4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
38 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/slab.h>
42 #include <linux/pagemap.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
48 #include <lustre/lustre_idl.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
57 #include <lustre_fid.h>
58 #include <lustre_ioctl.h>
59 #include "lmv_internal.h"
61 /* This hash is only for testing purpose */
62 static inline unsigned int
63 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
66 const unsigned char *p = (const unsigned char *)name;
68 while (--namelen >= 0)
76 static inline unsigned int
77 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
81 hash = lustre_hash_fnv_1a_64(name, namelen);
88 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
89 const char *name, int namelen)
92 __u32 hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
95 if (stripe_count <= 1)
98 /* for migrating object, always start from 0 stripe */
99 if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
103 case LMV_HASH_TYPE_ALL_CHARS:
104 idx = lmv_hash_all_chars(stripe_count, name, namelen);
106 case LMV_HASH_TYPE_FNV_1A_64:
107 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
114 CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
120 static void lmv_activate_target(struct lmv_obd *lmv,
121 struct lmv_tgt_desc *tgt,
124 if (tgt->ltd_active == activate)
127 tgt->ltd_active = activate;
128 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
134 * -EINVAL : UUID can't be found in the LMV's target list
135 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
136 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
138 static int lmv_set_mdc_active(struct lmv_obd *lmv,
139 const struct obd_uuid *uuid,
142 struct lmv_tgt_desc *tgt = NULL;
143 struct obd_device *obd;
148 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
149 lmv, uuid->uuid, activate);
151 spin_lock(&lmv->lmv_lock);
152 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
154 if (tgt == NULL || tgt->ltd_exp == NULL)
157 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
158 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
160 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
164 if (i == lmv->desc.ld_tgt_count)
165 GOTO(out_lmv_lock, rc = -EINVAL);
167 obd = class_exp2obd(tgt->ltd_exp);
169 GOTO(out_lmv_lock, rc = -ENOTCONN);
171 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
172 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
173 obd->obd_type->typ_name, i);
174 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
176 if (tgt->ltd_active == activate) {
177 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
178 activate ? "" : "in");
179 GOTO(out_lmv_lock, rc);
182 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
183 activate ? "" : "in");
184 lmv_activate_target(lmv, tgt, activate);
188 spin_unlock(&lmv->lmv_lock);
192 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
194 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
195 struct lmv_tgt_desc *tgt = lmv->tgts[0];
197 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
200 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
201 enum obd_notify_event ev, void *data)
203 struct obd_connect_data *conn_data;
204 struct lmv_obd *lmv = &obd->u.lmv;
205 struct obd_uuid *uuid;
209 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
210 CERROR("unexpected notification of %s %s!\n",
211 watched->obd_type->typ_name,
216 uuid = &watched->u.cli.cl_target_uuid;
217 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
219 * Set MDC as active before notifying the observer, so the
220 * observer can use the MDC normally.
222 rc = lmv_set_mdc_active(lmv, uuid,
223 ev == OBD_NOTIFY_ACTIVE);
225 CERROR("%sactivation of %s failed: %d\n",
226 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
230 } else if (ev == OBD_NOTIFY_OCD) {
231 conn_data = &watched->u.cli.cl_import->imp_connect_data;
233 * XXX: Make sure that ocd_connect_flags from all targets are
234 * the same. Otherwise one of MDTs runs wrong version or
235 * something like this. --umka
237 obd->obd_self_export->exp_connect_data = *conn_data;
240 else if (ev == OBD_NOTIFY_DISCON) {
242 * For disconnect event, flush fld cache for failout MDS case.
244 fld_client_flush(&lmv->lmv_fld);
248 * Pass the notification up the chain.
250 if (obd->obd_observer)
251 rc = obd_notify(obd->obd_observer, watched, ev, data);
257 * This is fake connect function. Its purpose is to initialize lmv and say
258 * caller that everything is okay. Real connection will be performed later.
260 static int lmv_connect(const struct lu_env *env,
261 struct obd_export **exp, struct obd_device *obd,
262 struct obd_uuid *cluuid, struct obd_connect_data *data,
265 struct lmv_obd *lmv = &obd->u.lmv;
266 struct lustre_handle conn = { 0 };
271 * We don't want to actually do the underlying connections more than
272 * once, so keep track.
275 if (lmv->refcount > 1) {
280 rc = class_connect(&conn, obd, cluuid);
282 CERROR("class_connection() returned %d\n", rc);
286 *exp = class_conn2export(&conn);
287 class_export_get(*exp);
291 lmv->cluuid = *cluuid;
294 lmv->conn_data = *data;
296 if (lmv->targets_proc_entry == NULL) {
297 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
300 if (IS_ERR(lmv->targets_proc_entry)) {
301 CERROR("%s: cannot register "
302 "/proc/fs/lustre/%s/%s/target_obds\n",
303 obd->obd_name, obd->obd_type->typ_name,
305 lmv->targets_proc_entry = NULL;
310 * All real clients should perform actual connection right away, because
311 * it is possible, that LMV will not have opportunity to connect targets
312 * and MDC stuff will be called directly, for instance while reading
313 * ../mdc/../kbytesfree procfs file, etc.
315 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
316 rc = lmv_check_connect(obd);
318 if (rc && lmv->targets_proc_entry != NULL)
319 lprocfs_remove(&lmv->targets_proc_entry);
323 static void lmv_set_timeouts(struct obd_device *obd)
329 if (lmv->server_timeout == 0)
332 if (lmv->connected == 0)
335 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
336 struct lmv_tgt_desc *tgt = lmv->tgts[i];
338 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
341 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
342 KEY_INTERMDS, 0, NULL, NULL);
346 static int lmv_init_ea_size(struct obd_export *exp, int easize,
347 int def_easize, int cookiesize, int def_cookiesize)
349 struct obd_device *obd = exp->exp_obd;
350 struct lmv_obd *lmv = &obd->u.lmv;
356 if (lmv->max_easize < easize) {
357 lmv->max_easize = easize;
360 if (lmv->max_def_easize < def_easize) {
361 lmv->max_def_easize = def_easize;
364 if (lmv->max_cookiesize < cookiesize) {
365 lmv->max_cookiesize = cookiesize;
368 if (lmv->max_def_cookiesize < def_cookiesize) {
369 lmv->max_def_cookiesize = def_cookiesize;
375 if (lmv->connected == 0)
378 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
379 struct lmv_tgt_desc *tgt = lmv->tgts[i];
381 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
382 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
386 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
387 cookiesize, def_cookiesize);
389 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
390 " rc = %d\n", obd->obd_name, i, rc);
397 #define MAX_STRING_SIZE 128
399 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
401 struct lmv_obd *lmv = &obd->u.lmv;
402 struct obd_uuid *cluuid = &lmv->cluuid;
403 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
404 struct obd_device *mdc_obd;
405 struct obd_export *mdc_exp;
406 struct lu_fld_target target;
410 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
413 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
417 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
418 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
419 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
422 if (!mdc_obd->obd_set_up) {
423 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
427 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
428 &lmv->conn_data, NULL);
430 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
435 * Init fid sequence client for this mdc and add new fld target.
437 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
441 target.ft_srv = NULL;
442 target.ft_exp = mdc_exp;
443 target.ft_idx = tgt->ltd_idx;
445 fld_client_add_target(&lmv->lmv_fld, &target);
447 rc = obd_register_observer(mdc_obd, obd);
449 obd_disconnect(mdc_exp);
450 CERROR("target %s register_observer error %d\n",
451 tgt->ltd_uuid.uuid, rc);
455 if (obd->obd_observer) {
457 * Tell the observer about the new target.
