx != tgt2->ltd_idx)
1219 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1222 case LL_IOC_HSM_CT_START: {
1223 struct lustre_kernelcomm *lk = karg;
1224 if (lk->lk_flags & LK_FLG_STOP)
1225 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1227 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1231 for (i = 0; i < count; i++) {
1232 struct obd_device *mdc_obd;
1236 if (tgt == NULL || tgt->ltd_exp == NULL)
1238 /* ll_umount_begin() sets force flag but for lmv, not
1239 * mdc. Let's pass it through */
1240 mdc_obd = class_exp2obd(tgt->ltd_exp);
1241 mdc_obd->obd_force = obddev->obd_force;
1242 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1243 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1246 if (tgt->ltd_active) {
1247 CERROR("error: iocontrol MDC %s on MDT"
1248 " idx %d cmd %x: err = %d\n",
1249 tgt->ltd_uuid.uuid, i, cmd, err);
1263 static int lmv_all_chars_policy(int count, const char *name,
1274 static int lmv_nid_policy(struct lmv_obd *lmv)
1276 struct obd_import *imp;
1280 * XXX: To get nid we assume that underlying obd device is mdc.
1282 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1283 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1284 return id % lmv->desc.ld_tgt_count;
1287 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1288 placement_policy_t placement)
1290 switch (placement) {
1291 case PLACEMENT_CHAR_POLICY:
1292 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1294 op_data->op_namelen);
1295 case PLACEMENT_NID_POLICY:
1296 return lmv_nid_policy(lmv);
1302 CERROR("Unsupported placement policy %x\n", placement);
1308 * This is _inode_ placement policy function (not name).
1310 static int lmv_placement_policy(struct obd_device *obd,
1311 struct md_op_data *op_data,
1314 struct lmv_obd *lmv = &obd->u.lmv;
1317 LASSERT(mds != NULL);
1319 if (lmv->desc.ld_tgt_count == 1) {
1325 * If stripe_offset is provided during setdirstripe
1326 * (setdirstripe -i xx), xx MDS will be choosen.
1328 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1329 struct lmv_user_md *lum;
1331 lum = op_data->op_data;
1333 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1334 *mds = le32_to_cpu(lum->lum_stripe_offset);
1336 /* -1 means default, which will be in the same MDT with
1338 *mds = op_data->op_mds;
1339 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1342 /* Allocate new fid on target according to operation type and
1343 * parent home mds. */
1344 *mds = op_data->op_mds;
1350 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1353 struct lmv_tgt_desc *tgt;
1357 tgt = lmv_get_target(lmv, mds);
1359 RETURN(PTR_ERR(tgt));
1362 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1363 * on server that seq in new allocated fid is not yet known.
1365 mutex_lock(&tgt->ltd_fid_mutex);
1367 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1368 GOTO(out, rc = -ENODEV);
1371 * Asking underlying tgt layer to allocate new fid.
1373 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1375 LASSERT(fid_is_sane(fid));
1381 mutex_unlock(&tgt->ltd_fid_mutex);
1385 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1386 struct lu_fid *fid, struct md_op_data *op_data)
1388 struct obd_device *obd = class_exp2obd(exp);
1389 struct lmv_obd *lmv = &obd->u.lmv;
1394 LASSERT(op_data != NULL);
1395 LASSERT(fid != NULL);
1397 rc = lmv_placement_policy(obd, op_data, &mds);
1399 CERROR("Can't get target for allocating fid, "
1404 rc = __lmv_fid_alloc(lmv, fid, mds);
1406 CERROR("Can't alloc new fid, rc %d\n", rc);
1413 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1415 struct lmv_obd *lmv = &obd->u.lmv;
1416 struct lmv_desc *desc;
1420 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1421 CERROR("LMV setup requires a descriptor\n");
1425 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1426 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1427 CERROR("Lmv descriptor size wrong: %d > %d\n",
1428 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1432 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1433 if (lmv->tgts == NULL)
1435 lmv->tgts_size = 32;
1437 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1438 lmv->desc.ld_tgt_count = 0;
1439 lmv->desc.ld_active_tgt_count = 0;
1440 lmv->max_cookiesize = 0;
1441 lmv->max_def_easize = 0;
1442 lmv->max_easize = 0;
1443 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1445 spin_lock_init(&lmv->lmv_lock);
1446 mutex_init(&lmv->init_mutex);
1449 obd->obd_vars = lprocfs_lmv_obd_vars;
1450 lprocfs_seq_obd_setup(obd);
1451 lprocfs_alloc_md_stats(obd, 0);
1452 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1453 0444, &lmv_proc_target_fops, obd);
1455 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1458 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1459 LUSTRE_CLI_FLD_HASH_DHT);
1461 CERROR("Can't init FLD, err %d\n", rc);
1471 static int lmv_cleanup(struct obd_device *obd)
1473 struct lmv_obd *lmv = &obd->u.lmv;
1476 fld_client_fini(&lmv->lmv_fld);
1477 if (lmv->tgts != NULL) {
1479 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1480 if (lmv->tgts[i] == NULL)
1482 lmv_del_target(lmv, i);
1484 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1490 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1492 struct lustre_cfg *lcfg = buf;
1493 struct obd_uuid obd_uuid;
1499 switch (lcfg->lcfg_command) {
1501 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1502 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1503 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1504 GOTO(out, rc = -EINVAL);
1506 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1508 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1509 GOTO(out, rc = -EINVAL);
1510 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1511 GOTO(out, rc = -EINVAL);
1512 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1515 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1516 GOTO(out, rc = -EINVAL);
1522 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1523 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1525 struct obd_device *obd = class_exp2obd(exp);
1526 struct lmv_obd *lmv = &obd->u.lmv;
1527 struct obd_statfs *temp;
1532 rc = lmv_check_connect(obd);
1536 OBD_ALLOC(temp, sizeof(*temp));
1540 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1541 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1544 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1547 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1548 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1550 GOTO(out_free_temp, rc);
1555 /* If the statfs is from mount, it will needs
1556 * retrieve necessary information from MDT0.
