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_lite.h>
58 #include <lustre_fid.h>
59 #include <lustre_ioctl.h>
60 #include "lmv_internal.h"
62 /* This hash is only for testing purpose */
63 static inline unsigned int
64 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
67 const unsigned char *p = (const unsigned char *)name;
69 while (--namelen >= 0)
77 static inline unsigned int
78 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
82 hash = lustre_hash_fnv_1a_64(name, namelen);
89 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
90 const char *name, int namelen)
93 __u32 hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
96 if (stripe_count <= 1)
99 /* for migrating object, always start from 0 stripe */
100 if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
104 case LMV_HASH_TYPE_ALL_CHARS:
105 idx = lmv_hash_all_chars(stripe_count, name, namelen);
107 case LMV_HASH_TYPE_FNV_1A_64:
108 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
115 CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
121 static void lmv_activate_target(struct lmv_obd *lmv,
122 struct lmv_tgt_desc *tgt,
125 if (tgt->ltd_active == activate)
128 tgt->ltd_active = activate;
129 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
135 * -EINVAL : UUID can't be found in the LMV's target list
136 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
137 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
139 static int lmv_set_mdc_active(struct lmv_obd *lmv,
140 const struct obd_uuid *uuid,
143 struct lmv_tgt_desc *tgt = NULL;
144 struct obd_device *obd;
149 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
150 lmv, uuid->uuid, activate);
152 spin_lock(&lmv->lmv_lock);
153 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
155 if (tgt == NULL || tgt->ltd_exp == NULL)
158 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
159 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
161 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
165 if (i == lmv->desc.ld_tgt_count)
166 GOTO(out_lmv_lock, rc = -EINVAL);
168 obd = class_exp2obd(tgt->ltd_exp);
170 GOTO(out_lmv_lock, rc = -ENOTCONN);
172 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
173 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
174 obd->obd_type->typ_name, i);
175 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
177 if (tgt->ltd_active == activate) {
178 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
179 activate ? "" : "in");
180 GOTO(out_lmv_lock, rc);
183 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
184 activate ? "" : "in");
185 lmv_activate_target(lmv, tgt, activate);
189 spin_unlock(&lmv->lmv_lock);
193 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
195 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
196 struct lmv_tgt_desc *tgt = lmv->tgts[0];
198 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
201 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
202 enum obd_notify_event ev, void *data)
204 struct obd_connect_data *conn_data;
205 struct lmv_obd *lmv = &obd->u.lmv;
206 struct obd_uuid *uuid;
210 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
211 CERROR("unexpected notification of %s %s!\n",
212 watched->obd_type->typ_name,
217 uuid = &watched->u.cli.cl_target_uuid;
218 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
220 * Set MDC as active before notifying the observer, so the
221 * observer can use the MDC normally.
223 rc = lmv_set_mdc_active(lmv, uuid,
224 ev == OBD_NOTIFY_ACTIVE);
226 CERROR("%sactivation of %s failed: %d\n",
227 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
231 } else if (ev == OBD_NOTIFY_OCD) {
232 conn_data = &watched->u.cli.cl_import->imp_connect_data;
234 * XXX: Make sure that ocd_connect_flags from all targets are
235 * the same. Otherwise one of MDTs runs wrong version or
236 * something like this. --umka
238 obd->obd_self_export->exp_connect_data = *conn_data;
241 else if (ev == OBD_NOTIFY_DISCON) {
243 * For disconnect event, flush fld cache for failout MDS case.
245 fld_client_flush(&lmv->lmv_fld);
249 * Pass the notification up the chain.
251 if (obd->obd_observer)
252 rc = obd_notify(obd->obd_observer, watched, ev, data);
258 * This is fake connect function. Its purpose is to initialize lmv and say
259 * caller that everything is okay. Real connection will be performed later.
261 static int lmv_connect(const struct lu_env *env,
262 struct obd_export **exp, struct obd_device *obd,
263 struct obd_uuid *cluuid, struct obd_connect_data *data,
266 struct lmv_obd *lmv = &obd->u.lmv;
267 struct lustre_handle conn = { 0 };
272 * We don't want to actually do the underlying connections more than
273 * once, so keep track.
276 if (lmv->refcount > 1) {
281 rc = class_connect(&conn, obd, cluuid);
283 CERROR("class_connection() returned %d\n", rc);
287 *exp = class_conn2export(&conn);
288 class_export_get(*exp);
292 lmv->cluuid = *cluuid;
295 lmv->conn_data = *data;
297 if (lmv->targets_proc_entry == NULL) {
298 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
301 if (IS_ERR(lmv->targets_proc_entry)) {
302 CERROR("%s: cannot register "
303 "/proc/fs/lustre/%s/%s/target_obds\n",
304 obd->obd_name, obd->obd_type->typ_name,
306 lmv->targets_proc_entry = NULL;
311 * All real clients should perform actual connection right away, because
312 * it is possible, that LMV will not have opportunity to connect targets
313 * and MDC stuff will be called directly, for instance while reading
314 * ../mdc/../kbytesfree procfs file, etc.
316 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
317 rc = lmv_check_connect(obd);
319 if (rc && lmv->targets_proc_entry != NULL)
320 lprocfs_remove(&lmv->targets_proc_entry);
324 static void lmv_set_timeouts(struct obd_device *obd)
330 if (lmv->server_timeout == 0)
333 if (lmv->connected == 0)
336 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
337 struct lmv_tgt_desc *tgt = lmv->tgts[i];
339 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
342 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
343 KEY_INTERMDS, 0, NULL, NULL);
347 static int lmv_init_ea_size(struct obd_export *exp, int easize,
348 int def_easize, int cookiesize, int def_cookiesize)
350 struct obd_device *obd = exp->exp_obd;
351 struct lmv_obd *lmv = &obd->u.lmv;
357 if (lmv->max_easize < easize) {
358 lmv->max_easize = easize;
361 if (lmv->max_def_easize < def_easize) {
362 lmv->max_def_easize = def_easize;
365 if (lmv->max_cookiesize < cookiesize) {
366 lmv->max_cookiesize = cookiesize;
369 if (lmv->max_def_cookiesize < def_cookiesize) {
370 lmv->max_def_cookiesize = def_cookiesize;
376 if (lmv->connected == 0)
379 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
380 struct lmv_tgt_desc *tgt = lmv->tgts[i];
382 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
383 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
387 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
388 cookiesize, def_cookiesize);
390 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
391 " rc = %d\n", obd->obd_name, i, rc);
398 #define MAX_STRING_SIZE 128
400 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
402 struct lmv_obd *lmv = &obd->u.lmv;
403 struct obd_uuid *cluuid = &lmv->cluuid;
404 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
405 struct obd_device *mdc_obd;
406 struct obd_export *mdc_exp;
407 struct lu_fld_target target;
411 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
414 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
418 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
419 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
420 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
423 if (!mdc_obd->obd_set_up) {
424 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
428 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
429 &lmv->conn_data, NULL);
431 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
436 * Init fid sequence client for this mdc and add new fld target.
438 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
442 target.ft_srv = NULL;
443 target.ft_exp = mdc_exp;
444 target.ft_idx = tgt->ltd_idx;
446 fld_client_add_target(&lmv->lmv_fld, &target);
448 rc = obd_register_observer(mdc_obd, obd);
450 obd_disconnect(mdc_exp);
451 CERROR("target %s register_observer error %d\n",
452 tgt->ltd_uuid.uuid, rc);
456 if (obd->obd_observer) {
458 * Tell the observer about the new target.
