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(enum lmv_hash_type hashtype,
94 unsigned int max_mdt_index,
95 const char *name, int namelen)
100 if (max_mdt_index <= 1)
104 case LMV_HASH_TYPE_ALL_CHARS:
105 idx = lmv_hash_all_chars(max_mdt_index, name, namelen);
107 case LMV_HASH_TYPE_FNV_1A_64:
108 idx = lmv_hash_fnv1a(max_mdt_index, name, namelen);
110 /* LMV_HASH_TYPE_MIGRATION means the file is being migrated,
111 * and the file should be accessed by client, except for
112 * lookup(see lmv_intent_lookup), return -EACCES here */
113 case LMV_HASH_TYPE_MIGRATION:
114 CERROR("%.*s is being migrated: rc = %d\n", namelen,
118 CERROR("Unknown hash type 0x%x\n", hashtype);
122 CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
125 LASSERT(idx < max_mdt_index);
129 static void lmv_activate_target(struct lmv_obd *lmv,
130 struct lmv_tgt_desc *tgt,
133 if (tgt->ltd_active == activate)
136 tgt->ltd_active = activate;
137 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
143 * -EINVAL : UUID can't be found in the LMV's target list
144 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
145 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
147 static int lmv_set_mdc_active(struct lmv_obd *lmv,
148 const struct obd_uuid *uuid,
151 struct lmv_tgt_desc *tgt = NULL;
152 struct obd_device *obd;
157 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
158 lmv, uuid->uuid, activate);
160 spin_lock(&lmv->lmv_lock);
161 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
163 if (tgt == NULL || tgt->ltd_exp == NULL)
166 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
167 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
169 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
173 if (i == lmv->desc.ld_tgt_count)
174 GOTO(out_lmv_lock, rc = -EINVAL);
176 obd = class_exp2obd(tgt->ltd_exp);
178 GOTO(out_lmv_lock, rc = -ENOTCONN);
180 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
181 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
182 obd->obd_type->typ_name, i);
183 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
185 if (tgt->ltd_active == activate) {
186 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
187 activate ? "" : "in");
188 GOTO(out_lmv_lock, rc);
191 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
192 activate ? "" : "in");
193 lmv_activate_target(lmv, tgt, activate);
197 spin_unlock(&lmv->lmv_lock);
201 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
203 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
204 struct lmv_tgt_desc *tgt = lmv->tgts[0];
206 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
209 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
210 enum obd_notify_event ev, void *data)
212 struct obd_connect_data *conn_data;
213 struct lmv_obd *lmv = &obd->u.lmv;
214 struct obd_uuid *uuid;
218 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
219 CERROR("unexpected notification of %s %s!\n",
220 watched->obd_type->typ_name,
225 uuid = &watched->u.cli.cl_target_uuid;
226 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
228 * Set MDC as active before notifying the observer, so the
229 * observer can use the MDC normally.
231 rc = lmv_set_mdc_active(lmv, uuid,
232 ev == OBD_NOTIFY_ACTIVE);
234 CERROR("%sactivation of %s failed: %d\n",
235 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
239 } else if (ev == OBD_NOTIFY_OCD) {
240 conn_data = &watched->u.cli.cl_import->imp_connect_data;
242 * XXX: Make sure that ocd_connect_flags from all targets are
243 * the same. Otherwise one of MDTs runs wrong version or
244 * something like this. --umka
246 obd->obd_self_export->exp_connect_data = *conn_data;
249 else if (ev == OBD_NOTIFY_DISCON) {
251 * For disconnect event, flush fld cache for failout MDS case.
253 fld_client_flush(&lmv->lmv_fld);
257 * Pass the notification up the chain.
259 if (obd->obd_observer)
260 rc = obd_notify(obd->obd_observer, watched, ev, data);
266 * This is fake connect function. Its purpose is to initialize lmv and say
267 * caller that everything is okay. Real connection will be performed later.
269 static int lmv_connect(const struct lu_env *env,
270 struct obd_export **exp, struct obd_device *obd,
271 struct obd_uuid *cluuid, struct obd_connect_data *data,
274 struct lmv_obd *lmv = &obd->u.lmv;
275 struct lustre_handle conn = { 0 };
280 * We don't want to actually do the underlying connections more than
281 * once, so keep track.
284 if (lmv->refcount > 1) {
289 rc = class_connect(&conn, obd, cluuid);
291 CERROR("class_connection() returned %d\n", rc);
295 *exp = class_conn2export(&conn);
296 class_export_get(*exp);
300 lmv->cluuid = *cluuid;
303 lmv->conn_data = *data;
305 if (obd->obd_type->typ_procsym == NULL) {
306 obd->obd_type->typ_procsym = lprocfs_seq_register("target_obds",
309 if (IS_ERR(obd->obd_type->typ_procsym)) {
310 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
311 obd->obd_type->typ_name, obd->obd_name);
312 obd->obd_type->typ_procsym = NULL;
317 * All real clients should perform actual connection right away, because
318 * it is possible, that LMV will not have opportunity to connect targets
319 * and MDC stuff will be called directly, for instance while reading
320 * ../mdc/../kbytesfree procfs file, etc.
322 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
323 rc = lmv_check_connect(obd);
325 if (rc && obd->obd_type->typ_procsym != NULL)
326 lprocfs_remove(&obd->obd_type->typ_procsym);
330 static void lmv_set_timeouts(struct obd_device *obd)
336 if (lmv->server_timeout == 0)
339 if (lmv->connected == 0)
342 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
343 struct lmv_tgt_desc *tgt = lmv->tgts[i];
345 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
348 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
349 KEY_INTERMDS, 0, NULL, NULL);
353 static int lmv_init_ea_size(struct obd_export *exp, int easize,
354 int def_easize, int cookiesize, int def_cookiesize)
356 struct obd_device *obd = exp->exp_obd;
357 struct lmv_obd *lmv = &obd->u.lmv;
363 if (lmv->max_easize < easize) {
364 lmv->max_easize = easize;
367 if (lmv->max_def_easize < def_easize) {
368 lmv->max_def_easize = def_easize;
371 if (lmv->max_cookiesize < cookiesize) {
372 lmv->max_cookiesize = cookiesize;
375 if (lmv->max_def_cookiesize < def_cookiesize) {
376 lmv->max_def_cookiesize = def_cookiesize;
382 if (lmv->connected == 0)
385 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
386 struct lmv_tgt_desc *tgt = lmv->tgts[i];
388 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
389 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
393 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
394 cookiesize, def_cookiesize);
396 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
397 " rc = %d.\n", obd->obd_name, i, rc);
404 #define MAX_STRING_SIZE 128
406 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
408 struct lmv_obd *lmv = &obd->u.lmv;
409 struct obd_uuid *cluuid = &lmv->cluuid;
410 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
411 struct obd_device *mdc_obd;
412 struct obd_export *mdc_exp;
413 struct lu_fld_target target;
417 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
420 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
424 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
425 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
426 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
429 if (!mdc_obd->obd_set_up) {
430 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
434 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
435 &lmv->conn_data, NULL);
437 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
442 * Init fid sequence client for this mdc and add new fld target.
444 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
448 target.ft_srv = NULL;
449 target.ft_exp = mdc_exp;
450 target.ft_idx = tgt->ltd_idx;
452 fld_client_add_target(&lmv->lmv_fld, &target);
454 rc = obd_register_observer(mdc_obd, obd);
456 obd_disconnect(mdc_exp);
457 CERROR("target %s register_observer error %d\n",
458 tgt->ltd_uuid.uuid, rc);
462 if (obd->obd_observer) {
464 * Tell the observer about the new target.
