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 OBD_IOC_FID2PATH: {
1124 rc = lmv_fid2path(exp, len, karg, uarg);
1127 case LL_IOC_HSM_STATE_GET:
1128 case LL_IOC_HSM_STATE_SET:
1129 case LL_IOC_HSM_ACTION: {
1130 struct md_op_data *op_data = karg;
1132 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1134 RETURN(PTR_ERR(tgt));
1136 if (tgt->ltd_exp == NULL)
1139 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1142 case LL_IOC_HSM_PROGRESS: {
1143 const struct hsm_progress_kernel *hpk = karg;
1145 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1147 RETURN(PTR_ERR(tgt));
1148 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1151 case LL_IOC_HSM_REQUEST: {
1152 struct hsm_user_request *hur = karg;
1153 unsigned int reqcount = hur->hur_request.hr_itemcount;
1158 /* if the request is about a single fid
1159 * or if there is a single MDS, no need to split
1161 if (reqcount == 1 || count == 1) {
1162 tgt = lmv_find_target(lmv,
1163 &hur->hur_user_item[0].hui_fid);
1165 RETURN(PTR_ERR(tgt));
1166 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1168 /* split fid list to their respective MDS */
1169 for (i = 0; i < count; i++) {
1170 unsigned int nr, reqlen;
1172 struct hsm_user_request *req;
1175 if (tgt == NULL || tgt->ltd_exp == NULL)
1178 nr = lmv_hsm_req_count(lmv, hur, tgt);
1179 if (nr == 0) /* nothing for this MDS */
1182 /* build a request with fids for this MDS */
1183 reqlen = offsetof(typeof(*hur),
1185 + hur->hur_request.hr_data_len;
1186 OBD_ALLOC_LARGE(req, reqlen);
1190 lmv_hsm_req_build(lmv, hur, tgt, req);
1192 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1194 if (rc1 != 0 && rc == 0)
1196 OBD_FREE_LARGE(req, reqlen);
1201 case LL_IOC_LOV_SWAP_LAYOUTS: {
1202 struct md_op_data *op_data = karg;
1203 struct lmv_tgt_desc *tgt1, *tgt2;
1205 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1207 RETURN(PTR_ERR(tgt1));
1209 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1211 RETURN(PTR_ERR(tgt2));
1213 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1216 /* only files on same MDT can have their layouts swapped */
1217 if (tgt1->ltd_idx != tgt2->ltd_idx)
1220 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1223 case LL_IOC_HSM_CT_START: {
1224 struct lustre_kernelcomm *lk = karg;
1225 if (lk->lk_flags & LK_FLG_STOP)
1226 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1228 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1232 for (i = 0; i < count; i++) {
1233 struct obd_device *mdc_obd;
1237 if (tgt == NULL || tgt->ltd_exp == NULL)
1239 /* ll_umount_begin() sets force flag but for lmv, not
1240 * mdc. Let's pass it through */
1241 mdc_obd = class_exp2obd(tgt->ltd_exp);
1242 mdc_obd->obd_force = obddev->obd_force;
1243 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1244 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1247 if (tgt->ltd_active) {
1248 CERROR("error: iocontrol MDC %s on MDT"
1249 " idx %d cmd %x: err = %d\n",
1250 tgt->ltd_uuid.uuid, i, cmd, err);
1264 static int lmv_all_chars_policy(int count, const char *name,
1275 static int lmv_nid_policy(struct lmv_obd *lmv)
1277 struct obd_import *imp;
1281 * XXX: To get nid we assume that underlying obd device is mdc.
1283 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1284 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1285 return id % lmv->desc.ld_tgt_count;
1288 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1289 placement_policy_t placement)
1291 switch (placement) {
1292 case PLACEMENT_CHAR_POLICY:
1293 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1295 op_data->op_namelen);
1296 case PLACEMENT_NID_POLICY:
1297 return lmv_nid_policy(lmv);
1303 CERROR("Unsupported placement policy %x\n", placement);
1309 * This is _inode_ placement policy function (not name).
1311 static int lmv_placement_policy(struct obd_device *obd,
1312 struct md_op_data *op_data,
1315 struct lmv_obd *lmv = &obd->u.lmv;
1318 LASSERT(mds != NULL);
1320 if (lmv->desc.ld_tgt_count == 1) {
1326 * If stripe_offset is provided during setdirstripe
1327 * (setdirstripe -i xx), xx MDS will be choosen.
1329 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1330 struct lmv_user_md *lum;
1332 lum = op_data->op_data;
1334 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1335 *mds = le32_to_cpu(lum->lum_stripe_offset);
1337 /* -1 means default, which will be in the same MDT with
1339 *mds = op_data->op_mds;
1340 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1343 /* Allocate new fid on target according to operation type and
1344 * parent home mds. */
1345 *mds = op_data->op_mds;
1351 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1354 struct lmv_tgt_desc *tgt;
1358 tgt = lmv_get_target(lmv, mds, NULL);
1360 RETURN(PTR_ERR(tgt));
1363 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1364 * on server that seq in new allocated fid is not yet known.
1366 mutex_lock(&tgt->ltd_fid_mutex);
1368 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1369 GOTO(out, rc = -ENODEV);
1372 * Asking underlying tgt layer to allocate new fid.
1374 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1376 LASSERT(fid_is_sane(fid));
1382 mutex_unlock(&tgt->ltd_fid_mutex);
1386 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1387 struct lu_fid *fid, struct md_op_data *op_data)
1389 struct obd_device *obd = class_exp2obd(exp);
1390 struct lmv_obd *lmv = &obd->u.lmv;
1395 LASSERT(op_data != NULL);
1396 LASSERT(fid != NULL);
1398 rc = lmv_placement_policy(obd, op_data, &mds);
1400 CERROR("Can't get target for allocating fid, "
1405 rc = __lmv_fid_alloc(lmv, fid, mds);
1407 CERROR("Can't alloc new fid, rc %d\n", rc);
1414 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1416 struct lmv_obd *lmv = &obd->u.lmv;
1417 struct lmv_desc *desc;
1421 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1422 CERROR("LMV setup requires a descriptor\n");
1426 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1427 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1428 CERROR("Lmv descriptor size wrong: %d > %d\n",
1429 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1433 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1434 if (lmv->tgts == NULL)
1436 lmv->tgts_size = 32;
1438 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1439 lmv->desc.ld_tgt_count = 0;
1440 lmv->desc.ld_active_tgt_count = 0;
1441 lmv->max_cookiesize = 0;
1442 lmv->max_def_easize = 0;
1443 lmv->max_easize = 0;
1444 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1446 spin_lock_init(&lmv->lmv_lock);
1447 mutex_init(&lmv->init_mutex);
1450 obd->obd_vars = lprocfs_lmv_obd_vars;
1451 lprocfs_seq_obd_setup(obd);
1452 lprocfs_alloc_md_stats(obd, 0);
1453 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1454 0444, &lmv_proc_target_fops, obd);
1456 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1459 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1460 LUSTRE_CLI_FLD_HASH_DHT);
1462 CERROR("Can't init FLD, err %d\n", rc);
1472 static int lmv_cleanup(struct obd_device *obd)
1474 struct lmv_obd *lmv = &obd->u.lmv;
1477 fld_client_fini(&lmv->lmv_fld);
1478 if (lmv->tgts != NULL) {
1480 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1481 if (lmv->tgts[i] == NULL)
1483 lmv_del_target(lmv, i);
1485 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1491 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1493 struct lustre_cfg *lcfg = buf;
1494 struct obd_uuid obd_uuid;
1500 switch (lcfg->lcfg_command) {
1502 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1503 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1504 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1505 GOTO(out, rc = -EINVAL);
1507 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1509 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1510 GOTO(out, rc = -EINVAL);
1511 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1512 GOTO(out, rc = -EINVAL);
1513 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1516 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1517 GOTO(out, rc = -EINVAL);
1523 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1524 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1526 struct obd_device *obd = class_exp2obd(exp);
1527 struct lmv_obd *lmv = &obd->u.lmv;
1528 struct obd_statfs *temp;
1533 rc = lmv_check_connect(obd);
1537 OBD_ALLOC(temp, sizeof(*temp));
1541 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1542 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1545 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1548 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1549 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1551 GOTO(out_free_temp, rc);
1556 /* If the statfs is from mount, it will needs
1557 * retrieve necessary information from MDT0.