460 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
462 (void *)(tgt - lmv->tgts[0]));
464 obd_disconnect(mdc_exp);
470 tgt->ltd_exp = mdc_exp;
471 lmv->desc.ld_active_tgt_count++;
473 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
474 lmv->max_cookiesize, lmv->max_def_cookiesize);
476 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
477 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
478 atomic_read(&obd->obd_refcount));
480 if (lmv->targets_proc_entry != NULL) {
481 struct proc_dir_entry *mdc_symlink;
483 LASSERT(mdc_obd->obd_type != NULL);
484 LASSERT(mdc_obd->obd_type->typ_name != NULL);
485 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
486 lmv->targets_proc_entry,
488 mdc_obd->obd_type->typ_name,
490 if (mdc_symlink == NULL) {
491 CERROR("cannot register LMV target "
492 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
493 obd->obd_type->typ_name, obd->obd_name,
500 static void lmv_del_target(struct lmv_obd *lmv, int index)
502 if (lmv->tgts[index] == NULL)
505 OBD_FREE_PTR(lmv->tgts[index]);
506 lmv->tgts[index] = NULL;
510 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
511 __u32 index, int gen)
513 struct lmv_obd *lmv = &obd->u.lmv;
514 struct lmv_tgt_desc *tgt;
515 int orig_tgt_count = 0;
519 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
523 if (lmv->desc.ld_tgt_count == 0) {
524 struct obd_device *mdc_obd;
526 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
529 lmv_init_unlock(lmv);
530 CERROR("%s: Target %s not attached: rc = %d\n",
531 obd->obd_name, uuidp->uuid, -EINVAL);
536 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
537 tgt = lmv->tgts[index];
538 CERROR("%s: UUID %s already assigned at LOV target index %d:"
539 " rc = %d\n", obd->obd_name,
540 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
541 lmv_init_unlock(lmv);
545 if (index >= lmv->tgts_size) {
546 /* We need to reallocate the lmv target array. */
547 struct lmv_tgt_desc **newtgts, **old = NULL;
551 while (newsize < index + 1)
552 newsize = newsize << 1;
553 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
554 if (newtgts == NULL) {
555 lmv_init_unlock(lmv);
559 if (lmv->tgts_size) {
560 memcpy(newtgts, lmv->tgts,
561 sizeof(*newtgts) * lmv->tgts_size);
563 oldsize = lmv->tgts_size;
567 lmv->tgts_size = newsize;
570 OBD_FREE(old, sizeof(*old) * oldsize);
572 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
578 lmv_init_unlock(lmv);
582 mutex_init(&tgt->ltd_fid_mutex);
583 tgt->ltd_idx = index;
584 tgt->ltd_uuid = *uuidp;
586 lmv->tgts[index] = tgt;
587 if (index >= lmv->desc.ld_tgt_count) {
588 orig_tgt_count = lmv->desc.ld_tgt_count;
589 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 if (lmv->desc.ld_tgt_count == index + 1)
597 lmv->desc.ld_tgt_count = orig_tgt_count;
598 memset(tgt, 0, sizeof(*tgt));
599 spin_unlock(&lmv->lmv_lock);
601 int easize = sizeof(struct lmv_stripe_md) +
602 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
603 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
607 lmv_init_unlock(lmv);
611 int lmv_check_connect(struct obd_device *obd)
613 struct lmv_obd *lmv = &obd->u.lmv;
614 struct lmv_tgt_desc *tgt;
624 if (lmv->connected) {
625 lmv_init_unlock(lmv);
629 if (lmv->desc.ld_tgt_count == 0) {
630 lmv_init_unlock(lmv);
631 CERROR("%s: no targets configured.\n", obd->obd_name);
635 LASSERT(lmv->tgts != NULL);
637 if (lmv->tgts[0] == NULL) {
638 lmv_init_unlock(lmv);
639 CERROR("%s: no target configured for index 0.\n",
644 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
645 lmv->cluuid.uuid, obd->obd_name);
647 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
651 rc = lmv_connect_mdc(obd, tgt);
656 lmv_set_timeouts(obd);
657 class_export_put(lmv->exp);
659 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
660 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
661 lmv_init_unlock(lmv);
672 --lmv->desc.ld_active_tgt_count;
673 rc2 = obd_disconnect(tgt->ltd_exp);
675 CERROR("LMV target %s disconnect on "
676 "MDC idx %d: error %d\n",
677 tgt->ltd_uuid.uuid, i, rc2);
681 class_disconnect(lmv->exp);
682 lmv_init_unlock(lmv);
686 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
688 struct lmv_obd *lmv = &obd->u.lmv;
689 struct obd_device *mdc_obd;
693 LASSERT(tgt != NULL);
694 LASSERT(obd != NULL);
696 mdc_obd = class_exp2obd(tgt->ltd_exp);
699 mdc_obd->obd_force = obd->obd_force;
700 mdc_obd->obd_fail = obd->obd_fail;
701 mdc_obd->obd_no_recov = obd->obd_no_recov;
704 if (lmv->targets_proc_entry != NULL)
705 lprocfs_remove_proc_entry(mdc_obd->obd_name,
706 lmv->targets_proc_entry);
708 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
710 CERROR("Can't finanize fids factory\n");
712 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
713 tgt->ltd_exp->exp_obd->obd_name,
714 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
716 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
717 rc = obd_disconnect(tgt->ltd_exp);
719 if (tgt->ltd_active) {
720 CERROR("Target %s disconnect error %d\n",
721 tgt->ltd_uuid.uuid, rc);
725 lmv_activate_target(lmv, tgt, 0);
730 static int lmv_disconnect(struct obd_export *exp)
732 struct obd_device *obd = class_exp2obd(exp);
733 struct lmv_obd *lmv = &obd->u.lmv;
742 * Only disconnect the underlying layers on the final disconnect.
745 if (lmv->refcount != 0)
748 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
749 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
752 lmv_disconnect_mdc(obd, lmv->tgts[i]);
755 if (lmv->targets_proc_entry != NULL)
756 lprocfs_remove(&lmv->targets_proc_entry);
758 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
759 obd->obd_type->typ_name, obd->obd_name);
763 * This is the case when no real connection is established by
764 * lmv_check_connect().
767 class_export_put(exp);
768 rc = class_disconnect(exp);
769 if (lmv->refcount == 0)
774 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
776 struct obd_device *obddev = class_exp2obd(exp);
777 struct lmv_obd *lmv = &obddev->u.lmv;
778 struct getinfo_fid2path *gf;
779 struct lmv_tgt_desc *tgt;
780 struct getinfo_fid2path *remote_gf = NULL;
781 int remote_gf_size = 0;
784 gf = (struct getinfo_fid2path *)karg;
785 tgt = lmv_find_target(lmv, &gf->gf_fid);
787 RETURN(PTR_ERR(tgt));
790 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
791 if (rc != 0 && rc != -EREMOTE)
792 GOTO(out_fid2path, rc);
794 /* If remote_gf != NULL, it means just building the
795 * path on the remote MDT, copy this path segement to gf */
796 if (remote_gf != NULL) {
797 struct getinfo_fid2path *ori_gf;
800 ori_gf = (struct getinfo_fid2path *)karg;
801 if (strlen(ori_gf->gf_path) +
802 strlen(gf->gf_path) > ori_gf->gf_pathlen)
803 GOTO(out_fid2path, rc = -EOVERFLOW);
805 ptr = ori_gf->gf_path;
807 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
808 strlen(ori_gf->gf_path));
810 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
811 ptr += strlen(gf->gf_path);
815 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
816 tgt->ltd_exp->exp_obd->obd_name,
817 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
821 GOTO(out_fid2path, rc);
823 /* sigh, has to go to another MDT to do path building further */
824 if (remote_gf == NULL) {
825 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
826 OBD_ALLOC(remote_gf, remote_gf_size);
827 if (remote_gf == NULL)
828 GOTO(out_fid2path, rc = -ENOMEM);
829 remote_gf->gf_pathlen = PATH_MAX;
832 if (!fid_is_sane(&gf->gf_fid)) {
833 CERROR("%s: invalid FID "DFID": rc = %d\n",
834 tgt->ltd_exp->exp_obd->obd_name,
835 PFID(&gf->gf_fid), -EINVAL);
836 GOTO(out_fid2path, rc = -EINVAL);
839 tgt = lmv_find_target(lmv, &gf->gf_fid);
841 GOTO(out_fid2path, rc = -EINVAL);
843 remote_gf->gf_fid = gf->gf_fid;
844 remote_gf->gf_recno = -1;
845 remote_gf->gf_linkno = -1;
846 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
848 goto repeat_fid2path;
851 if (remote_gf != NULL)
852 OBD_FREE(remote_gf, remote_gf_size);
856 static int lmv_hsm_req_count(struct lmv_obd *lmv,
857 const struct hsm_user_request *hur,
858 const struct lmv_tgt_desc *tgt_mds)
862 struct lmv_tgt_desc *curr_tgt;
864 /* count how many requests must be sent to the given target */
865 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
866 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
867 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
873 static void lmv_hsm_req_build(struct lmv_obd *lmv,
874 struct hsm_user_request *hur_in,
875 const struct lmv_tgt_desc *tgt_mds,
876 struct hsm_user_request *hur_out)
879 struct lmv_tgt_desc *curr_tgt;
881 /* build the hsm_user_request for the given target */
882 hur_out->hur_request = hur_in->hur_request;
884 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
885 curr_tgt = lmv_find_target(lmv,
886 &hur_in->hur_user_item[i].hui_fid);
887 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
888 hur_out->hur_user_item[nr_out] =
889 hur_in->hur_user_item[i];
893 hur_out->hur_request.hr_itemcount = nr_out;
894 memcpy(hur_data(hur_out), hur_data(hur_in),
895 hur_in->hur_request.hr_data_len);
898 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
899 struct lustre_kernelcomm *lk, void *uarg)
903 struct kkuc_ct_data *kcd = NULL;
906 /* unregister request (call from llapi_hsm_copytool_fini) */
907 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
908 struct lmv_tgt_desc *tgt = lmv->tgts[i];
910 if (tgt == NULL || tgt->ltd_exp == NULL)
912 /* best effort: try to clean as much as possible
913 * (continue on error) */
914 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
917 /* Whatever the result, remove copytool from kuc groups.
918 * Unreached coordinators will get EPIPE on next requests
919 * and will unregister automatically.
921 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
928 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
929 struct lustre_kernelcomm *lk, void *uarg)
934 bool any_set = false;
935 struct kkuc_ct_data *kcd;
938 /* All or nothing: try to register to all MDS.
939 * In case of failure, unregister from previous MDS,
940 * except if it because of inactive target. */
941 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
942 struct lmv_tgt_desc *tgt = lmv->tgts[i];
944 if (tgt == NULL || tgt->ltd_exp == NULL)
946 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
948 if (tgt->ltd_active) {
949 /* permanent error */
950 CERROR("%s: iocontrol MDC %s on MDT"
951 " idx %d cmd %x: err = %d\n",
952 class_exp2obd(lmv->exp)->obd_name,
953 tgt->ltd_uuid.uuid, i, cmd, err);
955 lk->lk_flags |= LK_FLG_STOP;
956 /* unregister from previous MDS */
957 for (j = 0; j < i; j++) {
959 if (tgt == NULL || tgt->ltd_exp == NULL)
961 obd_iocontrol(cmd, tgt->ltd_exp, len,
966 /* else: transient error.