466 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
468 (void *)(tgt - lmv->tgts[0]));
470 obd_disconnect(mdc_exp);
476 tgt->ltd_exp = mdc_exp;
477 lmv->desc.ld_active_tgt_count++;
479 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
480 lmv->max_cookiesize, lmv->max_def_cookiesize);
482 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
483 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
484 atomic_read(&obd->obd_refcount));
486 if (obd->obd_type->typ_procsym != NULL) {
487 struct proc_dir_entry *mdc_symlink;
489 LASSERT(mdc_obd->obd_type != NULL);
490 LASSERT(mdc_obd->obd_type->typ_name != NULL);
491 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
492 obd->obd_type->typ_procsym,
494 mdc_obd->obd_type->typ_name,
496 if (mdc_symlink == NULL) {
497 CERROR("Could not register LMV target "
498 "/proc/fs/lustre/%s/%s/target_obds/%s.",
499 obd->obd_type->typ_name, obd->obd_name,
501 lprocfs_remove(&obd->obd_type->typ_procsym);
507 static void lmv_del_target(struct lmv_obd *lmv, int index)
509 if (lmv->tgts[index] == NULL)
512 OBD_FREE_PTR(lmv->tgts[index]);
513 lmv->tgts[index] = NULL;
517 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
518 __u32 index, int gen)
520 struct lmv_obd *lmv = &obd->u.lmv;
521 struct lmv_tgt_desc *tgt;
525 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
529 if (lmv->desc.ld_tgt_count == 0) {
530 struct obd_device *mdc_obd;
532 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
535 lmv_init_unlock(lmv);
536 CERROR("%s: Target %s not attached: rc = %d\n",
537 obd->obd_name, uuidp->uuid, -EINVAL);
542 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
543 tgt = lmv->tgts[index];
544 CERROR("%s: UUID %s already assigned at LOV target index %d:"
545 " rc = %d\n", obd->obd_name,
546 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
547 lmv_init_unlock(lmv);
551 if (index >= lmv->tgts_size) {
552 /* We need to reallocate the lmv target array. */
553 struct lmv_tgt_desc **newtgts, **old = NULL;
557 while (newsize < index + 1)
558 newsize = newsize << 1;
559 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
560 if (newtgts == NULL) {
561 lmv_init_unlock(lmv);
565 if (lmv->tgts_size) {
566 memcpy(newtgts, lmv->tgts,
567 sizeof(*newtgts) * lmv->tgts_size);
569 oldsize = lmv->tgts_size;
573 lmv->tgts_size = newsize;
576 OBD_FREE(old, sizeof(*old) * oldsize);
578 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
584 lmv_init_unlock(lmv);
588 mutex_init(&tgt->ltd_fid_mutex);
589 tgt->ltd_idx = index;
590 tgt->ltd_uuid = *uuidp;
592 lmv->tgts[index] = tgt;
593 if (index >= lmv->desc.ld_tgt_count)
594 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 lmv->desc.ld_tgt_count--;
601 memset(tgt, 0, sizeof(*tgt));
602 spin_unlock(&lmv->lmv_lock);
604 int easize = sizeof(struct lmv_stripe_md) +
605 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
606 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
610 lmv_init_unlock(lmv);
614 int lmv_check_connect(struct obd_device *obd)
616 struct lmv_obd *lmv = &obd->u.lmv;
617 struct lmv_tgt_desc *tgt;
627 if (lmv->connected) {
628 lmv_init_unlock(lmv);
632 if (lmv->desc.ld_tgt_count == 0) {
633 lmv_init_unlock(lmv);
634 CERROR("%s: no targets configured.\n", obd->obd_name);
638 LASSERT(lmv->tgts != NULL);
640 if (lmv->tgts[0] == NULL) {
641 lmv_init_unlock(lmv);
642 CERROR("%s: no target configured for index 0.\n",
647 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
648 lmv->cluuid.uuid, obd->obd_name);
650 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
654 rc = lmv_connect_mdc(obd, tgt);
659 lmv_set_timeouts(obd);
660 class_export_put(lmv->exp);
662 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
663 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
664 lmv_init_unlock(lmv);
675 --lmv->desc.ld_active_tgt_count;
676 rc2 = obd_disconnect(tgt->ltd_exp);
678 CERROR("LMV target %s disconnect on "
679 "MDC idx %d: error %d\n",
680 tgt->ltd_uuid.uuid, i, rc2);
684 class_disconnect(lmv->exp);
685 lmv_init_unlock(lmv);
689 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
691 struct lmv_obd *lmv = &obd->u.lmv;
692 struct obd_device *mdc_obd;
696 LASSERT(tgt != NULL);
697 LASSERT(obd != NULL);
699 mdc_obd = class_exp2obd(tgt->ltd_exp);
702 mdc_obd->obd_force = obd->obd_force;
703 mdc_obd->obd_fail = obd->obd_fail;
704 mdc_obd->obd_no_recov = obd->obd_no_recov;
707 if (obd->obd_type->typ_procsym != NULL)
708 lprocfs_remove_proc_entry(mdc_obd->obd_name,
709 obd->obd_type->typ_procsym);
711 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
713 CERROR("Can't finanize fids factory\n");
715 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
716 tgt->ltd_exp->exp_obd->obd_name,
717 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
719 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
720 rc = obd_disconnect(tgt->ltd_exp);
722 if (tgt->ltd_active) {
723 CERROR("Target %s disconnect error %d\n",
724 tgt->ltd_uuid.uuid, rc);
728 lmv_activate_target(lmv, tgt, 0);
733 static int lmv_disconnect(struct obd_export *exp)
735 struct obd_device *obd = class_exp2obd(exp);
736 struct lmv_obd *lmv = &obd->u.lmv;
745 * Only disconnect the underlying layers on the final disconnect.
748 if (lmv->refcount != 0)
751 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
752 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
755 lmv_disconnect_mdc(obd, lmv->tgts[i]);
758 if (obd->obd_type->typ_procsym != NULL)
759 lprocfs_remove(&obd->obd_type->typ_procsym);
761 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
762 obd->obd_type->typ_name, obd->obd_name);
766 * This is the case when no real connection is established by
767 * lmv_check_connect().
770 class_export_put(exp);
771 rc = class_disconnect(exp);
772 if (lmv->refcount == 0)
777 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
779 struct obd_device *obddev = class_exp2obd(exp);
780 struct lmv_obd *lmv = &obddev->u.lmv;
781 struct getinfo_fid2path *gf;
782 struct lmv_tgt_desc *tgt;
783 struct getinfo_fid2path *remote_gf = NULL;
784 int remote_gf_size = 0;
787 gf = (struct getinfo_fid2path *)karg;
788 tgt = lmv_find_target(lmv, &gf->gf_fid);
790 RETURN(PTR_ERR(tgt));
793 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
794 if (rc != 0 && rc != -EREMOTE)
795 GOTO(out_fid2path, rc);
797 /* If remote_gf != NULL, it means just building the
798 * path on the remote MDT, copy this path segement to gf */
799 if (remote_gf != NULL) {
800 struct getinfo_fid2path *ori_gf;
803 ori_gf = (struct getinfo_fid2path *)karg;
804 if (strlen(ori_gf->gf_path) +
805 strlen(gf->gf_path) > ori_gf->gf_pathlen)
806 GOTO(out_fid2path, rc = -EOVERFLOW);
808 ptr = ori_gf->gf_path;
810 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
811 strlen(ori_gf->gf_path));
813 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
814 ptr += strlen(gf->gf_path);
818 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
819 tgt->ltd_exp->exp_obd->obd_name,
820 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
824 GOTO(out_fid2path, rc);
826 /* sigh, has to go to another MDT to do path building further */
827 if (remote_gf == NULL) {
828 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
829 OBD_ALLOC(remote_gf, remote_gf_size);
830 if (remote_gf == NULL)
831 GOTO(out_fid2path, rc = -ENOMEM);
832 remote_gf->gf_pathlen = PATH_MAX;
835 if (!fid_is_sane(&gf->gf_fid)) {
836 CERROR("%s: invalid FID "DFID": rc = %d\n",
837 tgt->ltd_exp->exp_obd->obd_name,
838 PFID(&gf->gf_fid), -EINVAL);
839 GOTO(out_fid2path, rc = -EINVAL);
842 tgt = lmv_find_target(lmv, &gf->gf_fid);
844 GOTO(out_fid2path, rc = -EINVAL);
846 remote_gf->gf_fid = gf->gf_fid;
847 remote_gf->gf_recno = -1;
848 remote_gf->gf_linkno = -1;
849 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
851 goto repeat_fid2path;
854 if (remote_gf != NULL)
855 OBD_FREE(remote_gf, remote_gf_size);
859 static int lmv_hsm_req_count(struct lmv_obd *lmv,
860 const struct hsm_user_request *hur,
861 const struct lmv_tgt_desc *tgt_mds)
865 struct lmv_tgt_desc *curr_tgt;
867 /* count how many requests must be sent to the given target */
868 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
869 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
870 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
876 static void lmv_hsm_req_build(struct lmv_obd *lmv,
877 struct hsm_user_request *hur_in,
878 const struct lmv_tgt_desc *tgt_mds,
879 struct hsm_user_request *hur_out)
882 struct lmv_tgt_desc *curr_tgt;
884 /* build the hsm_user_request for the given target */
885 hur_out->hur_request = hur_in->hur_request;
887 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
888 curr_tgt = lmv_find_target(lmv,
889 &hur_in->hur_user_item[i].hui_fid);
890 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
891 hur_out->hur_user_item[nr_out] =
892 hur_in->hur_user_item[i];
896 hur_out->hur_request.hr_itemcount = nr_out;
897 memcpy(hur_data(hur_out), hur_data(hur_in),
898 hur_in->hur_request.hr_data_len);
901 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
902 struct lustre_kernelcomm *lk, void *uarg)
906 struct kkuc_ct_data *kcd = NULL;
909 /* unregister request (call from llapi_hsm_copytool_fini) */
910 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
911 struct lmv_tgt_desc *tgt = lmv->tgts[i];
913 if (tgt == NULL || tgt->ltd_exp == NULL)
915 /* best effort: try to clean as much as possible
916 * (continue on error) */
917 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
920 /* Whatever the result, remove copytool from kuc groups.
