4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
39 #include <linux/slab.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
45 #include <linux/math64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
49 #include <liblustre.h>
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_lmv.h>
58 #include <lprocfs_status.h>
59 #include <cl_object.h>
61 #include <lustre_lite.h>
62 #include <lustre_fid.h>
63 #include <lustre_ioctl.h>
64 #include "lmv_internal.h"
66 /* This hash is only for testing purpose */
67 static inline unsigned int
68 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
71 const unsigned char *p = (const unsigned char *)name;
73 while (--namelen >= 0)
81 static inline unsigned int
82 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
86 hash = lustre_hash_fnv_1a_64(name, namelen);
93 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
94 const char *name, int namelen)
97 __u32 hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
100 if (stripe_count <= 1)
103 /* for migrating object, always start from 0 stripe */
104 if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
108 case LMV_HASH_TYPE_ALL_CHARS:
109 idx = lmv_hash_all_chars(stripe_count, name, namelen);
111 case LMV_HASH_TYPE_FNV_1A_64:
112 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
119 CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
125 static void lmv_activate_target(struct lmv_obd *lmv,
126 struct lmv_tgt_desc *tgt,
129 if (tgt->ltd_active == activate)
132 tgt->ltd_active = activate;
133 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
139 * -EINVAL : UUID can't be found in the LMV's target list
140 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
141 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
143 static int lmv_set_mdc_active(struct lmv_obd *lmv,
144 const struct obd_uuid *uuid,
147 struct lmv_tgt_desc *tgt = NULL;
148 struct obd_device *obd;
153 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
154 lmv, uuid->uuid, activate);
156 spin_lock(&lmv->lmv_lock);
157 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
159 if (tgt == NULL || tgt->ltd_exp == NULL)
162 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
163 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
165 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
169 if (i == lmv->desc.ld_tgt_count)
170 GOTO(out_lmv_lock, rc = -EINVAL);
172 obd = class_exp2obd(tgt->ltd_exp);
174 GOTO(out_lmv_lock, rc = -ENOTCONN);
176 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
177 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
178 obd->obd_type->typ_name, i);
179 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
181 if (tgt->ltd_active == activate) {
182 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
183 activate ? "" : "in");
184 GOTO(out_lmv_lock, rc);
187 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
188 activate ? "" : "in");
189 lmv_activate_target(lmv, tgt, activate);
193 spin_unlock(&lmv->lmv_lock);
197 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
199 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
200 struct lmv_tgt_desc *tgt = lmv->tgts[0];
202 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
205 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
206 enum obd_notify_event ev, void *data)
208 struct obd_connect_data *conn_data;
209 struct lmv_obd *lmv = &obd->u.lmv;
210 struct obd_uuid *uuid;
214 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
215 CERROR("unexpected notification of %s %s!\n",
216 watched->obd_type->typ_name,
221 uuid = &watched->u.cli.cl_target_uuid;
222 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
224 * Set MDC as active before notifying the observer, so the
225 * observer can use the MDC normally.
227 rc = lmv_set_mdc_active(lmv, uuid,
228 ev == OBD_NOTIFY_ACTIVE);
230 CERROR("%sactivation of %s failed: %d\n",
231 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
235 } else if (ev == OBD_NOTIFY_OCD) {
236 conn_data = &watched->u.cli.cl_import->imp_connect_data;
238 * XXX: Make sure that ocd_connect_flags from all targets are
239 * the same. Otherwise one of MDTs runs wrong version or
240 * something like this. --umka
242 obd->obd_self_export->exp_connect_data = *conn_data;
245 else if (ev == OBD_NOTIFY_DISCON) {
247 * For disconnect event, flush fld cache for failout MDS case.
249 fld_client_flush(&lmv->lmv_fld);
253 * Pass the notification up the chain.
255 if (obd->obd_observer)
256 rc = obd_notify(obd->obd_observer, watched, ev, data);
262 * This is fake connect function. Its purpose is to initialize lmv and say
263 * caller that everything is okay. Real connection will be performed later.
265 static int lmv_connect(const struct lu_env *env,
266 struct obd_export **exp, struct obd_device *obd,
267 struct obd_uuid *cluuid, struct obd_connect_data *data,
270 struct lmv_obd *lmv = &obd->u.lmv;
271 struct lustre_handle conn = { 0 };
276 * We don't want to actually do the underlying connections more than
277 * once, so keep track.
280 if (lmv->refcount > 1) {
285 rc = class_connect(&conn, obd, cluuid);
287 CERROR("class_connection() returned %d\n", rc);
291 *exp = class_conn2export(&conn);
292 class_export_get(*exp);
296 lmv->cluuid = *cluuid;
299 lmv->conn_data = *data;
301 if (lmv->targets_proc_entry == NULL) {
302 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
305 if (IS_ERR(lmv->targets_proc_entry)) {
306 CERROR("%s: cannot register "
307 "/proc/fs/lustre/%s/%s/target_obds\n",
308 obd->obd_name, obd->obd_type->typ_name,
310 lmv->targets_proc_entry = NULL;
315 * All real clients should perform actual connection right away, because
316 * it is possible, that LMV will not have opportunity to connect targets
317 * and MDC stuff will be called directly, for instance while reading
318 * ../mdc/../kbytesfree procfs file, etc.
320 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
321 rc = lmv_check_connect(obd);
323 if (rc && lmv->targets_proc_entry != NULL)
324 lprocfs_remove(&lmv->targets_proc_entry);
328 static void lmv_set_timeouts(struct obd_device *obd)
334 if (lmv->server_timeout == 0)
337 if (lmv->connected == 0)
340 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
341 struct lmv_tgt_desc *tgt = lmv->tgts[i];
343 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
346 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
347 KEY_INTERMDS, 0, NULL, NULL);
351 static int lmv_init_ea_size(struct obd_export *exp, int easize,
352 int def_easize, int cookiesize, int def_cookiesize)
354 struct obd_device *obd = exp->exp_obd;
355 struct lmv_obd *lmv = &obd->u.lmv;
361 if (lmv->max_easize < easize) {
362 lmv->max_easize = easize;
365 if (lmv->max_def_easize < def_easize) {
366 lmv->max_def_easize = def_easize;
369 if (lmv->max_cookiesize < cookiesize) {
370 lmv->max_cookiesize = cookiesize;
373 if (lmv->max_def_cookiesize < def_cookiesize) {
374 lmv->max_def_cookiesize = def_cookiesize;
380 if (lmv->connected == 0)
383 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
384 struct lmv_tgt_desc *tgt = lmv->tgts[i];
386 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
387 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
391 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
392 cookiesize, def_cookiesize);
394 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
395 " rc = %d\n", obd->obd_name, i, rc);
402 #define MAX_STRING_SIZE 128
404 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
406 struct lmv_obd *lmv = &obd->u.lmv;
407 struct obd_uuid *cluuid = &lmv->cluuid;
408 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
409 struct obd_device *mdc_obd;
410 struct obd_export *mdc_exp;
411 struct lu_fld_target target;
415 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
418 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
422 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
423 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
424 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
427 if (!mdc_obd->obd_set_up) {
428 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
432 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
433 &lmv->conn_data, NULL);
435 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
440 * Init fid sequence client for this mdc and add new fld target.
442 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
446 target.ft_srv = NULL;
447 target.ft_exp = mdc_exp;
448 target.ft_idx = tgt->ltd_idx;
450 fld_client_add_target(&lmv->lmv_fld, &target);
452 rc = obd_register_observer(mdc_obd, obd);
454 obd_disconnect(mdc_exp);
455 CERROR("target %s register_observer error %d\n",
456 tgt->ltd_uuid.uuid, rc);
460 if (obd->obd_observer) {
462 * Tell the observer about the new target.