967 * kuc will register to the missing MDT
975 /* no registration done: return error */
978 /* at least one registration done, with no failure */
979 filp = fget(lk->lk_wfd);
988 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
989 kcd->kcd_uuid = lmv->cluuid;
990 kcd->kcd_archive = lk->lk_data;
992 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
1005 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1006 int len, void *karg, void *uarg)
1008 struct obd_device *obddev = class_exp2obd(exp);
1009 struct lmv_obd *lmv = &obddev->u.lmv;
1010 struct lmv_tgt_desc *tgt = NULL;
1014 __u32 count = lmv->desc.ld_tgt_count;
1021 case IOC_OBD_STATFS: {
1022 struct obd_ioctl_data *data = karg;
1023 struct obd_device *mdc_obd;
1024 struct obd_statfs stat_buf = {0};
1027 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1028 if ((index >= count))
1031 tgt = lmv->tgts[index];
1032 if (tgt == NULL || !tgt->ltd_active)
1035 mdc_obd = class_exp2obd(tgt->ltd_exp);
1040 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1041 min((int) data->ioc_plen2,
1042 (int) sizeof(struct obd_uuid))))
1045 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1046 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1050 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1051 min((int) data->ioc_plen1,
1052 (int) sizeof(stat_buf))))
1056 case OBD_IOC_QUOTACTL: {
1057 struct if_quotactl *qctl = karg;
1058 struct obd_quotactl *oqctl;
1060 if (qctl->qc_valid == QC_MDTIDX) {
1061 if (count <= qctl->qc_idx)
1064 tgt = lmv->tgts[qctl->qc_idx];
1065 if (tgt == NULL || tgt->ltd_exp == NULL)
1067 } else if (qctl->qc_valid == QC_UUID) {
1068 for (i = 0; i < count; i++) {
1072 if (!obd_uuid_equals(&tgt->ltd_uuid,
1076 if (tgt->ltd_exp == NULL)
1088 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1089 OBD_ALLOC_PTR(oqctl);
1093 QCTL_COPY(oqctl, qctl);
1094 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1096 QCTL_COPY(qctl, oqctl);
1097 qctl->qc_valid = QC_MDTIDX;
1098 qctl->obd_uuid = tgt->ltd_uuid;
1100 OBD_FREE_PTR(oqctl);
1103 case OBD_IOC_CHANGELOG_SEND:
1104 case OBD_IOC_CHANGELOG_CLEAR: {
1105 struct ioc_changelog *icc = karg;
1107 if (icc->icc_mdtindex >= count)
1110 tgt = lmv->tgts[icc->icc_mdtindex];
1111 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1113 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1116 case LL_IOC_GET_CONNECT_FLAGS: {
1118 if (tgt == NULL || tgt->ltd_exp == NULL)
1120 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1123 case LL_IOC_FID2MDTIDX: {
1124 struct lu_fid *fid = karg;
1127 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1131 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1132 * point to user space memory for FID2MDTIDX. */
1133 *(__u32 *)uarg = mdt_index;
1136 case OBD_IOC_FID2PATH: {
1137 rc = lmv_fid2path(exp, len, karg, uarg);
1140 case LL_IOC_HSM_STATE_GET:
1141 case LL_IOC_HSM_STATE_SET:
1142 case LL_IOC_HSM_ACTION: {
1143 struct md_op_data *op_data = karg;
1145 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1147 RETURN(PTR_ERR(tgt));
1149 if (tgt->ltd_exp == NULL)
1152 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1155 case LL_IOC_HSM_PROGRESS: {
1156 const struct hsm_progress_kernel *hpk = karg;
1158 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1160 RETURN(PTR_ERR(tgt));
1161 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1164 case LL_IOC_HSM_REQUEST: {
1165 struct hsm_user_request *hur = karg;
1166 unsigned int reqcount = hur->hur_request.hr_itemcount;
1171 /* if the request is about a single fid
1172 * or if there is a single MDS, no need to split
1174 if (reqcount == 1 || count == 1) {
1175 tgt = lmv_find_target(lmv,
1176 &hur->hur_user_item[0].hui_fid);
1178 RETURN(PTR_ERR(tgt));
1179 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1181 /* split fid list to their respective MDS */
1182 for (i = 0; i < count; i++) {
1183 unsigned int nr, reqlen;
1185 struct hsm_user_request *req;
1188 if (tgt == NULL || tgt->ltd_exp == NULL)
1191 nr = lmv_hsm_req_count(lmv, hur, tgt);
1192 if (nr == 0) /* nothing for this MDS */
1195 /* build a request with fids for this MDS */
1196 reqlen = offsetof(typeof(*hur),
1198 + hur->hur_request.hr_data_len;
1199 OBD_ALLOC_LARGE(req, reqlen);
1203 lmv_hsm_req_build(lmv, hur, tgt, req);
1205 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1207 if (rc1 != 0 && rc == 0)
1209 OBD_FREE_LARGE(req, reqlen);
1214 case LL_IOC_LOV_SWAP_LAYOUTS: {
1215 struct md_op_data *op_data = karg;
1216 struct lmv_tgt_desc *tgt1, *tgt2;
1218 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1220 RETURN(PTR_ERR(tgt1));
1222 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1224 RETURN(PTR_ERR(tgt2));
1226 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1229 /* only files on same MDT can have their layouts swapped */
1230 if (tgt1->ltd_idx != tgt2->ltd_idx)
1233 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1236 case LL_IOC_HSM_CT_START: {
1237 struct lustre_kernelcomm *lk = karg;
1238 if (lk->lk_flags & LK_FLG_STOP)
1239 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1241 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1245 for (i = 0; i < count; i++) {
1246 struct obd_device *mdc_obd;
1250 if (tgt == NULL || tgt->ltd_exp == NULL)
1252 /* ll_umount_begin() sets force flag but for lmv, not
1253 * mdc. Let's pass it through */
1254 mdc_obd = class_exp2obd(tgt->ltd_exp);
1255 mdc_obd->obd_force = obddev->obd_force;
1256 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1257 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1260 if (tgt->ltd_active) {
1261 CERROR("error: iocontrol MDC %s on MDT"
1262 " idx %d cmd %x: err = %d\n",
1263 tgt->ltd_uuid.uuid, i, cmd, err);
1277 static int lmv_all_chars_policy(int count, const char *name,
1288 static int lmv_nid_policy(struct lmv_obd *lmv)
1290 struct obd_import *imp;
1294 * XXX: To get nid we assume that underlying obd device is mdc.
1296 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1297 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1298 return id % lmv->desc.ld_tgt_count;
1301 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1302 placement_policy_t placement)
1304 switch (placement) {
1305 case PLACEMENT_CHAR_POLICY:
1306 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1308 op_data->op_namelen);
1309 case PLACEMENT_NID_POLICY:
1310 return lmv_nid_policy(lmv);
1316 CERROR("Unsupported placement policy %x\n", placement);
1322 * This is _inode_ placement policy function (not name).
1324 static int lmv_placement_policy(struct obd_device *obd,
1325 struct md_op_data *op_data,
1328 struct lmv_obd *lmv = &obd->u.lmv;
1331 LASSERT(mds != NULL);
1333 if (lmv->desc.ld_tgt_count == 1) {
1339 * If stripe_offset is provided during setdirstripe
1340 * (setdirstripe -i xx), xx MDS will be choosen.
1342 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1343 struct lmv_user_md *lum;
1345 lum = op_data->op_data;
1347 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1348 *mds = le32_to_cpu(lum->lum_stripe_offset);
1350 /* -1 means default, which will be in the same MDT with
1352 *mds = op_data->op_mds;
1353 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1356 /* Allocate new fid on target according to operation type and
1357 * parent home mds. */
1358 *mds = op_data->op_mds;
1364 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1367 struct lmv_tgt_desc *tgt;
1371 tgt = lmv_get_target(lmv, mds, NULL);
1373 RETURN(PTR_ERR(tgt));
1376 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1377 * on server that seq in new allocated fid is not yet known.
1379 mutex_lock(&tgt->ltd_fid_mutex);
1381 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1382 GOTO(out, rc = -ENODEV);
1385 * Asking underlying tgt layer to allocate new fid.