1557 * i.e. mount does not need the merged osfs
1559 * And also clients can be mounted as long as
1560 * MDT0 is in service*/
1561 if (flags & OBD_STATFS_FOR_MDT0)
1562 GOTO(out_free_temp, rc);
1564 osfs->os_bavail += temp->os_bavail;
1565 osfs->os_blocks += temp->os_blocks;
1566 osfs->os_ffree += temp->os_ffree;
1567 osfs->os_files += temp->os_files;
1573 OBD_FREE(temp, sizeof(*temp));
1577 static int lmv_getstatus(struct obd_export *exp,
1579 struct obd_capa **pc)
1581 struct obd_device *obd = exp->exp_obd;
1582 struct lmv_obd *lmv = &obd->u.lmv;
1586 rc = lmv_check_connect(obd);
1590 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1594 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1595 struct obd_capa *oc, obd_valid valid, const char *name,
1596 const char *input, int input_size, int output_size,
1597 int flags, struct ptlrpc_request **request)
1599 struct obd_device *obd = exp->exp_obd;
1600 struct lmv_obd *lmv = &obd->u.lmv;
1601 struct lmv_tgt_desc *tgt;
1605 rc = lmv_check_connect(obd);
1609 tgt = lmv_find_target(lmv, fid);
1611 RETURN(PTR_ERR(tgt));
1613 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1614 input_size, output_size, flags, request);
1619 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1620 struct obd_capa *oc, obd_valid valid, const char *name,
1621 const char *input, int input_size, int output_size,
1622 int flags, __u32 suppgid,
1623 struct ptlrpc_request **request)
1625 struct obd_device *obd = exp->exp_obd;
1626 struct lmv_obd *lmv = &obd->u.lmv;
1627 struct lmv_tgt_desc *tgt;
1631 rc = lmv_check_connect(obd);
1635 tgt = lmv_find_target(lmv, fid);
1637 RETURN(PTR_ERR(tgt));
1639 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1640 input_size, output_size, flags, suppgid,
1646 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1647 struct ptlrpc_request **request)
1649 struct obd_device *obd = exp->exp_obd;
1650 struct lmv_obd *lmv = &obd->u.lmv;
1651 struct lmv_tgt_desc *tgt;
1655 rc = lmv_check_connect(obd);
1659 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1661 RETURN(PTR_ERR(tgt));
1663 if (op_data->op_flags & MF_GET_MDT_IDX) {
1664 op_data->op_mds = tgt->ltd_idx;
1668 rc = md_getattr(tgt->ltd_exp, op_data, request);
1673 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1675 struct obd_device *obd = exp->exp_obd;
1676 struct lmv_obd *lmv = &obd->u.lmv;
1681 rc = lmv_check_connect(obd);
1685 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1688 * With DNE every object can have two locks in different namespaces:
1689 * lookup lock in space of MDT storing direntry and update/open lock in
1690 * space of MDT storing inode.
1692 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1693 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1695 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1701 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1702 ldlm_iterator_t it, void *data)
1704 struct obd_device *obd = exp->exp_obd;
1705 struct lmv_obd *lmv = &obd->u.lmv;
1710 rc = lmv_check_connect(obd);
1714 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1717 * With DNE every object can have two locks in different namespaces:
1718 * lookup lock in space of MDT storing direntry and update/open lock in
1719 * space of MDT storing inode.
1721 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1722 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1724 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1733 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1734 struct md_open_data *mod, struct ptlrpc_request **request)
1736 struct obd_device *obd = exp->exp_obd;
1737 struct lmv_obd *lmv = &obd->u.lmv;
1738 struct lmv_tgt_desc *tgt;
1742 rc = lmv_check_connect(obd);
1746 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1748 RETURN(PTR_ERR(tgt));
1750 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1751 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1756 * Choosing the MDT by name or FID in @op_data.
1757 * For non-striped directory, it will locate MDT by fid.
1758 * For striped-directory, it will locate MDT by name. And also
1759 * it will reset op_fid1 with the FID of the choosen stripe.