1558 * i.e. mount does not need the merged osfs
1560 * And also clients can be mounted as long as
1561 * MDT0 is in service*/
1562 if (flags & OBD_STATFS_FOR_MDT0)
1563 GOTO(out_free_temp, rc);
1565 osfs->os_bavail += temp->os_bavail;
1566 osfs->os_blocks += temp->os_blocks;
1567 osfs->os_ffree += temp->os_ffree;
1568 osfs->os_files += temp->os_files;
1574 OBD_FREE(temp, sizeof(*temp));
1578 static int lmv_getstatus(struct obd_export *exp,
1580 struct obd_capa **pc)
1582 struct obd_device *obd = exp->exp_obd;
1583 struct lmv_obd *lmv = &obd->u.lmv;
1587 rc = lmv_check_connect(obd);
1591 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1595 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1596 struct obd_capa *oc, obd_valid valid, const char *name,
1597 const char *input, int input_size, int output_size,
1598 int flags, struct ptlrpc_request **request)
1600 struct obd_device *obd = exp->exp_obd;
1601 struct lmv_obd *lmv = &obd->u.lmv;
1602 struct lmv_tgt_desc *tgt;
1606 rc = lmv_check_connect(obd);
1610 tgt = lmv_find_target(lmv, fid);
1612 RETURN(PTR_ERR(tgt));
1614 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1615 input_size, output_size, flags, request);
1620 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1621 struct obd_capa *oc, obd_valid valid, const char *name,
1622 const char *input, int input_size, int output_size,
1623 int flags, __u32 suppgid,
1624 struct ptlrpc_request **request)
1626 struct obd_device *obd = exp->exp_obd;
1627 struct lmv_obd *lmv = &obd->u.lmv;
1628 struct lmv_tgt_desc *tgt;
1632 rc = lmv_check_connect(obd);
1636 tgt = lmv_find_target(lmv, fid);
1638 RETURN(PTR_ERR(tgt));
1640 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1641 input_size, output_size, flags, suppgid,
1647 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1648 struct ptlrpc_request **request)
1650 struct obd_device *obd = exp->exp_obd;
1651 struct lmv_obd *lmv = &obd->u.lmv;
1652 struct lmv_tgt_desc *tgt;
1656 rc = lmv_check_connect(obd);
1660 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1662 RETURN(PTR_ERR(tgt));
1664 if (op_data->op_flags & MF_GET_MDT_IDX) {
1665 op_data->op_mds = tgt->ltd_idx;
1669 rc = md_getattr(tgt->ltd_exp, op_data, request);
1674 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1676 struct obd_device *obd = exp->exp_obd;
1677 struct lmv_obd *lmv = &obd->u.lmv;
1682 rc = lmv_check_connect(obd);
1686 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1689 * With DNE every object can have two locks in different namespaces:
1690 * lookup lock in space of MDT storing direntry and update/open lock in
1691 * space of MDT storing inode.
1693 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1694 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1696 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1702 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1703 ldlm_iterator_t it, void *data)
1705 struct obd_device *obd = exp->exp_obd;
1706 struct lmv_obd *lmv = &obd->u.lmv;
1712 rc = lmv_check_connect(obd);
1716 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1719 * With DNE every object can have two locks in different namespaces:
1720 * lookup lock in space of MDT storing direntry and update/open lock in
1721 * space of MDT storing inode. Try the MDT that the FID maps to first,
1722 * since this can be easily found, and only try others if that fails.
1724 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1725 i < lmv->desc.ld_tgt_count;
1726 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1728 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1729 obd->obd_name, PFID(fid), tgt);
1733 if (lmv->tgts[tgt] == NULL ||
1734 lmv->tgts[tgt]->ltd_exp == NULL)
1737 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1746 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1747 struct md_open_data *mod, struct ptlrpc_request **request)
1749 struct obd_device *obd = exp->exp_obd;
1750 struct lmv_obd *lmv = &obd->u.lmv;
1751 struct lmv_tgt_desc *tgt;
1755 rc = lmv_check_connect(obd);
1759 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1761 RETURN(PTR_ERR(tgt));
1763 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1764 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1769 * Choosing the MDT by name or FID in @op_data.
1770 * For non-striped directory, it will locate MDT by fid.
1771 * For striped-directory, it will locate MDT by name. And also
1772 * it will reset op_fid1 with the FID of the choosen stripe.
1774 struct lmv_tgt_desc *
1775 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1776 const char *name, int namelen, struct lu_fid *fid,
1779 struct lmv_tgt_desc *tgt;
1780 const struct lmv_oinfo *oinfo;
1782 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1784 RETURN(ERR_CAST(oinfo));
1785 *fid = oinfo->lmo_fid;
1786 *mds = oinfo->lmo_mds;
1787 tgt = lmv_get_target(lmv, *mds, NULL);
1789 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1794 * Locate mds by fid or name
1796 * For striped directory (lsm != NULL), it will locate the stripe
1797 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1798 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1799 * walk through all of stripes to locate the entry.
1801 * For normal direcotry, it will locate MDS by FID directly.
1802 * \param[in] lmv LMV device
1803 * \param[in] op_data client MD stack parameters, name, namelen
1805 * \param[in] fid object FID used to locate MDS.
1807 * retval pointer to the lmv_tgt_desc if succeed.
1808 * ERR_PTR(errno) if failed.