464 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
466 (void *)(tgt - lmv->tgts[0]));
468 obd_disconnect(mdc_exp);
474 tgt->ltd_exp = mdc_exp;
475 lmv->desc.ld_active_tgt_count++;
477 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
478 lmv->max_cookiesize, lmv->max_def_cookiesize);
480 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
481 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
482 atomic_read(&obd->obd_refcount));
484 if (lmv->targets_proc_entry != NULL) {
485 struct proc_dir_entry *mdc_symlink;
487 LASSERT(mdc_obd->obd_type != NULL);
488 LASSERT(mdc_obd->obd_type->typ_name != NULL);
489 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
490 lmv->targets_proc_entry,
492 mdc_obd->obd_type->typ_name,
494 if (mdc_symlink == NULL) {
495 CERROR("cannot register LMV target "
496 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
497 obd->obd_type->typ_name, obd->obd_name,
504 static void lmv_del_target(struct lmv_obd *lmv, int index)
506 if (lmv->tgts[index] == NULL)
509 OBD_FREE_PTR(lmv->tgts[index]);
510 lmv->tgts[index] = NULL;
514 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
515 __u32 index, int gen)
517 struct lmv_obd *lmv = &obd->u.lmv;
518 struct lmv_tgt_desc *tgt;
519 int orig_tgt_count = 0;
523 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
527 if (lmv->desc.ld_tgt_count == 0) {
528 struct obd_device *mdc_obd;
530 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
533 lmv_init_unlock(lmv);
534 CERROR("%s: Target %s not attached: rc = %d\n",
535 obd->obd_name, uuidp->uuid, -EINVAL);
540 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
541 tgt = lmv->tgts[index];
542 CERROR("%s: UUID %s already assigned at LOV target index %d:"
543 " rc = %d\n", obd->obd_name,
544 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
545 lmv_init_unlock(lmv);
549 if (index >= lmv->tgts_size) {
550 /* We need to reallocate the lmv target array. */
551 struct lmv_tgt_desc **newtgts, **old = NULL;
555 while (newsize < index + 1)
556 newsize = newsize << 1;
557 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
558 if (newtgts == NULL) {
559 lmv_init_unlock(lmv);
563 if (lmv->tgts_size) {
564 memcpy(newtgts, lmv->tgts,
565 sizeof(*newtgts) * lmv->tgts_size);
567 oldsize = lmv->tgts_size;
571 lmv->tgts_size = newsize;
574 OBD_FREE(old, sizeof(*old) * oldsize);
576 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
582 lmv_init_unlock(lmv);
586 mutex_init(&tgt->ltd_fid_mutex);
587 tgt->ltd_idx = index;
588 tgt->ltd_uuid = *uuidp;
590 lmv->tgts[index] = tgt;
591 if (index >= lmv->desc.ld_tgt_count) {
592 orig_tgt_count = lmv->desc.ld_tgt_count;
593 lmv->desc.ld_tgt_count = index + 1;
596 if (lmv->connected) {
597 rc = lmv_connect_mdc(obd, tgt);
599 spin_lock(&lmv->lmv_lock);
600 if (lmv->desc.ld_tgt_count == index + 1)
601 lmv->desc.ld_tgt_count = orig_tgt_count;
602 memset(tgt, 0, sizeof(*tgt));
603 spin_unlock(&lmv->lmv_lock);
605 int easize = sizeof(struct lmv_stripe_md) +
606 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
607 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
611 lmv_init_unlock(lmv);
615 int lmv_check_connect(struct obd_device *obd)
617 struct lmv_obd *lmv = &obd->u.lmv;
618 struct lmv_tgt_desc *tgt;
628 if (lmv->connected) {
629 lmv_init_unlock(lmv);
633 if (lmv->desc.ld_tgt_count == 0) {
634 lmv_init_unlock(lmv);
635 CERROR("%s: no targets configured.\n", obd->obd_name);
639 LASSERT(lmv->tgts != NULL);
641 if (lmv->tgts[0] == NULL) {
642 lmv_init_unlock(lmv);
643 CERROR("%s: no target configured for index 0.\n",
648 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
649 lmv->cluuid.uuid, obd->obd_name);
651 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
655 rc = lmv_connect_mdc(obd, tgt);
660 lmv_set_timeouts(obd);
661 class_export_put(lmv->exp);
663 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
664 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
665 lmv_init_unlock(lmv);
676 --lmv->desc.ld_active_tgt_count;
677 rc2 = obd_disconnect(tgt->ltd_exp);
679 CERROR("LMV target %s disconnect on "
680 "MDC idx %d: error %d\n",
681 tgt->ltd_uuid.uuid, i, rc2);
685 class_disconnect(lmv->exp);
686 lmv_init_unlock(lmv);
690 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
692 struct lmv_obd *lmv = &obd->u.lmv;
693 struct obd_device *mdc_obd;
697 LASSERT(tgt != NULL);
698 LASSERT(obd != NULL);
700 mdc_obd = class_exp2obd(tgt->ltd_exp);
703 mdc_obd->obd_force = obd->obd_force;
704 mdc_obd->obd_fail = obd->obd_fail;
705 mdc_obd->obd_no_recov = obd->obd_no_recov;
708 if (lmv->targets_proc_entry != NULL)
709 lprocfs_remove_proc_entry(mdc_obd->obd_name,
710 lmv->targets_proc_entry);
712 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
714 CERROR("Can't finanize fids factory\n");
716 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
717 tgt->ltd_exp->exp_obd->obd_name,
718 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
720 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
721 rc = obd_disconnect(tgt->ltd_exp);
723 if (tgt->ltd_active) {
724 CERROR("Target %s disconnect error %d\n",
725 tgt->ltd_uuid.uuid, rc);
729 lmv_activate_target(lmv, tgt, 0);
734 static int lmv_disconnect(struct obd_export *exp)
736 struct obd_device *obd = class_exp2obd(exp);
737 struct lmv_obd *lmv = &obd->u.lmv;
746 * Only disconnect the underlying layers on the final disconnect.
749 if (lmv->refcount != 0)
752 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
753 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
756 lmv_disconnect_mdc(obd, lmv->tgts[i]);
759 if (lmv->targets_proc_entry != NULL)
760 lprocfs_remove(&lmv->targets_proc_entry);
762 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
763 obd->obd_type->typ_name, obd->obd_name);
767 * This is the case when no real connection is established by
768 * lmv_check_connect().
771 class_export_put(exp);
772 rc = class_disconnect(exp);
773 if (lmv->refcount == 0)
778 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
780 struct obd_device *obddev = class_exp2obd(exp);
781 struct lmv_obd *lmv = &obddev->u.lmv;
782 struct getinfo_fid2path *gf;
783 struct lmv_tgt_desc *tgt;
784 struct getinfo_fid2path *remote_gf = NULL;
785 int remote_gf_size = 0;
788 gf = (struct getinfo_fid2path *)karg;
789 tgt = lmv_find_target(lmv, &gf->gf_fid);
791 RETURN(PTR_ERR(tgt));
794 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
795 if (rc != 0 && rc != -EREMOTE)
796 GOTO(out_fid2path, rc);
798 /* If remote_gf != NULL, it means just building the
799 * path on the remote MDT, copy this path segement to gf */
800 if (remote_gf != NULL) {
801 struct getinfo_fid2path *ori_gf;
804 ori_gf = (struct getinfo_fid2path *)karg;
805 if (strlen(ori_gf->gf_path) +
806 strlen(gf->gf_path) > ori_gf->gf_pathlen)
807 GOTO(out_fid2path, rc = -EOVERFLOW);
809 ptr = ori_gf->gf_path;
811 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
812 strlen(ori_gf->gf_path));
814 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
815 ptr += strlen(gf->gf_path);
819 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
820 tgt->ltd_exp->exp_obd->obd_name,
821 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
825 GOTO(out_fid2path, rc);
827 /* sigh, has to go to another MDT to do path building further */
828 if (remote_gf == NULL) {
829 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
830 OBD_ALLOC(remote_gf, remote_gf_size);
831 if (remote_gf == NULL)
832 GOTO(out_fid2path, rc = -ENOMEM);
833 remote_gf->gf_pathlen = PATH_MAX;
836 if (!fid_is_sane(&gf->gf_fid)) {
837 CERROR("%s: invalid FID "DFID": rc = %d\n",
838 tgt->ltd_exp->exp_obd->obd_name,
839 PFID(&gf->gf_fid), -EINVAL);
840 GOTO(out_fid2path, rc = -EINVAL);
843 tgt = lmv_find_target(lmv, &gf->gf_fid);
845 GOTO(out_fid2path, rc = -EINVAL);
847 remote_gf->gf_fid = gf->gf_fid;
848 remote_gf->gf_recno = -1;
849 remote_gf->gf_linkno = -1;
850 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
852 goto repeat_fid2path;
855 if (remote_gf != NULL)
856 OBD_FREE(remote_gf, remote_gf_size);
860 static int lmv_hsm_req_count(struct lmv_obd *lmv,
861 const struct hsm_user_request *hur,
862 const struct lmv_tgt_desc *tgt_mds)
866 struct lmv_tgt_desc *curr_tgt;
868 /* count how many requests must be sent to the given target */
869 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
870 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
871 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
877 static void lmv_hsm_req_build(struct lmv_obd *lmv,
878 struct hsm_user_request *hur_in,
879 const struct lmv_tgt_desc *tgt_mds,
880 struct hsm_user_request *hur_out)
883 struct lmv_tgt_desc *curr_tgt;
885 /* build the hsm_user_request for the given target */
886 hur_out->hur_request = hur_in->hur_request;
888 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
889 curr_tgt = lmv_find_target(lmv,
890 &hur_in->hur_user_item[i].hui_fid);
891 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
892 hur_out->hur_user_item[nr_out] =
893 hur_in->hur_user_item[i];
897 hur_out->hur_request.hr_itemcount = nr_out;
898 memcpy(hur_data(hur_out), hur_data(hur_in),
899 hur_in->hur_request.hr_data_len);
902 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
903 struct lustre_kernelcomm *lk, void *uarg)
907 struct kkuc_ct_data *kcd = NULL;
910 /* unregister request (call from llapi_hsm_copytool_fini) */
911 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
912 struct lmv_tgt_desc *tgt = lmv->tgts[i];
914 if (tgt == NULL || tgt->ltd_exp == NULL)
916 /* best effort: try to clean as much as possible
917 * (continue on error) */
918 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
921 /* Whatever the result, remove copytool from kuc groups.
922 * Unreached coordinators will get EPIPE on next requests
923 * and will unregister automatically.
925 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
932 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
933 struct lustre_kernelcomm *lk, void *uarg)
938 bool any_set = false;
939 struct kkuc_ct_data *kcd;
942 /* All or nothing: try to register to all MDS.
943 * In case of failure, unregister from previous MDS,
944 * except if it because of inactive target. */
945 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
946 struct lmv_tgt_desc *tgt = lmv->tgts[i];
948 if (tgt == NULL || tgt->ltd_exp == NULL)
950 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
952 if (tgt->ltd_active) {
953 /* permanent error */
954 CERROR("%s: iocontrol MDC %s on MDT"
955 " idx %d cmd %x: err = %d\n",
956 class_exp2obd(lmv->exp)->obd_name,
957 tgt->ltd_uuid.uuid, i, cmd, err);
959 lk->lk_flags |= LK_FLG_STOP;
960 /* unregister from previous MDS */
961 for (j = 0; j < i; j++) {
963 if (tgt == NULL || tgt->ltd_exp == NULL)
965 obd_iocontrol(cmd, tgt->ltd_exp, len,
970 /* else: transient error.