1761 struct lmv_tgt_desc *
1762 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1763 const char *name, int namelen, struct lu_fid *fid,
1766 struct lmv_tgt_desc *tgt;
1767 const struct lmv_oinfo *oinfo;
1769 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1771 RETURN((void *)oinfo);
1772 *fid = oinfo->lmo_fid;
1773 *mds = oinfo->lmo_mds;
1774 tgt = lmv_get_target(lmv, *mds);
1776 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1781 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1784 struct lmv_stripe_md *lsm = op_data->op_mea1;
1785 struct lmv_tgt_desc *tgt;
1787 if (lsm == NULL || op_data->op_namelen == 0) {
1788 tgt = lmv_find_target(lmv, fid);
1792 op_data->op_mds = tgt->ltd_idx;
1796 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1797 op_data->op_namelen, fid,
1801 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1802 const void *data, int datalen, int mode, __u32 uid,
1803 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1804 struct ptlrpc_request **request)
1806 struct obd_device *obd = exp->exp_obd;
1807 struct lmv_obd *lmv = &obd->u.lmv;
1808 struct lmv_tgt_desc *tgt;
1812 rc = lmv_check_connect(obd);
1816 if (!lmv->desc.ld_active_tgt_count)
1819 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1821 RETURN(PTR_ERR(tgt));
1823 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1824 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1827 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1831 /* Send the create request to the MDT where the object
1832 * will be located */
1833 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1835 RETURN(PTR_ERR(tgt));
1837 op_data->op_mds = tgt->ltd_idx;
1839 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1840 PFID(&op_data->op_fid2), op_data->op_mds);
1842 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1843 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1844 cap_effective, rdev, request);
1846 if (*request == NULL)
1848 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1853 static int lmv_done_writing(struct obd_export *exp,
1854 struct md_op_data *op_data,
1855 struct md_open_data *mod)
1857 struct obd_device *obd = exp->exp_obd;
1858 struct lmv_obd *lmv = &obd->u.lmv;
1859 struct lmv_tgt_desc *tgt;
1863 rc = lmv_check_connect(obd);
1867 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1869 RETURN(PTR_ERR(tgt));
1871 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1876 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1877 struct lookup_intent *it, struct md_op_data *op_data,
1878 struct lustre_handle *lockh, void *lmm, int lmmsize,
1879 __u64 extra_lock_flags)
1881 struct ptlrpc_request *req = it->d.lustre.it_data;
1882 struct obd_device *obd = exp->exp_obd;
1883 struct lmv_obd *lmv = &obd->u.lmv;
1884 struct lustre_handle plock;
1885 struct lmv_tgt_desc *tgt;
1886 struct md_op_data *rdata;
1888 struct mdt_body *body;
1893 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1894 LASSERT(body != NULL);
1896 if (!(body->valid & OBD_MD_MDS))
1899 CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1900 LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1903 * We got LOOKUP lock, but we really need attrs.
1905 pmode = it->d.lustre.it_lock_mode;
1906 LASSERT(pmode != 0);
1907 memcpy(&plock, lockh, sizeof(plock));
1908 it->d.lustre.it_lock_mode = 0;
1909 it->d.lustre.it_data = NULL;
1912 ptlrpc_req_finished(req);
1914 tgt = lmv_find_target(lmv, &fid1);
1916 GOTO(out, rc = PTR_ERR(tgt));
1918 OBD_ALLOC_PTR(rdata);
1920 GOTO(out, rc = -ENOMEM);
1922 rdata->op_fid1 = fid1;
1923 rdata->op_bias = MDS_CROSS_REF;
1925 rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1926 lmm, lmmsize, NULL, extra_lock_flags);
1927 OBD_FREE_PTR(rdata);
1930 ldlm_lock_decref(&plock, pmode);
1935 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1936 struct lookup_intent *it, struct md_op_data *op_data,
1937 struct lustre_handle *lockh, void *lmm, int lmmsize,
1938 struct ptlrpc_request **req, __u64 extra_lock_flags)
1940 struct obd_device *obd = exp->exp_obd;
1941 struct lmv_obd *lmv = &obd->u.lmv;
1942 struct lmv_tgt_desc *tgt;
1946 rc = lmv_check_connect(obd);
1950 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1951 LL_IT2STR(it), PFID(&op_data->op_fid1));
1953 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1955 RETURN(PTR_ERR(tgt));
1957 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1958 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1960 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1961 lmm, lmmsize, req, extra_lock_flags);
1963 if (rc == 0 && it && it->it_op == IT_OPEN) {
1964 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1965 lmm, lmmsize, extra_lock_flags);
1971 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1972 struct ptlrpc_request **preq)
1974 struct ptlrpc_request *req = NULL;
1975 struct obd_device *obd = exp->exp_obd;
1976 struct lmv_obd *lmv = &obd->u.lmv;
1977 struct lmv_tgt_desc *tgt;
1978 struct mdt_body *body;
1982 rc = lmv_check_connect(obd);
1986 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1988 RETURN(PTR_ERR(tgt));
1990 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1991 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1994 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1998 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1999 LASSERT(body != NULL);
2001 if (body->valid & OBD_MD_MDS) {
2002 struct lu_fid rid = body->fid1;
2003 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2006 tgt = lmv_find_target(lmv, &rid);
2008 ptlrpc_req_finished(*preq);
2010 RETURN(PTR_ERR(tgt));
2013 op_data->op_fid1 = rid;
2014 op_data->op_valid |= OBD_MD_FLCROSSREF;
2015 op_data->op_namelen = 0;
2016 op_data->op_name = NULL;
2017 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2018 ptlrpc_req_finished(*preq);
2025 #define md_op_data_fid(op_data, fl) \
2026 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2027 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2028 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2029 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2032 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2033 struct md_op_data *op_data,
2034 int op_tgt, ldlm_mode_t mode, int bits, int flag)
2036 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2037 struct obd_device *obd = exp->exp_obd;
2038 struct lmv_obd *lmv = &obd->u.lmv;
2039 ldlm_policy_data_t policy = {{ 0 }};
2043 if (!fid_is_sane(fid))
2047 tgt = lmv_find_target(lmv, fid);
2049 RETURN(PTR_ERR(tgt));
2052 if (tgt->ltd_idx != op_tgt) {
2053 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2054 policy.l_inodebits.bits = bits;
2055 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2056 mode, LCF_ASYNC, NULL);
2059 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2061 op_data->op_flags |= flag;
2069 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2072 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2073 struct ptlrpc_request **request)
2075 struct obd_device *obd = exp->exp_obd;
2076 struct lmv_obd *lmv = &obd->u.lmv;
2077 struct lmv_tgt_desc *tgt;
2081 rc = lmv_check_connect(obd);
2085 LASSERT(op_data->op_namelen != 0);
2087 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2088 PFID(&op_data->op_fid2), op_data->op_namelen,
2089 op_data->op_name, PFID(&op_data->op_fid1));
2091 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2092 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2093 op_data->op_cap = cfs_curproc_cap_pack();
2094 if (op_data->op_mea2 != NULL) {
2095 struct lmv_stripe_md *lsm = op_data->op_mea2;
2096 const struct lmv_oinfo *oinfo;
2098 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2099 op_data->op_namelen);
2101 RETURN(PTR_ERR(oinfo));
2103 op_data->op_fid2 = oinfo->lmo_fid;
2106 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2108 RETURN(PTR_ERR(tgt));
2111 * Cancel UPDATE lock on child (fid1).