459 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
461 (void *)(tgt - lmv->tgts[0]));
463 obd_disconnect(mdc_exp);
469 tgt->ltd_exp = mdc_exp;
470 lmv->desc.ld_active_tgt_count++;
472 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
473 lmv->max_cookiesize, lmv->max_def_cookiesize);
475 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
476 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
477 atomic_read(&obd->obd_refcount));
479 if (lmv->targets_proc_entry != NULL) {
480 struct proc_dir_entry *mdc_symlink;
482 LASSERT(mdc_obd->obd_type != NULL);
483 LASSERT(mdc_obd->obd_type->typ_name != NULL);
484 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
485 lmv->targets_proc_entry,
487 mdc_obd->obd_type->typ_name,
489 if (mdc_symlink == NULL) {
490 CERROR("cannot register LMV target "
491 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
492 obd->obd_type->typ_name, obd->obd_name,
499 static void lmv_del_target(struct lmv_obd *lmv, int index)
501 if (lmv->tgts[index] == NULL)
504 OBD_FREE_PTR(lmv->tgts[index]);
505 lmv->tgts[index] = NULL;
509 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
510 __u32 index, int gen)
512 struct lmv_obd *lmv = &obd->u.lmv;
513 struct lmv_tgt_desc *tgt;
514 int orig_tgt_count = 0;
518 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
522 if (lmv->desc.ld_tgt_count == 0) {
523 struct obd_device *mdc_obd;
525 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
528 lmv_init_unlock(lmv);
529 CERROR("%s: Target %s not attached: rc = %d\n",
530 obd->obd_name, uuidp->uuid, -EINVAL);
535 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
536 tgt = lmv->tgts[index];
537 CERROR("%s: UUID %s already assigned at LOV target index %d:"
538 " rc = %d\n", obd->obd_name,
539 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
540 lmv_init_unlock(lmv);
544 if (index >= lmv->tgts_size) {
545 /* We need to reallocate the lmv target array. */
546 struct lmv_tgt_desc **newtgts, **old = NULL;
550 while (newsize < index + 1)
551 newsize = newsize << 1;
552 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
553 if (newtgts == NULL) {
554 lmv_init_unlock(lmv);
558 if (lmv->tgts_size) {
559 memcpy(newtgts, lmv->tgts,
560 sizeof(*newtgts) * lmv->tgts_size);
562 oldsize = lmv->tgts_size;
566 lmv->tgts_size = newsize;
569 OBD_FREE(old, sizeof(*old) * oldsize);
571 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
577 lmv_init_unlock(lmv);
581 mutex_init(&tgt->ltd_fid_mutex);
582 tgt->ltd_idx = index;
583 tgt->ltd_uuid = *uuidp;
585 lmv->tgts[index] = tgt;
586 if (index >= lmv->desc.ld_tgt_count) {
587 orig_tgt_count = lmv->desc.ld_tgt_count;
588 lmv->desc.ld_tgt_count = index + 1;
591 if (lmv->connected) {
592 rc = lmv_connect_mdc(obd, tgt);
594 spin_lock(&lmv->lmv_lock);
595 if (lmv->desc.ld_tgt_count == index + 1)
596 lmv->desc.ld_tgt_count = orig_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 (lmv->targets_proc_entry != NULL)
704 lprocfs_remove_proc_entry(mdc_obd->obd_name,
705 lmv->targets_proc_entry);
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 (lmv->targets_proc_entry != NULL)
755 lprocfs_remove(&lmv->targets_proc_entry);
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 LL_IOC_FID2MDTIDX: {
1123 struct lu_fid *fid = karg;
1126 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1130 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1131 * point to user space memory for FID2MDTIDX. */
1132 *(__u32 *)uarg = mdt_index;
1135 case OBD_IOC_FID2PATH: {
1136 rc = lmv_fid2path(exp, len, karg, uarg);
1139 case LL_IOC_HSM_STATE_GET:
1140 case LL_IOC_HSM_STATE_SET:
1141 case LL_IOC_HSM_ACTION: {
1142 struct md_op_data *op_data = karg;
1144 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1146 RETURN(PTR_ERR(tgt));
1148 if (tgt->ltd_exp == NULL)
1151 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1154 case LL_IOC_HSM_PROGRESS: {
1155 const struct hsm_progress_kernel *hpk = karg;
1157 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1159 RETURN(PTR_ERR(tgt));
1160 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1163 case LL_IOC_HSM_REQUEST: {
1164 struct hsm_user_request *hur = karg;
1165 unsigned int reqcount = hur->hur_request.hr_itemcount;
1170 /* if the request is about a single fid
1171 * or if there is a single MDS, no need to split
1173 if (reqcount == 1 || count == 1) {
1174 tgt = lmv_find_target(lmv,
1175 &hur->hur_user_item[0].hui_fid);
1177 RETURN(PTR_ERR(tgt));
1178 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1180 /* split fid list to their respective MDS */
1181 for (i = 0; i < count; i++) {
1182 unsigned int nr, reqlen;
1184 struct hsm_user_request *req;
1187 if (tgt == NULL || tgt->ltd_exp == NULL)
1190 nr = lmv_hsm_req_count(lmv, hur, tgt);
1191 if (nr == 0) /* nothing for this MDS */
1194 /* build a request with fids for this MDS */
1195 reqlen = offsetof(typeof(*hur),
1197 + hur->hur_request.hr_data_len;
1198 OBD_ALLOC_LARGE(req, reqlen);
1202 lmv_hsm_req_build(lmv, hur, tgt, req);
1204 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1206 if (rc1 != 0 && rc == 0)
1208 OBD_FREE_LARGE(req, reqlen);
1213 case LL_IOC_LOV_SWAP_LAYOUTS: {
1214 struct md_op_data *op_data = karg;
1215 struct lmv_tgt_desc *tgt1, *tgt2;
1217 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1219 RETURN(PTR_ERR(tgt1));
1221 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1223 RETURN(PTR_ERR(tgt2));
1225 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1228 /* only files on same MDT can have their layouts swapped */
1229 if (tgt1->ltd_idx != tgt2->ltd_idx)
1232 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1235 case LL_IOC_HSM_CT_START: {
1236 struct lustre_kernelcomm *lk = karg;
1237 if (lk->lk_flags & LK_FLG_STOP)
1238 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1240 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1244 for (i = 0; i < count; i++) {
1245 struct obd_device *mdc_obd;
1249 if (tgt == NULL || tgt->ltd_exp == NULL)
1251 /* ll_umount_begin() sets force flag but for lmv, not
1252 * mdc. Let's pass it through */
1253 mdc_obd = class_exp2obd(tgt->ltd_exp);
1254 mdc_obd->obd_force = obddev->obd_force;
1255 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1256 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1259 if (tgt->ltd_active) {
1260 CERROR("error: iocontrol MDC %s on MDT"
1261 " idx %d cmd %x: err = %d\n",
1262 tgt->ltd_uuid.uuid, i, cmd, err);
1276 static int lmv_all_chars_policy(int count, const char *name,
1287 static int lmv_nid_policy(struct lmv_obd *lmv)
1289 struct obd_import *imp;
1293 * XXX: To get nid we assume that underlying obd device is mdc.
1295 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1296 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1297 return id % lmv->desc.ld_tgt_count;
1300 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1301 placement_policy_t placement)
1303 switch (placement) {
1304 case PLACEMENT_CHAR_POLICY:
1305 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1307 op_data->op_namelen);
1308 case PLACEMENT_NID_POLICY:
1309 return lmv_nid_policy(lmv);
1315 CERROR("Unsupported placement policy %x\n", placement);
1321 * This is _inode_ placement policy function (not name).
1323 static int lmv_placement_policy(struct obd_device *obd,
1324 struct md_op_data *op_data, u32 *mds)
1326 struct lmv_obd *lmv = &obd->u.lmv;
1329 LASSERT(mds != NULL);
1331 if (lmv->desc.ld_tgt_count == 1) {
1337 * If stripe_offset is provided during setdirstripe
1338 * (setdirstripe -i xx), xx MDS will be choosen.