1811 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1814 struct lmv_stripe_md *lsm = op_data->op_mea1;
1815 struct lmv_tgt_desc *tgt;
1817 if (lsm == NULL || op_data->op_namelen == 0) {
1818 tgt = lmv_find_target(lmv, fid);
1822 op_data->op_mds = tgt->ltd_idx;
1826 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1827 op_data->op_namelen, fid,
1831 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1832 const void *data, int datalen, int mode, __u32 uid,
1833 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1834 struct ptlrpc_request **request)
1836 struct obd_device *obd = exp->exp_obd;
1837 struct lmv_obd *lmv = &obd->u.lmv;
1838 struct lmv_tgt_desc *tgt;
1842 rc = lmv_check_connect(obd);
1846 if (!lmv->desc.ld_active_tgt_count)
1849 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1851 RETURN(PTR_ERR(tgt));
1853 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1854 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1857 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1861 /* Send the create request to the MDT where the object
1862 * will be located */
1863 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1865 RETURN(PTR_ERR(tgt));
1867 op_data->op_mds = tgt->ltd_idx;
1869 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1870 PFID(&op_data->op_fid2), op_data->op_mds);
1872 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1873 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1874 cap_effective, rdev, request);
1876 if (*request == NULL)
1878 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1883 static int lmv_done_writing(struct obd_export *exp,
1884 struct md_op_data *op_data,
1885 struct md_open_data *mod)
1887 struct obd_device *obd = exp->exp_obd;
1888 struct lmv_obd *lmv = &obd->u.lmv;
1889 struct lmv_tgt_desc *tgt;
1893 rc = lmv_check_connect(obd);
1897 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1899 RETURN(PTR_ERR(tgt));
1901 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1906 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1907 const union ldlm_policy_data *policy,
1908 struct lookup_intent *it, struct md_op_data *op_data,
1909 struct lustre_handle *lockh, __u64 extra_lock_flags)
1911 struct obd_device *obd = exp->exp_obd;
1912 struct lmv_obd *lmv = &obd->u.lmv;
1913 struct lmv_tgt_desc *tgt;
1917 rc = lmv_check_connect(obd);
1921 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1922 LL_IT2STR(it), PFID(&op_data->op_fid1));
1924 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1926 RETURN(PTR_ERR(tgt));
1928 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1929 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1931 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1938 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1939 struct ptlrpc_request **preq)
1941 struct ptlrpc_request *req = NULL;
1942 struct obd_device *obd = exp->exp_obd;
1943 struct lmv_obd *lmv = &obd->u.lmv;
1944 struct lmv_tgt_desc *tgt;
1945 struct mdt_body *body;
1949 rc = lmv_check_connect(obd);
1953 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1955 RETURN(PTR_ERR(tgt));
1957 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1958 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1961 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1965 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1966 LASSERT(body != NULL);
1968 if (body->mbo_valid & OBD_MD_MDS) {
1969 struct lu_fid rid = body->mbo_fid1;
1970 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1973 tgt = lmv_find_target(lmv, &rid);
1975 ptlrpc_req_finished(*preq);
1977 RETURN(PTR_ERR(tgt));
1980 op_data->op_fid1 = rid;
1981 op_data->op_valid |= OBD_MD_FLCROSSREF;
1982 op_data->op_namelen = 0;
1983 op_data->op_name = NULL;
1984 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1985 ptlrpc_req_finished(*preq);
1992 #define md_op_data_fid(op_data, fl) \
1993 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1994 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1995 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1996 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1999 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2000 struct md_op_data *op_data,
2001 int op_tgt, ldlm_mode_t mode, int bits, int flag)
2003 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2004 struct obd_device *obd = exp->exp_obd;
2005 struct lmv_obd *lmv = &obd->u.lmv;
2006 ldlm_policy_data_t policy = {{ 0 }};
2010 if (!fid_is_sane(fid))
2014 tgt = lmv_find_target(lmv, fid);
2016 RETURN(PTR_ERR(tgt));
2019 if (tgt->ltd_idx != op_tgt) {
2020 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2021 policy.l_inodebits.bits = bits;
2022 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2023 mode, LCF_ASYNC, NULL);
2026 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2028 op_data->op_flags |= flag;
2036 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2039 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2040 struct ptlrpc_request **request)
2042 struct obd_device *obd = exp->exp_obd;
2043 struct lmv_obd *lmv = &obd->u.lmv;
2044 struct lmv_tgt_desc *tgt;
2048 rc = lmv_check_connect(obd);
2052 LASSERT(op_data->op_namelen != 0);
2054 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2055 PFID(&op_data->op_fid2), op_data->op_namelen,
2056 op_data->op_name, PFID(&op_data->op_fid1));
2058 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2059 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2060 op_data->op_cap = cfs_curproc_cap_pack();
2061 if (op_data->op_mea2 != NULL) {
2062 struct lmv_stripe_md *lsm = op_data->op_mea2;
2063 const struct lmv_oinfo *oinfo;
2065 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2066 op_data->op_namelen);
2068 RETURN(PTR_ERR(oinfo));
2070 op_data->op_fid2 = oinfo->lmo_fid;
2073 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2075 RETURN(PTR_ERR(tgt));
2078 * Cancel UPDATE lock on child (fid1).
2080 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2081 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2082 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2086 rc = md_link(tgt->ltd_exp, op_data, request);
2091 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2092 const char *old, int oldlen, const char *new, int newlen,
2093 struct ptlrpc_request **request)
2095 struct obd_device *obd = exp->exp_obd;
2096 struct lmv_obd *lmv = &obd->u.lmv;
2097 struct lmv_tgt_desc *src_tgt;
2101 LASSERT(oldlen != 0);
2103 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2104 oldlen, old, PFID(&op_data->op_fid1),
2105 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2106 newlen, new, PFID(&op_data->op_fid2),
2107 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2109 rc = lmv_check_connect(obd);
2113 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2114 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2115 op_data->op_cap = cfs_curproc_cap_pack();
2116 if (op_data->op_cli_flags & CLI_MIGRATE) {
2117 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2118 PFID(&op_data->op_fid3));
2119 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2122 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2124 if (op_data->op_mea1 != NULL) {
2125 struct lmv_stripe_md *lsm = op_data->op_mea1;
2127 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2131 if (IS_ERR(src_tgt))
2132 RETURN(PTR_ERR(src_tgt));
2134 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2135 if (IS_ERR(src_tgt))
2136 RETURN(PTR_ERR(src_tgt));
2138 op_data->op_mds = src_tgt->ltd_idx;
2141 if (op_data->op_mea2) {
2142 struct lmv_stripe_md *lsm = op_data->op_mea2;
2143 const struct lmv_oinfo *oinfo;
2145 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2147 RETURN(PTR_ERR(oinfo));
2149 op_data->op_fid2 = oinfo->lmo_fid;
2152 if (IS_ERR(src_tgt))
2153 RETURN(PTR_ERR(src_tgt));
2156 * LOOKUP lock on src child (fid3) should also be cancelled for
2157 * src_tgt in mdc_rename.