2113 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2114 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2115 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2119 rc = md_link(tgt->ltd_exp, op_data, request);
2124 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2125 const char *old, int oldlen, const char *new, int newlen,
2126 struct ptlrpc_request **request)
2128 struct obd_device *obd = exp->exp_obd;
2129 struct lmv_obd *lmv = &obd->u.lmv;
2130 struct lmv_tgt_desc *src_tgt;
2134 LASSERT(oldlen != 0);
2136 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2137 oldlen, old, PFID(&op_data->op_fid1),
2138 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2139 newlen, new, PFID(&op_data->op_fid2),
2140 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2142 rc = lmv_check_connect(obd);
2146 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2147 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2148 op_data->op_cap = cfs_curproc_cap_pack();
2149 if (op_data->op_cli_flags & CLI_MIGRATE) {
2150 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2151 PFID(&op_data->op_fid3));
2152 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2155 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2157 if (op_data->op_mea1 != NULL) {
2158 struct lmv_stripe_md *lsm = op_data->op_mea1;
2160 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2164 if (IS_ERR(src_tgt))
2165 RETURN(PTR_ERR(src_tgt));
2167 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2168 if (IS_ERR(src_tgt))
2169 RETURN(PTR_ERR(src_tgt));
2171 op_data->op_mds = src_tgt->ltd_idx;
2174 if (op_data->op_mea2) {
2175 struct lmv_stripe_md *lsm = op_data->op_mea2;
2176 const struct lmv_oinfo *oinfo;
2178 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2180 RETURN(PTR_ERR(oinfo));
2182 op_data->op_fid2 = oinfo->lmo_fid;
2185 if (IS_ERR(src_tgt))
2186 RETURN(PTR_ERR(src_tgt));
2189 * LOOKUP lock on src child (fid3) should also be cancelled for
2190 * src_tgt in mdc_rename.
2192 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2195 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2198 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2199 LCK_EX, MDS_INODELOCK_UPDATE,
2200 MF_MDC_CANCEL_FID2);
2205 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2207 if (fid_is_sane(&op_data->op_fid3)) {
2208 struct lmv_tgt_desc *tgt;
2210 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2212 RETURN(PTR_ERR(tgt));
2214 /* Cancel LOOKUP lock on its parent */
2215 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2216 LCK_EX, MDS_INODELOCK_LOOKUP,
2217 MF_MDC_CANCEL_FID3);
2221 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2222 LCK_EX, MDS_INODELOCK_FULL,
2223 MF_MDC_CANCEL_FID3);
2229 * Cancel all the locks on tgt child (fid4).
2231 if (fid_is_sane(&op_data->op_fid4))
2232 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2233 LCK_EX, MDS_INODELOCK_FULL,
2234 MF_MDC_CANCEL_FID4);
2236 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2237 op_data->op_mds, PFID(&op_data->op_fid2));
2239 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2245 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2246 void *ea, int ealen, void *ea2, int ea2len,
2247 struct ptlrpc_request **request,
2248 struct md_open_data **mod)
2250 struct obd_device *obd = exp->exp_obd;
2251 struct lmv_obd *lmv = &obd->u.lmv;
2252 struct lmv_tgt_desc *tgt;
2256 rc = lmv_check_connect(obd);
2260 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2261 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2263 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2264 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2266 RETURN(PTR_ERR(tgt));
2268 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2269 ea2len, request, mod);
2274 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2275 struct obd_capa *oc, struct ptlrpc_request **request)
2277 struct obd_device *obd = exp->exp_obd;
2278 struct lmv_obd *lmv = &obd->u.lmv;
2279 struct lmv_tgt_desc *tgt;
2283 rc = lmv_check_connect(obd);
2287 tgt = lmv_find_target(lmv, fid);
2289 RETURN(PTR_ERR(tgt));
2291 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2296 * Adjust a set of pages, each page containing an array of lu_dirpages,
2297 * so that each page can be used as a single logical lu_dirpage.
2299 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2300 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2301 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
2302 * value is used as a cookie to request the next lu_dirpage in a
2303 * directory listing that spans multiple pages (two in this example):
2306 * .|--------v------- -----.
2307 * |s|e|f|p|ent|ent| ... |ent|
2308 * '--|-------------- -----' Each CFS_PAGE contains a single
2309 * '------. lu_dirpage.
2310 * .---------v------- -----.
2311 * |s|e|f|p|ent| 0 | ... | 0 |
2312 * '----------------- -----'
2314 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2315 * larger than LU_PAGE_SIZE, a single host page may contain multiple
2316 * lu_dirpages. After reading the lu_dirpages from the MDS, the
2317 * ldp_hash_end of the first lu_dirpage refers to the one immediately
2318 * after it in the same CFS_PAGE (arrows simplified for brevity, but
2319 * in general e0==s1, e1==s2, etc.):
2321 * .-------------------- -----.
2322 * |s0|e0|f0|p|ent|ent| ... |ent|
2323 * |---v---------------- -----|
2324 * |s1|e1|f1|p|ent|ent| ... |ent|
2325 * |---v---------------- -----| Here, each CFS_PAGE contains
2326 * ... multiple lu_dirpages.
2327 * |---v---------------- -----|
2328 * |s'|e'|f'|p|ent|ent| ... |ent|
2329 * '---|---------------- -----'
2331 * .----------------------------.