1340 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1341 struct lmv_user_md *lum;
1343 lum = op_data->op_data;
1345 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1346 *mds = le32_to_cpu(lum->lum_stripe_offset);
1348 /* -1 means default, which will be in the same MDT with
1350 *mds = op_data->op_mds;
1351 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1354 /* Allocate new fid on target according to operation type and
1355 * parent home mds. */
1356 *mds = op_data->op_mds;
1362 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1364 struct lmv_tgt_desc *tgt;
1368 tgt = lmv_get_target(lmv, mds, NULL);
1370 RETURN(PTR_ERR(tgt));
1373 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1374 * on server that seq in new allocated fid is not yet known.
1376 mutex_lock(&tgt->ltd_fid_mutex);
1378 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1379 GOTO(out, rc = -ENODEV);
1382 * Asking underlying tgt layer to allocate new fid.
1384 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1386 LASSERT(fid_is_sane(fid));
1392 mutex_unlock(&tgt->ltd_fid_mutex);
1396 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1397 struct lu_fid *fid, struct md_op_data *op_data)
1399 struct obd_device *obd = class_exp2obd(exp);
1400 struct lmv_obd *lmv = &obd->u.lmv;
1405 LASSERT(op_data != NULL);
1406 LASSERT(fid != NULL);
1408 rc = lmv_placement_policy(obd, op_data, &mds);
1410 CERROR("Can't get target for allocating fid, "
1415 rc = __lmv_fid_alloc(lmv, fid, mds);
1417 CERROR("Can't alloc new fid, rc %d\n", rc);
1424 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1426 struct lmv_obd *lmv = &obd->u.lmv;
1427 struct lmv_desc *desc;
1431 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1432 CERROR("LMV setup requires a descriptor\n");
1436 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1437 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1438 CERROR("Lmv descriptor size wrong: %d > %d\n",
1439 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1443 lmv->tgts_size = 32U;
1444 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1445 if (lmv->tgts == NULL)
1448 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1449 lmv->desc.ld_tgt_count = 0;
1450 lmv->desc.ld_active_tgt_count = 0;
1451 lmv->max_cookiesize = 0;
1452 lmv->max_def_easize = 0;
1453 lmv->max_easize = 0;
1454 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1456 spin_lock_init(&lmv->lmv_lock);
1457 mutex_init(&lmv->init_mutex);
1460 obd->obd_vars = lprocfs_lmv_obd_vars;
1461 lprocfs_obd_setup(obd);
1462 lprocfs_alloc_md_stats(obd, 0);
1463 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1464 0444, &lmv_proc_target_fops, obd);
1466 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1469 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1470 LUSTRE_CLI_FLD_HASH_DHT);
1472 CERROR("Can't init FLD, err %d\n", rc);
1482 static int lmv_cleanup(struct obd_device *obd)
1484 struct lmv_obd *lmv = &obd->u.lmv;
1487 fld_client_fini(&lmv->lmv_fld);
1488 if (lmv->tgts != NULL) {
1490 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1491 if (lmv->tgts[i] == NULL)
1493 lmv_del_target(lmv, i);
1495 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1501 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1503 struct lustre_cfg *lcfg = buf;
1504 struct obd_uuid obd_uuid;
1510 switch (lcfg->lcfg_command) {
1512 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1513 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1514 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1515 GOTO(out, rc = -EINVAL);
1517 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1519 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1520 GOTO(out, rc = -EINVAL);
1521 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1522 GOTO(out, rc = -EINVAL);
1523 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1526 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1527 GOTO(out, rc = -EINVAL);
1533 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1534 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1536 struct obd_device *obd = class_exp2obd(exp);
1537 struct lmv_obd *lmv = &obd->u.lmv;
1538 struct obd_statfs *temp;
1543 rc = lmv_check_connect(obd);
1547 OBD_ALLOC(temp, sizeof(*temp));
1551 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1552 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1555 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1558 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1559 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1561 GOTO(out_free_temp, rc);
1566 /* If the statfs is from mount, it will needs
1567 * retrieve necessary information from MDT0.
1568 * i.e. mount does not need the merged osfs
1570 * And also clients can be mounted as long as
1571 * MDT0 is in service*/
1572 if (flags & OBD_STATFS_FOR_MDT0)
1573 GOTO(out_free_temp, rc);
1575 osfs->os_bavail += temp->os_bavail;
1576 osfs->os_blocks += temp->os_blocks;
1577 osfs->os_ffree += temp->os_ffree;
1578 osfs->os_files += temp->os_files;
1584 OBD_FREE(temp, sizeof(*temp));
1588 static int lmv_getstatus(struct obd_export *exp,
1590 struct obd_capa **pc)
1592 struct obd_device *obd = exp->exp_obd;
1593 struct lmv_obd *lmv = &obd->u.lmv;
1597 rc = lmv_check_connect(obd);
1601 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1605 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1606 struct obd_capa *oc, obd_valid valid, const char *name,
1607 const char *input, int input_size, int output_size,
1608 int flags, struct ptlrpc_request **request)
1610 struct obd_device *obd = exp->exp_obd;
1611 struct lmv_obd *lmv = &obd->u.lmv;
1612 struct lmv_tgt_desc *tgt;
1616 rc = lmv_check_connect(obd);
1620 tgt = lmv_find_target(lmv, fid);
1622 RETURN(PTR_ERR(tgt));
1624 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1625 input_size, output_size, flags, request);
1630 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1631 struct obd_capa *oc, obd_valid valid, const char *name,
1632 const char *input, int input_size, int output_size,
1633 int flags, __u32 suppgid,
1634 struct ptlrpc_request **request)
1636 struct obd_device *obd = exp->exp_obd;
1637 struct lmv_obd *lmv = &obd->u.lmv;
1638 struct lmv_tgt_desc *tgt;
1642 rc = lmv_check_connect(obd);
1646 tgt = lmv_find_target(lmv, fid);
1648 RETURN(PTR_ERR(tgt));
1650 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1651 input_size, output_size, flags, suppgid,
1657 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1658 struct ptlrpc_request **request)
1660 struct obd_device *obd = exp->exp_obd;
1661 struct lmv_obd *lmv = &obd->u.lmv;
1662 struct lmv_tgt_desc *tgt;
1666 rc = lmv_check_connect(obd);
1670 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1672 RETURN(PTR_ERR(tgt));
1674 if (op_data->op_flags & MF_GET_MDT_IDX) {
1675 op_data->op_mds = tgt->ltd_idx;
1679 rc = md_getattr(tgt->ltd_exp, op_data, request);
1684 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1686 struct obd_device *obd = exp->exp_obd;
1687 struct lmv_obd *lmv = &obd->u.lmv;
1692 rc = lmv_check_connect(obd);
1696 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1699 * With DNE every object can have two locks in different namespaces:
1700 * lookup lock in space of MDT storing direntry and update/open lock in
1701 * space of MDT storing inode.
1703 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1704 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1706 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1712 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1713 ldlm_iterator_t it, void *data)
1715 struct obd_device *obd = exp->exp_obd;
1716 struct lmv_obd *lmv = &obd->u.lmv;
1722 rc = lmv_check_connect(obd);
1726 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1729 * With DNE every object can have two locks in different namespaces:
1730 * lookup lock in space of MDT storing direntry and update/open lock in
1731 * space of MDT storing inode. Try the MDT that the FID maps to first,
1732 * since this can be easily found, and only try others if that fails.
1734 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1735 i < lmv->desc.ld_tgt_count;
1736 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1738 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1739 obd->obd_name, PFID(fid), tgt);
1743 if (lmv->tgts[tgt] == NULL ||
1744 lmv->tgts[tgt]->ltd_exp == NULL)
1747 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1756 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1757 struct md_open_data *mod, struct ptlrpc_request **request)
1759 struct obd_device *obd = exp->exp_obd;
1760 struct lmv_obd *lmv = &obd->u.lmv;
1761 struct lmv_tgt_desc *tgt;
1765 rc = lmv_check_connect(obd);
1769 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1771 RETURN(PTR_ERR(tgt));
1773 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1774 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1779 * Choosing the MDT by name or FID in @op_data.
1780 * For non-striped directory, it will locate MDT by fid.