1387 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1389 LASSERT(fid_is_sane(fid));
1395 mutex_unlock(&tgt->ltd_fid_mutex);
1399 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1400 struct lu_fid *fid, struct md_op_data *op_data)
1402 struct obd_device *obd = class_exp2obd(exp);
1403 struct lmv_obd *lmv = &obd->u.lmv;
1408 LASSERT(op_data != NULL);
1409 LASSERT(fid != NULL);
1411 rc = lmv_placement_policy(obd, op_data, &mds);
1413 CERROR("Can't get target for allocating fid, "
1418 rc = __lmv_fid_alloc(lmv, fid, mds);
1420 CERROR("Can't alloc new fid, rc %d\n", rc);
1427 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1429 struct lmv_obd *lmv = &obd->u.lmv;
1430 struct lmv_desc *desc;
1434 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1435 CERROR("LMV setup requires a descriptor\n");
1439 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1440 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1441 CERROR("Lmv descriptor size wrong: %d > %d\n",
1442 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1446 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1447 if (lmv->tgts == NULL)
1449 lmv->tgts_size = 32;
1451 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1452 lmv->desc.ld_tgt_count = 0;
1453 lmv->desc.ld_active_tgt_count = 0;
1454 lmv->max_cookiesize = 0;
1455 lmv->max_def_easize = 0;
1456 lmv->max_easize = 0;
1457 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1459 spin_lock_init(&lmv->lmv_lock);
1460 mutex_init(&lmv->init_mutex);
1463 obd->obd_vars = lprocfs_lmv_obd_vars;
1464 lprocfs_seq_obd_setup(obd);
1465 lprocfs_alloc_md_stats(obd, 0);
1466 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1467 0444, &lmv_proc_target_fops, obd);
1469 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1472 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1473 LUSTRE_CLI_FLD_HASH_DHT);
1475 CERROR("Can't init FLD, err %d\n", rc);
1485 static int lmv_cleanup(struct obd_device *obd)
1487 struct lmv_obd *lmv = &obd->u.lmv;
1490 fld_client_fini(&lmv->lmv_fld);
1491 if (lmv->tgts != NULL) {
1493 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1494 if (lmv->tgts[i] == NULL)
1496 lmv_del_target(lmv, i);
1498 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1504 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1506 struct lustre_cfg *lcfg = buf;
1507 struct obd_uuid obd_uuid;
1513 switch (lcfg->lcfg_command) {
1515 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1516 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1517 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1518 GOTO(out, rc = -EINVAL);
1520 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1522 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1523 GOTO(out, rc = -EINVAL);
1524 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1525 GOTO(out, rc = -EINVAL);
1526 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1529 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1530 GOTO(out, rc = -EINVAL);
1536 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1537 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1539 struct obd_device *obd = class_exp2obd(exp);
1540 struct lmv_obd *lmv = &obd->u.lmv;
1541 struct obd_statfs *temp;
1546 rc = lmv_check_connect(obd);
1550 OBD_ALLOC(temp, sizeof(*temp));
1554 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1555 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1558 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1561 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1562 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1564 GOTO(out_free_temp, rc);
1569 /* If the statfs is from mount, it will needs
1570 * retrieve necessary information from MDT0.
1571 * i.e. mount does not need the merged osfs
1573 * And also clients can be mounted as long as
1574 * MDT0 is in service*/
1575 if (flags & OBD_STATFS_FOR_MDT0)
1576 GOTO(out_free_temp, rc);
1578 osfs->os_bavail += temp->os_bavail;
1579 osfs->os_blocks += temp->os_blocks;
1580 osfs->os_ffree += temp->os_ffree;
1581 osfs->os_files += temp->os_files;
1587 OBD_FREE(temp, sizeof(*temp));
1591 static int lmv_getstatus(struct obd_export *exp,
1593 struct obd_capa **pc)
1595 struct obd_device *obd = exp->exp_obd;
1596 struct lmv_obd *lmv = &obd->u.lmv;
1600 rc = lmv_check_connect(obd);
1604 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1608 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1609 struct obd_capa *oc, obd_valid valid, const char *name,
1610 const char *input, int input_size, int output_size,
1611 int flags, struct ptlrpc_request **request)
1613 struct obd_device *obd = exp->exp_obd;
1614 struct lmv_obd *lmv = &obd->u.lmv;
1615 struct lmv_tgt_desc *tgt;
1619 rc = lmv_check_connect(obd);
1623 tgt = lmv_find_target(lmv, fid);
1625 RETURN(PTR_ERR(tgt));
1627 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1628 input_size, output_size, flags, request);
1633 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1634 struct obd_capa *oc, obd_valid valid, const char *name,
1635 const char *input, int input_size, int output_size,
1636 int flags, __u32 suppgid,
1637 struct ptlrpc_request **request)
1639 struct obd_device *obd = exp->exp_obd;
1640 struct lmv_obd *lmv = &obd->u.lmv;
1641 struct lmv_tgt_desc *tgt;
1645 rc = lmv_check_connect(obd);
1649 tgt = lmv_find_target(lmv, fid);
1651 RETURN(PTR_ERR(tgt));
1653 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1654 input_size, output_size, flags, suppgid,
1660 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1661 struct ptlrpc_request **request)
1663 struct obd_device *obd = exp->exp_obd;
1664 struct lmv_obd *lmv = &obd->u.lmv;
1665 struct lmv_tgt_desc *tgt;
1669 rc = lmv_check_connect(obd);
1673 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1675 RETURN(PTR_ERR(tgt));
1677 if (op_data->op_flags & MF_GET_MDT_IDX) {
1678 op_data->op_mds = tgt->ltd_idx;
1682 rc = md_getattr(tgt->ltd_exp, op_data, request);
1687 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1689 struct obd_device *obd = exp->exp_obd;
1690 struct lmv_obd *lmv = &obd->u.lmv;
1695 rc = lmv_check_connect(obd);
1699 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1702 * With DNE every object can have two locks in different namespaces:
1703 * lookup lock in space of MDT storing direntry and update/open lock in
1704 * space of MDT storing inode.
1706 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1707 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1709 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1715 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1716 ldlm_iterator_t it, void *data)
1718 struct obd_device *obd = exp->exp_obd;
1719 struct lmv_obd *lmv = &obd->u.lmv;
1725 rc = lmv_check_connect(obd);
1729 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1732 * With DNE every object can have two locks in different namespaces:
1733 * lookup lock in space of MDT storing direntry and update/open lock in
1734 * space of MDT storing inode. Try the MDT that the FID maps to first,
1735 * since this can be easily found, and only try others if that fails.
1737 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1738 i < lmv->desc.ld_tgt_count;
1739 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1741 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1742 obd->obd_name, PFID(fid), tgt);
1746 if (lmv->tgts[tgt] == NULL ||
1747 lmv->tgts[tgt]->ltd_exp == NULL)
1750 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1759 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1760 struct md_open_data *mod, struct ptlrpc_request **request)
1762 struct obd_device *obd = exp->exp_obd;
1763 struct lmv_obd *lmv = &obd->u.lmv;
1764 struct lmv_tgt_desc *tgt;
1768 rc = lmv_check_connect(obd);
1772 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1774 RETURN(PTR_ERR(tgt));
1776 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1777 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1782 * Choosing the MDT by name or FID in @op_data.
1783 * For non-striped directory, it will locate MDT by fid.
1784 * For striped-directory, it will locate MDT by name. And also
1785 * it will reset op_fid1 with the FID of the choosen stripe.
1787 struct lmv_tgt_desc *
1788 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1789 const char *name, int namelen, struct lu_fid *fid,
1792 struct lmv_tgt_desc *tgt;
1793 const struct lmv_oinfo *oinfo;
1795 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1797 RETURN(ERR_CAST(oinfo));
1798 *fid = oinfo->lmo_fid;
1799 *mds = oinfo->lmo_mds;
1800 tgt = lmv_get_target(lmv, *mds, NULL);
1802 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1807 * Locate mds by fid or name
1809 * For striped directory (lsm != NULL), it will locate the stripe
1810 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1811 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1812 * walk through all of stripes to locate the entry.
1814 * For normal direcotry, it will locate MDS by FID directly.
1815 * \param[in] lmv LMV device
1816 * \param[in] op_data client MD stack parameters, name, namelen
1818 * \param[in] fid object FID used to locate MDS.
1820 * retval pointer to the lmv_tgt_desc if succeed.
1821 * ERR_PTR(errno) if failed.