2334 * This structure is transformed into a single logical lu_dirpage as follows:
2336 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2337 * labeled 'next CFS_PAGE'.
2339 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2340 * a hash collision with the next page exists.
2342 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2343 * to the first entry of the next lu_dirpage.
2345 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2346 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2350 for (i = 0; i < ncfspgs; i++) {
2351 struct lu_dirpage *dp = kmap(pages[i]);
2352 struct lu_dirpage *first = dp;
2353 struct lu_dirent *end_dirent = NULL;
2354 struct lu_dirent *ent;
2355 __u64 hash_end = dp->ldp_hash_end;
2356 __u32 flags = dp->ldp_flags;
2358 while (--nlupgs > 0) {
2359 ent = lu_dirent_start(dp);
2360 for (end_dirent = ent; ent != NULL;
2361 end_dirent = ent, ent = lu_dirent_next(ent));
2363 /* Advance dp to next lu_dirpage. */
2364 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2366 /* Check if we've reached the end of the CFS_PAGE. */
2367 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2370 /* Save the hash and flags of this lu_dirpage. */
2371 hash_end = dp->ldp_hash_end;
2372 flags = dp->ldp_flags;
2374 /* Check if lu_dirpage contains no entries. */
2378 /* Enlarge the end entry lde_reclen from 0 to
2379 * first entry of next lu_dirpage. */
2380 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2381 end_dirent->lde_reclen =
2382 cpu_to_le16((char *)(dp->ldp_entries) -
2383 (char *)end_dirent);
2386 first->ldp_hash_end = hash_end;
2387 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2388 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2392 LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2395 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2396 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2398 #define NORMAL_MAX_STRIPES 4
2399 int lmv_read_entry(struct obd_export *exp, struct md_op_data *op_data,
2400 struct md_callback *cb_op, struct lu_dirent **ldp,
2401 struct page **ppage)
2403 struct obd_device *obd = exp->exp_obd;
2404 struct lmv_obd *lmv = &obd->u.lmv;
2405 struct lmv_stripe_md *lsm = op_data->op_mea1;
2406 struct lu_dirent *tmp_ents[NORMAL_MAX_STRIPES];
2407 struct lu_dirent **ents = NULL;
2411 struct page *min_page = NULL;
2416 rc = lmv_check_connect(obd);
2423 stripe_count = lsm->lsm_md_stripe_count;
2425 if (stripe_count > NORMAL_MAX_STRIPES) {
2426 OBD_ALLOC(ents, sizeof(ents[0]) * stripe_count);
2428 GOTO(out, rc = -ENOMEM);
2431 memset(ents, 0, sizeof(ents[0]) * stripe_count);
2434 min_hash = MDS_DIR_END_OFF;
2435 for (i = 0; i < stripe_count; i++) {
2436 struct lmv_tgt_desc *tgt;
2437 struct page *page = NULL;
2439 if (likely(lsm == NULL)) {
2440 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2442 GOTO(out, rc = PTR_ERR(tgt));
2443 LASSERT(op_data->op_data != NULL);
2445 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds);
2447 GOTO(out, rc = PTR_ERR(tgt));
2448 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2449 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2450 op_data->op_stripe_offset = i;
2453 rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, &ents[i],
2458 if (ents[i] != NULL &&
2459 le64_to_cpu(ents[i]->lde_hash) <= min_hash) {
2460 if (min_page != NULL)
2461 page_cache_release(min_page);
2463 min_hash = le64_to_cpu(ents[i]->lde_hash);
2468 if (min_hash != MDS_DIR_END_OFF)
2469 *ldp = ents[min_idx];
2473 if (stripe_count > NORMAL_MAX_STRIPES && ents != NULL)
2474 OBD_FREE(ents, sizeof(ents[0]) * stripe_count);
2476 if (rc != 0 && min_page != NULL) {
2478 page_cache_release(min_page);
2486 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2487 struct ptlrpc_request **request)
2489 struct obd_device *obd = exp->exp_obd;
2490 struct lmv_obd *lmv = &obd->u.lmv;
2491 struct lmv_tgt_desc *tgt = NULL;
2492 struct lmv_tgt_desc *parent_tgt = NULL;
2493 struct mdt_body *body;
2497 rc = lmv_check_connect(obd);
2501 /* Send unlink requests to the MDT where the child is located */
2502 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2503 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2505 RETURN(PTR_ERR(tgt));
2507 /* For striped dir, we need to locate the parent as well */
2508 if (op_data->op_mea1 != NULL) {
2509 struct lmv_tgt_desc *tmp;
2511 LASSERT(op_data->op_name != NULL &&
2512 op_data->op_namelen != 0);
2513 tmp = lmv_locate_target_for_name(lmv,
2516 op_data->op_namelen,
2520 RETURN(PTR_ERR(tmp));
2523 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2525 RETURN(PTR_ERR(tgt));
2528 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2529 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2530 op_data->op_cap = cfs_curproc_cap_pack();
2533 * If child's fid is given, cancel unused locks for it if it is from
2534 * another export than parent.
2536 * LOOKUP lock for child (fid3) should also be cancelled on parent
2537 * tgt_tgt in mdc_unlink().
2539 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2542 * Cancel FULL locks on child (fid3).