971 * kuc will register to the missing MDT
979 /* no registration done: return error */
982 /* at least one registration done, with no failure */
983 filp = fget(lk->lk_wfd);
992 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
993 kcd->kcd_uuid = lmv->cluuid;
994 kcd->kcd_archive = lk->lk_data;
996 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
1009 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1010 int len, void *karg, void *uarg)
1012 struct obd_device *obddev = class_exp2obd(exp);
1013 struct lmv_obd *lmv = &obddev->u.lmv;
1014 struct lmv_tgt_desc *tgt = NULL;
1018 __u32 count = lmv->desc.ld_tgt_count;
1025 case IOC_OBD_STATFS: {
1026 struct obd_ioctl_data *data = karg;
1027 struct obd_device *mdc_obd;
1028 struct obd_statfs stat_buf = {0};
1031 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1032 if ((index >= count))
1035 tgt = lmv->tgts[index];
1036 if (tgt == NULL || !tgt->ltd_active)
1039 mdc_obd = class_exp2obd(tgt->ltd_exp);
1044 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1045 min((int) data->ioc_plen2,
1046 (int) sizeof(struct obd_uuid))))
1049 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1050 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1054 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1055 min((int) data->ioc_plen1,
1056 (int) sizeof(stat_buf))))
1060 case OBD_IOC_QUOTACTL: {
1061 struct if_quotactl *qctl = karg;
1062 struct obd_quotactl *oqctl;
1064 if (qctl->qc_valid == QC_MDTIDX) {
1065 if (count <= qctl->qc_idx)
1068 tgt = lmv->tgts[qctl->qc_idx];
1069 if (tgt == NULL || tgt->ltd_exp == NULL)
1071 } else if (qctl->qc_valid == QC_UUID) {
1072 for (i = 0; i < count; i++) {
1076 if (!obd_uuid_equals(&tgt->ltd_uuid,
1080 if (tgt->ltd_exp == NULL)
1092 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1093 OBD_ALLOC_PTR(oqctl);
1097 QCTL_COPY(oqctl, qctl);
1098 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1100 QCTL_COPY(qctl, oqctl);
1101 qctl->qc_valid = QC_MDTIDX;
1102 qctl->obd_uuid = tgt->ltd_uuid;
1104 OBD_FREE_PTR(oqctl);
1107 case OBD_IOC_CHANGELOG_SEND:
1108 case OBD_IOC_CHANGELOG_CLEAR: {
1109 struct ioc_changelog *icc = karg;
1111 if (icc->icc_mdtindex >= count)
1114 tgt = lmv->tgts[icc->icc_mdtindex];
1115 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1117 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1120 case LL_IOC_GET_CONNECT_FLAGS: {
1122 if (tgt == NULL || tgt->ltd_exp == NULL)
1124 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1127 case OBD_IOC_FID2PATH: {
1128 rc = lmv_fid2path(exp, len, karg, uarg);
1131 case LL_IOC_HSM_STATE_GET:
1132 case LL_IOC_HSM_STATE_SET:
1133 case LL_IOC_HSM_ACTION: {
1134 struct md_op_data *op_data = karg;
1136 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1138 RETURN(PTR_ERR(tgt));
1140 if (tgt->ltd_exp == NULL)
1143 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1146 case LL_IOC_HSM_PROGRESS: {
1147 const struct hsm_progress_kernel *hpk = karg;
1149 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1151 RETURN(PTR_ERR(tgt));
1152 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1155 case LL_IOC_HSM_REQUEST: {
1156 struct hsm_user_request *hur = karg;
1157 unsigned int reqcount = hur->hur_request.hr_itemcount;
1162 /* if the request is about a single fid
1163 * or if there is a single MDS, no need to split
1165 if (reqcount == 1 || count == 1) {
1166 tgt = lmv_find_target(lmv,
1167 &hur->hur_user_item[0].hui_fid);
1169 RETURN(PTR_ERR(tgt));
1170 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1172 /* split fid list to their respective MDS */
1173 for (i = 0; i < count; i++) {
1174 unsigned int nr, reqlen;
1176 struct hsm_user_request *req;
1179 if (tgt == NULL || tgt->ltd_exp == NULL)
1182 nr = lmv_hsm_req_count(lmv, hur, tgt);
1183 if (nr == 0) /* nothing for this MDS */
1186 /* build a request with fids for this MDS */
1187 reqlen = offsetof(typeof(*hur),
1189 + hur->hur_request.hr_data_len;
1190 OBD_ALLOC_LARGE(req, reqlen);
1194 lmv_hsm_req_build(lmv, hur, tgt, req);
1196 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1198 if (rc1 != 0 && rc == 0)
1200 OBD_FREE_LARGE(req, reqlen);
1205 case LL_IOC_LOV_SWAP_LAYOUTS: {
1206 struct md_op_data *op_data = karg;
1207 struct lmv_tgt_desc *tgt1, *tgt2;
1209 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1211 RETURN(PTR_ERR(tgt1));
1213 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1215 RETURN(PTR_ERR(tgt2));
1217 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1220 /* only files on same MDT can have their layouts swapped */
1221 if (tgt1->ltd_idx != tgt2->ltd_idx)
1224 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1227 case LL_IOC_HSM_CT_START: {
1228 struct lustre_kernelcomm *lk = karg;
1229 if (lk->lk_flags & LK_FLG_STOP)
1230 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1232 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1236 for (i = 0; i < count; i++) {
1237 struct obd_device *mdc_obd;
1241 if (tgt == NULL || tgt->ltd_exp == NULL)
1243 /* ll_umount_begin() sets force flag but for lmv, not
1244 * mdc. Let's pass it through */
1245 mdc_obd = class_exp2obd(tgt->ltd_exp);
1246 mdc_obd->obd_force = obddev->obd_force;
1247 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1248 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1251 if (tgt->ltd_active) {
1252 CERROR("error: iocontrol MDC %s on MDT"
1253 " idx %d cmd %x: err = %d\n",
1254 tgt->ltd_uuid.uuid, i, cmd, err);
1268 static int lmv_all_chars_policy(int count, const char *name,
1279 static int lmv_nid_policy(struct lmv_obd *lmv)
1281 struct obd_import *imp;
1285 * XXX: To get nid we assume that underlying obd device is mdc.
1287 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1288 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1289 return id % lmv->desc.ld_tgt_count;
1292 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1293 placement_policy_t placement)
1295 switch (placement) {
1296 case PLACEMENT_CHAR_POLICY:
1297 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1299 op_data->op_namelen);
1300 case PLACEMENT_NID_POLICY:
1301 return lmv_nid_policy(lmv);
1307 CERROR("Unsupported placement policy %x\n", placement);
1313 * This is _inode_ placement policy function (not name).
1315 static int lmv_placement_policy(struct obd_device *obd,
1316 struct md_op_data *op_data,
1319 struct lmv_obd *lmv = &obd->u.lmv;
1322 LASSERT(mds != NULL);
1324 if (lmv->desc.ld_tgt_count == 1) {
1330 * If stripe_offset is provided during setdirstripe
1331 * (setdirstripe -i xx), xx MDS will be choosen.
1333 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1334 struct lmv_user_md *lum;
1336 lum = op_data->op_data;
1338 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1339 *mds = le32_to_cpu(lum->lum_stripe_offset);
1341 /* -1 means default, which will be in the same MDT with
1343 *mds = op_data->op_mds;
1344 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1347 /* Allocate new fid on target according to operation type and
1348 * parent home mds. */
1349 *mds = op_data->op_mds;
1355 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1358 struct lmv_tgt_desc *tgt;
1362 tgt = lmv_get_target(lmv, mds, NULL);
1364 RETURN(PTR_ERR(tgt));
1367 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1368 * on server that seq in new allocated fid is not yet known.
1370 mutex_lock(&tgt->ltd_fid_mutex);
1372 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1373 GOTO(out, rc = -ENODEV);
1376 * Asking underlying tgt layer to allocate new fid.