2159 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2162 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2165 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2166 LCK_EX, MDS_INODELOCK_UPDATE,
2167 MF_MDC_CANCEL_FID2);
2172 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2174 if (fid_is_sane(&op_data->op_fid3)) {
2175 struct lmv_tgt_desc *tgt;
2177 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2179 RETURN(PTR_ERR(tgt));
2181 /* Cancel LOOKUP lock on its parent */
2182 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2183 LCK_EX, MDS_INODELOCK_LOOKUP,
2184 MF_MDC_CANCEL_FID3);
2188 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2189 LCK_EX, MDS_INODELOCK_FULL,
2190 MF_MDC_CANCEL_FID3);
2196 * Cancel all the locks on tgt child (fid4).
2198 if (fid_is_sane(&op_data->op_fid4))
2199 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2200 LCK_EX, MDS_INODELOCK_FULL,
2201 MF_MDC_CANCEL_FID4);
2203 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2204 op_data->op_mds, PFID(&op_data->op_fid2));
2206 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2212 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2213 void *ea, int ealen, void *ea2, int ea2len,
2214 struct ptlrpc_request **request,
2215 struct md_open_data **mod)
2217 struct obd_device *obd = exp->exp_obd;
2218 struct lmv_obd *lmv = &obd->u.lmv;
2219 struct lmv_tgt_desc *tgt;
2223 rc = lmv_check_connect(obd);
2227 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2228 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2230 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2231 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2233 RETURN(PTR_ERR(tgt));
2235 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2236 ea2len, request, mod);
2241 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2242 struct obd_capa *oc, struct ptlrpc_request **request)
2244 struct obd_device *obd = exp->exp_obd;
2245 struct lmv_obd *lmv = &obd->u.lmv;
2246 struct lmv_tgt_desc *tgt;
2250 rc = lmv_check_connect(obd);
2254 tgt = lmv_find_target(lmv, fid);
2256 RETURN(PTR_ERR(tgt));
2258 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2263 * Adjust a set of pages, each page containing an array of lu_dirpages,
2264 * so that each page can be used as a single logical lu_dirpage.
2266 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2267 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2268 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
2269 * value is used as a cookie to request the next lu_dirpage in a
2270 * directory listing that spans multiple pages (two in this example):
2273 * .|--------v------- -----.
2274 * |s|e|f|p|ent|ent| ... |ent|
2275 * '--|-------------- -----' Each CFS_PAGE contains a single
2276 * '------. lu_dirpage.
2277 * .---------v------- -----.
2278 * |s|e|f|p|ent| 0 | ... | 0 |
2279 * '----------------- -----'
2281 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2282 * larger than LU_PAGE_SIZE, a single host page may contain multiple
2283 * lu_dirpages. After reading the lu_dirpages from the MDS, the
2284 * ldp_hash_end of the first lu_dirpage refers to the one immediately
2285 * after it in the same CFS_PAGE (arrows simplified for brevity, but
2286 * in general e0==s1, e1==s2, etc.):
2288 * .-------------------- -----.
2289 * |s0|e0|f0|p|ent|ent| ... |ent|
2290 * |---v---------------- -----|
2291 * |s1|e1|f1|p|ent|ent| ... |ent|
2292 * |---v---------------- -----| Here, each CFS_PAGE contains
2293 * ... multiple lu_dirpages.
2294 * |---v---------------- -----|
2295 * |s'|e'|f'|p|ent|ent| ... |ent|
2296 * '---|---------------- -----'
2298 * .----------------------------.
2301 * This structure is transformed into a single logical lu_dirpage as follows:
2303 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2304 * labeled 'next CFS_PAGE'.
2306 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2307 * a hash collision with the next page exists.
2309 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2310 * to the first entry of the next lu_dirpage.
2312 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2313 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2317 for (i = 0; i < ncfspgs; i++) {
2318 struct lu_dirpage *dp = kmap(pages[i]);
2319 struct lu_dirpage *first = dp;
2320 struct lu_dirent *end_dirent = NULL;
2321 struct lu_dirent *ent;
2322 __u64 hash_end = dp->ldp_hash_end;
2323 __u32 flags = dp->ldp_flags;
2325 while (--nlupgs > 0) {
2326 ent = lu_dirent_start(dp);
2327 for (end_dirent = ent; ent != NULL;
2328 end_dirent = ent, ent = lu_dirent_next(ent));
2330 /* Advance dp to next lu_dirpage. */
2331 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2333 /* Check if we've reached the end of the CFS_PAGE. */
2334 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2337 /* Save the hash and flags of this lu_dirpage. */
2338 hash_end = dp->ldp_hash_end;
2339 flags = dp->ldp_flags;
2341 /* Check if lu_dirpage contains no entries. */
2345 /* Enlarge the end entry lde_reclen from 0 to
2346 * first entry of next lu_dirpage. */
2347 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2348 end_dirent->lde_reclen =
2349 cpu_to_le16((char *)(dp->ldp_entries) -
2350 (char *)end_dirent);
2353 first->ldp_hash_end = hash_end;
2354 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2355 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2359 LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2362 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2363 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2366 * Get current minimum entry from striped directory
2368 * This function will search the dir entry, whose hash value is the
2369 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2370 * only being called for striped directory.
2372 * \param[in] exp export of LMV
2373 * \param[in] op_data parameters transferred beween client MD stack
2374 * stripe_information will be included in this
2376 * \param[in] cb_op ldlm callback being used in enqueue in
2378 * \param[in] hash_offset the hash value, which is used to locate
2379 * minum(closet) dir entry
2380 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2381 * index of last entry, so to avoid hash conflict
2382 * between stripes. It will also be used to
2383 * return the stripe index of current dir entry.