2544 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2545 if (IS_ERR(parent_tgt))
2546 RETURN(PTR_ERR(parent_tgt));
2548 if (parent_tgt != tgt) {
2549 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2550 LCK_EX, MDS_INODELOCK_LOOKUP,
2551 MF_MDC_CANCEL_FID3);
2554 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2555 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2559 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2560 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2562 rc = md_unlink(tgt->ltd_exp, op_data, request);
2563 if (rc != 0 && rc != -EREMOTE)
2566 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2570 /* Not cross-ref case, just get out of here. */
2571 if (likely(!(body->valid & OBD_MD_MDS)))
2574 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2575 exp->exp_obd->obd_name, PFID(&body->fid1));
2577 /* This is a remote object, try remote MDT, Note: it may
2578 * try more than 1 time here, Considering following case
2579 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2580 * 1. Initially A does not know where remote1 is, it send
2581 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2582 * resend unlink RPC to MDT1 (retry 1st time).
2584 * 2. During the unlink RPC in flight,
2585 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2586 * and create new remote1, but on MDT0
2588 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2589 * /mnt/lustre, then lookup get fid of remote1, and find
2590 * it is remote dir again, and replay -EREMOTE again.
2592 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2594 * In theory, it might try unlimited time here, but it should
2595 * be very rare case. */
2596 op_data->op_fid2 = body->fid1;
2597 ptlrpc_req_finished(*request);
2603 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2605 struct lmv_obd *lmv = &obd->u.lmv;
2609 case OBD_CLEANUP_EARLY:
2610 /* XXX: here should be calling obd_precleanup() down to
2613 case OBD_CLEANUP_EXPORTS:
2614 fld_client_proc_fini(&lmv->lmv_fld);
2615 lprocfs_obd_cleanup(obd);
2616 lprocfs_free_md_stats(obd);
2624 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2625 __u32 keylen, void *key, __u32 *vallen, void *val,
2626 struct lov_stripe_md *lsm)
2628 struct obd_device *obd;
2629 struct lmv_obd *lmv;
2633 obd = class_exp2obd(exp);
2635 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2636 exp->exp_handle.h_cookie);
2641 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2644 rc = lmv_check_connect(obd);
2648 LASSERT(*vallen == sizeof(__u32));
2649 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2650 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2652 * All tgts should be connected when this gets called.
2654 if (tgt == NULL || tgt->ltd_exp == NULL)
2657 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2662 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2663 KEY_IS(KEY_DEFAULT_EASIZE) ||
2664 KEY_IS(KEY_MAX_COOKIESIZE) ||
2665 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2666 KEY_IS(KEY_CONN_DATA)) {
2667 rc = lmv_check_connect(obd);
2672 * Forwarding this request to first MDS, it should know LOV
2675 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2677 if (!rc && KEY_IS(KEY_CONN_DATA))
2678 exp->exp_connect_data = *(struct obd_connect_data *)val;
2680 } else if (KEY_IS(KEY_TGT_COUNT)) {
2681 *((int *)val) = lmv->desc.ld_tgt_count;
2685 CDEBUG(D_IOCTL, "Invalid key\n");
2689 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2690 obd_count keylen, void *key, obd_count vallen,
2691 void *val, struct ptlrpc_request_set *set)
2693 struct lmv_tgt_desc *tgt = NULL;
2694 struct obd_device *obd;
2695 struct lmv_obd *lmv;
2699 obd = class_exp2obd(exp);
2701 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2702 exp->exp_handle.h_cookie);
2707 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2710 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2713 if (tgt == NULL || tgt->ltd_exp == NULL)
2716 err = obd_set_info_async(env, tgt->ltd_exp,
2717 keylen, key, vallen, val, set);
2728 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2729 struct lmv_mds_md_v1 *lmm1)
2734 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2735 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2736 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2737 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2738 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2739 sizeof(lmm1->lmv_pool_name));
2740 if (cplen >= sizeof(lmm1->lmv_pool_name))
2743 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2744 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2745 &lsm->lsm_md_oinfo[i].lmo_fid);
2749 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2753 bool allocated = false;
2757 LASSERT(lmmp != NULL);
2759 if (*lmmp != NULL && lsm == NULL) {
2762 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2763 lmm_size = lmv_mds_md_size(stripe_count,
2764 le32_to_cpu((*lmmp)->lmv_magic));
2767 OBD_FREE(*lmmp, lmm_size);
2773 if (*lmmp == NULL && lsm == NULL) {
2774 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2775 LASSERT(lmm_size > 0);
2776 OBD_ALLOC(*lmmp, lmm_size);
2779 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2780 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2785 LASSERT(lsm != NULL);
2786 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2787 if (*lmmp == NULL) {
2788 OBD_ALLOC(*lmmp, lmm_size);
2794 switch (lsm->lsm_md_magic) {
2796 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2803 if (rc != 0 && allocated) {
2804 OBD_FREE(*lmmp, lmm_size);
2810 EXPORT_SYMBOL(lmv_pack_md);
2812 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2813 const struct lmv_mds_md_v1 *lmm1)
2815 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2822 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2823 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2824 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2825 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2826 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2827 fid_le_to_cpu(&lsm->lsm_md_master_fid, &lmm1->lmv_master_fid);
2828 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2829 sizeof(lsm->lsm_md_pool_name));
2831 if (!fid_is_sane(&lsm->lsm_md_master_fid))
2834 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2837 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2838 "layout_version %d\n", lsm->lsm_md_stripe_count,
2839 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2840 lsm->lsm_md_layout_version);
2842 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2843 for (i = 0; i < le32_to_cpu(stripe_count); i++) {
2844 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2845 &lmm1->lmv_stripe_fids[i]);
2846 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2847 &lsm->lsm_md_oinfo[i].lmo_mds);
2850 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2851 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2857 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2858 const union lmv_mds_md *lmm, int stripe_count)
2860 struct lmv_stripe_md *lsm;
2863 bool allocated = false;
2866 LASSERT(lsmp != NULL);
2870 if (lsm != NULL && lmm == NULL) {
2873 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2874 /* For migrating inode, the master stripe and master
2875 * object will be the same, so do not need iput, see
2876 * ll_update_lsm_md */
2877 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2878 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2879 iput(lsm->lsm_md_oinfo[i].lmo_root);
2882 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2883 OBD_FREE(lsm, lsm_size);
2889 if (lsm == NULL && lmm == NULL) {
2890 lsm_size = lmv_stripe_md_size(stripe_count);
2891 OBD_ALLOC(lsm, lsm_size);
2894 lsm->lsm_md_stripe_count = stripe_count;
2900 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2901 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2902 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2903 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2908 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2909 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2912 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2913 * stripecount should be 0 then.