1378 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1380 LASSERT(fid_is_sane(fid));
1386 mutex_unlock(&tgt->ltd_fid_mutex);
1390 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1391 struct lu_fid *fid, struct md_op_data *op_data)
1393 struct obd_device *obd = class_exp2obd(exp);
1394 struct lmv_obd *lmv = &obd->u.lmv;
1399 LASSERT(op_data != NULL);
1400 LASSERT(fid != NULL);
1402 rc = lmv_placement_policy(obd, op_data, &mds);
1404 CERROR("Can't get target for allocating fid, "
1409 rc = __lmv_fid_alloc(lmv, fid, mds);
1411 CERROR("Can't alloc new fid, rc %d\n", rc);
1418 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1420 struct lmv_obd *lmv = &obd->u.lmv;
1421 struct lmv_desc *desc;
1425 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1426 CERROR("LMV setup requires a descriptor\n");
1430 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1431 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1432 CERROR("Lmv descriptor size wrong: %d > %d\n",
1433 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1437 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1438 if (lmv->tgts == NULL)
1440 lmv->tgts_size = 32;
1442 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1443 lmv->desc.ld_tgt_count = 0;
1444 lmv->desc.ld_active_tgt_count = 0;
1445 lmv->max_cookiesize = 0;
1446 lmv->max_def_easize = 0;
1447 lmv->max_easize = 0;
1448 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1450 spin_lock_init(&lmv->lmv_lock);
1451 mutex_init(&lmv->init_mutex);
1454 obd->obd_vars = lprocfs_lmv_obd_vars;
1455 lprocfs_seq_obd_setup(obd);
1456 lprocfs_alloc_md_stats(obd, 0);
1457 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1458 0444, &lmv_proc_target_fops, obd);
1460 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1463 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1464 LUSTRE_CLI_FLD_HASH_DHT);
1466 CERROR("Can't init FLD, err %d\n", rc);
1476 static int lmv_cleanup(struct obd_device *obd)
1478 struct lmv_obd *lmv = &obd->u.lmv;
1481 fld_client_fini(&lmv->lmv_fld);
1482 if (lmv->tgts != NULL) {
1484 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1485 if (lmv->tgts[i] == NULL)
1487 lmv_del_target(lmv, i);
1489 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1495 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1497 struct lustre_cfg *lcfg = buf;
1498 struct obd_uuid obd_uuid;
1504 switch (lcfg->lcfg_command) {
1506 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1507 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1508 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1509 GOTO(out, rc = -EINVAL);
1511 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1513 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1514 GOTO(out, rc = -EINVAL);
1515 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1516 GOTO(out, rc = -EINVAL);
1517 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1520 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1521 GOTO(out, rc = -EINVAL);
1527 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1528 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1530 struct obd_device *obd = class_exp2obd(exp);
1531 struct lmv_obd *lmv = &obd->u.lmv;
1532 struct obd_statfs *temp;
1537 rc = lmv_check_connect(obd);
1541 OBD_ALLOC(temp, sizeof(*temp));
1545 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1546 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1549 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1552 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1553 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1555 GOTO(out_free_temp, rc);
1560 /* If the statfs is from mount, it will needs
1561 * retrieve necessary information from MDT0.
1562 * i.e. mount does not need the merged osfs
1564 * And also clients can be mounted as long as
1565 * MDT0 is in service*/
1566 if (flags & OBD_STATFS_FOR_MDT0)
1567 GOTO(out_free_temp, rc);
1569 osfs->os_bavail += temp->os_bavail;
1570 osfs->os_blocks += temp->os_blocks;
1571 osfs->os_ffree += temp->os_ffree;
1572 osfs->os_files += temp->os_files;
1578 OBD_FREE(temp, sizeof(*temp));
1582 static int lmv_getstatus(struct obd_export *exp,
1584 struct obd_capa **pc)
1586 struct obd_device *obd = exp->exp_obd;
1587 struct lmv_obd *lmv = &obd->u.lmv;
1591 rc = lmv_check_connect(obd);
1595 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1599 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1600 struct obd_capa *oc, obd_valid valid, const char *name,
1601 const char *input, int input_size, int output_size,
1602 int flags, struct ptlrpc_request **request)
1604 struct obd_device *obd = exp->exp_obd;
1605 struct lmv_obd *lmv = &obd->u.lmv;
1606 struct lmv_tgt_desc *tgt;
1610 rc = lmv_check_connect(obd);
1614 tgt = lmv_find_target(lmv, fid);
1616 RETURN(PTR_ERR(tgt));
1618 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1619 input_size, output_size, flags, request);
1624 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1625 struct obd_capa *oc, obd_valid valid, const char *name,
1626 const char *input, int input_size, int output_size,
1627 int flags, __u32 suppgid,
1628 struct ptlrpc_request **request)
1630 struct obd_device *obd = exp->exp_obd;
1631 struct lmv_obd *lmv = &obd->u.lmv;
1632 struct lmv_tgt_desc *tgt;
1636 rc = lmv_check_connect(obd);
1640 tgt = lmv_find_target(lmv, fid);
1642 RETURN(PTR_ERR(tgt));
1644 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1645 input_size, output_size, flags, suppgid,
1651 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1652 struct ptlrpc_request **request)
1654 struct obd_device *obd = exp->exp_obd;
1655 struct lmv_obd *lmv = &obd->u.lmv;
1656 struct lmv_tgt_desc *tgt;
1660 rc = lmv_check_connect(obd);
1664 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1666 RETURN(PTR_ERR(tgt));
1668 if (op_data->op_flags & MF_GET_MDT_IDX) {
1669 op_data->op_mds = tgt->ltd_idx;
1673 rc = md_getattr(tgt->ltd_exp, op_data, request);
1678 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1680 struct obd_device *obd = exp->exp_obd;
1681 struct lmv_obd *lmv = &obd->u.lmv;
1686 rc = lmv_check_connect(obd);
1690 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1693 * With DNE every object can have two locks in different namespaces:
1694 * lookup lock in space of MDT storing direntry and update/open lock in
1695 * space of MDT storing inode.
1697 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1698 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1700 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1706 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1707 ldlm_iterator_t it, void *data)
1709 struct obd_device *obd = exp->exp_obd;
1710 struct lmv_obd *lmv = &obd->u.lmv;
1716 rc = lmv_check_connect(obd);
1720 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1723 * With DNE every object can have two locks in different namespaces:
1724 * lookup lock in space of MDT storing direntry and update/open lock in
1725 * space of MDT storing inode. Try the MDT that the FID maps to first,
1726 * since this can be easily found, and only try others if that fails.
1728 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1729 i < lmv->desc.ld_tgt_count;
1730 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1732 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1733 obd->obd_name, PFID(fid), tgt);
1737 if (lmv->tgts[tgt] == NULL ||
1738 lmv->tgts[tgt]->ltd_exp == NULL)
1741 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1750 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1751 struct md_open_data *mod, struct ptlrpc_request **request)
1753 struct obd_device *obd = exp->exp_obd;
1754 struct lmv_obd *lmv = &obd->u.lmv;
1755 struct lmv_tgt_desc *tgt;
1759 rc = lmv_check_connect(obd);
1763 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1765 RETURN(PTR_ERR(tgt));
1767 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1768 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1773 * Choosing the MDT by name or FID in @op_data.
1774 * For non-striped directory, it will locate MDT by fid.
1775 * For striped-directory, it will locate MDT by name. And also
1776 * it will reset op_fid1 with the FID of the choosen stripe.
1778 struct lmv_tgt_desc *
1779 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1780 const char *name, int namelen, struct lu_fid *fid,
1783 struct lmv_tgt_desc *tgt;
1784 const struct lmv_oinfo *oinfo;
1786 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1788 RETURN(ERR_CAST(oinfo));
1789 *fid = oinfo->lmo_fid;
1790 *mds = oinfo->lmo_mds;
1791 tgt = lmv_get_target(lmv, *mds, NULL);
1793 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1798 * Locate mds by fid or name
1800 * For striped directory (lsm != NULL), it will locate the stripe
1801 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1802 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1803 * walk through all of stripes to locate the entry.
1805 * For normal direcotry, it will locate MDS by FID directly.
1806 * \param[in] lmv LMV device
1807 * \param[in] op_data client MD stack parameters, name, namelen
1809 * \param[in] fid object FID used to locate MDS.
1811 * retval pointer to the lmv_tgt_desc if succeed.
1812 * ERR_PTR(errno) if failed.
1815 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1818 struct lmv_stripe_md *lsm = op_data->op_mea1;
1819 struct lmv_tgt_desc *tgt;
1821 if (lsm == NULL || op_data->op_namelen == 0) {
1822 tgt = lmv_find_target(lmv, fid);
1826 op_data->op_mds = tgt->ltd_idx;
1830 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1831 op_data->op_namelen, fid,
1835 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1836 const void *data, int datalen, int mode, __u32 uid,
1837 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1838 struct ptlrpc_request **request)
1840 struct obd_device *obd = exp->exp_obd;
1841 struct lmv_obd *lmv = &obd->u.lmv;
1842 struct lmv_tgt_desc *tgt;
1846 rc = lmv_check_connect(obd);
1850 if (!lmv->desc.ld_active_tgt_count)
1853 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1855 RETURN(PTR_ERR(tgt));
1857 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1858 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1861 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1865 /* Send the create request to the MDT where the object
1866 * will be located */
1867 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1869 RETURN(PTR_ERR(tgt));
1871 op_data->op_mds = tgt->ltd_idx;
1873 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1874 PFID(&op_data->op_fid2), op_data->op_mds);
1876 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1877 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1878 cap_effective, rdev, request);
1880 if (*request == NULL)
1882 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1887 static int lmv_done_writing(struct obd_export *exp,
1888 struct md_op_data *op_data,
1889 struct md_open_data *mod)
1891 struct obd_device *obd = exp->exp_obd;
1892 struct lmv_obd *lmv = &obd->u.lmv;
1893 struct lmv_tgt_desc *tgt;
1897 rc = lmv_check_connect(obd);
1901 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1903 RETURN(PTR_ERR(tgt));
1905 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1910 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1911 const union ldlm_policy_data *policy,
1912 struct lookup_intent *it, struct md_op_data *op_data,
1913 struct lustre_handle *lockh, __u64 extra_lock_flags)
1915 struct obd_device *obd = exp->exp_obd;
1916 struct lmv_obd *lmv = &obd->u.lmv;
1917 struct lmv_tgt_desc *tgt;
1921 rc = lmv_check_connect(obd);
1925 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1926 LL_IT2STR(it), PFID(&op_data->op_fid1));
1928 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1930 RETURN(PTR_ERR(tgt));
1932 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1933 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1935 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1942 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1943 struct ptlrpc_request **preq)
1945 struct ptlrpc_request *req = NULL;
1946 struct obd_device *obd = exp->exp_obd;
1947 struct lmv_obd *lmv = &obd->u.lmv;
1948 struct lmv_tgt_desc *tgt;
1949 struct mdt_body *body;
1953 rc = lmv_check_connect(obd);
1957 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1959 RETURN(PTR_ERR(tgt));
1961 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1962 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1965 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1969 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1970 LASSERT(body != NULL);
1972 if (body->mbo_valid & OBD_MD_MDS) {
1973 struct lu_fid rid = body->mbo_fid1;
1974 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1977 tgt = lmv_find_target(lmv, &rid);
1979 ptlrpc_req_finished(*preq);
1981 RETURN(PTR_ERR(tgt));
1984 op_data->op_fid1 = rid;
1985 op_data->op_valid |= OBD_MD_FLCROSSREF;
1986 op_data->op_namelen = 0;
1987 op_data->op_name = NULL;
1988 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1989 ptlrpc_req_finished(*preq);
1996 #define md_op_data_fid(op_data, fl) \
1997 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1998 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1999 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2000 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2003 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2004 struct md_op_data *op_data,
2005 int op_tgt, ldlm_mode_t mode, int bits, int flag)
2007 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2008 struct obd_device *obd = exp->exp_obd;
2009 struct lmv_obd *lmv = &obd->u.lmv;
2010 ldlm_policy_data_t policy = {{ 0 }};
2014 if (!fid_is_sane(fid))
2018 tgt = lmv_find_target(lmv, fid);
2020 RETURN(PTR_ERR(tgt));
2023 if (tgt->ltd_idx != op_tgt) {
2024 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2025 policy.l_inodebits.bits = bits;
2026 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2027 mode, LCF_ASYNC, NULL);
2030 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2032 op_data->op_flags |= flag;
2040 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2043 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2044 struct ptlrpc_request **request)
2046 struct obd_device *obd = exp->exp_obd;
2047 struct lmv_obd *lmv = &obd->u.lmv;
2048 struct lmv_tgt_desc *tgt;
2052 rc = lmv_check_connect(obd);
2056 LASSERT(op_data->op_namelen != 0);
2058 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2059 PFID(&op_data->op_fid2), op_data->op_namelen,
2060 op_data->op_name, PFID(&op_data->op_fid1));
2062 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2063 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2064 op_data->op_cap = cfs_curproc_cap_pack();
2065 if (op_data->op_mea2 != NULL) {
2066 struct lmv_stripe_md *lsm = op_data->op_mea2;
2067 const struct lmv_oinfo *oinfo;
2069 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2070 op_data->op_namelen);
2072 RETURN(PTR_ERR(oinfo));
2074 op_data->op_fid2 = oinfo->lmo_fid;
2077 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2079 RETURN(PTR_ERR(tgt));
2082 * Cancel UPDATE lock on child (fid1).