921 * Unreached coordinators will get EPIPE on next requests
922 * and will unregister automatically.
924 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
931 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
932 struct lustre_kernelcomm *lk, void *uarg)
937 bool any_set = false;
938 struct kkuc_ct_data *kcd;
941 /* All or nothing: try to register to all MDS.
942 * In case of failure, unregister from previous MDS,
943 * except if it because of inactive target. */
944 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
945 struct lmv_tgt_desc *tgt = lmv->tgts[i];
947 if (tgt == NULL || tgt->ltd_exp == NULL)
949 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
951 if (tgt->ltd_active) {
952 /* permanent error */
953 CERROR("%s: iocontrol MDC %s on MDT"
954 " idx %d cmd %x: err = %d\n",
955 class_exp2obd(lmv->exp)->obd_name,
956 tgt->ltd_uuid.uuid, i, cmd, err);
958 lk->lk_flags |= LK_FLG_STOP;
959 /* unregister from previous MDS */
960 for (j = 0; j < i; j++) {
962 if (tgt == NULL || tgt->ltd_exp == NULL)
964 obd_iocontrol(cmd, tgt->ltd_exp, len,
969 /* else: transient error.
970 * kuc will register to the missing MDT
978 /* no registration done: return error */
981 /* at least one registration done, with no failure */
982 filp = fget(lk->lk_wfd);
991 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
992 kcd->kcd_uuid = lmv->cluuid;
993 kcd->kcd_archive = lk->lk_data;
995 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
1008 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1009 int len, void *karg, void *uarg)
1011 struct obd_device *obddev = class_exp2obd(exp);
1012 struct lmv_obd *lmv = &obddev->u.lmv;
1013 struct lmv_tgt_desc *tgt = NULL;
1017 __u32 count = lmv->desc.ld_tgt_count;
1024 case IOC_OBD_STATFS: {
1025 struct obd_ioctl_data *data = karg;
1026 struct obd_device *mdc_obd;
1027 struct obd_statfs stat_buf = {0};
1030 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1031 if ((index >= count))
1034 tgt = lmv->tgts[index];
1035 if (tgt == NULL || !tgt->ltd_active)
1038 mdc_obd = class_exp2obd(tgt->ltd_exp);
1043 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1044 min((int) data->ioc_plen2,
1045 (int) sizeof(struct obd_uuid))))
1048 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1049 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1053 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1054 min((int) data->ioc_plen1,
1055 (int) sizeof(stat_buf))))
1059 case OBD_IOC_QUOTACTL: {
1060 struct if_quotactl *qctl = karg;
1061 struct obd_quotactl *oqctl;
1063 if (qctl->qc_valid == QC_MDTIDX) {
1064 if (count <= qctl->qc_idx)
1067 tgt = lmv->tgts[qctl->qc_idx];
1068 if (tgt == NULL || tgt->ltd_exp == NULL)
1070 } else if (qctl->qc_valid == QC_UUID) {
1071 for (i = 0; i < count; i++) {
1075 if (!obd_uuid_equals(&tgt->ltd_uuid,
1079 if (tgt->ltd_exp == NULL)
1091 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1092 OBD_ALLOC_PTR(oqctl);
1096 QCTL_COPY(oqctl, qctl);
1097 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1099 QCTL_COPY(qctl, oqctl);
1100 qctl->qc_valid = QC_MDTIDX;
1101 qctl->obd_uuid = tgt->ltd_uuid;
1103 OBD_FREE_PTR(oqctl);
1106 case OBD_IOC_CHANGELOG_SEND:
1107 case OBD_IOC_CHANGELOG_CLEAR: {
1108 struct ioc_changelog *icc = karg;
1110 if (icc->icc_mdtindex >= count)
1113 tgt = lmv->tgts[icc->icc_mdtindex];
1114 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1116 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1119 case LL_IOC_GET_CONNECT_FLAGS: {
1121 if (tgt == NULL || tgt->ltd_exp == NULL)
1123 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1126 case OBD_IOC_FID2PATH: {
1127 rc = lmv_fid2path(exp, len, karg, uarg);
1130 case LL_IOC_HSM_STATE_GET:
1131 case LL_IOC_HSM_STATE_SET:
1132 case LL_IOC_HSM_ACTION: {
1133 struct md_op_data *op_data = karg;
1135 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1137 RETURN(PTR_ERR(tgt));
1139 if (tgt->ltd_exp == NULL)
1142 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1145 case LL_IOC_HSM_PROGRESS: {
1146 const struct hsm_progress_kernel *hpk = karg;
1148 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1150 RETURN(PTR_ERR(tgt));
1151 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1154 case LL_IOC_HSM_REQUEST: {
1155 struct hsm_user_request *hur = karg;
1156 unsigned int reqcount = hur->hur_request.hr_itemcount;
1161 /* if the request is about a single fid
1162 * or if there is a single MDS, no need to split
1164 if (reqcount == 1 || count == 1) {
1165 tgt = lmv_find_target(lmv,
1166 &hur->hur_user_item[0].hui_fid);
1168 RETURN(PTR_ERR(tgt));
1169 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1171 /* split fid list to their respective MDS */
1172 for (i = 0; i < count; i++) {
1173 unsigned int nr, reqlen;
1175 struct hsm_user_request *req;
1178 if (tgt == NULL || tgt->ltd_exp == NULL)
1181 nr = lmv_hsm_req_count(lmv, hur, tgt);
1182 if (nr == 0) /* nothing for this MDS */
1185 /* build a request with fids for this MDS */
1186 reqlen = offsetof(typeof(*hur),
1188 + hur->hur_request.hr_data_len;
1189 OBD_ALLOC_LARGE(req, reqlen);
1193 lmv_hsm_req_build(lmv, hur, tgt, req);
1195 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1197 if (rc1 != 0 && rc == 0)
1199 OBD_FREE_LARGE(req, reqlen);
1204 case LL_IOC_LOV_SWAP_LAYOUTS: {
1205 struct md_op_data *op_data = karg;
1206 struct lmv_tgt_desc *tgt1, *tgt2;
1208 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1210 RETURN(PTR_ERR(tgt1));
1212 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1214 RETURN(PTR_ERR(tgt2));
1216 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1219 /* only files on same MDT can have their layouts swapped */
1220 if (tgt1->ltd_idx != tgt2->ltd_idx)
1223 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1226 case LL_IOC_HSM_CT_START: {
1227 struct lustre_kernelcomm *lk = karg;
1228 if (lk->lk_flags & LK_FLG_STOP)
1229 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1231 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1235 for (i = 0; i < count; i++) {
1236 struct obd_device *mdc_obd;
1240 if (tgt == NULL || tgt->ltd_exp == NULL)
1242 /* ll_umount_begin() sets force flag but for lmv, not
1243 * mdc. Let's pass it through */
1244 mdc_obd = class_exp2obd(tgt->ltd_exp);
1245 mdc_obd->obd_force = obddev->obd_force;
1246 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1247 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1250 if (tgt->ltd_active) {
1251 CERROR("error: iocontrol MDC %s on MDT"
1252 " idx %d cmd %x: err = %d\n",
1253 tgt->ltd_uuid.uuid, i, cmd, err);
1267 static int lmv_all_chars_policy(int count, const char *name,
1278 static int lmv_nid_policy(struct lmv_obd *lmv)
1280 struct obd_import *imp;
1284 * XXX: To get nid we assume that underlying obd device is mdc.
1286 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1287 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1288 return id % lmv->desc.ld_tgt_count;
1291 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1292 placement_policy_t placement)
1294 switch (placement) {
1295 case PLACEMENT_CHAR_POLICY:
1296 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1298 op_data->op_namelen);
1299 case PLACEMENT_NID_POLICY:
1300 return lmv_nid_policy(lmv);
1306 CERROR("Unsupported placement policy %x\n", placement);
1312 * This is _inode_ placement policy function (not name).
1314 static int lmv_placement_policy(struct obd_device *obd,
1315 struct md_op_data *op_data,
1318 struct lmv_obd *lmv = &obd->u.lmv;
1321 LASSERT(mds != NULL);
1323 if (lmv->desc.ld_tgt_count == 1) {
1329 * If stripe_offset is provided during setdirstripe
1330 * (setdirstripe -i xx), xx MDS will be choosen.