1823 struct lmv_tgt_desc*
1824 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1827 struct lmv_stripe_md *lsm = op_data->op_mea1;
1828 struct lmv_tgt_desc *tgt;
1830 /* During creating VOLATILE file, it should honor the mdt
1831 * index if the file under striped dir is being restored, see
1833 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1834 (int)op_data->op_mds != -1 && lsm != NULL) {
1836 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1840 /* refill the right parent fid */
1841 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1842 struct lmv_oinfo *oinfo;
1844 oinfo = &lsm->lsm_md_oinfo[i];
1845 if (oinfo->lmo_mds == op_data->op_mds) {
1846 *fid = oinfo->lmo_fid;
1851 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1852 if (i == lsm->lsm_md_stripe_count)
1853 tgt = ERR_PTR(-EINVAL);
1858 if (lsm == NULL || op_data->op_namelen == 0) {
1859 tgt = lmv_find_target(lmv, fid);
1863 op_data->op_mds = tgt->ltd_idx;
1867 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1868 op_data->op_namelen, fid,
1872 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1873 const void *data, int datalen, int mode, __u32 uid,
1874 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1875 struct ptlrpc_request **request)
1877 struct obd_device *obd = exp->exp_obd;
1878 struct lmv_obd *lmv = &obd->u.lmv;
1879 struct lmv_tgt_desc *tgt;
1883 rc = lmv_check_connect(obd);
1887 if (!lmv->desc.ld_active_tgt_count)
1890 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1892 RETURN(PTR_ERR(tgt));
1894 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1895 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1898 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1902 /* Send the create request to the MDT where the object
1903 * will be located */
1904 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1906 RETURN(PTR_ERR(tgt));
1908 op_data->op_mds = tgt->ltd_idx;
1910 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1911 PFID(&op_data->op_fid2), op_data->op_mds);
1913 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1914 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1915 cap_effective, rdev, request);
1917 if (*request == NULL)
1919 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1924 static int lmv_done_writing(struct obd_export *exp,
1925 struct md_op_data *op_data,
1926 struct md_open_data *mod)
1928 struct obd_device *obd = exp->exp_obd;
1929 struct lmv_obd *lmv = &obd->u.lmv;
1930 struct lmv_tgt_desc *tgt;
1934 rc = lmv_check_connect(obd);
1938 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1940 RETURN(PTR_ERR(tgt));
1942 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1947 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1948 const union ldlm_policy_data *policy,
1949 struct lookup_intent *it, struct md_op_data *op_data,
1950 struct lustre_handle *lockh, __u64 extra_lock_flags)
1952 struct obd_device *obd = exp->exp_obd;
1953 struct lmv_obd *lmv = &obd->u.lmv;
1954 struct lmv_tgt_desc *tgt;
1958 rc = lmv_check_connect(obd);
1962 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1963 LL_IT2STR(it), PFID(&op_data->op_fid1));
1965 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1967 RETURN(PTR_ERR(tgt));
1969 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1970 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1972 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1979 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1980 struct ptlrpc_request **preq)
1982 struct ptlrpc_request *req = NULL;
1983 struct obd_device *obd = exp->exp_obd;
1984 struct lmv_obd *lmv = &obd->u.lmv;
1985 struct lmv_tgt_desc *tgt;
1986 struct mdt_body *body;
1990 rc = lmv_check_connect(obd);
1994 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1996 RETURN(PTR_ERR(tgt));
1998 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1999 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
2002 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2006 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2007 LASSERT(body != NULL);
2009 if (body->mbo_valid & OBD_MD_MDS) {
2010 struct lu_fid rid = body->mbo_fid1;
2011 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2014 tgt = lmv_find_target(lmv, &rid);
2016 ptlrpc_req_finished(*preq);
2018 RETURN(PTR_ERR(tgt));
2021 op_data->op_fid1 = rid;
2022 op_data->op_valid |= OBD_MD_FLCROSSREF;
2023 op_data->op_namelen = 0;
2024 op_data->op_name = NULL;
2025 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2026 ptlrpc_req_finished(*preq);
2033 #define md_op_data_fid(op_data, fl) \
2034 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2035 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2036 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2037 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2040 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2041 struct md_op_data *op_data,
2042 int op_tgt, ldlm_mode_t mode, int bits, int flag)
2044 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2045 struct obd_device *obd = exp->exp_obd;
2046 struct lmv_obd *lmv = &obd->u.lmv;
2047 ldlm_policy_data_t policy = {{ 0 }};
2051 if (!fid_is_sane(fid))
2055 tgt = lmv_find_target(lmv, fid);
2057 RETURN(PTR_ERR(tgt));
2060 if (tgt->ltd_idx != op_tgt) {
2061 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2062 policy.l_inodebits.bits = bits;
2063 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2064 mode, LCF_ASYNC, NULL);
2067 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2069 op_data->op_flags |= flag;
2077 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2080 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2081 struct ptlrpc_request **request)
2083 struct obd_device *obd = exp->exp_obd;
2084 struct lmv_obd *lmv = &obd->u.lmv;
2085 struct lmv_tgt_desc *tgt;
2089 rc = lmv_check_connect(obd);
2093 LASSERT(op_data->op_namelen != 0);
2095 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2096 PFID(&op_data->op_fid2), op_data->op_namelen,
2097 op_data->op_name, PFID(&op_data->op_fid1));
2099 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2100 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2101 op_data->op_cap = cfs_curproc_cap_pack();
2102 if (op_data->op_mea2 != NULL) {
2103 struct lmv_stripe_md *lsm = op_data->op_mea2;
2104 const struct lmv_oinfo *oinfo;
2106 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2107 op_data->op_namelen);
2109 RETURN(PTR_ERR(oinfo));
2111 op_data->op_fid2 = oinfo->lmo_fid;
2114 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2116 RETURN(PTR_ERR(tgt));
2119 * Cancel UPDATE lock on child (fid1).
2121 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2122 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2123 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2127 rc = md_link(tgt->ltd_exp, op_data, request);
2132 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2133 const char *old, int oldlen, const char *new, int newlen,
2134 struct ptlrpc_request **request)
2136 struct obd_device *obd = exp->exp_obd;
2137 struct lmv_obd *lmv = &obd->u.lmv;
2138 struct lmv_tgt_desc *src_tgt;
2142 LASSERT(oldlen != 0);
2144 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2145 oldlen, old, PFID(&op_data->op_fid1),
2146 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2147 newlen, new, PFID(&op_data->op_fid2),
2148 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2150 rc = lmv_check_connect(obd);
2154 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2155 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2156 op_data->op_cap = cfs_curproc_cap_pack();
2157 if (op_data->op_cli_flags & CLI_MIGRATE) {
2158 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2159 PFID(&op_data->op_fid3));
2160 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2163 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2165 if (op_data->op_mea1 != NULL) {
2166 struct lmv_stripe_md *lsm = op_data->op_mea1;
2168 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2172 if (IS_ERR(src_tgt))
2173 RETURN(PTR_ERR(src_tgt));
2175 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2176 if (IS_ERR(src_tgt))
2177 RETURN(PTR_ERR(src_tgt));
2179 op_data->op_mds = src_tgt->ltd_idx;
2182 if (op_data->op_mea2) {
2183 struct lmv_stripe_md *lsm = op_data->op_mea2;
2184 const struct lmv_oinfo *oinfo;
2186 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2188 RETURN(PTR_ERR(oinfo));
2190 op_data->op_fid2 = oinfo->lmo_fid;
2193 if (IS_ERR(src_tgt))
2194 RETURN(PTR_ERR(src_tgt));
2197 * LOOKUP lock on src child (fid3) should also be cancelled for
2198 * src_tgt in mdc_rename.
2200 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2203 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2206 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2207 LCK_EX, MDS_INODELOCK_UPDATE,
2208 MF_MDC_CANCEL_FID2);
2213 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2215 if (fid_is_sane(&op_data->op_fid3)) {
2216 struct lmv_tgt_desc *tgt;
2218 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2220 RETURN(PTR_ERR(tgt));
2222 /* Cancel LOOKUP lock on its parent */
2223 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2224 LCK_EX, MDS_INODELOCK_LOOKUP,
2225 MF_MDC_CANCEL_FID3);
2229 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2230 LCK_EX, MDS_INODELOCK_FULL,
2231 MF_MDC_CANCEL_FID3);
2237 * Cancel all the locks on tgt child (fid4).
2239 if (fid_is_sane(&op_data->op_fid4))
2240 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2241 LCK_EX, MDS_INODELOCK_FULL,
2242 MF_MDC_CANCEL_FID4);
2244 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2245 op_data->op_mds, PFID(&op_data->op_fid2));
2247 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2253 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2254 void *ea, int ealen, void *ea2, int ea2len,
2255 struct ptlrpc_request **request,
2256 struct md_open_data **mod)
2258 struct obd_device *obd = exp->exp_obd;
2259 struct lmv_obd *lmv = &obd->u.lmv;
2260 struct lmv_tgt_desc *tgt;
2264 rc = lmv_check_connect(obd);
2268 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2269 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2271 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2272 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2274 RETURN(PTR_ERR(tgt));
2276 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2277 ea2len, request, mod);
2282 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2283 struct obd_capa *oc, struct ptlrpc_request **request)
2285 struct obd_device *obd = exp->exp_obd;
2286 struct lmv_obd *lmv = &obd->u.lmv;
2287 struct lmv_tgt_desc *tgt;
2291 rc = lmv_check_connect(obd);
2295 tgt = lmv_find_target(lmv, fid);
2297 RETURN(PTR_ERR(tgt));
2299 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2304 * Get current minimum entry from striped directory
2306 * This function will search the dir entry, whose hash value is the
2307 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2308 * only being called for striped directory.
2310 * \param[in] exp export of LMV
2311 * \param[in] op_data parameters transferred beween client MD stack
2312 * stripe_information will be included in this
2314 * \param[in] cb_op ldlm callback being used in enqueue in
2316 * \param[in] hash_offset the hash value, which is used to locate
2317 * minum(closet) dir entry
2318 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2319 * index of last entry, so to avoid hash conflict
2320 * between stripes. It will also be used to
2321 * return the stripe index of current dir entry.