2915 lsm_size = lmv_stripe_md_size(0);
2917 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2919 OBD_ALLOC(lsm, lsm_size);
2926 switch (le32_to_cpu(lmm->lmv_magic)) {
2928 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2931 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2932 le32_to_cpu(lmm->lmv_magic));
2937 if (rc != 0 && allocated) {
2938 OBD_FREE(lsm, lsm_size);
2945 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
2947 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
2949 EXPORT_SYMBOL(lmv_alloc_memmd);
2951 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2953 lmv_unpack_md(NULL, &lsm, NULL, 0);
2955 EXPORT_SYMBOL(lmv_free_memmd);
2957 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2958 struct lov_mds_md *lmm, int disk_len)
2960 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
2961 (union lmv_mds_md *)lmm, disk_len);
2964 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2965 struct lov_stripe_md *lsm)
2967 struct obd_device *obd = exp->exp_obd;
2968 struct lmv_obd *lmv_obd = &obd->u.lmv;
2969 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
2974 stripe_count = lmv->lsm_md_stripe_count;
2976 stripe_count = lmv_obd->desc.ld_tgt_count;
2978 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
2981 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
2984 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2985 ldlm_policy_data_t *policy, ldlm_mode_t mode,
2986 ldlm_cancel_flags_t flags, void *opaque)
2988 struct obd_device *obd = exp->exp_obd;
2989 struct lmv_obd *lmv = &obd->u.lmv;
2995 LASSERT(fid != NULL);
2997 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2998 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3000 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3003 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3011 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3014 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3015 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3019 if (tgt == NULL || tgt->ltd_exp == NULL)
3021 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3025 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3026 const struct lu_fid *fid, ldlm_type_t type,
3027 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3028 struct lustre_handle *lockh)
3030 struct obd_device *obd = exp->exp_obd;
3031 struct lmv_obd *lmv = &obd->u.lmv;
3036 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3039 * With CMD every object can have two locks in different namespaces:
3040 * lookup lock in space of mds storing direntry and update/open lock in
3041 * space of mds storing inode. Thus we check all targets, not only that
3042 * one fid was created in.
3044 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3045 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3047 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3050 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3059 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3060 struct obd_export *dt_exp, struct obd_export *md_exp,
3061 struct lustre_md *md)
3063 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3064 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3066 if (tgt == NULL || tgt->ltd_exp == NULL)
3069 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3072 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3074 struct obd_device *obd = exp->exp_obd;
3075 struct lmv_obd *lmv = &obd->u.lmv;
3076 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3079 if (md->lmv != NULL) {
3080 lmv_free_memmd(md->lmv);
3083 if (tgt == NULL || tgt->ltd_exp == NULL)
3085 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3088 int lmv_set_open_replay_data(struct obd_export *exp,
3089 struct obd_client_handle *och,
3090 struct lookup_intent *it)
3092 struct obd_device *obd = exp->exp_obd;
3093 struct lmv_obd *lmv = &obd->u.lmv;
3094 struct lmv_tgt_desc *tgt;
3097 tgt = lmv_find_target(lmv, &och->och_fid);
3099 RETURN(PTR_ERR(tgt));
3101 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3104 int lmv_clear_open_replay_data(struct obd_export *exp,
3105 struct obd_client_handle *och)
3107 struct obd_device *obd = exp->exp_obd;
3108 struct lmv_obd *lmv = &obd->u.lmv;
3109 struct lmv_tgt_desc *tgt;
3112 tgt = lmv_find_target(lmv, &och->och_fid);
3114 RETURN(PTR_ERR(tgt));
3116 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3119 static int lmv_get_remote_perm(struct obd_export *exp,
3120 const struct lu_fid *fid,
3121 struct obd_capa *oc, __u32 suppgid,
3122 struct ptlrpc_request **request)
3124 struct obd_device *obd = exp->exp_obd;
3125 struct lmv_obd *lmv = &obd->u.lmv;
3126 struct lmv_tgt_desc *tgt;
3130 rc = lmv_check_connect(obd);
3134 tgt = lmv_find_target(lmv, fid);
3136 RETURN(PTR_ERR(tgt));
3138 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3142 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3145 struct obd_device *obd = exp->exp_obd;
3146 struct lmv_obd *lmv = &obd->u.lmv;
3147 struct lmv_tgt_desc *tgt;
3151 rc = lmv_check_connect(obd);
3155 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3157 RETURN(PTR_ERR(tgt));
3159 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3163 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3164 const struct req_msg_field *field, struct obd_capa **oc)
3166 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3167 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3169 if (tgt == NULL || tgt->ltd_exp == NULL)
3171 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3174 int lmv_intent_getattr_async(struct obd_export *exp,
3175 struct md_enqueue_info *minfo,
3176 struct ldlm_enqueue_info *einfo)
3178 struct md_op_data *op_data = &minfo->mi_data;
3179 struct obd_device *obd = exp->exp_obd;
3180 struct lmv_obd *lmv = &obd->u.lmv;
3181 struct lmv_tgt_desc *tgt = NULL;
3185 rc = lmv_check_connect(obd);
3189 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3191 RETURN(PTR_ERR(tgt));
3193 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3197 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3198 struct lu_fid *fid, __u64 *bits)
3200 struct obd_device *obd = exp->exp_obd;
3201 struct lmv_obd *lmv = &obd->u.lmv;
3202 struct lmv_tgt_desc *tgt;
3206 rc = lmv_check_connect(obd);
3210 tgt = lmv_find_target(lmv, fid);
3212 RETURN(PTR_ERR(tgt));
3214 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3219 * For lmv, only need to send request to master MDT, and the master MDT will
3220 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3221 * we directly fetch data from the slave MDTs.