1332 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1333 struct lmv_user_md *lum;
1335 lum = op_data->op_data;
1337 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1338 *mds = le32_to_cpu(lum->lum_stripe_offset);
1340 /* -1 means default, which will be in the same MDT with
1342 *mds = op_data->op_mds;
1343 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1346 /* Allocate new fid on target according to operation type and
1347 * parent home mds. */
1348 *mds = op_data->op_mds;
1354 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1357 struct lmv_tgt_desc *tgt;
1361 tgt = lmv_get_target(lmv, mds);
1363 RETURN(PTR_ERR(tgt));
1366 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1367 * on server that seq in new allocated fid is not yet known.
1369 mutex_lock(&tgt->ltd_fid_mutex);
1371 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1372 GOTO(out, rc = -ENODEV);
1375 * Asking underlaying tgt layer to allocate new fid.
1377 rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1379 LASSERT(fid_is_sane(fid));
1385 mutex_unlock(&tgt->ltd_fid_mutex);
1389 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1390 struct md_op_data *op_data)
1392 struct obd_device *obd = class_exp2obd(exp);
1393 struct lmv_obd *lmv = &obd->u.lmv;
1398 LASSERT(op_data != NULL);
1399 LASSERT(fid != NULL);
1401 rc = lmv_placement_policy(obd, op_data, &mds);
1403 CERROR("Can't get target for allocating fid, "
1408 rc = __lmv_fid_alloc(lmv, fid, mds);
1410 CERROR("Can't alloc new fid, rc %d\n", rc);
1417 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1419 struct lmv_obd *lmv = &obd->u.lmv;
1420 struct lmv_desc *desc;
1424 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1425 CERROR("LMV setup requires a descriptor\n");
1429 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1430 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1431 CERROR("Lmv descriptor size wrong: %d > %d\n",
1432 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1436 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1437 if (lmv->tgts == NULL)
1439 lmv->tgts_size = 32;
1441 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1442 lmv->desc.ld_tgt_count = 0;
1443 lmv->desc.ld_active_tgt_count = 0;
1444 lmv->max_cookiesize = 0;
1445 lmv->max_def_easize = 0;
1446 lmv->max_easize = 0;
1447 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1449 spin_lock_init(&lmv->lmv_lock);
1450 mutex_init(&lmv->init_mutex);
1453 obd->obd_vars = lprocfs_lmv_obd_vars;
1454 lprocfs_seq_obd_setup(obd);
1455 lprocfs_alloc_md_stats(obd, 0);
1456 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1457 0444, &lmv_proc_target_fops, obd);
1459 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1462 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1463 LUSTRE_CLI_FLD_HASH_DHT);
1465 CERROR("Can't init FLD, err %d\n", rc);
1475 static int lmv_cleanup(struct obd_device *obd)
1477 struct lmv_obd *lmv = &obd->u.lmv;
1480 fld_client_fini(&lmv->lmv_fld);
1481 if (lmv->tgts != NULL) {
1483 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1484 if (lmv->tgts[i] == NULL)
1486 lmv_del_target(lmv, i);
1488 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1494 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1496 struct lustre_cfg *lcfg = buf;
1497 struct obd_uuid obd_uuid;
1503 switch (lcfg->lcfg_command) {
1505 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1506 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1507 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1508 GOTO(out, rc = -EINVAL);
1510 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1512 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1513 GOTO(out, rc = -EINVAL);
1514 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1515 GOTO(out, rc = -EINVAL);
1516 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1519 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1520 GOTO(out, rc = -EINVAL);
1526 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1527 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1529 struct obd_device *obd = class_exp2obd(exp);
1530 struct lmv_obd *lmv = &obd->u.lmv;
1531 struct obd_statfs *temp;
1536 rc = lmv_check_connect(obd);
1540 OBD_ALLOC(temp, sizeof(*temp));
1544 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1545 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1548 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1551 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1552 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1554 GOTO(out_free_temp, rc);
1559 /* If the statfs is from mount, it will needs
1560 * retrieve necessary information from MDT0.
1561 * i.e. mount does not need the merged osfs
1563 * And also clients can be mounted as long as
1564 * MDT0 is in service*/
1565 if (flags & OBD_STATFS_FOR_MDT0)
1566 GOTO(out_free_temp, rc);
1568 osfs->os_bavail += temp->os_bavail;
1569 osfs->os_blocks += temp->os_blocks;
1570 osfs->os_ffree += temp->os_ffree;
1571 osfs->os_files += temp->os_files;
1577 OBD_FREE(temp, sizeof(*temp));
1581 static int lmv_getstatus(struct obd_export *exp,
1583 struct obd_capa **pc)
1585 struct obd_device *obd = exp->exp_obd;
1586 struct lmv_obd *lmv = &obd->u.lmv;
1590 rc = lmv_check_connect(obd);
1594 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1598 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1599 struct obd_capa *oc, obd_valid valid, const char *name,
1600 const char *input, int input_size, int output_size,
1601 int flags, struct ptlrpc_request **request)
1603 struct obd_device *obd = exp->exp_obd;
1604 struct lmv_obd *lmv = &obd->u.lmv;
1605 struct lmv_tgt_desc *tgt;
1609 rc = lmv_check_connect(obd);
1613 tgt = lmv_find_target(lmv, fid);
1615 RETURN(PTR_ERR(tgt));
1617 rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1618 input_size, output_size, flags, request);
1623 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1624 struct obd_capa *oc, obd_valid valid, const char *name,
1625 const char *input, int input_size, int output_size,
1626 int flags, __u32 suppgid,
1627 struct ptlrpc_request **request)
1629 struct obd_device *obd = exp->exp_obd;
1630 struct lmv_obd *lmv = &obd->u.lmv;
1631 struct lmv_tgt_desc *tgt;
1635 rc = lmv_check_connect(obd);
1639 tgt = lmv_find_target(lmv, fid);
1641 RETURN(PTR_ERR(tgt));
1643 rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1644 input_size, output_size, flags, suppgid,
1650 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1651 struct ptlrpc_request **request)
1653 struct obd_device *obd = exp->exp_obd;
1654 struct lmv_obd *lmv = &obd->u.lmv;
1655 struct lmv_tgt_desc *tgt;
1659 rc = lmv_check_connect(obd);
1663 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1665 RETURN(PTR_ERR(tgt));
1667 if (op_data->op_flags & MF_GET_MDT_IDX) {
1668 op_data->op_mds = tgt->ltd_idx;
1672 rc = md_getattr(tgt->ltd_exp, op_data, request);
1677 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1679 struct obd_device *obd = exp->exp_obd;
1680 struct lmv_obd *lmv = &obd->u.lmv;
1685 rc = lmv_check_connect(obd);
1689 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1692 * With DNE every object can have two locks in different namespaces:
1693 * lookup lock in space of MDT storing direntry and update/open lock in
1694 * space of MDT storing inode.
1696 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1697 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1699 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1705 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1706 ldlm_iterator_t it, void *data)
1708 struct obd_device *obd = exp->exp_obd;
1709 struct lmv_obd *lmv = &obd->u.lmv;
1714 rc = lmv_check_connect(obd);
1718 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1721 * With DNE every object can have two locks in different namespaces:
1722 * lookup lock in space of MDT storing direntry and update/open lock in
1723 * space of MDT storing inode.
1725 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1726 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1728 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1737 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1738 struct md_open_data *mod, struct ptlrpc_request **request)
1740 struct obd_device *obd = exp->exp_obd;
1741 struct lmv_obd *lmv = &obd->u.lmv;
1742 struct lmv_tgt_desc *tgt;
1746 rc = lmv_check_connect(obd);
1750 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1752 RETURN(PTR_ERR(tgt));
1754 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1755 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1760 * Choosing the MDT by name or FID in @op_data.
1761 * For non-striped directory, it will locate MDT by fid.
1762 * For striped-directory, it will locate MDT by name. And also
1763 * it will reset op_fid1 with the FID of the choosen stripe.