2084 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2085 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2086 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2090 rc = md_link(tgt->ltd_exp, op_data, request);
2095 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2096 const char *old, int oldlen, const char *new, int newlen,
2097 struct ptlrpc_request **request)
2099 struct obd_device *obd = exp->exp_obd;
2100 struct lmv_obd *lmv = &obd->u.lmv;
2101 struct lmv_tgt_desc *src_tgt;
2105 LASSERT(oldlen != 0);
2107 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2108 oldlen, old, PFID(&op_data->op_fid1),
2109 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2110 newlen, new, PFID(&op_data->op_fid2),
2111 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2113 rc = lmv_check_connect(obd);
2117 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2118 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2119 op_data->op_cap = cfs_curproc_cap_pack();
2120 if (op_data->op_cli_flags & CLI_MIGRATE) {
2121 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2122 PFID(&op_data->op_fid3));
2123 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2126 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2128 if (op_data->op_mea1 != NULL) {
2129 struct lmv_stripe_md *lsm = op_data->op_mea1;
2131 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2135 if (IS_ERR(src_tgt))
2136 RETURN(PTR_ERR(src_tgt));
2138 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2139 if (IS_ERR(src_tgt))
2140 RETURN(PTR_ERR(src_tgt));
2142 op_data->op_mds = src_tgt->ltd_idx;
2145 if (op_data->op_mea2) {
2146 struct lmv_stripe_md *lsm = op_data->op_mea2;
2147 const struct lmv_oinfo *oinfo;
2149 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2151 RETURN(PTR_ERR(oinfo));
2153 op_data->op_fid2 = oinfo->lmo_fid;
2156 if (IS_ERR(src_tgt))
2157 RETURN(PTR_ERR(src_tgt));
2160 * LOOKUP lock on src child (fid3) should also be cancelled for
2161 * src_tgt in mdc_rename.
2163 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2166 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2169 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2170 LCK_EX, MDS_INODELOCK_UPDATE,
2171 MF_MDC_CANCEL_FID2);
2176 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2178 if (fid_is_sane(&op_data->op_fid3)) {
2179 struct lmv_tgt_desc *tgt;
2181 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2183 RETURN(PTR_ERR(tgt));
2185 /* Cancel LOOKUP lock on its parent */
2186 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2187 LCK_EX, MDS_INODELOCK_LOOKUP,
2188 MF_MDC_CANCEL_FID3);
2192 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2193 LCK_EX, MDS_INODELOCK_FULL,
2194 MF_MDC_CANCEL_FID3);
2200 * Cancel all the locks on tgt child (fid4).
2202 if (fid_is_sane(&op_data->op_fid4))
2203 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2204 LCK_EX, MDS_INODELOCK_FULL,
2205 MF_MDC_CANCEL_FID4);
2207 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2208 op_data->op_mds, PFID(&op_data->op_fid2));
2210 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2216 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2217 void *ea, int ealen, void *ea2, int ea2len,
2218 struct ptlrpc_request **request,
2219 struct md_open_data **mod)
2221 struct obd_device *obd = exp->exp_obd;
2222 struct lmv_obd *lmv = &obd->u.lmv;
2223 struct lmv_tgt_desc *tgt;
2227 rc = lmv_check_connect(obd);
2231 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2232 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2234 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2235 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2237 RETURN(PTR_ERR(tgt));
2239 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2240 ea2len, request, mod);
2245 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2246 struct obd_capa *oc, struct ptlrpc_request **request)
2248 struct obd_device *obd = exp->exp_obd;
2249 struct lmv_obd *lmv = &obd->u.lmv;
2250 struct lmv_tgt_desc *tgt;
2254 rc = lmv_check_connect(obd);
2258 tgt = lmv_find_target(lmv, fid);
2260 RETURN(PTR_ERR(tgt));
2262 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2267 * Adjust a set of pages, each page containing an array of lu_dirpages,
2268 * so that each page can be used as a single logical lu_dirpage.
2270 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2271 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2272 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
2273 * value is used as a cookie to request the next lu_dirpage in a
2274 * directory listing that spans multiple pages (two in this example):
2277 * .|--------v------- -----.
2278 * |s|e|f|p|ent|ent| ... |ent|
2279 * '--|-------------- -----' Each CFS_PAGE contains a single
2280 * '------. lu_dirpage.
2281 * .---------v------- -----.
2282 * |s|e|f|p|ent| 0 | ... | 0 |
2283 * '----------------- -----'
2285 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2286 * larger than LU_PAGE_SIZE, a single host page may contain multiple
2287 * lu_dirpages. After reading the lu_dirpages from the MDS, the
2288 * ldp_hash_end of the first lu_dirpage refers to the one immediately
2289 * after it in the same CFS_PAGE (arrows simplified for brevity, but
2290 * in general e0==s1, e1==s2, etc.):
2292 * .-------------------- -----.
2293 * |s0|e0|f0|p|ent|ent| ... |ent|
2294 * |---v---------------- -----|
2295 * |s1|e1|f1|p|ent|ent| ... |ent|
2296 * |---v---------------- -----| Here, each CFS_PAGE contains
2297 * ... multiple lu_dirpages.
2298 * |---v---------------- -----|
2299 * |s'|e'|f'|p|ent|ent| ... |ent|
2300 * '---|---------------- -----'
2302 * .----------------------------.
2305 * This structure is transformed into a single logical lu_dirpage as follows:
2307 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2308 * labeled 'next CFS_PAGE'.
2310 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2311 * a hash collision with the next page exists.
2313 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2314 * to the first entry of the next lu_dirpage.
2316 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2317 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2321 for (i = 0; i < ncfspgs; i++) {
2322 struct lu_dirpage *dp = kmap(pages[i]);
2323 struct lu_dirpage *first = dp;
2324 struct lu_dirent *end_dirent = NULL;
2325 struct lu_dirent *ent;
2326 __u64 hash_end = dp->ldp_hash_end;
2327 __u32 flags = dp->ldp_flags;
2329 while (--nlupgs > 0) {
2330 ent = lu_dirent_start(dp);
2331 for (end_dirent = ent; ent != NULL;
2332 end_dirent = ent, ent = lu_dirent_next(ent));
2334 /* Advance dp to next lu_dirpage. */
2335 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2337 /* Check if we've reached the end of the CFS_PAGE. */
2338 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2341 /* Save the hash and flags of this lu_dirpage. */
2342 hash_end = dp->ldp_hash_end;
2343 flags = dp->ldp_flags;
2345 /* Check if lu_dirpage contains no entries. */
2349 /* Enlarge the end entry lde_reclen from 0 to
2350 * first entry of next lu_dirpage. */
2351 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2352 end_dirent->lde_reclen =
2353 cpu_to_le16((char *)(dp->ldp_entries) -
2354 (char *)end_dirent);
2357 first->ldp_hash_end = hash_end;
2358 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2359 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2363 LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2366 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2367 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2370 * Get current minimum entry from striped directory
2372 * This function will search the dir entry, whose hash value is the
2373 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2374 * only being called for striped directory.