1781 * For striped-directory, it will locate MDT by name. And also
1782 * it will reset op_fid1 with the FID of the choosen stripe.
1784 struct lmv_tgt_desc *
1785 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1786 const char *name, int namelen, struct lu_fid *fid,
1789 struct lmv_tgt_desc *tgt;
1790 const struct lmv_oinfo *oinfo;
1792 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1794 RETURN(ERR_CAST(oinfo));
1795 *fid = oinfo->lmo_fid;
1796 *mds = oinfo->lmo_mds;
1797 tgt = lmv_get_target(lmv, *mds, NULL);
1799 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1804 * Locate mds by fid or name
1806 * For striped directory (lsm != NULL), it will locate the stripe
1807 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1808 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1809 * walk through all of stripes to locate the entry.
1811 * For normal direcotry, it will locate MDS by FID directly.
1812 * \param[in] lmv LMV device
1813 * \param[in] op_data client MD stack parameters, name, namelen
1815 * \param[in] fid object FID used to locate MDS.
1817 * retval pointer to the lmv_tgt_desc if succeed.
1818 * ERR_PTR(errno) if failed.
1820 struct lmv_tgt_desc*
1821 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1824 struct lmv_stripe_md *lsm = op_data->op_mea1;
1825 struct lmv_tgt_desc *tgt;
1827 /* During creating VOLATILE file, it should honor the mdt
1828 * index if the file under striped dir is being restored, see
1830 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1831 (int)op_data->op_mds != -1 && lsm != NULL) {
1833 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1837 /* refill the right parent fid */
1838 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1839 struct lmv_oinfo *oinfo;
1841 oinfo = &lsm->lsm_md_oinfo[i];
1842 if (oinfo->lmo_mds == op_data->op_mds) {
1843 *fid = oinfo->lmo_fid;
1848 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1849 if (i == lsm->lsm_md_stripe_count)
1850 tgt = ERR_PTR(-EINVAL);
1855 if (lsm == NULL || op_data->op_namelen == 0) {
1856 tgt = lmv_find_target(lmv, fid);
1860 op_data->op_mds = tgt->ltd_idx;
1864 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1865 op_data->op_namelen, fid,
1869 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1870 const void *data, int datalen, int mode, __u32 uid,
1871 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1872 struct ptlrpc_request **request)
1874 struct obd_device *obd = exp->exp_obd;
1875 struct lmv_obd *lmv = &obd->u.lmv;
1876 struct lmv_tgt_desc *tgt;
1880 rc = lmv_check_connect(obd);
1884 if (!lmv->desc.ld_active_tgt_count)
1887 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1889 RETURN(PTR_ERR(tgt));
1891 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1892 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1895 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1899 /* Send the create request to the MDT where the object
1900 * will be located */
1901 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1903 RETURN(PTR_ERR(tgt));
1905 op_data->op_mds = tgt->ltd_idx;
1907 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1908 PFID(&op_data->op_fid2), op_data->op_mds);
1910 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1911 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1912 cap_effective, rdev, request);
1914 if (*request == NULL)
1916 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1921 static int lmv_done_writing(struct obd_export *exp,
1922 struct md_op_data *op_data,
1923 struct md_open_data *mod)
1925 struct obd_device *obd = exp->exp_obd;
1926 struct lmv_obd *lmv = &obd->u.lmv;
1927 struct lmv_tgt_desc *tgt;
1931 rc = lmv_check_connect(obd);
1935 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1937 RETURN(PTR_ERR(tgt));
1939 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1944 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1945 const union ldlm_policy_data *policy,
1946 struct lookup_intent *it, struct md_op_data *op_data,
1947 struct lustre_handle *lockh, __u64 extra_lock_flags)
1949 struct obd_device *obd = exp->exp_obd;
1950 struct lmv_obd *lmv = &obd->u.lmv;
1951 struct lmv_tgt_desc *tgt;
1955 rc = lmv_check_connect(obd);
1959 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1960 LL_IT2STR(it), PFID(&op_data->op_fid1));
1962 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1964 RETURN(PTR_ERR(tgt));
1966 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1967 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1969 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1976 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1977 struct ptlrpc_request **preq)
1979 struct ptlrpc_request *req = NULL;
1980 struct obd_device *obd = exp->exp_obd;
1981 struct lmv_obd *lmv = &obd->u.lmv;
1982 struct lmv_tgt_desc *tgt;
1983 struct mdt_body *body;
1987 rc = lmv_check_connect(obd);
1991 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1993 RETURN(PTR_ERR(tgt));
1995 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%u\n",
1996 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1999 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2003 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2004 LASSERT(body != NULL);
2006 if (body->mbo_valid & OBD_MD_MDS) {
2007 struct lu_fid rid = body->mbo_fid1;
2008 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2011 tgt = lmv_find_target(lmv, &rid);
2013 ptlrpc_req_finished(*preq);
2015 RETURN(PTR_ERR(tgt));
2018 op_data->op_fid1 = rid;
2019 op_data->op_valid |= OBD_MD_FLCROSSREF;
2020 op_data->op_namelen = 0;
2021 op_data->op_name = NULL;
2022 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2023 ptlrpc_req_finished(*preq);
2030 #define md_op_data_fid(op_data, fl) \
2031 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2032 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2033 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2034 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2037 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2038 struct md_op_data *op_data,
2039 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
2041 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2042 struct obd_device *obd = exp->exp_obd;
2043 struct lmv_obd *lmv = &obd->u.lmv;
2044 ldlm_policy_data_t policy = {{ 0 }};
2048 if (!fid_is_sane(fid))
2052 tgt = lmv_find_target(lmv, fid);
2054 RETURN(PTR_ERR(tgt));
2057 if (tgt->ltd_idx != op_tgt) {
2058 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2059 policy.l_inodebits.bits = bits;
2060 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2061 mode, LCF_ASYNC, NULL);
2064 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2066 op_data->op_flags |= flag;
2074 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2077 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2078 struct ptlrpc_request **request)
2080 struct obd_device *obd = exp->exp_obd;
2081 struct lmv_obd *lmv = &obd->u.lmv;
2082 struct lmv_tgt_desc *tgt;
2086 rc = lmv_check_connect(obd);
2090 LASSERT(op_data->op_namelen != 0);
2092 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2093 PFID(&op_data->op_fid2), op_data->op_namelen,
2094 op_data->op_name, PFID(&op_data->op_fid1));
2096 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2097 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2098 op_data->op_cap = cfs_curproc_cap_pack();
2099 if (op_data->op_mea2 != NULL) {
2100 struct lmv_stripe_md *lsm = op_data->op_mea2;
2101 const struct lmv_oinfo *oinfo;
2103 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2104 op_data->op_namelen);
2106 RETURN(PTR_ERR(oinfo));
2108 op_data->op_fid2 = oinfo->lmo_fid;
2111 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2113 RETURN(PTR_ERR(tgt));
2116 * Cancel UPDATE lock on child (fid1).
2118 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2119 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2120 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2124 rc = md_link(tgt->ltd_exp, op_data, request);
2129 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2130 const char *old, int oldlen, const char *new, int newlen,
2131 struct ptlrpc_request **request)
2133 struct obd_device *obd = exp->exp_obd;
2134 struct lmv_obd *lmv = &obd->u.lmv;
2135 struct lmv_tgt_desc *src_tgt;
2139 LASSERT(oldlen != 0);
2141 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2142 oldlen, old, PFID(&op_data->op_fid1),
2143 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2144 newlen, new, PFID(&op_data->op_fid2),
2145 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2147 rc = lmv_check_connect(obd);
2151 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2152 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2153 op_data->op_cap = cfs_curproc_cap_pack();
2154 if (op_data->op_cli_flags & CLI_MIGRATE) {
2155 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2156 PFID(&op_data->op_fid3));
2157 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2160 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2162 if (op_data->op_mea1 != NULL) {
2163 struct lmv_stripe_md *lsm = op_data->op_mea1;
2165 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2169 if (IS_ERR(src_tgt))
2170 RETURN(PTR_ERR(src_tgt));
2172 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2173 if (IS_ERR(src_tgt))
2174 RETURN(PTR_ERR(src_tgt));
2176 op_data->op_mds = src_tgt->ltd_idx;
2179 if (op_data->op_mea2) {
2180 struct lmv_stripe_md *lsm = op_data->op_mea2;
2181 const struct lmv_oinfo *oinfo;
2183 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2185 RETURN(PTR_ERR(oinfo));
2187 op_data->op_fid2 = oinfo->lmo_fid;
2190 if (IS_ERR(src_tgt))
2191 RETURN(PTR_ERR(src_tgt));
2194 * LOOKUP lock on src child (fid3) should also be cancelled for
2195 * src_tgt in mdc_rename.