1765 struct lmv_tgt_desc *
1766 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1767 const char *name, int namelen, struct lu_fid *fid,
1770 struct lmv_tgt_desc *tgt;
1771 const struct lmv_oinfo *oinfo;
1773 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1775 RETURN((void *)oinfo);
1776 *fid = oinfo->lmo_fid;
1777 *mds = oinfo->lmo_mds;
1778 tgt = lmv_get_target(lmv, *mds);
1780 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1785 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1788 struct lmv_stripe_md *lsm = op_data->op_mea1;
1789 struct lmv_tgt_desc *tgt;
1791 if (lsm == NULL || lsm->lsm_md_stripe_count <= 1 ||
1792 op_data->op_namelen == 0 ||
1793 lsm->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1794 tgt = lmv_find_target(lmv, fid);
1798 op_data->op_mds = tgt->ltd_idx;
1802 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1803 op_data->op_namelen, fid,
1807 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1808 const void *data, int datalen, int mode, __u32 uid,
1809 __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1810 struct ptlrpc_request **request)
1812 struct obd_device *obd = exp->exp_obd;
1813 struct lmv_obd *lmv = &obd->u.lmv;
1814 struct lmv_tgt_desc *tgt;
1818 rc = lmv_check_connect(obd);
1822 if (!lmv->desc.ld_active_tgt_count)
1825 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1827 RETURN(PTR_ERR(tgt));
1829 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1830 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1833 rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1837 /* Send the create request to the MDT where the object
1838 * will be located */
1839 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1841 RETURN(PTR_ERR(tgt));
1843 op_data->op_mds = tgt->ltd_idx;
1845 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1846 PFID(&op_data->op_fid2), op_data->op_mds);
1848 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1849 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1850 cap_effective, rdev, request);
1852 if (*request == NULL)
1854 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1859 static int lmv_done_writing(struct obd_export *exp,
1860 struct md_op_data *op_data,
1861 struct md_open_data *mod)
1863 struct obd_device *obd = exp->exp_obd;
1864 struct lmv_obd *lmv = &obd->u.lmv;
1865 struct lmv_tgt_desc *tgt;
1869 rc = lmv_check_connect(obd);
1873 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1875 RETURN(PTR_ERR(tgt));
1877 rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1882 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1883 struct lookup_intent *it, struct md_op_data *op_data,
1884 struct lustre_handle *lockh, void *lmm, int lmmsize,
1885 __u64 extra_lock_flags)
1887 struct ptlrpc_request *req = it->d.lustre.it_data;
1888 struct obd_device *obd = exp->exp_obd;
1889 struct lmv_obd *lmv = &obd->u.lmv;
1890 struct lustre_handle plock;
1891 struct lmv_tgt_desc *tgt;
1892 struct md_op_data *rdata;
1894 struct mdt_body *body;
1899 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1900 LASSERT(body != NULL);
1902 if (!(body->valid & OBD_MD_MDS))
1905 CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1906 LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1909 * We got LOOKUP lock, but we really need attrs.
1911 pmode = it->d.lustre.it_lock_mode;
1912 LASSERT(pmode != 0);
1913 memcpy(&plock, lockh, sizeof(plock));
1914 it->d.lustre.it_lock_mode = 0;
1915 it->d.lustre.it_data = NULL;
1918 ptlrpc_req_finished(req);
1920 tgt = lmv_find_target(lmv, &fid1);
1922 GOTO(out, rc = PTR_ERR(tgt));
1924 OBD_ALLOC_PTR(rdata);
1926 GOTO(out, rc = -ENOMEM);
1928 rdata->op_fid1 = fid1;
1929 rdata->op_bias = MDS_CROSS_REF;
1931 rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1932 lmm, lmmsize, NULL, extra_lock_flags);
1933 OBD_FREE_PTR(rdata);
1936 ldlm_lock_decref(&plock, pmode);
1941 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1942 struct lookup_intent *it, struct md_op_data *op_data,
1943 struct lustre_handle *lockh, void *lmm, int lmmsize,
1944 struct ptlrpc_request **req, __u64 extra_lock_flags)
1946 struct obd_device *obd = exp->exp_obd;
1947 struct lmv_obd *lmv = &obd->u.lmv;
1948 struct lmv_tgt_desc *tgt;
1952 rc = lmv_check_connect(obd);
1956 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1957 LL_IT2STR(it), PFID(&op_data->op_fid1));
1959 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1961 RETURN(PTR_ERR(tgt));
1963 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1964 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1966 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1967 lmm, lmmsize, req, extra_lock_flags);
1969 if (rc == 0 && it && it->it_op == IT_OPEN) {
1970 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1971 lmm, lmmsize, extra_lock_flags);
1977 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1978 struct ptlrpc_request **preq)
1980 struct ptlrpc_request *req = NULL;
1981 struct obd_device *obd = exp->exp_obd;
1982 struct lmv_obd *lmv = &obd->u.lmv;
1983 struct lmv_tgt_desc *tgt;
1984 struct mdt_body *body;
1988 rc = lmv_check_connect(obd);
1992 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1994 RETURN(PTR_ERR(tgt));
1996 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1997 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
2000 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2004 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2005 LASSERT(body != NULL);
2007 if (body->valid & OBD_MD_MDS) {
2008 struct lu_fid rid = body->fid1;
2009 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2012 tgt = lmv_find_target(lmv, &rid);
2014 ptlrpc_req_finished(*preq);
2016 RETURN(PTR_ERR(tgt));
2019 op_data->op_fid1 = rid;
2020 op_data->op_valid |= OBD_MD_FLCROSSREF;
2021 op_data->op_namelen = 0;
2022 op_data->op_name = NULL;
2023 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2024 ptlrpc_req_finished(*preq);
2031 #define md_op_data_fid(op_data, fl) \
2032 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2033 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2034 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2035 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2038 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2039 struct md_op_data *op_data,
2040 int op_tgt, ldlm_mode_t mode, int bits, int flag)
2042 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2043 struct obd_device *obd = exp->exp_obd;
2044 struct lmv_obd *lmv = &obd->u.lmv;
2045 ldlm_policy_data_t policy = {{ 0 }};
2049 if (!fid_is_sane(fid))
2053 tgt = lmv_find_target(lmv, fid);
2055 RETURN(PTR_ERR(tgt));
2058 if (tgt->ltd_idx != op_tgt) {
2059 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2060 policy.l_inodebits.bits = bits;
2061 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2062 mode, LCF_ASYNC, NULL);
2065 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2067 op_data->op_flags |= flag;
2075 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2078 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2079 struct ptlrpc_request **request)
2081 struct obd_device *obd = exp->exp_obd;
2082 struct lmv_obd *lmv = &obd->u.lmv;
2083 struct lmv_tgt_desc *tgt;
2087 rc = lmv_check_connect(obd);
2091 LASSERT(op_data->op_namelen != 0);
2093 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2094 PFID(&op_data->op_fid2), op_data->op_namelen,
2095 op_data->op_name, PFID(&op_data->op_fid1));
2097 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2098 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2099 op_data->op_cap = cfs_curproc_cap_pack();
2100 if (op_data->op_mea2 != NULL) {
2101 struct lmv_stripe_md *lsm = op_data->op_mea2;
2102 const struct lmv_oinfo *oinfo;
2104 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2105 op_data->op_namelen);
2107 RETURN(PTR_ERR(oinfo));
2109 op_data->op_fid2 = oinfo->lmo_fid;
2112 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2114 RETURN(PTR_ERR(tgt));
2117 * Cancel UPDATE lock on child (fid1).
2119 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2120 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2121 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2125 rc = md_link(tgt->ltd_exp, op_data, request);
2130 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2131 const char *old, int oldlen, const char *new, int newlen,
2132 struct ptlrpc_request **request)
2134 struct obd_device *obd = exp->exp_obd;
2135 struct lmv_obd *lmv = &obd->u.lmv;
2136 struct lmv_tgt_desc *src_tgt;
2140 LASSERT(oldlen != 0);
2142 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2143 oldlen, old, PFID(&op_data->op_fid1),
2144 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2145 newlen, new, PFID(&op_data->op_fid2),
2146 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2148 rc = lmv_check_connect(obd);
2152 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2153 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2154 op_data->op_cap = cfs_curproc_cap_pack();
2155 if (op_data->op_cli_flags & CLI_MIGRATE) {
2156 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2157 PFID(&op_data->op_fid3));
2158 rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
2161 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2163 if (op_data->op_mea1 != NULL) {
2164 struct lmv_stripe_md *lsm = op_data->op_mea1;
2166 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2170 if (IS_ERR(src_tgt))
2171 RETURN(PTR_ERR(src_tgt));
2173 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2174 if (IS_ERR(src_tgt))
2175 RETURN(PTR_ERR(src_tgt));
2177 op_data->op_mds = src_tgt->ltd_idx;
2180 if (op_data->op_mea2) {
2181 struct lmv_stripe_md *lsm = op_data->op_mea2;
2182 const struct lmv_oinfo *oinfo;
2184 oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2186 RETURN(PTR_ERR(oinfo));
2188 op_data->op_fid2 = oinfo->lmo_fid;
2191 if (IS_ERR(src_tgt))
2192 RETURN(PTR_ERR(src_tgt));
2195 * LOOKUP lock on src child (fid3) should also be cancelled for
2196 * src_tgt in mdc_rename.