2376 * \param[in] exp export of LMV
2377 * \param[in] op_data parameters transferred beween client MD stack
2378 * stripe_information will be included in this
2380 * \param[in] cb_op ldlm callback being used in enqueue in
2382 * \param[in] hash_offset the hash value, which is used to locate
2383 * minum(closet) dir entry
2384 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2385 * index of last entry, so to avoid hash conflict
2386 * between stripes. It will also be used to
2387 * return the stripe index of current dir entry.
2388 * \param[in|out] entp the minum entry and it also is being used
2389 * to input the last dir entry to resolve the
2392 * \param[out] ppage the page which holds the minum entry
2394 * \retval = 0 get the entry successfully
2395 * negative errno (< 0) does not get the entry
2397 static int lmv_get_min_striped_entry(struct obd_export *exp,
2398 struct md_op_data *op_data,
2399 struct md_callback *cb_op,
2400 __u64 hash_offset, int *stripe_offset,
2401 struct lu_dirent **entp,
2402 struct page **ppage)
2404 struct obd_device *obd = exp->exp_obd;
2405 struct lmv_obd *lmv = &obd->u.lmv;
2406 struct lmv_stripe_md *lsm = op_data->op_mea1;
2407 struct lmv_tgt_desc *tgt;
2409 struct lu_dirent *min_ent = NULL;
2410 struct page *min_page = NULL;
2416 stripe_count = lsm->lsm_md_stripe_count;
2417 for (i = 0; i < stripe_count; i++) {
2418 struct lu_dirent *ent = NULL;
2419 struct page *page = NULL;
2420 struct lu_dirpage *dp;
2421 __u64 stripe_hash = hash_offset;
2423 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2425 GOTO(out, rc = PTR_ERR(tgt));
2427 /* op_data will be shared by each stripe, so we need
2428 * reset these value for each stripe */
2429 op_data->op_stripe_offset = i;
2430 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2431 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2432 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2434 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2439 dp = page_address(page);
2440 for (ent = lu_dirent_start(dp); ent != NULL;
2441 ent = lu_dirent_next(ent)) {
2442 /* Skip dummy entry */
2443 if (le16_to_cpu(ent->lde_namelen) == 0)
2446 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2449 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2450 (*entp == ent || i < *stripe_offset))
2453 /* skip . and .. for other stripes */
2455 (strncmp(ent->lde_name, ".",
2456 le16_to_cpu(ent->lde_namelen)) == 0 ||
2457 strncmp(ent->lde_name, "..",
2458 le16_to_cpu(ent->lde_namelen)) == 0))
2464 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2467 page_cache_release(page);
2470 /* reach the end of current stripe, go to next stripe */
2471 if (stripe_hash == MDS_DIR_END_OFF)
2477 if (min_ent != NULL) {
2478 if (le64_to_cpu(min_ent->lde_hash) >
2479 le64_to_cpu(ent->lde_hash)) {
2482 page_cache_release(min_page);
2487 page_cache_release(page);
2498 if (*ppage != NULL) {
2500 page_cache_release(*ppage);
2502 *stripe_offset = min_idx;
2509 * Build dir entry page from a striped directory
2511 * This function gets one entry by @offset from a striped directory. It will
2512 * read entries from all of stripes, and choose one closest to the required
2513 * offset(&offset). A few notes
2514 * 1. skip . and .. for non-zero stripes, because there can only have one .
2515 * and .. in a directory.
2516 * 2. op_data will be shared by all of stripes, instead of allocating new
2517 * one, so need to restore before reusing.
2518 * 3. release the entry page if that is not being chosen.
2520 * \param[in] exp obd export refer to LMV
2521 * \param[in] op_data hold those MD parameters of read_entry
2522 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2523 * \param[out] ldp the entry being read
2524 * \param[out] ppage the page holding the entry. Note: because the entry
2525 * will be accessed in upper layer, so we need hold the
2526 * page until the usages of entry is finished, see
2527 * ll_dir_entry_next.
2529 * retval =0 if get entry successfully
2530 * <0 cannot get entry
2532 static int lmv_read_striped_page(struct obd_export *exp,
2533 struct md_op_data *op_data,
2534 struct md_callback *cb_op,
2535 __u64 offset, struct page **ppage)
2537 struct obd_device *obd = exp->exp_obd;
2538 struct lu_fid master_fid = op_data->op_fid1;
2539 struct inode *master_inode = op_data->op_data;
2540 __u64 hash_offset = offset;
2541 struct lu_dirpage *dp;
2542 struct page *min_ent_page = NULL;
2543 struct page *ent_page = NULL;
2544 struct lu_dirent *ent;
2547 struct lu_dirent *min_ent = NULL;
2548 struct lu_dirent *last_ent;
2553 rc = lmv_check_connect(obd);
2557 /* Allocate a page and read entries from all of stripes and fill
2558 * the page by hash order */
2559 ent_page = alloc_page(GFP_KERNEL);
2560 if (ent_page == NULL)
2563 /* Initialize the entry page */
2564 dp = kmap(ent_page);
2565 memset(dp, 0, sizeof(*dp));
2566 dp->ldp_hash_start = cpu_to_le64(offset);
2567 dp->ldp_flags |= LDF_COLLIDE;
2570 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2576 /* Find the minum entry from all sub-stripes */
2577 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2583 /* If it can not get minum entry, it means it already reaches
2584 * the end of this directory */
2585 if (min_ent == NULL) {
2586 last_ent->lde_reclen = 0;
2587 hash_offset = MDS_DIR_END_OFF;
2591 ent_size = le16_to_cpu(min_ent->lde_reclen);
2593 /* the last entry lde_reclen is 0, but it might not
2594 * the end of this entry of this temporay entry */
2596 ent_size = lu_dirent_calc_size(
2597 le16_to_cpu(min_ent->lde_namelen),
2598 le32_to_cpu(min_ent->lde_attrs));
2599 if (ent_size > left_bytes) {
2600 last_ent->lde_reclen = cpu_to_le16(0);
2601 hash_offset = le64_to_cpu(min_ent->lde_hash);
2605 memcpy(ent, min_ent, ent_size);
2607 /* Replace . with master FID and Replace .. with the parent FID
2608 * of master object */
2609 if (strncmp(ent->lde_name, ".",
2610 le16_to_cpu(ent->lde_namelen)) == 0 &&
2611 le16_to_cpu(ent->lde_namelen) == 1)
2612 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2613 else if (strncmp(ent->lde_name, "..",
2614 le16_to_cpu(ent->lde_namelen)) == 0 &&
2615 le16_to_cpu(ent->lde_namelen) == 2)
2616 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2618 left_bytes -= ent_size;
2619 ent->lde_reclen = cpu_to_le16(ent_size);
2621 ent = (void *)ent + ent_size;
2622 hash_offset = le64_to_cpu(min_ent->lde_hash);
2623 if (hash_offset == MDS_DIR_END_OFF) {
2624 last_ent->lde_reclen = 0;
2629 if (min_ent_page != NULL) {
2630 kunmap(min_ent_page);
2631 page_cache_release(min_ent_page);
2634 if (unlikely(rc != 0)) {
2635 __free_page(ent_page);
2639 dp->ldp_flags |= LDF_EMPTY;
2640 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2641 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2644 /* We do not want to allocate md_op_data during each
2645 * dir entry reading, so op_data will be shared by every stripe,
2646 * then we need to restore it back to original value before
2647 * return to the upper layer */
2648 op_data->op_fid1 = master_fid;
2649 op_data->op_fid2 = master_fid;
2650 op_data->op_data = master_inode;
2657 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2658 struct md_callback *cb_op, __u64 offset,
2659 struct page **ppage)
2661 struct obd_device *obd = exp->exp_obd;
2662 struct lmv_obd *lmv = &obd->u.lmv;
2663 struct lmv_stripe_md *lsm = op_data->op_mea1;
2664 struct lmv_tgt_desc *tgt;
2668 rc = lmv_check_connect(obd);
2672 if (unlikely(lsm != NULL)) {
2673 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2677 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2679 RETURN(PTR_ERR(tgt));
2681 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2687 * Unlink a file/directory
2689 * Unlink a file or directory under the parent dir. The unlink request
2690 * usually will be sent to the MDT where the child is located, but if
2691 * the client does not have the child FID then request will be sent to the
2692 * MDT where the parent is located.
2694 * If the parent is a striped directory then it also needs to locate which
2695 * stripe the name of the child is located, and replace the parent FID
2696 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2697 * it will walk through all of sub-stripes until the child is being
2700 * \param[in] exp export refer to LMV
2701 * \param[in] op_data different parameters transferred beween client
2702 * MD stacks, name, namelen, FIDs etc.
2703 * op_fid1 is the parent FID, op_fid2 is the child
2705 * \param[out] request point to the request of unlink.
2707 * retval 0 if succeed
2708 * negative errno if failed.