2322 * \param[in|out] entp the minum entry and it also is being used
2323 * to input the last dir entry to resolve the
2326 * \param[out] ppage the page which holds the minum entry
2328 * \retval = 0 get the entry successfully
2329 * negative errno (< 0) does not get the entry
2331 static int lmv_get_min_striped_entry(struct obd_export *exp,
2332 struct md_op_data *op_data,
2333 struct md_callback *cb_op,
2334 __u64 hash_offset, int *stripe_offset,
2335 struct lu_dirent **entp,
2336 struct page **ppage)
2338 struct obd_device *obd = exp->exp_obd;
2339 struct lmv_obd *lmv = &obd->u.lmv;
2340 struct lmv_stripe_md *lsm = op_data->op_mea1;
2341 struct lmv_tgt_desc *tgt;
2343 struct lu_dirent *min_ent = NULL;
2344 struct page *min_page = NULL;
2350 stripe_count = lsm->lsm_md_stripe_count;
2351 for (i = 0; i < stripe_count; i++) {
2352 struct lu_dirent *ent = NULL;
2353 struct page *page = NULL;
2354 struct lu_dirpage *dp;
2355 __u64 stripe_hash = hash_offset;
2357 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2359 GOTO(out, rc = PTR_ERR(tgt));
2361 /* op_data will be shared by each stripe, so we need
2362 * reset these value for each stripe */
2363 op_data->op_stripe_offset = i;
2364 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2365 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2366 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2368 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2373 dp = page_address(page);
2374 for (ent = lu_dirent_start(dp); ent != NULL;
2375 ent = lu_dirent_next(ent)) {
2376 /* Skip dummy entry */
2377 if (le16_to_cpu(ent->lde_namelen) == 0)
2380 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2383 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2384 (*entp == ent || i < *stripe_offset))
2387 /* skip . and .. for other stripes */
2389 (strncmp(ent->lde_name, ".",
2390 le16_to_cpu(ent->lde_namelen)) == 0 ||
2391 strncmp(ent->lde_name, "..",
2392 le16_to_cpu(ent->lde_namelen)) == 0))
2398 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2401 page_cache_release(page);
2404 /* reach the end of current stripe, go to next stripe */
2405 if (stripe_hash == MDS_DIR_END_OFF)
2411 if (min_ent != NULL) {
2412 if (le64_to_cpu(min_ent->lde_hash) >
2413 le64_to_cpu(ent->lde_hash)) {
2416 page_cache_release(min_page);
2421 page_cache_release(page);
2432 if (*ppage != NULL) {
2434 page_cache_release(*ppage);
2436 *stripe_offset = min_idx;
2443 * Build dir entry page from a striped directory
2445 * This function gets one entry by @offset from a striped directory. It will
2446 * read entries from all of stripes, and choose one closest to the required
2447 * offset(&offset). A few notes
2448 * 1. skip . and .. for non-zero stripes, because there can only have one .
2449 * and .. in a directory.
2450 * 2. op_data will be shared by all of stripes, instead of allocating new
2451 * one, so need to restore before reusing.
2452 * 3. release the entry page if that is not being chosen.
2454 * \param[in] exp obd export refer to LMV
2455 * \param[in] op_data hold those MD parameters of read_entry
2456 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2457 * \param[out] ldp the entry being read
2458 * \param[out] ppage the page holding the entry. Note: because the entry
2459 * will be accessed in upper layer, so we need hold the
2460 * page until the usages of entry is finished, see
2461 * ll_dir_entry_next.
2463 * retval =0 if get entry successfully
2464 * <0 cannot get entry
2466 static int lmv_read_striped_page(struct obd_export *exp,
2467 struct md_op_data *op_data,
2468 struct md_callback *cb_op,
2469 __u64 offset, struct page **ppage)
2471 struct obd_device *obd = exp->exp_obd;
2472 struct lu_fid master_fid = op_data->op_fid1;
2473 struct inode *master_inode = op_data->op_data;
2474 __u64 hash_offset = offset;
2475 struct lu_dirpage *dp;
2476 struct page *min_ent_page = NULL;
2477 struct page *ent_page = NULL;
2478 struct lu_dirent *ent;
2481 struct lu_dirent *min_ent = NULL;
2482 struct lu_dirent *last_ent;
2487 rc = lmv_check_connect(obd);
2491 /* Allocate a page and read entries from all of stripes and fill
2492 * the page by hash order */
2493 ent_page = alloc_page(GFP_KERNEL);
2494 if (ent_page == NULL)
2497 /* Initialize the entry page */
2498 dp = kmap(ent_page);
2499 memset(dp, 0, sizeof(*dp));
2500 dp->ldp_hash_start = cpu_to_le64(offset);
2501 dp->ldp_flags |= LDF_COLLIDE;
2504 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2510 /* Find the minum entry from all sub-stripes */
2511 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2517 /* If it can not get minum entry, it means it already reaches
2518 * the end of this directory */
2519 if (min_ent == NULL) {
2520 last_ent->lde_reclen = 0;
2521 hash_offset = MDS_DIR_END_OFF;
2525 ent_size = le16_to_cpu(min_ent->lde_reclen);
2527 /* the last entry lde_reclen is 0, but it might not
2528 * the end of this entry of this temporay entry */
2530 ent_size = lu_dirent_calc_size(
2531 le16_to_cpu(min_ent->lde_namelen),
2532 le32_to_cpu(min_ent->lde_attrs));
2533 if (ent_size > left_bytes) {
2534 last_ent->lde_reclen = cpu_to_le16(0);
2535 hash_offset = le64_to_cpu(min_ent->lde_hash);
2539 memcpy(ent, min_ent, ent_size);
2541 /* Replace . with master FID and Replace .. with the parent FID
2542 * of master object */
2543 if (strncmp(ent->lde_name, ".",
2544 le16_to_cpu(ent->lde_namelen)) == 0 &&
2545 le16_to_cpu(ent->lde_namelen) == 1)
2546 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2547 else if (strncmp(ent->lde_name, "..",
2548 le16_to_cpu(ent->lde_namelen)) == 0 &&
2549 le16_to_cpu(ent->lde_namelen) == 2)
2550 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2552 left_bytes -= ent_size;
2553 ent->lde_reclen = cpu_to_le16(ent_size);
2555 ent = (void *)ent + ent_size;
2556 hash_offset = le64_to_cpu(min_ent->lde_hash);
2557 if (hash_offset == MDS_DIR_END_OFF) {
2558 last_ent->lde_reclen = 0;
2563 if (min_ent_page != NULL) {
2564 kunmap(min_ent_page);
2565 page_cache_release(min_ent_page);
2568 if (unlikely(rc != 0)) {
2569 __free_page(ent_page);
2573 dp->ldp_flags |= LDF_EMPTY;
2574 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2575 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2578 /* We do not want to allocate md_op_data during each
2579 * dir entry reading, so op_data will be shared by every stripe,
2580 * then we need to restore it back to original value before
2581 * return to the upper layer */
2582 op_data->op_fid1 = master_fid;
2583 op_data->op_fid2 = master_fid;
2584 op_data->op_data = master_inode;
2591 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2592 struct md_callback *cb_op, __u64 offset,
2593 struct page **ppage)
2595 struct obd_device *obd = exp->exp_obd;
2596 struct lmv_obd *lmv = &obd->u.lmv;
2597 struct lmv_stripe_md *lsm = op_data->op_mea1;
2598 struct lmv_tgt_desc *tgt;
2602 rc = lmv_check_connect(obd);
2606 if (unlikely(lsm != NULL)) {
2607 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2611 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2613 RETURN(PTR_ERR(tgt));
2615 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2621 * Unlink a file/directory
2623 * Unlink a file or directory under the parent dir. The unlink request
2624 * usually will be sent to the MDT where the child is located, but if
2625 * the client does not have the child FID then request will be sent to the
2626 * MDT where the parent is located.
2628 * If the parent is a striped directory then it also needs to locate which
2629 * stripe the name of the child is located, and replace the parent FID
2630 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2631 * it will walk through all of sub-stripes until the child is being
2634 * \param[in] exp export refer to LMV
2635 * \param[in] op_data different parameters transferred beween client
2636 * MD stacks, name, namelen, FIDs etc.
2637 * op_fid1 is the parent FID, op_fid2 is the child
2639 * \param[out] request point to the request of unlink.
2641 * retval 0 if succeed
2642 * negative errno if failed.
2644 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2645 struct ptlrpc_request **request)
2647 struct obd_device *obd = exp->exp_obd;
2648 struct lmv_obd *lmv = &obd->u.lmv;
2649 struct lmv_tgt_desc *tgt = NULL;
2650 struct lmv_tgt_desc *parent_tgt = NULL;
2651 struct mdt_body *body;
2653 int stripe_index = 0;
2654 struct lmv_stripe_md *lsm = op_data->op_mea1;
2657 rc = lmv_check_connect(obd);
2661 /* For striped dir, we need to locate the parent as well */
2663 struct lmv_tgt_desc *tmp;
2665 LASSERT(op_data->op_name != NULL &&
2666 op_data->op_namelen != 0);
2668 tmp = lmv_locate_target_for_name(lmv, lsm,
2670 op_data->op_namelen,
2674 /* return -EBADFD means unknown hash type, might
2675 * need try all sub-stripe here */
2676 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2677 RETURN(PTR_ERR(tmp));
2679 /* Note: both migrating dir and unknown hash dir need to
2680 * try all of sub-stripes, so we need start search the
2681 * name from stripe 0, but migrating dir is already handled
2682 * inside lmv_locate_target_for_name(), so we only check
2683 * unknown hash type directory here */
2684 if (!lmv_is_known_hash_type(lsm)) {
2685 struct lmv_oinfo *oinfo;
2687 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2689 op_data->op_fid1 = oinfo->lmo_fid;
2690 op_data->op_mds = oinfo->lmo_mds;
2695 /* Send unlink requests to the MDT where the child is located */
2696 if (likely(!fid_is_zero(&op_data->op_fid2)))
2697 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2698 else if (lsm != NULL)
2699 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2701 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2704 RETURN(PTR_ERR(tgt));
2706 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2707 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2708 op_data->op_cap = cfs_curproc_cap_pack();
2711 * If child's fid is given, cancel unused locks for it if it is from
2712 * another export than parent.