2197 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2200 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2203 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2204 LCK_EX, MDS_INODELOCK_UPDATE,
2205 MF_MDC_CANCEL_FID2);
2210 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2212 if (fid_is_sane(&op_data->op_fid3)) {
2213 struct lmv_tgt_desc *tgt;
2215 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2217 RETURN(PTR_ERR(tgt));
2219 /* Cancel LOOKUP lock on its parent */
2220 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2221 LCK_EX, MDS_INODELOCK_LOOKUP,
2222 MF_MDC_CANCEL_FID3);
2226 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2227 LCK_EX, MDS_INODELOCK_FULL,
2228 MF_MDC_CANCEL_FID3);
2234 * Cancel all the locks on tgt child (fid4).
2236 if (fid_is_sane(&op_data->op_fid4))
2237 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2238 LCK_EX, MDS_INODELOCK_FULL,
2239 MF_MDC_CANCEL_FID4);
2241 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2242 op_data->op_mds, PFID(&op_data->op_fid2));
2244 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2250 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2251 void *ea, int ealen, void *ea2, int ea2len,
2252 struct ptlrpc_request **request,
2253 struct md_open_data **mod)
2255 struct obd_device *obd = exp->exp_obd;
2256 struct lmv_obd *lmv = &obd->u.lmv;
2257 struct lmv_tgt_desc *tgt;
2261 rc = lmv_check_connect(obd);
2265 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2266 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2268 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2269 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2271 RETURN(PTR_ERR(tgt));
2273 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2274 ea2len, request, mod);
2279 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2280 struct obd_capa *oc, struct ptlrpc_request **request)
2282 struct obd_device *obd = exp->exp_obd;
2283 struct lmv_obd *lmv = &obd->u.lmv;
2284 struct lmv_tgt_desc *tgt;
2288 rc = lmv_check_connect(obd);
2292 tgt = lmv_find_target(lmv, fid);
2294 RETURN(PTR_ERR(tgt));
2296 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2301 * Get current minimum entry from striped directory
2303 * This function will search the dir entry, whose hash value is the
2304 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2305 * only being called for striped directory.
2307 * \param[in] exp export of LMV
2308 * \param[in] op_data parameters transferred beween client MD stack
2309 * stripe_information will be included in this
2311 * \param[in] cb_op ldlm callback being used in enqueue in
2313 * \param[in] hash_offset the hash value, which is used to locate
2314 * minum(closet) dir entry
2315 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2316 * index of last entry, so to avoid hash conflict
2317 * between stripes. It will also be used to
2318 * return the stripe index of current dir entry.
2319 * \param[in|out] entp the minum entry and it also is being used
2320 * to input the last dir entry to resolve the
2323 * \param[out] ppage the page which holds the minum entry
2325 * \retval = 0 get the entry successfully
2326 * negative errno (< 0) does not get the entry
2328 static int lmv_get_min_striped_entry(struct obd_export *exp,
2329 struct md_op_data *op_data,
2330 struct md_callback *cb_op,
2331 __u64 hash_offset, int *stripe_offset,
2332 struct lu_dirent **entp,
2333 struct page **ppage)
2335 struct obd_device *obd = exp->exp_obd;
2336 struct lmv_obd *lmv = &obd->u.lmv;
2337 struct lmv_stripe_md *lsm = op_data->op_mea1;
2338 struct lmv_tgt_desc *tgt;
2340 struct lu_dirent *min_ent = NULL;
2341 struct page *min_page = NULL;
2347 stripe_count = lsm->lsm_md_stripe_count;
2348 for (i = 0; i < stripe_count; i++) {
2349 struct lu_dirent *ent = NULL;
2350 struct page *page = NULL;
2351 struct lu_dirpage *dp;
2352 __u64 stripe_hash = hash_offset;
2354 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2356 GOTO(out, rc = PTR_ERR(tgt));
2358 /* op_data will be shared by each stripe, so we need
2359 * reset these value for each stripe */
2360 op_data->op_stripe_offset = i;
2361 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2362 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2363 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2365 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2370 dp = page_address(page);
2371 for (ent = lu_dirent_start(dp); ent != NULL;
2372 ent = lu_dirent_next(ent)) {
2373 /* Skip dummy entry */
2374 if (le16_to_cpu(ent->lde_namelen) == 0)
2377 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2380 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2381 (*entp == ent || i < *stripe_offset))
2384 /* skip . and .. for other stripes */
2386 (strncmp(ent->lde_name, ".",
2387 le16_to_cpu(ent->lde_namelen)) == 0 ||
2388 strncmp(ent->lde_name, "..",
2389 le16_to_cpu(ent->lde_namelen)) == 0))
2395 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2398 page_cache_release(page);
2401 /* reach the end of current stripe, go to next stripe */
2402 if (stripe_hash == MDS_DIR_END_OFF)
2408 if (min_ent != NULL) {
2409 if (le64_to_cpu(min_ent->lde_hash) >
2410 le64_to_cpu(ent->lde_hash)) {
2413 page_cache_release(min_page);
2418 page_cache_release(page);
2429 if (*ppage != NULL) {
2431 page_cache_release(*ppage);
2433 *stripe_offset = min_idx;
2440 * Build dir entry page from a striped directory
2442 * This function gets one entry by @offset from a striped directory. It will
2443 * read entries from all of stripes, and choose one closest to the required
2444 * offset(&offset). A few notes
2445 * 1. skip . and .. for non-zero stripes, because there can only have one .
2446 * and .. in a directory.
2447 * 2. op_data will be shared by all of stripes, instead of allocating new
2448 * one, so need to restore before reusing.
2449 * 3. release the entry page if that is not being chosen.
2451 * \param[in] exp obd export refer to LMV
2452 * \param[in] op_data hold those MD parameters of read_entry
2453 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2454 * \param[out] ldp the entry being read
2455 * \param[out] ppage the page holding the entry. Note: because the entry
2456 * will be accessed in upper layer, so we need hold the
2457 * page until the usages of entry is finished, see
2458 * ll_dir_entry_next.