2384 * \param[in|out] entp the minum entry and it also is being used
2385 * to input the last dir entry to resolve the
2388 * \param[out] ppage the page which holds the minum entry
2390 * \retval = 0 get the entry successfully
2391 * negative errno (< 0) does not get the entry
2393 static int lmv_get_min_striped_entry(struct obd_export *exp,
2394 struct md_op_data *op_data,
2395 struct md_callback *cb_op,
2396 __u64 hash_offset, int *stripe_offset,
2397 struct lu_dirent **entp,
2398 struct page **ppage)
2400 struct obd_device *obd = exp->exp_obd;
2401 struct lmv_obd *lmv = &obd->u.lmv;
2402 struct lmv_stripe_md *lsm = op_data->op_mea1;
2403 struct lmv_tgt_desc *tgt;
2405 struct lu_dirent *min_ent = NULL;
2406 struct page *min_page = NULL;
2412 stripe_count = lsm->lsm_md_stripe_count;
2413 for (i = 0; i < stripe_count; i++) {
2414 struct lu_dirent *ent = NULL;
2415 struct page *page = NULL;
2416 struct lu_dirpage *dp;
2417 __u64 stripe_hash = hash_offset;
2419 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2421 GOTO(out, rc = PTR_ERR(tgt));
2423 /* op_data will be shared by each stripe, so we need
2424 * reset these value for each stripe */
2425 op_data->op_stripe_offset = i;
2426 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2427 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2428 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2430 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2435 dp = page_address(page);
2436 for (ent = lu_dirent_start(dp); ent != NULL;
2437 ent = lu_dirent_next(ent)) {
2438 /* Skip dummy entry */
2439 if (le16_to_cpu(ent->lde_namelen) == 0)
2442 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2445 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2446 (*entp == ent || i < *stripe_offset))
2449 /* skip . and .. for other stripes */
2451 (strncmp(ent->lde_name, ".",
2452 le16_to_cpu(ent->lde_namelen)) == 0 ||
2453 strncmp(ent->lde_name, "..",
2454 le16_to_cpu(ent->lde_namelen)) == 0))
2460 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2463 page_cache_release(page);
2466 /* reach the end of current stripe, go to next stripe */
2467 if (stripe_hash == MDS_DIR_END_OFF)
2473 if (min_ent != NULL) {
2474 if (le64_to_cpu(min_ent->lde_hash) >
2475 le64_to_cpu(ent->lde_hash)) {
2478 page_cache_release(min_page);
2483 page_cache_release(page);
2494 if (*ppage != NULL) {
2496 page_cache_release(*ppage);
2498 *stripe_offset = min_idx;
2505 * Build dir entry page from a striped directory
2507 * This function gets one entry by @offset from a striped directory. It will
2508 * read entries from all of stripes, and choose one closest to the required
2509 * offset(&offset). A few notes
2510 * 1. skip . and .. for non-zero stripes, because there can only have one .
2511 * and .. in a directory.
2512 * 2. op_data will be shared by all of stripes, instead of allocating new
2513 * one, so need to restore before reusing.
2514 * 3. release the entry page if that is not being chosen.
2516 * \param[in] exp obd export refer to LMV
2517 * \param[in] op_data hold those MD parameters of read_entry
2518 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2519 * \param[out] ldp the entry being read
2520 * \param[out] ppage the page holding the entry. Note: because the entry
2521 * will be accessed in upper layer, so we need hold the
2522 * page until the usages of entry is finished, see
2523 * ll_dir_entry_next.
2525 * retval =0 if get entry successfully
2526 * <0 cannot get entry
2528 static int lmv_read_striped_page(struct obd_export *exp,
2529 struct md_op_data *op_data,
2530 struct md_callback *cb_op,
2531 __u64 offset, struct page **ppage)
2533 struct obd_device *obd = exp->exp_obd;
2534 struct lu_fid master_fid = op_data->op_fid1;
2535 struct inode *master_inode = op_data->op_data;
2536 __u64 hash_offset = offset;
2537 struct lu_dirpage *dp;
2538 struct page *min_ent_page = NULL;
2539 struct page *ent_page = NULL;
2540 struct lu_dirent *ent;
2543 struct lu_dirent *min_ent = NULL;
2544 struct lu_dirent *last_ent;
2549 rc = lmv_check_connect(obd);
2553 /* Allocate a page and read entries from all of stripes and fill
2554 * the page by hash order */
2555 ent_page = alloc_page(GFP_KERNEL);
2556 if (ent_page == NULL)
2559 /* Initialize the entry page */
2560 dp = kmap(ent_page);
2561 memset(dp, 0, sizeof(*dp));
2562 dp->ldp_hash_start = cpu_to_le64(offset);
2563 dp->ldp_flags |= LDF_COLLIDE;
2566 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2572 /* Find the minum entry from all sub-stripes */
2573 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2579 /* If it can not get minum entry, it means it already reaches
2580 * the end of this directory */
2581 if (min_ent == NULL) {
2582 last_ent->lde_reclen = 0;
2583 hash_offset = MDS_DIR_END_OFF;
2587 ent_size = le16_to_cpu(min_ent->lde_reclen);
2589 /* the last entry lde_reclen is 0, but it might not
2590 * the end of this entry of this temporay entry */
2592 ent_size = lu_dirent_calc_size(
2593 le16_to_cpu(min_ent->lde_namelen),
2594 le32_to_cpu(min_ent->lde_attrs));
2595 if (ent_size > left_bytes) {
2596 last_ent->lde_reclen = cpu_to_le16(0);
2597 hash_offset = le64_to_cpu(min_ent->lde_hash);
2601 memcpy(ent, min_ent, ent_size);
2603 /* Replace . with master FID and Replace .. with the parent FID
2604 * of master object */
2605 if (strncmp(ent->lde_name, ".",
2606 le16_to_cpu(ent->lde_namelen)) == 0 &&
2607 le16_to_cpu(ent->lde_namelen) == 1)
2608 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2609 else if (strncmp(ent->lde_name, "..",
2610 le16_to_cpu(ent->lde_namelen)) == 0 &&
2611 le16_to_cpu(ent->lde_namelen) == 2)
2612 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2614 left_bytes -= ent_size;
2615 ent->lde_reclen = cpu_to_le16(ent_size);
2617 ent = (void *)ent + ent_size;
2618 hash_offset = le64_to_cpu(min_ent->lde_hash);
2619 if (hash_offset == MDS_DIR_END_OFF) {
2620 last_ent->lde_reclen = 0;
2625 if (min_ent_page != NULL) {
2626 kunmap(min_ent_page);
2627 page_cache_release(min_ent_page);
2630 if (unlikely(rc != 0)) {
2631 __free_page(ent_page);
2635 dp->ldp_flags |= LDF_EMPTY;
2636 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2637 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2640 /* We do not want to allocate md_op_data during each
2641 * dir entry reading, so op_data will be shared by every stripe,
2642 * then we need to restore it back to original value before
2643 * return to the upper layer */
2644 op_data->op_fid1 = master_fid;
2645 op_data->op_fid2 = master_fid;
2646 op_data->op_data = master_inode;
2653 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2654 struct md_callback *cb_op, __u64 offset,
2655 struct page **ppage)
2657 struct obd_device *obd = exp->exp_obd;
2658 struct lmv_obd *lmv = &obd->u.lmv;
2659 struct lmv_stripe_md *lsm = op_data->op_mea1;
2660 struct lmv_tgt_desc *tgt;
2664 rc = lmv_check_connect(obd);
2668 if (unlikely(lsm != NULL)) {
2669 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2673 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2675 RETURN(PTR_ERR(tgt));
2677 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2683 * Unlink a file/directory
2685 * Unlink a file or directory under the parent dir. The unlink request
2686 * usually will be sent to the MDT where the child is located, but if
2687 * the client does not have the child FID then request will be sent to the
2688 * MDT where the parent is located.