2198 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2201 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2204 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2205 LCK_EX, MDS_INODELOCK_UPDATE,
2206 MF_MDC_CANCEL_FID2);
2211 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2213 if (fid_is_sane(&op_data->op_fid3)) {
2214 struct lmv_tgt_desc *tgt;
2216 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2218 RETURN(PTR_ERR(tgt));
2220 /* Cancel LOOKUP lock on its parent */
2221 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2222 LCK_EX, MDS_INODELOCK_LOOKUP,
2223 MF_MDC_CANCEL_FID3);
2227 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2228 LCK_EX, MDS_INODELOCK_FULL,
2229 MF_MDC_CANCEL_FID3);
2235 * Cancel all the locks on tgt child (fid4).
2237 if (fid_is_sane(&op_data->op_fid4))
2238 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2239 LCK_EX, MDS_INODELOCK_FULL,
2240 MF_MDC_CANCEL_FID4);
2242 CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2243 op_data->op_mds, PFID(&op_data->op_fid2));
2245 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2251 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2252 void *ea, int ealen, void *ea2, int ea2len,
2253 struct ptlrpc_request **request,
2254 struct md_open_data **mod)
2256 struct obd_device *obd = exp->exp_obd;
2257 struct lmv_obd *lmv = &obd->u.lmv;
2258 struct lmv_tgt_desc *tgt;
2262 rc = lmv_check_connect(obd);
2266 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2267 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2269 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2270 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2272 RETURN(PTR_ERR(tgt));
2274 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2275 ea2len, request, mod);
2280 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2281 struct obd_capa *oc, struct ptlrpc_request **request)
2283 struct obd_device *obd = exp->exp_obd;
2284 struct lmv_obd *lmv = &obd->u.lmv;
2285 struct lmv_tgt_desc *tgt;
2289 rc = lmv_check_connect(obd);
2293 tgt = lmv_find_target(lmv, fid);
2295 RETURN(PTR_ERR(tgt));
2297 rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2302 * Adjust a set of pages, each page containing an array of lu_dirpages,
2303 * so that each page can be used as a single logical lu_dirpage.
2305 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2306 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2307 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
2308 * value is used as a cookie to request the next lu_dirpage in a
2309 * directory listing that spans multiple pages (two in this example):
2312 * .|--------v------- -----.
2313 * |s|e|f|p|ent|ent| ... |ent|
2314 * '--|-------------- -----' Each CFS_PAGE contains a single
2315 * '------. lu_dirpage.
2316 * .---------v------- -----.
2317 * |s|e|f|p|ent| 0 | ... | 0 |
2318 * '----------------- -----'
2320 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2321 * larger than LU_PAGE_SIZE, a single host page may contain multiple
2322 * lu_dirpages. After reading the lu_dirpages from the MDS, the
2323 * ldp_hash_end of the first lu_dirpage refers to the one immediately
2324 * after it in the same CFS_PAGE (arrows simplified for brevity, but
2325 * in general e0==s1, e1==s2, etc.):
2327 * .-------------------- -----.
2328 * |s0|e0|f0|p|ent|ent| ... |ent|
2329 * |---v---------------- -----|
2330 * |s1|e1|f1|p|ent|ent| ... |ent|
2331 * |---v---------------- -----| Here, each CFS_PAGE contains
2332 * ... multiple lu_dirpages.
2333 * |---v---------------- -----|
2334 * |s'|e'|f'|p|ent|ent| ... |ent|
2335 * '---|---------------- -----'
2337 * .----------------------------.
2340 * This structure is transformed into a single logical lu_dirpage as follows:
2342 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2343 * labeled 'next CFS_PAGE'.
2345 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2346 * a hash collision with the next page exists.
2348 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2349 * to the first entry of the next lu_dirpage.
2351 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2352 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2356 for (i = 0; i < ncfspgs; i++) {
2357 struct lu_dirpage *dp = kmap(pages[i]);
2358 struct lu_dirpage *first = dp;
2359 struct lu_dirent *end_dirent = NULL;
2360 struct lu_dirent *ent;
2361 __u64 hash_end = dp->ldp_hash_end;
2362 __u32 flags = dp->ldp_flags;
2364 while (--nlupgs > 0) {
2365 ent = lu_dirent_start(dp);
2366 for (end_dirent = ent; ent != NULL;
2367 end_dirent = ent, ent = lu_dirent_next(ent));
2369 /* Advance dp to next lu_dirpage. */
2370 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2372 /* Check if we've reached the end of the CFS_PAGE. */
2373 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2376 /* Save the hash and flags of this lu_dirpage. */
2377 hash_end = dp->ldp_hash_end;
2378 flags = dp->ldp_flags;
2380 /* Check if lu_dirpage contains no entries. */
2384 /* Enlarge the end entry lde_reclen from 0 to
2385 * first entry of next lu_dirpage. */
2386 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2387 end_dirent->lde_reclen =
2388 cpu_to_le16((char *)(dp->ldp_entries) -
2389 (char *)end_dirent);
2392 first->ldp_hash_end = hash_end;
2393 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2394 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2398 LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2401 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2402 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2404 #define NORMAL_MAX_STRIPES 4
2405 int lmv_read_entry(struct obd_export *exp, struct md_op_data *op_data,
2406 struct md_callback *cb_op, struct lu_dirent **ldp,
2407 struct page **ppage)
2409 struct obd_device *obd = exp->exp_obd;
2410 struct lmv_obd *lmv = &obd->u.lmv;
2411 struct lmv_stripe_md *lsm = op_data->op_mea1;
2412 struct lu_dirent *tmp_ents[NORMAL_MAX_STRIPES];
2413 struct lu_dirent **ents = NULL;
2417 struct page *min_page = NULL;
2422 rc = lmv_check_connect(obd);
2429 stripe_count = lsm->lsm_md_stripe_count;
2431 if (stripe_count > NORMAL_MAX_STRIPES) {
2432 OBD_ALLOC(ents, sizeof(ents[0]) * stripe_count);
2434 GOTO(out, rc = -ENOMEM);
2437 memset(ents, 0, sizeof(ents[0]) * stripe_count);
2440 min_hash = MDS_DIR_END_OFF;
2441 for (i = 0; i < stripe_count; i++) {
2442 struct lmv_tgt_desc *tgt;
2443 struct page *page = NULL;
2445 if (likely(lsm == NULL)) {
2446 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2448 GOTO(out, rc = PTR_ERR(tgt));
2449 LASSERT(op_data->op_data != NULL);
2451 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds);
2453 GOTO(out, rc = PTR_ERR(tgt));
2454 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2455 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2456 op_data->op_stripe_offset = i;
2459 rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, &ents[i],
2464 if (ents[i] != NULL &&
2465 le64_to_cpu(ents[i]->lde_hash) <= min_hash) {
2466 if (min_page != NULL)
2467 page_cache_release(min_page);
2469 min_hash = le64_to_cpu(ents[i]->lde_hash);
2474 if (min_hash != MDS_DIR_END_OFF)
2475 *ldp = ents[min_idx];
2479 if (stripe_count > NORMAL_MAX_STRIPES && ents != NULL)
2480 OBD_FREE(ents, sizeof(ents[0]) * stripe_count);
2482 if (rc != 0 && min_page != NULL) {
2484 page_cache_release(min_page);
2492 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2493 struct ptlrpc_request **request)
2495 struct obd_device *obd = exp->exp_obd;
2496 struct lmv_obd *lmv = &obd->u.lmv;
2497 struct lmv_tgt_desc *tgt = NULL;
2498 struct lmv_tgt_desc *parent_tgt = NULL;
2499 struct mdt_body *body;
2503 rc = lmv_check_connect(obd);
2507 /* Send unlink requests to the MDT where the child is located */
2508 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2509 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2511 RETURN(PTR_ERR(tgt));
2513 /* For striped dir, we need to locate the parent as well */
2514 if (op_data->op_mea1 != NULL &&
2515 op_data->op_mea1->lsm_md_stripe_count > 1) {
2516 struct lmv_tgt_desc *tmp;
2518 LASSERT(op_data->op_name != NULL &&
2519 op_data->op_namelen != 0);
2520 tmp = lmv_locate_target_for_name(lmv,
2523 op_data->op_namelen,
2527 RETURN(PTR_ERR(tmp));
2530 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2532 RETURN(PTR_ERR(tgt));
2535 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2536 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2537 op_data->op_cap = cfs_curproc_cap_pack();
2540 * If child's fid is given, cancel unused locks for it if it is from
2541 * another export than parent.
2543 * LOOKUP lock for child (fid3) should also be cancelled on parent
2544 * tgt_tgt in mdc_unlink().
2546 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2549 * Cancel FULL locks on child (fid3).