2460 * retval =0 if get entry successfully
2461 * <0 cannot get entry
2463 static int lmv_read_striped_page(struct obd_export *exp,
2464 struct md_op_data *op_data,
2465 struct md_callback *cb_op,
2466 __u64 offset, struct page **ppage)
2468 struct obd_device *obd = exp->exp_obd;
2469 struct lu_fid master_fid = op_data->op_fid1;
2470 struct inode *master_inode = op_data->op_data;
2471 __u64 hash_offset = offset;
2472 struct lu_dirpage *dp;
2473 struct page *min_ent_page = NULL;
2474 struct page *ent_page = NULL;
2475 struct lu_dirent *ent;
2478 struct lu_dirent *min_ent = NULL;
2479 struct lu_dirent *last_ent;
2484 rc = lmv_check_connect(obd);
2488 /* Allocate a page and read entries from all of stripes and fill
2489 * the page by hash order */
2490 ent_page = alloc_page(GFP_KERNEL);
2491 if (ent_page == NULL)
2494 /* Initialize the entry page */
2495 dp = kmap(ent_page);
2496 memset(dp, 0, sizeof(*dp));
2497 dp->ldp_hash_start = cpu_to_le64(offset);
2498 dp->ldp_flags |= LDF_COLLIDE;
2501 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2507 /* Find the minum entry from all sub-stripes */
2508 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2514 /* If it can not get minum entry, it means it already reaches
2515 * the end of this directory */
2516 if (min_ent == NULL) {
2517 last_ent->lde_reclen = 0;
2518 hash_offset = MDS_DIR_END_OFF;
2522 ent_size = le16_to_cpu(min_ent->lde_reclen);
2524 /* the last entry lde_reclen is 0, but it might not
2525 * the end of this entry of this temporay entry */
2527 ent_size = lu_dirent_calc_size(
2528 le16_to_cpu(min_ent->lde_namelen),
2529 le32_to_cpu(min_ent->lde_attrs));
2530 if (ent_size > left_bytes) {
2531 last_ent->lde_reclen = cpu_to_le16(0);
2532 hash_offset = le64_to_cpu(min_ent->lde_hash);
2536 memcpy(ent, min_ent, ent_size);
2538 /* Replace . with master FID and Replace .. with the parent FID
2539 * of master object */
2540 if (strncmp(ent->lde_name, ".",
2541 le16_to_cpu(ent->lde_namelen)) == 0 &&
2542 le16_to_cpu(ent->lde_namelen) == 1)
2543 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2544 else if (strncmp(ent->lde_name, "..",
2545 le16_to_cpu(ent->lde_namelen)) == 0 &&
2546 le16_to_cpu(ent->lde_namelen) == 2)
2547 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2549 left_bytes -= ent_size;
2550 ent->lde_reclen = cpu_to_le16(ent_size);
2552 ent = (void *)ent + ent_size;
2553 hash_offset = le64_to_cpu(min_ent->lde_hash);
2554 if (hash_offset == MDS_DIR_END_OFF) {
2555 last_ent->lde_reclen = 0;
2560 if (min_ent_page != NULL) {
2561 kunmap(min_ent_page);
2562 page_cache_release(min_ent_page);
2565 if (unlikely(rc != 0)) {
2566 __free_page(ent_page);
2570 dp->ldp_flags |= LDF_EMPTY;
2571 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2572 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2575 /* We do not want to allocate md_op_data during each
2576 * dir entry reading, so op_data will be shared by every stripe,
2577 * then we need to restore it back to original value before
2578 * return to the upper layer */
2579 op_data->op_fid1 = master_fid;
2580 op_data->op_fid2 = master_fid;
2581 op_data->op_data = master_inode;
2588 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2589 struct md_callback *cb_op, __u64 offset,
2590 struct page **ppage)
2592 struct obd_device *obd = exp->exp_obd;
2593 struct lmv_obd *lmv = &obd->u.lmv;
2594 struct lmv_stripe_md *lsm = op_data->op_mea1;
2595 struct lmv_tgt_desc *tgt;
2599 rc = lmv_check_connect(obd);
2603 if (unlikely(lsm != NULL)) {
2604 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2608 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2610 RETURN(PTR_ERR(tgt));
2612 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2618 * Unlink a file/directory
2620 * Unlink a file or directory under the parent dir. The unlink request
2621 * usually will be sent to the MDT where the child is located, but if
2622 * the client does not have the child FID then request will be sent to the
2623 * MDT where the parent is located.
2625 * If the parent is a striped directory then it also needs to locate which
2626 * stripe the name of the child is located, and replace the parent FID
2627 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2628 * it will walk through all of sub-stripes until the child is being
2631 * \param[in] exp export refer to LMV
2632 * \param[in] op_data different parameters transferred beween client
2633 * MD stacks, name, namelen, FIDs etc.
2634 * op_fid1 is the parent FID, op_fid2 is the child
2636 * \param[out] request point to the request of unlink.
2638 * retval 0 if succeed
2639 * negative errno if failed.
2641 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2642 struct ptlrpc_request **request)
2644 struct obd_device *obd = exp->exp_obd;
2645 struct lmv_obd *lmv = &obd->u.lmv;
2646 struct lmv_tgt_desc *tgt = NULL;
2647 struct lmv_tgt_desc *parent_tgt = NULL;
2648 struct mdt_body *body;
2650 int stripe_index = 0;
2651 struct lmv_stripe_md *lsm = op_data->op_mea1;
2654 rc = lmv_check_connect(obd);
2658 /* For striped dir, we need to locate the parent as well */
2660 struct lmv_tgt_desc *tmp;
2662 LASSERT(op_data->op_name != NULL &&
2663 op_data->op_namelen != 0);
2665 tmp = lmv_locate_target_for_name(lmv, lsm,
2667 op_data->op_namelen,
2671 /* return -EBADFD means unknown hash type, might
2672 * need try all sub-stripe here */
2673 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2674 RETURN(PTR_ERR(tmp));
2676 /* Note: both migrating dir and unknown hash dir need to
2677 * try all of sub-stripes, so we need start search the
2678 * name from stripe 0, but migrating dir is already handled
2679 * inside lmv_locate_target_for_name(), so we only check
2680 * unknown hash type directory here */
2681 if (!lmv_is_known_hash_type(lsm)) {
2682 struct lmv_oinfo *oinfo;
2684 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2686 op_data->op_fid1 = oinfo->lmo_fid;
2687 op_data->op_mds = oinfo->lmo_mds;
2692 /* Send unlink requests to the MDT where the child is located */
2693 if (likely(!fid_is_zero(&op_data->op_fid2)))
2694 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2695 else if (lsm != NULL)
2696 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2698 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2701 RETURN(PTR_ERR(tgt));
2703 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2704 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2705 op_data->op_cap = cfs_curproc_cap_pack();
2708 * If child's fid is given, cancel unused locks for it if it is from
2709 * another export than parent.
2711 * LOOKUP lock for child (fid3) should also be cancelled on parent
2712 * tgt_tgt in mdc_unlink().
2714 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2717 * Cancel FULL locks on child (fid3).
2719 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2720 if (IS_ERR(parent_tgt))
2721 RETURN(PTR_ERR(parent_tgt));
2723 if (parent_tgt != tgt) {
2724 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2725 LCK_EX, MDS_INODELOCK_LOOKUP,
2726 MF_MDC_CANCEL_FID3);
2729 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2730 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2734 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2735 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2737 rc = md_unlink(tgt->ltd_exp, op_data, request);
2738 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2741 /* Try next stripe if it is needed. */
2742 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2743 struct lmv_oinfo *oinfo;
2746 if (stripe_index >= lsm->lsm_md_stripe_count)
2749 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2751 op_data->op_fid1 = oinfo->lmo_fid;
2752 op_data->op_mds = oinfo->lmo_mds;
2754 ptlrpc_req_finished(*request);
2757 goto try_next_stripe;
2760 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2764 /* Not cross-ref case, just get out of here. */
2765 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2768 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2769 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2771 /* This is a remote object, try remote MDT, Note: it may
2772 * try more than 1 time here, Considering following case
2773 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2774 * 1. Initially A does not know where remote1 is, it send
2775 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2776 * resend unlink RPC to MDT1 (retry 1st time).
2778 * 2. During the unlink RPC in flight,
2779 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2780 * and create new remote1, but on MDT0
2782 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2783 * /mnt/lustre, then lookup get fid of remote1, and find
2784 * it is remote dir again, and replay -EREMOTE again.
2786 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2788 * In theory, it might try unlimited time here, but it should
2789 * be very rare case. */
2790 op_data->op_fid2 = body->mbo_fid1;
2791 ptlrpc_req_finished(*request);
2797 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2799 struct lmv_obd *lmv = &obd->u.lmv;
2803 case OBD_CLEANUP_EARLY:
2804 /* XXX: here should be calling obd_precleanup() down to
2807 case OBD_CLEANUP_EXPORTS:
2808 fld_client_proc_fini(&lmv->lmv_fld);
2809 lprocfs_obd_cleanup(obd);
2810 lprocfs_free_md_stats(obd);
2818 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2819 __u32 keylen, void *key, __u32 *vallen, void *val,
2820 struct lov_stripe_md *lsm)
2822 struct obd_device *obd;
2823 struct lmv_obd *lmv;
2827 obd = class_exp2obd(exp);
2829 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2830 exp->exp_handle.h_cookie);
2835 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2838 rc = lmv_check_connect(obd);
2842 LASSERT(*vallen == sizeof(__u32));
2843 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2844 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2846 * All tgts should be connected when this gets called.