2710 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2711 struct ptlrpc_request **request)
2713 struct obd_device *obd = exp->exp_obd;
2714 struct lmv_obd *lmv = &obd->u.lmv;
2715 struct lmv_tgt_desc *tgt = NULL;
2716 struct lmv_tgt_desc *parent_tgt = NULL;
2717 struct mdt_body *body;
2719 int stripe_index = 0;
2720 struct lmv_stripe_md *lsm = op_data->op_mea1;
2723 rc = lmv_check_connect(obd);
2727 /* For striped dir, we need to locate the parent as well */
2729 struct lmv_tgt_desc *tmp;
2731 LASSERT(op_data->op_name != NULL &&
2732 op_data->op_namelen != 0);
2734 tmp = lmv_locate_target_for_name(lmv, lsm,
2736 op_data->op_namelen,
2740 /* return -EBADFD means unknown hash type, might
2741 * need try all sub-stripe here */
2742 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2743 RETURN(PTR_ERR(tmp));
2745 /* Note: both migrating dir and unknown hash dir need to
2746 * try all of sub-stripes, so we need start search the
2747 * name from stripe 0, but migrating dir is already handled
2748 * inside lmv_locate_target_for_name(), so we only check
2749 * unknown hash type directory here */
2750 if (!lmv_is_known_hash_type(lsm)) {
2751 struct lmv_oinfo *oinfo;
2753 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2755 op_data->op_fid1 = oinfo->lmo_fid;
2756 op_data->op_mds = oinfo->lmo_mds;
2761 /* Send unlink requests to the MDT where the child is located */
2762 if (likely(!fid_is_zero(&op_data->op_fid2)))
2763 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2764 else if (lsm != NULL)
2765 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2767 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2770 RETURN(PTR_ERR(tgt));
2772 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2773 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2774 op_data->op_cap = cfs_curproc_cap_pack();
2777 * If child's fid is given, cancel unused locks for it if it is from
2778 * another export than parent.
2780 * LOOKUP lock for child (fid3) should also be cancelled on parent
2781 * tgt_tgt in mdc_unlink().
2783 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2786 * Cancel FULL locks on child (fid3).
2788 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2789 if (IS_ERR(parent_tgt))
2790 RETURN(PTR_ERR(parent_tgt));
2792 if (parent_tgt != tgt) {
2793 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2794 LCK_EX, MDS_INODELOCK_LOOKUP,
2795 MF_MDC_CANCEL_FID3);
2798 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2799 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2803 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2804 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2806 rc = md_unlink(tgt->ltd_exp, op_data, request);
2807 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2810 /* Try next stripe if it is needed. */
2811 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2812 struct lmv_oinfo *oinfo;
2815 if (stripe_index >= lsm->lsm_md_stripe_count)
2818 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2820 op_data->op_fid1 = oinfo->lmo_fid;
2821 op_data->op_mds = oinfo->lmo_mds;
2823 ptlrpc_req_finished(*request);
2826 goto try_next_stripe;
2829 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2833 /* Not cross-ref case, just get out of here. */
2834 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2837 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2838 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2840 /* This is a remote object, try remote MDT, Note: it may
2841 * try more than 1 time here, Considering following case
2842 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2843 * 1. Initially A does not know where remote1 is, it send
2844 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2845 * resend unlink RPC to MDT1 (retry 1st time).
2847 * 2. During the unlink RPC in flight,
2848 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2849 * and create new remote1, but on MDT0
2851 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2852 * /mnt/lustre, then lookup get fid of remote1, and find
2853 * it is remote dir again, and replay -EREMOTE again.
2855 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2857 * In theory, it might try unlimited time here, but it should
2858 * be very rare case. */
2859 op_data->op_fid2 = body->mbo_fid1;
2860 ptlrpc_req_finished(*request);
2866 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2868 struct lmv_obd *lmv = &obd->u.lmv;
2872 case OBD_CLEANUP_EARLY:
2873 /* XXX: here should be calling obd_precleanup() down to
2876 case OBD_CLEANUP_EXPORTS:
2877 fld_client_proc_fini(&lmv->lmv_fld);
2878 lprocfs_obd_cleanup(obd);
2879 lprocfs_free_md_stats(obd);
2887 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2888 __u32 keylen, void *key, __u32 *vallen, void *val,
2889 struct lov_stripe_md *lsm)
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 (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2907 rc = lmv_check_connect(obd);
2911 LASSERT(*vallen == sizeof(__u32));
2912 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2913 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2915 * All tgts should be connected when this gets called.
2917 if (tgt == NULL || tgt->ltd_exp == NULL)
2920 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2925 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2926 KEY_IS(KEY_DEFAULT_EASIZE) ||
2927 KEY_IS(KEY_MAX_COOKIESIZE) ||
2928 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2929 KEY_IS(KEY_CONN_DATA)) {
2930 rc = lmv_check_connect(obd);
2935 * Forwarding this request to first MDS, it should know LOV
2938 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2940 if (!rc && KEY_IS(KEY_CONN_DATA))
2941 exp->exp_connect_data = *(struct obd_connect_data *)val;
2943 } else if (KEY_IS(KEY_TGT_COUNT)) {
2944 *((int *)val) = lmv->desc.ld_tgt_count;
2948 CDEBUG(D_IOCTL, "Invalid key\n");
2952 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2953 obd_count keylen, void *key, obd_count vallen,
2954 void *val, struct ptlrpc_request_set *set)
2956 struct lmv_tgt_desc *tgt = NULL;
2957 struct obd_device *obd;
2958 struct lmv_obd *lmv;
2962 obd = class_exp2obd(exp);
2964 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2965 exp->exp_handle.h_cookie);
2970 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2973 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2976 if (tgt == NULL || tgt->ltd_exp == NULL)
2979 err = obd_set_info_async(env, tgt->ltd_exp,
2980 keylen, key, vallen, val, set);
2991 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2992 struct lmv_mds_md_v1 *lmm1)
2997 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2998 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2999 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
3000 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
3001 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
3002 sizeof(lmm1->lmv_pool_name));
3003 if (cplen >= sizeof(lmm1->lmv_pool_name))
3006 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3007 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
3008 &lsm->lsm_md_oinfo[i].lmo_fid);
3012 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
3016 bool allocated = false;
3020 LASSERT(lmmp != NULL);
3022 if (*lmmp != NULL && lsm == NULL) {
3025 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
3026 lmm_size = lmv_mds_md_size(stripe_count,
3027 le32_to_cpu((*lmmp)->lmv_magic));
3030 OBD_FREE(*lmmp, lmm_size);
3036 if (*lmmp == NULL && lsm == NULL) {
3037 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
3038 LASSERT(lmm_size > 0);
3039 OBD_ALLOC(*lmmp, lmm_size);
3042 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
3043 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
3048 LASSERT(lsm != NULL);
3049 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
3050 if (*lmmp == NULL) {
3051 OBD_ALLOC(*lmmp, lmm_size);
3057 switch (lsm->lsm_md_magic) {
3059 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
3066 if (rc != 0 && allocated) {
3067 OBD_FREE(*lmmp, lmm_size);
3073 EXPORT_SYMBOL(lmv_pack_md);
3075 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3076 const struct lmv_mds_md_v1 *lmm1)
3078 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3085 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3086 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3087 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3088 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3089 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3091 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3092 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3093 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3094 sizeof(lsm->lsm_md_pool_name));
3096 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3099 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3100 "layout_version %d\n", lsm->lsm_md_stripe_count,
3101 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3102 lsm->lsm_md_layout_version);
3104 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3105 for (i = 0; i < le32_to_cpu(stripe_count); i++) {
3106 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3107 &lmm1->lmv_stripe_fids[i]);
3108 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3109 &lsm->lsm_md_oinfo[i].lmo_mds);
3112 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3113 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3119 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3120 const union lmv_mds_md *lmm, int stripe_count)
3122 struct lmv_stripe_md *lsm;
3125 bool allocated = false;
3128 LASSERT(lsmp != NULL);
3132 if (lsm != NULL && lmm == NULL) {
3135 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3136 /* For migrating inode, the master stripe and master
3137 * object will be the same, so do not need iput, see
3138 * ll_update_lsm_md */
3139 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3140 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3141 iput(lsm->lsm_md_oinfo[i].lmo_root);
3144 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3145 OBD_FREE(lsm, lsm_size);
3151 if (lsm == NULL && lmm == NULL) {
3152 lsm_size = lmv_stripe_md_size(stripe_count);
3153 OBD_ALLOC(lsm, lsm_size);
3156 lsm->lsm_md_stripe_count = stripe_count;
3161 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3165 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3166 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3167 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3168 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3173 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3174 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3177 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3178 * stripecount should be 0 then.
3180 lsm_size = lmv_stripe_md_size(0);
3182 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3184 OBD_ALLOC(lsm, lsm_size);
3191 switch (le32_to_cpu(lmm->lmv_magic)) {
3193 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3196 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3197 le32_to_cpu(lmm->lmv_magic));
3202 if (rc != 0 && allocated) {
3203 OBD_FREE(lsm, lsm_size);
3210 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3212 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3214 EXPORT_SYMBOL(lmv_alloc_memmd);
3216 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3218 lmv_unpack_md(NULL, &lsm, NULL, 0);
3220 EXPORT_SYMBOL(lmv_free_memmd);
3222 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3223 struct lov_mds_md *lmm, int disk_len)
3225 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3226 (union lmv_mds_md *)lmm, disk_len);
3229 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3230 struct lov_stripe_md *lsm)
3232 struct obd_device *obd = exp->exp_obd;
3233 struct lmv_obd *lmv_obd = &obd->u.lmv;
3234 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3239 stripe_count = lmv->lsm_md_stripe_count;
3241 stripe_count = lmv_obd->desc.ld_tgt_count;
3243 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3246 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3249 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3250 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3251 ldlm_cancel_flags_t flags, void *opaque)
3253 struct obd_device *obd = exp->exp_obd;
3254 struct lmv_obd *lmv = &obd->u.lmv;
3260 LASSERT(fid != NULL);
3262 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3263 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3265 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3268 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3276 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3279 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3280 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3284 if (tgt == NULL || tgt->ltd_exp == NULL)
3286 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3290 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3291 const struct lu_fid *fid, ldlm_type_t type,
3292 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3293 struct lustre_handle *lockh)
3295 struct obd_device *obd = exp->exp_obd;
3296 struct lmv_obd *lmv = &obd->u.lmv;
3302 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3305 * With DNE every object can have two locks in different namespaces:
3306 * lookup lock in space of MDT storing direntry and update/open lock in
3307 * space of MDT storing inode. Try the MDT that the FID maps to first,
3308 * since this can be easily found, and only try others if that fails.