2551 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2552 if (IS_ERR(parent_tgt))
2553 RETURN(PTR_ERR(parent_tgt));
2555 if (parent_tgt != tgt) {
2556 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2557 LCK_EX, MDS_INODELOCK_LOOKUP,
2558 MF_MDC_CANCEL_FID3);
2561 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2562 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2566 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2567 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2569 rc = md_unlink(tgt->ltd_exp, op_data, request);
2570 if (rc != 0 && rc != -EREMOTE)
2573 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2577 /* Not cross-ref case, just get out of here. */
2578 if (likely(!(body->valid & OBD_MD_MDS)))
2581 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2582 exp->exp_obd->obd_name, PFID(&body->fid1));
2584 /* This is a remote object, try remote MDT, Note: it may
2585 * try more than 1 time here, Considering following case
2586 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2587 * 1. Initially A does not know where remote1 is, it send
2588 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2589 * resend unlink RPC to MDT1 (retry 1st time).
2591 * 2. During the unlink RPC in flight,
2592 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2593 * and create new remote1, but on MDT0
2595 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2596 * /mnt/lustre, then lookup get fid of remote1, and find
2597 * it is remote dir again, and replay -EREMOTE again.
2599 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2601 * In theory, it might try unlimited time here, but it should
2602 * be very rare case. */
2603 op_data->op_fid2 = body->fid1;
2604 ptlrpc_req_finished(*request);
2610 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2612 struct lmv_obd *lmv = &obd->u.lmv;
2616 case OBD_CLEANUP_EARLY:
2617 /* XXX: here should be calling obd_precleanup() down to
2620 case OBD_CLEANUP_EXPORTS:
2621 fld_client_proc_fini(&lmv->lmv_fld);
2622 lprocfs_obd_cleanup(obd);
2623 lprocfs_free_md_stats(obd);
2631 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2632 __u32 keylen, void *key, __u32 *vallen, void *val,
2633 struct lov_stripe_md *lsm)
2635 struct obd_device *obd;
2636 struct lmv_obd *lmv;
2640 obd = class_exp2obd(exp);
2642 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2643 exp->exp_handle.h_cookie);
2648 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2651 rc = lmv_check_connect(obd);
2655 LASSERT(*vallen == sizeof(__u32));
2656 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2657 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2659 * All tgts should be connected when this gets called.
2661 if (tgt == NULL || tgt->ltd_exp == NULL)
2664 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2669 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2670 KEY_IS(KEY_DEFAULT_EASIZE) ||
2671 KEY_IS(KEY_MAX_COOKIESIZE) ||
2672 KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2673 KEY_IS(KEY_CONN_DATA)) {
2674 rc = lmv_check_connect(obd);
2679 * Forwarding this request to first MDS, it should know LOV
2682 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2684 if (!rc && KEY_IS(KEY_CONN_DATA))
2685 exp->exp_connect_data = *(struct obd_connect_data *)val;
2687 } else if (KEY_IS(KEY_TGT_COUNT)) {
2688 *((int *)val) = lmv->desc.ld_tgt_count;
2692 CDEBUG(D_IOCTL, "Invalid key\n");
2696 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2697 obd_count keylen, void *key, obd_count vallen,
2698 void *val, struct ptlrpc_request_set *set)
2700 struct lmv_tgt_desc *tgt = NULL;
2701 struct obd_device *obd;
2702 struct lmv_obd *lmv;
2706 obd = class_exp2obd(exp);
2708 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2709 exp->exp_handle.h_cookie);
2714 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2717 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2720 if (tgt == NULL || tgt->ltd_exp == NULL)
2723 err = obd_set_info_async(env, tgt->ltd_exp,
2724 keylen, key, vallen, val, set);
2735 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2736 struct lmv_mds_md_v1 *lmm1)
2741 lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2742 lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2743 lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2744 lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2745 cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2746 sizeof(lmm1->lmv_pool_name));
2747 if (cplen >= sizeof(lmm1->lmv_pool_name))
2750 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2751 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2752 &lsm->lsm_md_oinfo[i].lmo_fid);
2756 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2760 bool allocated = false;
2764 LASSERT(lmmp != NULL);
2766 if (*lmmp != NULL && lsm == NULL) {
2769 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2770 lmm_size = lmv_mds_md_size(stripe_count,
2771 le32_to_cpu((*lmmp)->lmv_magic));
2774 OBD_FREE(*lmmp, lmm_size);
2780 if (*lmmp == NULL && lsm == NULL) {
2781 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2782 LASSERT(lmm_size > 0);
2783 OBD_ALLOC(*lmmp, lmm_size);
2786 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2787 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2792 LASSERT(lsm != NULL);
2793 lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2794 if (*lmmp == NULL) {
2795 OBD_ALLOC(*lmmp, lmm_size);
2801 switch (lsm->lsm_md_magic) {
2803 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2810 if (rc != 0 && allocated) {
2811 OBD_FREE(*lmmp, lmm_size);
2817 EXPORT_SYMBOL(lmv_pack_md);
2819 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2820 const struct lmv_mds_md_v1 *lmm1)
2822 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2829 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2830 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2831 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2832 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2833 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2834 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2835 sizeof(lsm->lsm_md_pool_name));
2837 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2840 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2841 "layout_version %d\n", lsm->lsm_md_stripe_count,
2842 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2843 lsm->lsm_md_layout_version);
2845 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2846 for (i = 0; i < le32_to_cpu(stripe_count); i++) {
2847 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2848 &lmm1->lmv_stripe_fids[i]);
2849 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2850 &lsm->lsm_md_oinfo[i].lmo_mds);
2853 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2854 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2860 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2861 const union lmv_mds_md *lmm, int stripe_count)
2863 struct lmv_stripe_md *lsm;
2866 bool allocated = false;
2869 LASSERT(lsmp != NULL);
2873 if (lsm != NULL && lmm == NULL) {
2876 for (i = 1; i < lsm->lsm_md_stripe_count; i++) {
2877 if (lsm->lsm_md_oinfo[i].lmo_root != NULL)
2878 iput(lsm->lsm_md_oinfo[i].lmo_root);
2881 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2882 OBD_FREE(lsm, lsm_size);
2888 if (lsm == NULL && lmm == NULL) {
2889 lsm_size = lmv_stripe_md_size(stripe_count);
2890 OBD_ALLOC(lsm, lsm_size);
2893 lsm->lsm_md_stripe_count = stripe_count;
2899 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2900 le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_MIGRATE &&
2901 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2902 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2903 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2908 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1 ||
2909 le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_MIGRATE)
2910 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2913 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2914 * stripecount should be 0 then.
2916 lsm_size = lmv_stripe_md_size(0);
2918 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2920 OBD_ALLOC(lsm, lsm_size);
2927 switch (le32_to_cpu(lmm->lmv_magic)) {
2929 case LMV_MAGIC_MIGRATE:
2930 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2933 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2934 le32_to_cpu(lmm->lmv_magic));
2939 if (rc != 0 && allocated) {
2940 OBD_FREE(lsm, lsm_size);
2947 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
2949 return lmv_unpack_md(NULL, lsmp, NULL, stripes);
2951 EXPORT_SYMBOL(lmv_alloc_memmd);
2953 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2955 lmv_unpack_md(NULL, &lsm, NULL, 0);
2957 EXPORT_SYMBOL(lmv_free_memmd);
2959 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2960 struct lov_mds_md *lmm, int disk_len)
2962 return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
2963 (union lmv_mds_md *)lmm, disk_len);
2966 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2967 struct lov_stripe_md *lsm)
2969 struct obd_device *obd = exp->exp_obd;
2970 struct lmv_obd *lmv_obd = &obd->u.lmv;
2971 const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm;
2976 stripe_count = lmv->lsm_md_stripe_count;
2978 stripe_count = lmv_obd->desc.ld_tgt_count;
2980 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
2983 return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
2986 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2987 ldlm_policy_data_t *policy, ldlm_mode_t mode,
2988 ldlm_cancel_flags_t flags, void *opaque)
2990 struct obd_device *obd = exp->exp_obd;
2991 struct lmv_obd *lmv = &obd->u.lmv;
2997 LASSERT(fid != NULL);
2999 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3000 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3002 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3005 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3013 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3016 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3017 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3021 if (tgt == NULL || tgt->ltd_exp == NULL)
3023 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3027 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3028 const struct lu_fid *fid, ldlm_type_t type,
3029 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3030 struct lustre_handle *lockh)
3032 struct obd_device *obd = exp->exp_obd;
3033 struct lmv_obd *lmv = &obd->u.lmv;
3038 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3041 * With CMD every object can have two locks in different namespaces:
3042 * lookup lock in space of mds storing direntry and update/open lock in
3043 * space of mds storing inode. Thus we check all targets, not only that
3044 * one fid was created in.