2714 * LOOKUP lock for child (fid3) should also be cancelled on parent
2715 * tgt_tgt in mdc_unlink().
2717 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2720 * Cancel FULL locks on child (fid3).
2722 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2723 if (IS_ERR(parent_tgt))
2724 RETURN(PTR_ERR(parent_tgt));
2726 if (parent_tgt != tgt) {
2727 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2728 LCK_EX, MDS_INODELOCK_LOOKUP,
2729 MF_MDC_CANCEL_FID3);
2732 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2733 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2737 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2738 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2740 rc = md_unlink(tgt->ltd_exp, op_data, request);
2741 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2744 /* Try next stripe if it is needed. */
2745 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2746 struct lmv_oinfo *oinfo;
2749 if (stripe_index >= lsm->lsm_md_stripe_count)
2752 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2754 op_data->op_fid1 = oinfo->lmo_fid;
2755 op_data->op_mds = oinfo->lmo_mds;
2757 ptlrpc_req_finished(*request);
2760 goto try_next_stripe;
2763 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2767 /* Not cross-ref case, just get out of here. */
2768 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2771 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2772 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2774 /* This is a remote object, try remote MDT, Note: it may
2775 * try more than 1 time here, Considering following case
2776 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2777 * 1. Initially A does not know where remote1 is, it send
2778 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2779 * resend unlink RPC to MDT1 (retry 1st time).
2781 * 2. During the unlink RPC in flight,
2782 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2783 * and create new remote1, but on MDT0
2785 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2786 * /mnt/lustre, then lookup get fid of remote1, and find
2787 * it is remote dir again, and replay -EREMOTE again.
2789 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2791 * In theory, it might try unlimited time here, but it should
2792 * be very rare case. */
2793 op_data->op_fid2 = body->mbo_fid1;
2794 ptlrpc_req_finished(*request);
2800 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2802 struct lmv_obd *lmv = &obd->u.lmv;
2806 case OBD_CLEANUP_EARLY:
2807 /* XXX: here should be calling obd_precleanup() down to
2810 case OBD_CLEANUP_EXPORTS:
2811 fld_client_proc_fini(&lmv->lmv_fld);
2812 lprocfs_obd_cleanup(obd);
2813 lprocfs_free_md_stats(obd);
2821 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2822 __u32 keylen, void *key, __u32 *vallen, void *val,
2823 struct lov_stripe_md *lsm)
2825 struct obd_device *obd;
2826 struct lmv_obd *lmv;
2830 obd = class_exp2obd(exp);
2832 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2833 exp->exp_handle.h_cookie);
2838 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2841 rc = lmv_check_connect(obd);
2845 LASSERT(*vallen == sizeof(__u32));
2846 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2847 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2849 * All tgts should be connected when this gets called.
2851 if (tgt == NULL || tgt->ltd_exp == NULL)
2854 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2859 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2860 KEY_IS(KEY_DEFAULT_EASIZE) ||
2861 KEY_IS(KEY_MAX_COOKIESIZE) ||
2862 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2863 KEY_IS(KEY_CONN_DATA)) {
2864 rc = lmv_check_connect(obd);
2869 * Forwarding this request to first MDS, it should know LOV
2872 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2874 if (!rc && KEY_IS(KEY_CONN_DATA))
2875 exp->exp_connect_data = *(struct obd_connect_data *)val;
2877 } else if (KEY_IS(KEY_TGT_COUNT)) {
2878 *((int *)val) = lmv->desc.ld_tgt_count;
2882 CDEBUG(D_IOCTL, "Invalid key\n");
2886 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2887 obd_count keylen, void *key, obd_count vallen,
2888 void *val, struct ptlrpc_request_set *set)
2890 struct lmv_tgt_desc *tgt = NULL;
2891 struct obd_device *obd;
2892 struct lmv_obd *lmv;
2896 obd = class_exp2obd(exp);
2898 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2899 exp->exp_handle.h_cookie);
2904 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2907 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2910 if (tgt == NULL || tgt->ltd_exp == NULL)
2913 err = obd_set_info_async(env, tgt->ltd_exp,
2914 keylen, key, vallen, val, set);
2925 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2926 struct lmv_mds_md_v1 *lmm1)
2931 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2932 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2933 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2934 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2935 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2936 sizeof(lmm1->lmv_pool_name));
2937 if (cplen >= sizeof(lmm1->lmv_pool_name))
2940 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2941 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2942 &lsm->lsm_md_oinfo[i].lmo_fid);
2946 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2950 bool allocated = false;
2954 LASSERT(lmmp != NULL);
2956 if (*lmmp != NULL && lsm == NULL) {
2959 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2960 lmm_size = lmv_mds_md_size(stripe_count,
2961 le32_to_cpu((*lmmp)->lmv_magic));
2964 OBD_FREE(*lmmp, lmm_size);
2970 if (*lmmp == NULL && lsm == NULL) {
2971 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2972 LASSERT(lmm_size > 0);
2973 OBD_ALLOC(*lmmp, lmm_size);
2976 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2977 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2982 LASSERT(lsm != NULL);
2983 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2984 if (*lmmp == NULL) {
2985 OBD_ALLOC(*lmmp, lmm_size);
2991 switch (lsm->lsm_md_magic) {
2993 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
3000 if (rc != 0 && allocated) {
3001 OBD_FREE(*lmmp, lmm_size);
3007 EXPORT_SYMBOL(lmv_pack_md);
3009 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3010 const struct lmv_mds_md_v1 *lmm1)
3012 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3019 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3020 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3021 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3022 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3023 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3025 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3026 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3027 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3028 sizeof(lsm->lsm_md_pool_name));
3030 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3033 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3034 "layout_version %d\n", lsm->lsm_md_stripe_count,
3035 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3036 lsm->lsm_md_layout_version);
3038 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3039 for (i = 0; i < stripe_count; i++) {
3040 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3041 &lmm1->lmv_stripe_fids[i]);
3042 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3043 &lsm->lsm_md_oinfo[i].lmo_mds);
3046 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3047 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3053 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3054 const union lmv_mds_md *lmm, int stripe_count)
3056 struct lmv_stripe_md *lsm;
3059 bool allocated = false;
3062 LASSERT(lsmp != NULL);
3066 if (lsm != NULL && lmm == NULL) {
3068 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3069 /* For migrating inode, the master stripe and master
3070 * object will be the same, so do not need iput, see
3071 * ll_update_lsm_md */
3072 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3073 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3074 iput(lsm->lsm_md_oinfo[i].lmo_root);
3076 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3077 OBD_FREE(lsm, lsm_size);
3083 if (lsm == NULL && lmm == NULL) {
3084 lsm_size = lmv_stripe_md_size(stripe_count);
3085 OBD_ALLOC(lsm, lsm_size);
3088 lsm->lsm_md_stripe_count = stripe_count;
3093 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3097 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3098 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3099 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3100 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3105 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3106 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3109 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3110 * stripecount should be 0 then.
3112 lsm_size = lmv_stripe_md_size(0);
3114 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3116 OBD_ALLOC(lsm, lsm_size);
3123 switch (le32_to_cpu(lmm->lmv_magic)) {
3125 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3128 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3129 le32_to_cpu(lmm->lmv_magic));
3134 if (rc != 0 && allocated) {
3135 OBD_FREE(lsm, lsm_size);
3142 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3144 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3146 EXPORT_SYMBOL(lmv_alloc_memmd);
3148 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3150 lmv_unpack_md(NULL, &lsm, NULL, 0);
3152 EXPORT_SYMBOL(lmv_free_memmd);
3154 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3155 struct lov_mds_md *lmm, int disk_len)
3157 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3158 (union lmv_mds_md *)lmm, disk_len);
3161 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3162 struct lov_stripe_md *lsm)
3164 struct obd_device *obd = exp->exp_obd;
3165 struct lmv_obd *lmv_obd = &obd->u.lmv;
3166 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3171 stripe_count = lmv->lsm_md_stripe_count;
3173 stripe_count = lmv_obd->desc.ld_tgt_count;
3175 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3178 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3181 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3182 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3183 ldlm_cancel_flags_t flags, void *opaque)
3185 struct obd_device *obd = exp->exp_obd;
3186 struct lmv_obd *lmv = &obd->u.lmv;
3192 LASSERT(fid != NULL);
3194 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3195 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3197 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3200 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3208 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3211 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3212 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3216 if (tgt == NULL || tgt->ltd_exp == NULL)
3218 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3222 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3223 const struct lu_fid *fid, ldlm_type_t type,
3224 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3225 struct lustre_handle *lockh)
3227 struct obd_device *obd = exp->exp_obd;
3228 struct lmv_obd *lmv = &obd->u.lmv;
3234 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3237 * With DNE every object can have two locks in different namespaces:
3238 * lookup lock in space of MDT storing direntry and update/open lock in
3239 * space of MDT storing inode. Try the MDT that the FID maps to first,
3240 * since this can be easily found, and only try others if that fails.