2690 * If the parent is a striped directory then it also needs to locate which
2691 * stripe the name of the child is located, and replace the parent FID
2692 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2693 * it will walk through all of sub-stripes until the child is being
2696 * \param[in] exp export refer to LMV
2697 * \param[in] op_data different parameters transferred beween client
2698 * MD stacks, name, namelen, FIDs etc.
2699 * op_fid1 is the parent FID, op_fid2 is the child
2701 * \param[out] request point to the request of unlink.
2703 * retval 0 if succeed
2704 * negative errno if failed.
2706 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2707 struct ptlrpc_request **request)
2709 struct obd_device *obd = exp->exp_obd;
2710 struct lmv_obd *lmv = &obd->u.lmv;
2711 struct lmv_tgt_desc *tgt = NULL;
2712 struct lmv_tgt_desc *parent_tgt = NULL;
2713 struct mdt_body *body;
2715 int stripe_index = 0;
2716 struct lmv_stripe_md *lsm = op_data->op_mea1;
2719 rc = lmv_check_connect(obd);
2723 /* For striped dir, we need to locate the parent as well */
2725 struct lmv_tgt_desc *tmp;
2727 LASSERT(op_data->op_name != NULL &&
2728 op_data->op_namelen != 0);
2730 tmp = lmv_locate_target_for_name(lmv, lsm,
2732 op_data->op_namelen,
2736 /* return -EBADFD means unknown hash type, might
2737 * need try all sub-stripe here */
2738 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2739 RETURN(PTR_ERR(tmp));
2741 /* Note: both migrating dir and unknown hash dir need to
2742 * try all of sub-stripes, so we need start search the
2743 * name from stripe 0, but migrating dir is already handled
2744 * inside lmv_locate_target_for_name(), so we only check
2745 * unknown hash type directory here */
2746 if (!lmv_is_known_hash_type(lsm)) {
2747 struct lmv_oinfo *oinfo;
2749 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2751 op_data->op_fid1 = oinfo->lmo_fid;
2752 op_data->op_mds = oinfo->lmo_mds;
2757 /* Send unlink requests to the MDT where the child is located */
2758 if (likely(!fid_is_zero(&op_data->op_fid2)))
2759 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2760 else if (lsm != NULL)
2761 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2763 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2766 RETURN(PTR_ERR(tgt));
2768 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2769 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2770 op_data->op_cap = cfs_curproc_cap_pack();
2773 * If child's fid is given, cancel unused locks for it if it is from
2774 * another export than parent.
2776 * LOOKUP lock for child (fid3) should also be cancelled on parent
2777 * tgt_tgt in mdc_unlink().
2779 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2782 * Cancel FULL locks on child (fid3).
2784 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2785 if (IS_ERR(parent_tgt))
2786 RETURN(PTR_ERR(parent_tgt));
2788 if (parent_tgt != tgt) {
2789 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2790 LCK_EX, MDS_INODELOCK_LOOKUP,
2791 MF_MDC_CANCEL_FID3);
2794 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2795 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2799 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2800 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2802 rc = md_unlink(tgt->ltd_exp, op_data, request);
2803 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2806 /* Try next stripe if it is needed. */
2807 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2808 struct lmv_oinfo *oinfo;
2811 if (stripe_index >= lsm->lsm_md_stripe_count)
2814 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2816 op_data->op_fid1 = oinfo->lmo_fid;
2817 op_data->op_mds = oinfo->lmo_mds;
2819 ptlrpc_req_finished(*request);
2822 goto try_next_stripe;
2825 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2829 /* Not cross-ref case, just get out of here. */
2830 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2833 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2834 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2836 /* This is a remote object, try remote MDT, Note: it may
2837 * try more than 1 time here, Considering following case
2838 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2839 * 1. Initially A does not know where remote1 is, it send
2840 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2841 * resend unlink RPC to MDT1 (retry 1st time).
2843 * 2. During the unlink RPC in flight,
2844 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2845 * and create new remote1, but on MDT0
2847 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2848 * /mnt/lustre, then lookup get fid of remote1, and find
2849 * it is remote dir again, and replay -EREMOTE again.
2851 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2853 * In theory, it might try unlimited time here, but it should
2854 * be very rare case. */
2855 op_data->op_fid2 = body->mbo_fid1;
2856 ptlrpc_req_finished(*request);
2862 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2864 struct lmv_obd *lmv = &obd->u.lmv;
2868 case OBD_CLEANUP_EARLY:
2869 /* XXX: here should be calling obd_precleanup() down to
2872 case OBD_CLEANUP_EXPORTS:
2873 fld_client_proc_fini(&lmv->lmv_fld);
2874 lprocfs_obd_cleanup(obd);
2875 lprocfs_free_md_stats(obd);
2883 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2884 __u32 keylen, void *key, __u32 *vallen, void *val,
2885 struct lov_stripe_md *lsm)
2887 struct obd_device *obd;
2888 struct lmv_obd *lmv;
2892 obd = class_exp2obd(exp);
2894 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2895 exp->exp_handle.h_cookie);
2900 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2903 rc = lmv_check_connect(obd);
2907 LASSERT(*vallen == sizeof(__u32));
2908 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2909 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2911 * All tgts should be connected when this gets called.