2848 if (tgt == NULL || tgt->ltd_exp == NULL)
2851 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2856 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2857 KEY_IS(KEY_DEFAULT_EASIZE) ||
2858 KEY_IS(KEY_MAX_COOKIESIZE) ||
2859 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2860 KEY_IS(KEY_CONN_DATA)) {
2861 rc = lmv_check_connect(obd);
2866 * Forwarding this request to first MDS, it should know LOV
2869 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2871 if (!rc && KEY_IS(KEY_CONN_DATA))
2872 exp->exp_connect_data = *(struct obd_connect_data *)val;
2874 } else if (KEY_IS(KEY_TGT_COUNT)) {
2875 *((int *)val) = lmv->desc.ld_tgt_count;
2879 CDEBUG(D_IOCTL, "Invalid key\n");
2883 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2884 obd_count keylen, void *key, obd_count vallen,
2885 void *val, struct ptlrpc_request_set *set)
2887 struct lmv_tgt_desc *tgt = NULL;
2888 struct obd_device *obd;
2889 struct lmv_obd *lmv;
2893 obd = class_exp2obd(exp);
2895 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2896 exp->exp_handle.h_cookie);
2901 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2904 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2907 if (tgt == NULL || tgt->ltd_exp == NULL)
2910 err = obd_set_info_async(env, tgt->ltd_exp,
2911 keylen, key, vallen, val, set);
2922 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2923 struct lmv_mds_md_v1 *lmm1)
2928 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2929 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2930 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2931 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2932 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2933 sizeof(lmm1->lmv_pool_name));
2934 if (cplen >= sizeof(lmm1->lmv_pool_name))
2937 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2938 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2939 &lsm->lsm_md_oinfo[i].lmo_fid);
2943 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2947 bool allocated = false;
2951 LASSERT(lmmp != NULL);
2953 if (*lmmp != NULL && lsm == NULL) {
2956 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2957 lmm_size = lmv_mds_md_size(stripe_count,
2958 le32_to_cpu((*lmmp)->lmv_magic));
2961 OBD_FREE(*lmmp, lmm_size);
2967 if (*lmmp == NULL && lsm == NULL) {
2968 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2969 LASSERT(lmm_size > 0);
2970 OBD_ALLOC(*lmmp, lmm_size);
2973 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2974 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2979 LASSERT(lsm != NULL);
2980 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2981 if (*lmmp == NULL) {
2982 OBD_ALLOC(*lmmp, lmm_size);
2988 switch (lsm->lsm_md_magic) {
2990 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2997 if (rc != 0 && allocated) {
2998 OBD_FREE(*lmmp, lmm_size);
3004 EXPORT_SYMBOL(lmv_pack_md);
3006 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3007 const struct lmv_mds_md_v1 *lmm1)
3009 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3016 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3017 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3018 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3019 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3020 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3022 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3023 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3024 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3025 sizeof(lsm->lsm_md_pool_name));
3027 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3030 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3031 "layout_version %d\n", lsm->lsm_md_stripe_count,
3032 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3033 lsm->lsm_md_layout_version);
3035 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3036 for (i = 0; i < stripe_count; i++) {
3037 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3038 &lmm1->lmv_stripe_fids[i]);
3039 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3040 &lsm->lsm_md_oinfo[i].lmo_mds);
3043 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3044 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3050 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3051 const union lmv_mds_md *lmm, int stripe_count)
3053 struct lmv_stripe_md *lsm;
3056 bool allocated = false;
3059 LASSERT(lsmp != NULL);
3063 if (lsm != NULL && lmm == NULL) {
3065 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3066 /* For migrating inode, the master stripe and master
3067 * object will be the same, so do not need iput, see
3068 * ll_update_lsm_md */
3069 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3070 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3071 iput(lsm->lsm_md_oinfo[i].lmo_root);
3073 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3074 OBD_FREE(lsm, lsm_size);
3080 if (lsm == NULL && lmm == NULL) {
3081 lsm_size = lmv_stripe_md_size(stripe_count);
3082 OBD_ALLOC(lsm, lsm_size);
3085 lsm->lsm_md_stripe_count = stripe_count;
3090 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3094 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3095 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3096 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3097 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3102 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3103 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3106 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3107 * stripecount should be 0 then.
3109 lsm_size = lmv_stripe_md_size(0);
3111 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3113 OBD_ALLOC(lsm, lsm_size);
3120 switch (le32_to_cpu(lmm->lmv_magic)) {
3122 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3125 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3126 le32_to_cpu(lmm->lmv_magic));
3131 if (rc != 0 && allocated) {
3132 OBD_FREE(lsm, lsm_size);
3139 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3141 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3143 EXPORT_SYMBOL(lmv_alloc_memmd);
3145 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3147 lmv_unpack_md(NULL, &lsm, NULL, 0);
3149 EXPORT_SYMBOL(lmv_free_memmd);
3151 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3152 struct lov_mds_md *lmm, int disk_len)
3154 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3155 (union lmv_mds_md *)lmm, disk_len);
3158 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3159 struct lov_stripe_md *lsm)
3161 struct obd_device *obd = exp->exp_obd;
3162 struct lmv_obd *lmv_obd = &obd->u.lmv;
3163 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3168 stripe_count = lmv->lsm_md_stripe_count;
3170 stripe_count = lmv_obd->desc.ld_tgt_count;
3172 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3175 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3178 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3179 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3180 ldlm_cancel_flags_t flags, void *opaque)
3182 struct obd_device *obd = exp->exp_obd;
3183 struct lmv_obd *lmv = &obd->u.lmv;
3189 LASSERT(fid != NULL);
3191 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3192 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3194 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3197 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3205 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3208 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3209 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3213 if (tgt == NULL || tgt->ltd_exp == NULL)
3215 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3219 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3220 const struct lu_fid *fid, ldlm_type_t type,
3221 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3222 struct lustre_handle *lockh)
3224 struct obd_device *obd = exp->exp_obd;
3225 struct lmv_obd *lmv = &obd->u.lmv;
3231 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3234 * With DNE every object can have two locks in different namespaces:
3235 * lookup lock in space of MDT storing direntry and update/open lock in
3236 * space of MDT storing inode. Try the MDT that the FID maps to first,
3237 * since this can be easily found, and only try others if that fails.