3310 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3311 i < lmv->desc.ld_tgt_count;
3312 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3314 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3315 obd->obd_name, PFID(fid), tgt);
3319 if (lmv->tgts[tgt] == NULL ||
3320 lmv->tgts[tgt]->ltd_exp == NULL ||
3321 lmv->tgts[tgt]->ltd_active == 0)
3324 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3325 type, policy, mode, lockh);
3333 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3334 struct obd_export *dt_exp, struct obd_export *md_exp,
3335 struct lustre_md *md)
3337 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3338 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3340 if (tgt == NULL || tgt->ltd_exp == NULL)
3343 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3346 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3348 struct obd_device *obd = exp->exp_obd;
3349 struct lmv_obd *lmv = &obd->u.lmv;
3350 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3353 if (md->lmv != NULL) {
3354 lmv_free_memmd(md->lmv);
3357 if (tgt == NULL || tgt->ltd_exp == NULL)
3359 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3362 int lmv_set_open_replay_data(struct obd_export *exp,
3363 struct obd_client_handle *och,
3364 struct lookup_intent *it)
3366 struct obd_device *obd = exp->exp_obd;
3367 struct lmv_obd *lmv = &obd->u.lmv;
3368 struct lmv_tgt_desc *tgt;
3371 tgt = lmv_find_target(lmv, &och->och_fid);
3373 RETURN(PTR_ERR(tgt));
3375 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3378 int lmv_clear_open_replay_data(struct obd_export *exp,
3379 struct obd_client_handle *och)
3381 struct obd_device *obd = exp->exp_obd;
3382 struct lmv_obd *lmv = &obd->u.lmv;
3383 struct lmv_tgt_desc *tgt;
3386 tgt = lmv_find_target(lmv, &och->och_fid);
3388 RETURN(PTR_ERR(tgt));
3390 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3393 static int lmv_get_remote_perm(struct obd_export *exp,
3394 const struct lu_fid *fid,
3395 struct obd_capa *oc, __u32 suppgid,
3396 struct ptlrpc_request **request)
3398 struct obd_device *obd = exp->exp_obd;
3399 struct lmv_obd *lmv = &obd->u.lmv;
3400 struct lmv_tgt_desc *tgt;
3404 rc = lmv_check_connect(obd);
3408 tgt = lmv_find_target(lmv, fid);
3410 RETURN(PTR_ERR(tgt));
3412 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3416 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3419 struct obd_device *obd = exp->exp_obd;
3420 struct lmv_obd *lmv = &obd->u.lmv;
3421 struct lmv_tgt_desc *tgt;
3425 rc = lmv_check_connect(obd);
3429 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3431 RETURN(PTR_ERR(tgt));
3433 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3437 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3438 const struct req_msg_field *field, struct obd_capa **oc)
3440 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3441 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3443 if (tgt == NULL || tgt->ltd_exp == NULL)
3445 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3448 int lmv_intent_getattr_async(struct obd_export *exp,
3449 struct md_enqueue_info *minfo,
3450 struct ldlm_enqueue_info *einfo)
3452 struct md_op_data *op_data = &minfo->mi_data;
3453 struct obd_device *obd = exp->exp_obd;
3454 struct lmv_obd *lmv = &obd->u.lmv;
3455 struct lmv_tgt_desc *tgt = NULL;
3459 rc = lmv_check_connect(obd);
3463 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3465 RETURN(PTR_ERR(tgt));
3467 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3471 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3472 struct lu_fid *fid, __u64 *bits)
3474 struct obd_device *obd = exp->exp_obd;
3475 struct lmv_obd *lmv = &obd->u.lmv;
3476 struct lmv_tgt_desc *tgt;
3480 rc = lmv_check_connect(obd);
3484 tgt = lmv_find_target(lmv, fid);
3486 RETURN(PTR_ERR(tgt));
3488 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3492 int lmv_get_fid_from_lsm(struct obd_export *exp,
3493 const struct lmv_stripe_md *lsm,
3494 const char *name, int namelen, struct lu_fid *fid)
3496 const struct lmv_oinfo *oinfo;
3498 LASSERT(lsm != NULL);
3499 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3501 return PTR_ERR(oinfo);
3503 *fid = oinfo->lmo_fid;
3509 * For lmv, only need to send request to master MDT, and the master MDT will
3510 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3511 * we directly fetch data from the slave MDTs.
3513 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3514 struct obd_quotactl *oqctl)
3516 struct obd_device *obd = class_exp2obd(exp);
3517 struct lmv_obd *lmv = &obd->u.lmv;
3518 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3521 __u64 curspace, curinodes;
3525 tgt->ltd_exp == NULL ||
3527 lmv->desc.ld_tgt_count == 0) {
3528 CERROR("master lmv inactive\n");
3532 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3533 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3537 curspace = curinodes = 0;
3538 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3542 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3545 err = obd_quotactl(tgt->ltd_exp, oqctl);
3547 CERROR("getquota on mdt %d failed. %d\n", i, err);
3551 curspace += oqctl->qc_dqblk.dqb_curspace;
3552 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3555 oqctl->qc_dqblk.dqb_curspace = curspace;
3556 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3561 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3562 struct obd_quotactl *oqctl)
3564 struct obd_device *obd = class_exp2obd(exp);
3565 struct lmv_obd *lmv = &obd->u.lmv;
3566 struct lmv_tgt_desc *tgt;
3571 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3574 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3575 CERROR("lmv idx %d inactive\n", i);
3579 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3587 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3588 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3590 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3593 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3594 struct cl_attr *attr)
3599 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3600 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3602 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3603 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3604 i_size_read(inode), inode->i_nlink,
3605 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3606 LTIME_S(inode->i_mtime));
3608 /* for slave stripe, it needs to subtract nlink for . and .. */
3610 attr->cat_nlink += inode->i_nlink - 2;
3612 attr->cat_nlink = inode->i_nlink;
3614 attr->cat_size += i_size_read(inode);
3616 if (attr->cat_atime < LTIME_S(inode->i_atime))
3617 attr->cat_atime = LTIME_S(inode->i_atime);
3619 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3620 attr->cat_ctime = LTIME_S(inode->i_ctime);
3622 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3623 attr->cat_mtime = LTIME_S(inode->i_mtime);
3629 struct obd_ops lmv_obd_ops = {
3630 .o_owner = THIS_MODULE,
3631 .o_setup = lmv_setup,
3632 .o_cleanup = lmv_cleanup,
3633 .o_precleanup = lmv_precleanup,
3634 .o_process_config = lmv_process_config,
3635 .o_connect = lmv_connect,
3636 .o_disconnect = lmv_disconnect,
3637 .o_statfs = lmv_statfs,
3638 .o_get_info = lmv_get_info,
3639 .o_set_info_async = lmv_set_info_async,
3640 .o_packmd = lmv_packmd,
3641 .o_unpackmd = lmv_unpackmd,
3642 .o_notify = lmv_notify,
3643 .o_get_uuid = lmv_get_uuid,
3644 .o_iocontrol = lmv_iocontrol,
3645 .o_quotacheck = lmv_quotacheck,
3646 .o_quotactl = lmv_quotactl
3649 struct md_ops lmv_md_ops = {
3650 .m_getstatus = lmv_getstatus,
3651 .m_null_inode = lmv_null_inode,
3652 .m_find_cbdata = lmv_find_cbdata,
3653 .m_close = lmv_close,
3654 .m_create = lmv_create,
3655 .m_done_writing = lmv_done_writing,
3656 .m_enqueue = lmv_enqueue,
3657 .m_getattr = lmv_getattr,
3658 .m_getxattr = lmv_getxattr,
3659 .m_getattr_name = lmv_getattr_name,
3660 .m_intent_lock = lmv_intent_lock,
3662 .m_rename = lmv_rename,
3663 .m_setattr = lmv_setattr,
3664 .m_setxattr = lmv_setxattr,
3665 .m_fsync = lmv_fsync,
3666 .m_read_page = lmv_read_page,
3667 .m_unlink = lmv_unlink,
3668 .m_init_ea_size = lmv_init_ea_size,
3669 .m_cancel_unused = lmv_cancel_unused,
3670 .m_set_lock_data = lmv_set_lock_data,
3671 .m_lock_match = lmv_lock_match,
3672 .m_get_lustre_md = lmv_get_lustre_md,
3673 .m_free_lustre_md = lmv_free_lustre_md,
3674 .m_update_lsm_md = lmv_update_lsm_md,
3675 .m_merge_attr = lmv_merge_attr,
3676 .m_set_open_replay_data = lmv_set_open_replay_data,
3677 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3678 .m_renew_capa = lmv_renew_capa,
3679 .m_unpack_capa = lmv_unpack_capa,
3680 .m_get_remote_perm = lmv_get_remote_perm,
3681 .m_intent_getattr_async = lmv_intent_getattr_async,
3682 .m_revalidate_lock = lmv_revalidate_lock,
3683 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3686 int __init lmv_init(void)
3688 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3689 #ifndef HAVE_ONLY_PROCFS_SEQ
3692 LUSTRE_LMV_NAME, NULL);
3696 static void lmv_exit(void)
3698 class_unregister_type(LUSTRE_LMV_NAME);
3701 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3702 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3703 MODULE_LICENSE("GPL");
3705 module_init(lmv_init);
3706 module_exit(lmv_exit);