2913 if (tgt == NULL || tgt->ltd_exp == NULL)
2916 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2921 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2922 KEY_IS(KEY_DEFAULT_EASIZE) ||
2923 KEY_IS(KEY_MAX_COOKIESIZE) ||
2924 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2925 KEY_IS(KEY_CONN_DATA)) {
2926 rc = lmv_check_connect(obd);
2931 * Forwarding this request to first MDS, it should know LOV
2934 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2936 if (!rc && KEY_IS(KEY_CONN_DATA))
2937 exp->exp_connect_data = *(struct obd_connect_data *)val;
2939 } else if (KEY_IS(KEY_TGT_COUNT)) {
2940 *((int *)val) = lmv->desc.ld_tgt_count;
2944 CDEBUG(D_IOCTL, "Invalid key\n");
2948 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2949 obd_count keylen, void *key, obd_count vallen,
2950 void *val, struct ptlrpc_request_set *set)
2952 struct lmv_tgt_desc *tgt = NULL;
2953 struct obd_device *obd;
2954 struct lmv_obd *lmv;
2958 obd = class_exp2obd(exp);
2960 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2961 exp->exp_handle.h_cookie);
2966 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2969 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2972 if (tgt == NULL || tgt->ltd_exp == NULL)
2975 err = obd_set_info_async(env, tgt->ltd_exp,
2976 keylen, key, vallen, val, set);
2987 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2988 struct lmv_mds_md_v1 *lmm1)
2993 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2994 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2995 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2996 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2997 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2998 sizeof(lmm1->lmv_pool_name));
2999 if (cplen >= sizeof(lmm1->lmv_pool_name))
3002 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3003 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
3004 &lsm->lsm_md_oinfo[i].lmo_fid);
3008 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
3012 bool allocated = false;
3016 LASSERT(lmmp != NULL);
3018 if (*lmmp != NULL && lsm == NULL) {
3021 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
3022 lmm_size = lmv_mds_md_size(stripe_count,
3023 le32_to_cpu((*lmmp)->lmv_magic));
3026 OBD_FREE(*lmmp, lmm_size);
3032 if (*lmmp == NULL && lsm == NULL) {
3033 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
3034 LASSERT(lmm_size > 0);
3035 OBD_ALLOC(*lmmp, lmm_size);
3038 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
3039 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
3044 LASSERT(lsm != NULL);
3045 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
3046 if (*lmmp == NULL) {
3047 OBD_ALLOC(*lmmp, lmm_size);
3053 switch (lsm->lsm_md_magic) {
3055 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
3062 if (rc != 0 && allocated) {
3063 OBD_FREE(*lmmp, lmm_size);
3069 EXPORT_SYMBOL(lmv_pack_md);
3071 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3072 const struct lmv_mds_md_v1 *lmm1)
3074 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3081 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3082 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3083 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3084 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3085 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3087 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3088 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3089 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3090 sizeof(lsm->lsm_md_pool_name));
3092 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3095 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3096 "layout_version %d\n", lsm->lsm_md_stripe_count,
3097 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3098 lsm->lsm_md_layout_version);
3100 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3101 for (i = 0; i < le32_to_cpu(stripe_count); i++) {
3102 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3103 &lmm1->lmv_stripe_fids[i]);
3104 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3105 &lsm->lsm_md_oinfo[i].lmo_mds);
3108 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3109 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3115 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3116 const union lmv_mds_md *lmm, int stripe_count)
3118 struct lmv_stripe_md *lsm;
3121 bool allocated = false;
3124 LASSERT(lsmp != NULL);
3128 if (lsm != NULL && lmm == NULL) {
3130 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3131 /* For migrating inode, the master stripe and master
3132 * object will be the same, so do not need iput, see
3133 * ll_update_lsm_md */
3134 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3135 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3136 iput(lsm->lsm_md_oinfo[i].lmo_root);
3138 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3139 OBD_FREE(lsm, lsm_size);
3145 if (lsm == NULL && lmm == NULL) {
3146 lsm_size = lmv_stripe_md_size(stripe_count);
3147 OBD_ALLOC(lsm, lsm_size);
3150 lsm->lsm_md_stripe_count = stripe_count;
3155 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3159 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3160 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3161 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3162 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3167 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3168 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3171 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3172 * stripecount should be 0 then.
3174 lsm_size = lmv_stripe_md_size(0);
3176 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3178 OBD_ALLOC(lsm, lsm_size);
3185 switch (le32_to_cpu(lmm->lmv_magic)) {
3187 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3190 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3191 le32_to_cpu(lmm->lmv_magic));
3196 if (rc != 0 && allocated) {
3197 OBD_FREE(lsm, lsm_size);
3204 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3206 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3208 EXPORT_SYMBOL(lmv_alloc_memmd);
3210 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3212 lmv_unpack_md(NULL, &lsm, NULL, 0);
3214 EXPORT_SYMBOL(lmv_free_memmd);
3216 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3217 struct lov_mds_md *lmm, int disk_len)
3219 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3220 (union lmv_mds_md *)lmm, disk_len);
3223 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3224 struct lov_stripe_md *lsm)
3226 struct obd_device *obd = exp->exp_obd;
3227 struct lmv_obd *lmv_obd = &obd->u.lmv;
3228 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
3233 stripe_count = lmv->lsm_md_stripe_count;
3235 stripe_count = lmv_obd->desc.ld_tgt_count;
3237 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3240 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3243 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3244 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3245 ldlm_cancel_flags_t flags, void *opaque)
3247 struct obd_device *obd = exp->exp_obd;
3248 struct lmv_obd *lmv = &obd->u.lmv;
3254 LASSERT(fid != NULL);
3256 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3257 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3259 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3262 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3270 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3273 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3274 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3278 if (tgt == NULL || tgt->ltd_exp == NULL)
3280 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3284 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3285 const struct lu_fid *fid, ldlm_type_t type,
3286 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3287 struct lustre_handle *lockh)
3289 struct obd_device *obd = exp->exp_obd;
3290 struct lmv_obd *lmv = &obd->u.lmv;
3296 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3299 * With DNE every object can have two locks in different namespaces:
3300 * lookup lock in space of MDT storing direntry and update/open lock in
3301 * space of MDT storing inode. Try the MDT that the FID maps to first,
3302 * since this can be easily found, and only try others if that fails.
3304 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3305 i < lmv->desc.ld_tgt_count;
3306 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3308 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3309 obd->obd_name, PFID(fid), tgt);
3313 if (lmv->tgts[tgt] == NULL ||
3314 lmv->tgts[tgt]->ltd_exp == NULL ||
3315 lmv->tgts[tgt]->ltd_active == 0)
3318 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3319 type, policy, mode, lockh);
3327 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3328 struct obd_export *dt_exp, struct obd_export *md_exp,
3329 struct lustre_md *md)
3331 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3332 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3334 if (tgt == NULL || tgt->ltd_exp == NULL)
3337 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3340 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3342 struct obd_device *obd = exp->exp_obd;
3343 struct lmv_obd *lmv = &obd->u.lmv;
3344 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3347 if (md->lmv != NULL) {
3348 lmv_free_memmd(md->lmv);
3351 if (tgt == NULL || tgt->ltd_exp == NULL)
3353 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3356 int lmv_set_open_replay_data(struct obd_export *exp,
3357 struct obd_client_handle *och,
3358 struct lookup_intent *it)
3360 struct obd_device *obd = exp->exp_obd;
3361 struct lmv_obd *lmv = &obd->u.lmv;
3362 struct lmv_tgt_desc *tgt;
3365 tgt = lmv_find_target(lmv, &och->och_fid);
3367 RETURN(PTR_ERR(tgt));
3369 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3372 int lmv_clear_open_replay_data(struct obd_export *exp,
3373 struct obd_client_handle *och)
3375 struct obd_device *obd = exp->exp_obd;
3376 struct lmv_obd *lmv = &obd->u.lmv;
3377 struct lmv_tgt_desc *tgt;
3380 tgt = lmv_find_target(lmv, &och->och_fid);
3382 RETURN(PTR_ERR(tgt));
3384 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3387 static int lmv_get_remote_perm(struct obd_export *exp,
3388 const struct lu_fid *fid,
3389 struct obd_capa *oc, __u32 suppgid,
3390 struct ptlrpc_request **request)
3392 struct obd_device *obd = exp->exp_obd;
3393 struct lmv_obd *lmv = &obd->u.lmv;
3394 struct lmv_tgt_desc *tgt;
3398 rc = lmv_check_connect(obd);
3402 tgt = lmv_find_target(lmv, fid);
3404 RETURN(PTR_ERR(tgt));
3406 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3410 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3413 struct obd_device *obd = exp->exp_obd;
3414 struct lmv_obd *lmv = &obd->u.lmv;
3415 struct lmv_tgt_desc *tgt;
3419 rc = lmv_check_connect(obd);
3423 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3425 RETURN(PTR_ERR(tgt));
3427 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3431 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3432 const struct req_msg_field *field, struct obd_capa **oc)
3434 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3435 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3437 if (tgt == NULL || tgt->ltd_exp == NULL)
3439 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3442 int lmv_intent_getattr_async(struct obd_export *exp,
3443 struct md_enqueue_info *minfo,
3444 struct ldlm_enqueue_info *einfo)
3446 struct md_op_data *op_data = &minfo->mi_data;
3447 struct obd_device *obd = exp->exp_obd;
3448 struct lmv_obd *lmv = &obd->u.lmv;
3449 struct lmv_tgt_desc *tgt = NULL;
3453 rc = lmv_check_connect(obd);
3457 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3459 RETURN(PTR_ERR(tgt));
3461 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3465 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3466 struct lu_fid *fid, __u64 *bits)
3468 struct obd_device *obd = exp->exp_obd;
3469 struct lmv_obd *lmv = &obd->u.lmv;
3470 struct lmv_tgt_desc *tgt;
3474 rc = lmv_check_connect(obd);
3478 tgt = lmv_find_target(lmv, fid);
3480 RETURN(PTR_ERR(tgt));
3482 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3486 int lmv_get_fid_from_lsm(struct obd_export *exp,
3487 const struct lmv_stripe_md *lsm,
3488 const char *name, int namelen, struct lu_fid *fid)
3490 const struct lmv_oinfo *oinfo;
3492 LASSERT(lsm != NULL);
3493 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3495 return PTR_ERR(oinfo);
3497 *fid = oinfo->lmo_fid;
3503 * For lmv, only need to send request to master MDT, and the master MDT will
3504 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3505 * we directly fetch data from the slave MDTs.