3239 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3240 i < lmv->desc.ld_tgt_count;
3241 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3243 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3244 obd->obd_name, PFID(fid), tgt);
3248 if (lmv->tgts[tgt] == NULL ||
3249 lmv->tgts[tgt]->ltd_exp == NULL ||
3250 lmv->tgts[tgt]->ltd_active == 0)
3253 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3254 type, policy, mode, lockh);
3262 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3263 struct obd_export *dt_exp, struct obd_export *md_exp,
3264 struct lustre_md *md)
3266 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3267 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3269 if (tgt == NULL || tgt->ltd_exp == NULL)
3272 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3275 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3277 struct obd_device *obd = exp->exp_obd;
3278 struct lmv_obd *lmv = &obd->u.lmv;
3279 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3282 if (md->lmv != NULL) {
3283 lmv_free_memmd(md->lmv);
3286 if (tgt == NULL || tgt->ltd_exp == NULL)
3288 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3291 int lmv_set_open_replay_data(struct obd_export *exp,
3292 struct obd_client_handle *och,
3293 struct lookup_intent *it)
3295 struct obd_device *obd = exp->exp_obd;
3296 struct lmv_obd *lmv = &obd->u.lmv;
3297 struct lmv_tgt_desc *tgt;
3300 tgt = lmv_find_target(lmv, &och->och_fid);
3302 RETURN(PTR_ERR(tgt));
3304 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3307 int lmv_clear_open_replay_data(struct obd_export *exp,
3308 struct obd_client_handle *och)
3310 struct obd_device *obd = exp->exp_obd;
3311 struct lmv_obd *lmv = &obd->u.lmv;
3312 struct lmv_tgt_desc *tgt;
3315 tgt = lmv_find_target(lmv, &och->och_fid);
3317 RETURN(PTR_ERR(tgt));
3319 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3322 static int lmv_get_remote_perm(struct obd_export *exp,
3323 const struct lu_fid *fid,
3324 struct obd_capa *oc, __u32 suppgid,
3325 struct ptlrpc_request **request)
3327 struct obd_device *obd = exp->exp_obd;
3328 struct lmv_obd *lmv = &obd->u.lmv;
3329 struct lmv_tgt_desc *tgt;
3333 rc = lmv_check_connect(obd);
3337 tgt = lmv_find_target(lmv, fid);
3339 RETURN(PTR_ERR(tgt));
3341 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3345 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3348 struct obd_device *obd = exp->exp_obd;
3349 struct lmv_obd *lmv = &obd->u.lmv;
3350 struct lmv_tgt_desc *tgt;
3354 rc = lmv_check_connect(obd);
3358 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3360 RETURN(PTR_ERR(tgt));
3362 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3366 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3367 const struct req_msg_field *field, struct obd_capa **oc)
3369 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3370 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3372 if (tgt == NULL || tgt->ltd_exp == NULL)
3374 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3377 int lmv_intent_getattr_async(struct obd_export *exp,
3378 struct md_enqueue_info *minfo,
3379 struct ldlm_enqueue_info *einfo)
3381 struct md_op_data *op_data = &minfo->mi_data;
3382 struct obd_device *obd = exp->exp_obd;
3383 struct lmv_obd *lmv = &obd->u.lmv;
3384 struct lmv_tgt_desc *tgt = NULL;
3388 rc = lmv_check_connect(obd);
3392 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3394 RETURN(PTR_ERR(tgt));
3396 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3400 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3401 struct lu_fid *fid, __u64 *bits)
3403 struct obd_device *obd = exp->exp_obd;
3404 struct lmv_obd *lmv = &obd->u.lmv;
3405 struct lmv_tgt_desc *tgt;
3409 rc = lmv_check_connect(obd);
3413 tgt = lmv_find_target(lmv, fid);
3415 RETURN(PTR_ERR(tgt));
3417 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3421 int lmv_get_fid_from_lsm(struct obd_export *exp,
3422 const struct lmv_stripe_md *lsm,
3423 const char *name, int namelen, struct lu_fid *fid)
3425 const struct lmv_oinfo *oinfo;
3427 LASSERT(lsm != NULL);
3428 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3430 return PTR_ERR(oinfo);
3432 *fid = oinfo->lmo_fid;
3438 * For lmv, only need to send request to master MDT, and the master MDT will
3439 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3440 * we directly fetch data from the slave MDTs.
3442 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3443 struct obd_quotactl *oqctl)
3445 struct obd_device *obd = class_exp2obd(exp);
3446 struct lmv_obd *lmv = &obd->u.lmv;
3447 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3450 __u64 curspace, curinodes;
3454 tgt->ltd_exp == NULL ||
3456 lmv->desc.ld_tgt_count == 0) {
3457 CERROR("master lmv inactive\n");
3461 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3462 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3466 curspace = curinodes = 0;
3467 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3471 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3474 err = obd_quotactl(tgt->ltd_exp, oqctl);
3476 CERROR("getquota on mdt %d failed. %d\n", i, err);
3480 curspace += oqctl->qc_dqblk.dqb_curspace;
3481 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3484 oqctl->qc_dqblk.dqb_curspace = curspace;
3485 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3490 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3491 struct obd_quotactl *oqctl)
3493 struct obd_device *obd = class_exp2obd(exp);
3494 struct lmv_obd *lmv = &obd->u.lmv;
3495 struct lmv_tgt_desc *tgt;
3500 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3503 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3504 CERROR("lmv idx %d inactive\n", i);
3508 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3516 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3517 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3519 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3522 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3523 struct cl_attr *attr)
3527 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3528 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3530 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3531 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3532 i_size_read(inode), inode->i_nlink,
3533 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3534 LTIME_S(inode->i_mtime));
3536 /* for slave stripe, it needs to subtract nlink for . and .. */
3538 attr->cat_nlink += inode->i_nlink - 2;
3540 attr->cat_nlink = inode->i_nlink;
3542 attr->cat_size += i_size_read(inode);
3544 if (attr->cat_atime < LTIME_S(inode->i_atime))
3545 attr->cat_atime = LTIME_S(inode->i_atime);
3547 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3548 attr->cat_ctime = LTIME_S(inode->i_ctime);
3550 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3551 attr->cat_mtime = LTIME_S(inode->i_mtime);
3556 struct obd_ops lmv_obd_ops = {
3557 .o_owner = THIS_MODULE,
3558 .o_setup = lmv_setup,
3559 .o_cleanup = lmv_cleanup,
3560 .o_precleanup = lmv_precleanup,
3561 .o_process_config = lmv_process_config,
3562 .o_connect = lmv_connect,
3563 .o_disconnect = lmv_disconnect,
3564 .o_statfs = lmv_statfs,
3565 .o_get_info = lmv_get_info,
3566 .o_set_info_async = lmv_set_info_async,
3567 .o_packmd = lmv_packmd,
3568 .o_unpackmd = lmv_unpackmd,
3569 .o_notify = lmv_notify,
3570 .o_get_uuid = lmv_get_uuid,
3571 .o_iocontrol = lmv_iocontrol,
3572 .o_quotacheck = lmv_quotacheck,
3573 .o_quotactl = lmv_quotactl
3576 struct md_ops lmv_md_ops = {
3577 .m_getstatus = lmv_getstatus,
3578 .m_null_inode = lmv_null_inode,
3579 .m_find_cbdata = lmv_find_cbdata,
3580 .m_close = lmv_close,
3581 .m_create = lmv_create,
3582 .m_done_writing = lmv_done_writing,
3583 .m_enqueue = lmv_enqueue,
3584 .m_getattr = lmv_getattr,
3585 .m_getxattr = lmv_getxattr,
3586 .m_getattr_name = lmv_getattr_name,
3587 .m_intent_lock = lmv_intent_lock,
3589 .m_rename = lmv_rename,
3590 .m_setattr = lmv_setattr,
3591 .m_setxattr = lmv_setxattr,
3592 .m_fsync = lmv_fsync,
3593 .m_read_page = lmv_read_page,
3594 .m_unlink = lmv_unlink,
3595 .m_init_ea_size = lmv_init_ea_size,
3596 .m_cancel_unused = lmv_cancel_unused,
3597 .m_set_lock_data = lmv_set_lock_data,
3598 .m_lock_match = lmv_lock_match,
3599 .m_get_lustre_md = lmv_get_lustre_md,
3600 .m_free_lustre_md = lmv_free_lustre_md,
3601 .m_update_lsm_md = lmv_update_lsm_md,
3602 .m_merge_attr = lmv_merge_attr,
3603 .m_set_open_replay_data = lmv_set_open_replay_data,
3604 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3605 .m_renew_capa = lmv_renew_capa,
3606 .m_unpack_capa = lmv_unpack_capa,
3607 .m_get_remote_perm = lmv_get_remote_perm,
3608 .m_intent_getattr_async = lmv_intent_getattr_async,
3609 .m_revalidate_lock = lmv_revalidate_lock,
3610 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3613 int __init lmv_init(void)
3615 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3616 LUSTRE_LMV_NAME, NULL);
3619 static void lmv_exit(void)
3621 class_unregister_type(LUSTRE_LMV_NAME);
3624 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3625 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3626 MODULE_LICENSE("GPL");
3628 module_init(lmv_init);
3629 module_exit(lmv_exit);