3046 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3047 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3049 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3052 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3061 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3062 struct obd_export *dt_exp, struct obd_export *md_exp,
3063 struct lustre_md *md)
3065 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3066 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3068 if (tgt == NULL || tgt->ltd_exp == NULL)
3071 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3074 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3076 struct obd_device *obd = exp->exp_obd;
3077 struct lmv_obd *lmv = &obd->u.lmv;
3078 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3081 if (md->lmv != NULL) {
3082 lmv_free_memmd(md->lmv);
3085 if (tgt == NULL || tgt->ltd_exp == NULL)
3087 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3090 int lmv_set_open_replay_data(struct obd_export *exp,
3091 struct obd_client_handle *och,
3092 struct lookup_intent *it)
3094 struct obd_device *obd = exp->exp_obd;
3095 struct lmv_obd *lmv = &obd->u.lmv;
3096 struct lmv_tgt_desc *tgt;
3099 tgt = lmv_find_target(lmv, &och->och_fid);
3101 RETURN(PTR_ERR(tgt));
3103 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3106 int lmv_clear_open_replay_data(struct obd_export *exp,
3107 struct obd_client_handle *och)
3109 struct obd_device *obd = exp->exp_obd;
3110 struct lmv_obd *lmv = &obd->u.lmv;
3111 struct lmv_tgt_desc *tgt;
3114 tgt = lmv_find_target(lmv, &och->och_fid);
3116 RETURN(PTR_ERR(tgt));
3118 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3121 static int lmv_get_remote_perm(struct obd_export *exp,
3122 const struct lu_fid *fid,
3123 struct obd_capa *oc, __u32 suppgid,
3124 struct ptlrpc_request **request)
3126 struct obd_device *obd = exp->exp_obd;
3127 struct lmv_obd *lmv = &obd->u.lmv;
3128 struct lmv_tgt_desc *tgt;
3132 rc = lmv_check_connect(obd);
3136 tgt = lmv_find_target(lmv, fid);
3138 RETURN(PTR_ERR(tgt));
3140 rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3144 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3147 struct obd_device *obd = exp->exp_obd;
3148 struct lmv_obd *lmv = &obd->u.lmv;
3149 struct lmv_tgt_desc *tgt;
3153 rc = lmv_check_connect(obd);
3157 tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3159 RETURN(PTR_ERR(tgt));
3161 rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3165 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3166 const struct req_msg_field *field, struct obd_capa **oc)
3168 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3169 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3171 if (tgt == NULL || tgt->ltd_exp == NULL)
3173 return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3176 int lmv_intent_getattr_async(struct obd_export *exp,
3177 struct md_enqueue_info *minfo,
3178 struct ldlm_enqueue_info *einfo)
3180 struct md_op_data *op_data = &minfo->mi_data;
3181 struct obd_device *obd = exp->exp_obd;
3182 struct lmv_obd *lmv = &obd->u.lmv;
3183 struct lmv_tgt_desc *tgt = NULL;
3187 rc = lmv_check_connect(obd);
3191 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3193 RETURN(PTR_ERR(tgt));
3195 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3199 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3200 struct lu_fid *fid, __u64 *bits)
3202 struct obd_device *obd = exp->exp_obd;
3203 struct lmv_obd *lmv = &obd->u.lmv;
3204 struct lmv_tgt_desc *tgt;
3208 rc = lmv_check_connect(obd);
3212 tgt = lmv_find_target(lmv, fid);
3214 RETURN(PTR_ERR(tgt));
3216 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3221 * For lmv, only need to send request to master MDT, and the master MDT will
3222 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3223 * we directly fetch data from the slave MDTs.
3225 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3226 struct obd_quotactl *oqctl)
3228 struct obd_device *obd = class_exp2obd(exp);
3229 struct lmv_obd *lmv = &obd->u.lmv;
3230 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3233 __u64 curspace, curinodes;
3237 tgt->ltd_exp == NULL ||
3239 lmv->desc.ld_tgt_count == 0) {
3240 CERROR("master lmv inactive\n");
3244 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3245 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3249 curspace = curinodes = 0;
3250 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3254 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3257 err = obd_quotactl(tgt->ltd_exp, oqctl);
3259 CERROR("getquota on mdt %d failed. %d\n", i, err);
3263 curspace += oqctl->qc_dqblk.dqb_curspace;
3264 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3267 oqctl->qc_dqblk.dqb_curspace = curspace;
3268 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3273 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3274 struct obd_quotactl *oqctl)
3276 struct obd_device *obd = class_exp2obd(exp);
3277 struct lmv_obd *lmv = &obd->u.lmv;
3278 struct lmv_tgt_desc *tgt;
3283 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3286 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3287 CERROR("lmv idx %d inactive\n", i);
3291 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3299 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3300 struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3302 if (lsm->lsm_md_stripe_count <= 1)
3305 return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3308 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3309 struct cl_attr *attr)
3314 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3315 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3317 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3318 "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3319 i_size_read(inode), inode->i_nlink,
3320 LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3321 LTIME_S(inode->i_mtime));
3323 /* for slave stripe, it needs to subtract nlink for . and .. */
3325 attr->cat_nlink += inode->i_nlink - 2;
3327 attr->cat_nlink = inode->i_nlink;
3329 attr->cat_size += i_size_read(inode);
3331 if (attr->cat_atime < LTIME_S(inode->i_atime))
3332 attr->cat_atime = LTIME_S(inode->i_atime);
3334 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3335 attr->cat_ctime = LTIME_S(inode->i_ctime);
3337 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3338 attr->cat_mtime = LTIME_S(inode->i_mtime);
3344 struct obd_ops lmv_obd_ops = {
3345 .o_owner = THIS_MODULE,
3346 .o_setup = lmv_setup,
3347 .o_cleanup = lmv_cleanup,
3348 .o_precleanup = lmv_precleanup,
3349 .o_process_config = lmv_process_config,
3350 .o_connect = lmv_connect,
3351 .o_disconnect = lmv_disconnect,
3352 .o_statfs = lmv_statfs,
3353 .o_get_info = lmv_get_info,
3354 .o_set_info_async = lmv_set_info_async,
3355 .o_packmd = lmv_packmd,
3356 .o_unpackmd = lmv_unpackmd,
3357 .o_notify = lmv_notify,
3358 .o_get_uuid = lmv_get_uuid,
3359 .o_iocontrol = lmv_iocontrol,
3360 .o_quotacheck = lmv_quotacheck,
3361 .o_quotactl = lmv_quotactl
3364 struct md_ops lmv_md_ops = {
3365 .m_getstatus = lmv_getstatus,
3366 .m_null_inode = lmv_null_inode,
3367 .m_find_cbdata = lmv_find_cbdata,
3368 .m_close = lmv_close,
3369 .m_create = lmv_create,
3370 .m_done_writing = lmv_done_writing,
3371 .m_enqueue = lmv_enqueue,
3372 .m_getattr = lmv_getattr,
3373 .m_getxattr = lmv_getxattr,
3374 .m_getattr_name = lmv_getattr_name,
3375 .m_intent_lock = lmv_intent_lock,
3377 .m_rename = lmv_rename,
3378 .m_setattr = lmv_setattr,
3379 .m_setxattr = lmv_setxattr,
3380 .m_fsync = lmv_fsync,
3381 .m_read_entry = lmv_read_entry,
3382 .m_unlink = lmv_unlink,
3383 .m_init_ea_size = lmv_init_ea_size,
3384 .m_cancel_unused = lmv_cancel_unused,
3385 .m_set_lock_data = lmv_set_lock_data,
3386 .m_lock_match = lmv_lock_match,
3387 .m_get_lustre_md = lmv_get_lustre_md,
3388 .m_free_lustre_md = lmv_free_lustre_md,
3389 .m_update_lsm_md = lmv_update_lsm_md,
3390 .m_merge_attr = lmv_merge_attr,
3391 .m_set_open_replay_data = lmv_set_open_replay_data,
3392 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3393 .m_renew_capa = lmv_renew_capa,
3394 .m_unpack_capa = lmv_unpack_capa,
3395 .m_get_remote_perm = lmv_get_remote_perm,
3396 .m_intent_getattr_async = lmv_intent_getattr_async,
3397 .m_revalidate_lock = lmv_revalidate_lock
3400 int __init lmv_init(void)
3402 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3403 #ifndef HAVE_ONLY_PROCFS_SEQ
3406 LUSTRE_LMV_NAME, NULL);
3410 static void lmv_exit(void)
3412 class_unregister_type(LUSTRE_LMV_NAME);
3415 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3416 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3417 MODULE_LICENSE("GPL");
3419 module_init(lmv_init);
3420 module_exit(lmv_exit);