3507 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3508 struct obd_quotactl *oqctl)
3510 struct obd_device *obd = class_exp2obd(exp);
3511 struct lmv_obd *lmv = &obd->u.lmv;
3512 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3515 __u64 curspace, curinodes;
3519 tgt->ltd_exp == NULL ||
3521 lmv->desc.ld_tgt_count == 0) {
3522 CERROR("master lmv inactive\n");
3526 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3527 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3531 curspace = curinodes = 0;
3532 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3536 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3539 err = obd_quotactl(tgt->ltd_exp, oqctl);
3541 CERROR("getquota on mdt %d failed. %d\n", i, err);
3545 curspace += oqctl->qc_dqblk.dqb_curspace;
3546 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3549 oqctl->qc_dqblk.dqb_curspace = curspace;
3550 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3555 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3556 struct obd_quotactl *oqctl)
3558 struct obd_device *obd = class_exp2obd(exp);
3559 struct lmv_obd *lmv = &obd->u.lmv;
3560 struct lmv_tgt_desc *tgt;
3565 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3568 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3569 CERROR("lmv idx %d inactive\n", i);
3573 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3581 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3582 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3584 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3587 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3588 struct cl_attr *attr)
3592 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3593 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3595 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3596 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3597 i_size_read(inode), inode->i_nlink,
3598 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3599 LTIME_S(inode->i_mtime));
3601 /* for slave stripe, it needs to subtract nlink for . and .. */
3603 attr->cat_nlink += inode->i_nlink - 2;
3605 attr->cat_nlink = inode->i_nlink;
3607 attr->cat_size += i_size_read(inode);
3609 if (attr->cat_atime < LTIME_S(inode->i_atime))
3610 attr->cat_atime = LTIME_S(inode->i_atime);
3612 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3613 attr->cat_ctime = LTIME_S(inode->i_ctime);
3615 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3616 attr->cat_mtime = LTIME_S(inode->i_mtime);
3621 struct obd_ops lmv_obd_ops = {
3622 .o_owner = THIS_MODULE,
3623 .o_setup = lmv_setup,
3624 .o_cleanup = lmv_cleanup,
3625 .o_precleanup = lmv_precleanup,
3626 .o_process_config = lmv_process_config,
3627 .o_connect = lmv_connect,
3628 .o_disconnect = lmv_disconnect,
3629 .o_statfs = lmv_statfs,
3630 .o_get_info = lmv_get_info,
3631 .o_set_info_async = lmv_set_info_async,
3632 .o_packmd = lmv_packmd,
3633 .o_unpackmd = lmv_unpackmd,
3634 .o_notify = lmv_notify,
3635 .o_get_uuid = lmv_get_uuid,
3636 .o_iocontrol = lmv_iocontrol,
3637 .o_quotacheck = lmv_quotacheck,
3638 .o_quotactl = lmv_quotactl
3641 struct md_ops lmv_md_ops = {
3642 .m_getstatus = lmv_getstatus,
3643 .m_null_inode = lmv_null_inode,
3644 .m_find_cbdata = lmv_find_cbdata,
3645 .m_close = lmv_close,
3646 .m_create = lmv_create,
3647 .m_done_writing = lmv_done_writing,
3648 .m_enqueue = lmv_enqueue,
3649 .m_getattr = lmv_getattr,
3650 .m_getxattr = lmv_getxattr,
3651 .m_getattr_name = lmv_getattr_name,
3652 .m_intent_lock = lmv_intent_lock,
3654 .m_rename = lmv_rename,
3655 .m_setattr = lmv_setattr,
3656 .m_setxattr = lmv_setxattr,
3657 .m_fsync = lmv_fsync,
3658 .m_read_page = lmv_read_page,
3659 .m_unlink = lmv_unlink,
3660 .m_init_ea_size = lmv_init_ea_size,
3661 .m_cancel_unused = lmv_cancel_unused,
3662 .m_set_lock_data = lmv_set_lock_data,
3663 .m_lock_match = lmv_lock_match,
3664 .m_get_lustre_md = lmv_get_lustre_md,
3665 .m_free_lustre_md = lmv_free_lustre_md,
3666 .m_update_lsm_md = lmv_update_lsm_md,
3667 .m_merge_attr = lmv_merge_attr,
3668 .m_set_open_replay_data = lmv_set_open_replay_data,
3669 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3670 .m_renew_capa = lmv_renew_capa,
3671 .m_unpack_capa = lmv_unpack_capa,
3672 .m_get_remote_perm = lmv_get_remote_perm,
3673 .m_intent_getattr_async = lmv_intent_getattr_async,
3674 .m_revalidate_lock = lmv_revalidate_lock,
3675 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3678 int __init lmv_init(void)
3680 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3681 #ifndef HAVE_ONLY_PROCFS_SEQ
3684 LUSTRE_LMV_NAME, NULL);
3687 static void lmv_exit(void)
3689 class_unregister_type(LUSTRE_LMV_NAME);
3692 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3693 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3694 MODULE_LICENSE("GPL");
3696 module_init(lmv_init);
3697 module_exit(lmv_exit);