3223 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3224 struct obd_quotactl *oqctl)
3226 struct obd_device *obd = class_exp2obd(exp);
3227 struct lmv_obd *lmv = &obd->u.lmv;
3228 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3231 __u64 curspace, curinodes;
3235 tgt->ltd_exp == NULL ||
3237 lmv->desc.ld_tgt_count == 0) {
3238 CERROR("master lmv inactive\n");
3242 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3243 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3247 curspace = curinodes = 0;
3248 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3252 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3255 err = obd_quotactl(tgt->ltd_exp, oqctl);
3257 CERROR("getquota on mdt %d failed. %d\n", i, err);
3261 curspace += oqctl->qc_dqblk.dqb_curspace;
3262 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3265 oqctl->qc_dqblk.dqb_curspace = curspace;
3266 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3271 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3272 struct obd_quotactl *oqctl)
3274 struct obd_device *obd = class_exp2obd(exp);
3275 struct lmv_obd *lmv = &obd->u.lmv;
3276 struct lmv_tgt_desc *tgt;
3281 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3284 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3285 CERROR("lmv idx %d inactive\n", i);
3289 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3297 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3298 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3300 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3303 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3304 struct cl_attr *attr)
3309 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3310 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3312 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3313 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3314 i_size_read(inode), inode->i_nlink,
3315 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3316 LTIME_S(inode->i_mtime));
3318 /* for slave stripe, it needs to subtract nlink for . and .. */
3320 attr->cat_nlink += inode->i_nlink - 2;
3322 attr->cat_nlink = inode->i_nlink;
3324 attr->cat_size += i_size_read(inode);
3326 if (attr->cat_atime < LTIME_S(inode->i_atime))
3327 attr->cat_atime = LTIME_S(inode->i_atime);
3329 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3330 attr->cat_ctime = LTIME_S(inode->i_ctime);
3332 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3333 attr->cat_mtime = LTIME_S(inode->i_mtime);
3339 struct obd_ops lmv_obd_ops = {
3340 .o_owner = THIS_MODULE,
3341 .o_setup = lmv_setup,
3342 .o_cleanup = lmv_cleanup,
3343 .o_precleanup = lmv_precleanup,
3344 .o_process_config = lmv_process_config,
3345 .o_connect = lmv_connect,
3346 .o_disconnect = lmv_disconnect,
3347 .o_statfs = lmv_statfs,
3348 .o_get_info = lmv_get_info,
3349 .o_set_info_async = lmv_set_info_async,
3350 .o_packmd = lmv_packmd,
3351 .o_unpackmd = lmv_unpackmd,
3352 .o_notify = lmv_notify,
3353 .o_get_uuid = lmv_get_uuid,
3354 .o_iocontrol = lmv_iocontrol,
3355 .o_quotacheck = lmv_quotacheck,
3356 .o_quotactl = lmv_quotactl
3359 struct md_ops lmv_md_ops = {
3360 .m_getstatus = lmv_getstatus,
3361 .m_null_inode = lmv_null_inode,
3362 .m_find_cbdata = lmv_find_cbdata,
3363 .m_close = lmv_close,
3364 .m_create = lmv_create,
3365 .m_done_writing = lmv_done_writing,
3366 .m_enqueue = lmv_enqueue,
3367 .m_getattr = lmv_getattr,
3368 .m_getxattr = lmv_getxattr,
3369 .m_getattr_name = lmv_getattr_name,
3370 .m_intent_lock = lmv_intent_lock,
3372 .m_rename = lmv_rename,
3373 .m_setattr = lmv_setattr,
3374 .m_setxattr = lmv_setxattr,
3375 .m_fsync = lmv_fsync,
3376 .m_read_entry = lmv_read_entry,
3377 .m_unlink = lmv_unlink,
3378 .m_init_ea_size = lmv_init_ea_size,
3379 .m_cancel_unused = lmv_cancel_unused,
3380 .m_set_lock_data = lmv_set_lock_data,
3381 .m_lock_match = lmv_lock_match,
3382 .m_get_lustre_md = lmv_get_lustre_md,
3383 .m_free_lustre_md = lmv_free_lustre_md,
3384 .m_update_lsm_md = lmv_update_lsm_md,
3385 .m_merge_attr = lmv_merge_attr,
3386 .m_set_open_replay_data = lmv_set_open_replay_data,
3387 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3388 .m_renew_capa = lmv_renew_capa,
3389 .m_unpack_capa = lmv_unpack_capa,
3390 .m_get_remote_perm = lmv_get_remote_perm,
3391 .m_intent_getattr_async = lmv_intent_getattr_async,
3392 .m_revalidate_lock = lmv_revalidate_lock
3395 int __init lmv_init(void)
3397 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3398 #ifndef HAVE_ONLY_PROCFS_SEQ
3401 LUSTRE_LMV_NAME, NULL);
3405 static void lmv_exit(void)
3407 class_unregister_type(LUSTRE_LMV_NAME);
3410 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3411 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3412 MODULE_LICENSE("GPL");
3414 module_init(lmv_init);
3415 module_exit(lmv_exit);