3242 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3243 i < lmv->desc.ld_tgt_count;
3244 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3246 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3247 obd->obd_name, PFID(fid), tgt);
3251 if (lmv->tgts[tgt] == NULL ||
3252 lmv->tgts[tgt]->ltd_exp == NULL ||
3253 lmv->tgts[tgt]->ltd_active == 0)
3256 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3257 type, policy, mode, lockh);
3265 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3266 struct obd_export *dt_exp, struct obd_export *md_exp,
3267 struct lustre_md *md)
3269 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3270 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3272 if (tgt == NULL || tgt->ltd_exp == NULL)
3275 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3278 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3280 struct obd_device *obd = exp->exp_obd;
3281 struct lmv_obd *lmv = &obd->u.lmv;
3282 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3285 if (md->lmv != NULL) {
3286 lmv_free_memmd(md->lmv);
3289 if (tgt == NULL || tgt->ltd_exp == NULL)
3291 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3294 int lmv_set_open_replay_data(struct obd_export *exp,
3295 struct obd_client_handle *och,
3296 struct lookup_intent *it)
3298 struct obd_device *obd = exp->exp_obd;
3299 struct lmv_obd *lmv = &obd->u.lmv;
3300 struct lmv_tgt_desc *tgt;
3303 tgt = lmv_find_target(lmv, &och->och_fid);
3305 RETURN(PTR_ERR(tgt));
3307 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3310 int lmv_clear_open_replay_data(struct obd_export *exp,
3311 struct obd_client_handle *och)
3313 struct obd_device *obd = exp->exp_obd;
3314 struct lmv_obd *lmv = &obd->u.lmv;
3315 struct lmv_tgt_desc *tgt;
3318 tgt = lmv_find_target(lmv, &och->och_fid);
3320 RETURN(PTR_ERR(tgt));
3322 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3325 static int lmv_get_remote_perm(struct obd_export *exp,
3326 const struct lu_fid *fid,
3327 struct obd_capa *oc, __u32 suppgid,
3328 struct ptlrpc_request **request)
3330 struct obd_device *obd = exp->exp_obd;
3331 struct lmv_obd *lmv = &obd->u.lmv;
3332 struct lmv_tgt_desc *tgt;
3336 rc = lmv_check_connect(obd);
3340 tgt = lmv_find_target(lmv, fid);
3342 RETURN(PTR_ERR(tgt));
3344 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3348 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3351 struct obd_device *obd = exp->exp_obd;
3352 struct lmv_obd *lmv = &obd->u.lmv;
3353 struct lmv_tgt_desc *tgt;
3357 rc = lmv_check_connect(obd);
3361 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3363 RETURN(PTR_ERR(tgt));
3365 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3369 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3370 const struct req_msg_field *field, struct obd_capa **oc)
3372 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3373 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3375 if (tgt == NULL || tgt->ltd_exp == NULL)
3377 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3380 int lmv_intent_getattr_async(struct obd_export *exp,
3381 struct md_enqueue_info *minfo,
3382 struct ldlm_enqueue_info *einfo)
3384 struct md_op_data *op_data = &minfo->mi_data;
3385 struct obd_device *obd = exp->exp_obd;
3386 struct lmv_obd *lmv = &obd->u.lmv;
3387 struct lmv_tgt_desc *tgt = NULL;
3391 rc = lmv_check_connect(obd);
3395 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3397 RETURN(PTR_ERR(tgt));
3399 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3403 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3404 struct lu_fid *fid, __u64 *bits)
3406 struct obd_device *obd = exp->exp_obd;
3407 struct lmv_obd *lmv = &obd->u.lmv;
3408 struct lmv_tgt_desc *tgt;
3412 rc = lmv_check_connect(obd);
3416 tgt = lmv_find_target(lmv, fid);
3418 RETURN(PTR_ERR(tgt));
3420 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3424 int lmv_get_fid_from_lsm(struct obd_export *exp,
3425 const struct lmv_stripe_md *lsm,
3426 const char *name, int namelen, struct lu_fid *fid)
3428 const struct lmv_oinfo *oinfo;
3430 LASSERT(lsm != NULL);
3431 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3433 return PTR_ERR(oinfo);
3435 *fid = oinfo->lmo_fid;
3441 * For lmv, only need to send request to master MDT, and the master MDT will
3442 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3443 * we directly fetch data from the slave MDTs.
3445 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3446 struct obd_quotactl *oqctl)
3448 struct obd_device *obd = class_exp2obd(exp);
3449 struct lmv_obd *lmv = &obd->u.lmv;
3450 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3453 __u64 curspace, curinodes;
3457 tgt->ltd_exp == NULL ||
3459 lmv->desc.ld_tgt_count == 0) {
3460 CERROR("master lmv inactive\n");
3464 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3465 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3469 curspace = curinodes = 0;
3470 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3474 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3477 err = obd_quotactl(tgt->ltd_exp, oqctl);
3479 CERROR("getquota on mdt %d failed. %d\n", i, err);
3483 curspace += oqctl->qc_dqblk.dqb_curspace;
3484 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3487 oqctl->qc_dqblk.dqb_curspace = curspace;
3488 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3493 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3494 struct obd_quotactl *oqctl)
3496 struct obd_device *obd = class_exp2obd(exp);
3497 struct lmv_obd *lmv = &obd->u.lmv;
3498 struct lmv_tgt_desc *tgt;
3503 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3506 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3507 CERROR("lmv idx %d inactive\n", i);
3511 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3519 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3520 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3522 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3525 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3526 struct cl_attr *attr)
3530 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3531 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3533 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3534 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3535 i_size_read(inode), inode->i_nlink,
3536 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3537 LTIME_S(inode->i_mtime));
3539 /* for slave stripe, it needs to subtract nlink for . and .. */
3541 attr->cat_nlink += inode->i_nlink - 2;
3543 attr->cat_nlink = inode->i_nlink;
3545 attr->cat_size += i_size_read(inode);
3547 if (attr->cat_atime < LTIME_S(inode->i_atime))
3548 attr->cat_atime = LTIME_S(inode->i_atime);
3550 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3551 attr->cat_ctime = LTIME_S(inode->i_ctime);
3553 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3554 attr->cat_mtime = LTIME_S(inode->i_mtime);
3559 struct obd_ops lmv_obd_ops = {
3560 .o_owner = THIS_MODULE,
3561 .o_setup = lmv_setup,
3562 .o_cleanup = lmv_cleanup,
3563 .o_precleanup = lmv_precleanup,
3564 .o_process_config = lmv_process_config,
3565 .o_connect = lmv_connect,
3566 .o_disconnect = lmv_disconnect,
3567 .o_statfs = lmv_statfs,
3568 .o_get_info = lmv_get_info,
3569 .o_set_info_async = lmv_set_info_async,
3570 .o_packmd = lmv_packmd,
3571 .o_unpackmd = lmv_unpackmd,
3572 .o_notify = lmv_notify,
3573 .o_get_uuid = lmv_get_uuid,
3574 .o_iocontrol = lmv_iocontrol,
3575 .o_quotacheck = lmv_quotacheck,
3576 .o_quotactl = lmv_quotactl
3579 struct md_ops lmv_md_ops = {
3580 .m_getstatus = lmv_getstatus,
3581 .m_null_inode = lmv_null_inode,
3582 .m_find_cbdata = lmv_find_cbdata,
3583 .m_close = lmv_close,
3584 .m_create = lmv_create,
3585 .m_done_writing = lmv_done_writing,
3586 .m_enqueue = lmv_enqueue,
3587 .m_getattr = lmv_getattr,
3588 .m_getxattr = lmv_getxattr,
3589 .m_getattr_name = lmv_getattr_name,
3590 .m_intent_lock = lmv_intent_lock,
3592 .m_rename = lmv_rename,
3593 .m_setattr = lmv_setattr,
3594 .m_setxattr = lmv_setxattr,
3595 .m_fsync = lmv_fsync,
3596 .m_read_page = lmv_read_page,
3597 .m_unlink = lmv_unlink,
3598 .m_init_ea_size = lmv_init_ea_size,
3599 .m_cancel_unused = lmv_cancel_unused,
3600 .m_set_lock_data = lmv_set_lock_data,
3601 .m_lock_match = lmv_lock_match,
3602 .m_get_lustre_md = lmv_get_lustre_md,
3603 .m_free_lustre_md = lmv_free_lustre_md,
3604 .m_update_lsm_md = lmv_update_lsm_md,
3605 .m_merge_attr = lmv_merge_attr,
3606 .m_set_open_replay_data = lmv_set_open_replay_data,
3607 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3608 .m_renew_capa = lmv_renew_capa,
3609 .m_unpack_capa = lmv_unpack_capa,
3610 .m_get_remote_perm = lmv_get_remote_perm,
3611 .m_intent_getattr_async = lmv_intent_getattr_async,
3612 .m_revalidate_lock = lmv_revalidate_lock,
3613 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3616 int __init lmv_init(void)
3618 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3619 #ifndef HAVE_ONLY_PROCFS_SEQ
3622 LUSTRE_LMV_NAME, NULL);
3625 static void lmv_exit(void)
3627 class_unregister_type(LUSTRE_LMV_NAME);
3630 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3631 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3632 MODULE_LICENSE("GPL");
3634 module_init(lmv_init);
3635 module_exit(lmv_exit);