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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LMV
35 #include <linux/file.h>
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/user_namespace.h>
39 #ifdef HAVE_UIDGID_HEADER
40 # include <linux/uidgid.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 <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>
56 #include <lustre_fid.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #include <lustre_kernelcomm.h>
59 #include "lmv_internal.h"
61 static int lmv_check_connect(struct obd_device *obd);
63 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
66 if (tgt->ltd_active == activate)
69 tgt->ltd_active = activate;
70 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
72 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
78 * -EINVAL : UUID can't be found in the LMV's target list
79 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
80 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
82 static int lmv_set_mdc_active(struct lmv_obd *lmv,
83 const struct obd_uuid *uuid,
86 struct lmv_tgt_desc *tgt = NULL;
87 struct obd_device *obd;
92 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
93 lmv, uuid->uuid, activate);
95 spin_lock(&lmv->lmv_lock);
96 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
98 if (tgt == NULL || tgt->ltd_exp == NULL)
101 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n", i,
102 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
104 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
108 if (i == lmv->desc.ld_tgt_count)
109 GOTO(out_lmv_lock, rc = -EINVAL);
111 obd = class_exp2obd(tgt->ltd_exp);
113 GOTO(out_lmv_lock, rc = -ENOTCONN);
115 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
116 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
117 obd->obd_type->typ_name, i);
118 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
120 if (tgt->ltd_active == activate) {
121 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
122 activate ? "" : "in");
123 GOTO(out_lmv_lock, rc);
126 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
127 activate ? "" : "in");
128 lmv_activate_target(lmv, tgt, activate);
132 spin_unlock(&lmv->lmv_lock);
136 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
138 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
139 struct lmv_tgt_desc *tgt = lmv->tgts[0];
141 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
144 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
145 enum obd_notify_event ev)
147 struct obd_connect_data *conn_data;
148 struct lmv_obd *lmv = &obd->u.lmv;
149 struct obd_uuid *uuid;
153 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
154 CERROR("unexpected notification of %s %s!\n",
155 watched->obd_type->typ_name,
160 uuid = &watched->u.cli.cl_target_uuid;
161 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
163 * Set MDC as active before notifying the observer, so the
164 * observer can use the MDC normally.
166 rc = lmv_set_mdc_active(lmv, uuid,
167 ev == OBD_NOTIFY_ACTIVE);
169 CERROR("%sactivation of %s failed: %d\n",
170 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
174 } else if (ev == OBD_NOTIFY_OCD) {
175 conn_data = &watched->u.cli.cl_import->imp_connect_data;
177 * XXX: Make sure that ocd_connect_flags from all targets are
178 * the same. Otherwise one of MDTs runs wrong version or
179 * something like this. --umka
181 obd->obd_self_export->exp_connect_data = *conn_data;
185 * Pass the notification up the chain.
187 if (obd->obd_observer)
188 rc = obd_notify(obd->obd_observer, watched, ev);
193 static int lmv_connect(const struct lu_env *env,
194 struct obd_export **pexp, struct obd_device *obd,
195 struct obd_uuid *cluuid, struct obd_connect_data *data,
198 struct lmv_obd *lmv = &obd->u.lmv;
199 struct lustre_handle conn = { 0 };
200 struct obd_export *exp;
204 rc = class_connect(&conn, obd, cluuid);
206 CERROR("class_connection() returned %d\n", rc);
210 exp = class_conn2export(&conn);
213 lmv->conn_data = *data;
214 lmv->lmv_cache = localdata;
216 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
217 &obd->obd_kset.kobj);
218 if (!lmv->lmv_tgts_kobj) {
219 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
220 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
223 rc = lmv_check_connect(obd);
232 if (lmv->lmv_tgts_kobj)
233 kobject_put(lmv->lmv_tgts_kobj);
235 class_disconnect(exp);
240 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
243 struct obd_device *obd = exp->exp_obd;
244 struct lmv_obd *lmv = &obd->u.lmv;
250 if (lmv->max_easize < easize) {
251 lmv->max_easize = easize;
254 if (lmv->max_def_easize < def_easize) {
255 lmv->max_def_easize = def_easize;
262 if (lmv->connected == 0)
265 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
266 struct lmv_tgt_desc *tgt = lmv->tgts[i];
268 if (tgt == NULL || tgt->ltd_exp == NULL) {
269 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
272 if (!tgt->ltd_active)
275 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
277 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
278 " rc = %d\n", obd->obd_name, i, rc);
285 #define MAX_STRING_SIZE 128
287 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
289 struct lmv_obd *lmv = &obd->u.lmv;
290 struct obd_device *mdc_obd;
291 struct obd_export *mdc_exp;
292 struct lu_fld_target target;
296 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
299 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
303 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
304 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
305 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
307 if (!mdc_obd->obd_set_up) {
308 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
312 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
313 &lmv->conn_data, lmv->lmv_cache);
315 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
320 * Init fid sequence client for this mdc and add new fld target.
322 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
326 target.ft_srv = NULL;
327 target.ft_exp = mdc_exp;
328 target.ft_idx = tgt->ltd_index;
330 fld_client_add_target(&lmv->lmv_fld, &target);
332 rc = obd_register_observer(mdc_obd, obd);
334 obd_disconnect(mdc_exp);
335 CERROR("target %s register_observer error %d\n",
336 tgt->ltd_uuid.uuid, rc);
340 if (obd->obd_observer) {
342 * Tell the observer about the new target.
344 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
347 obd_disconnect(mdc_exp);
353 tgt->ltd_exp = mdc_exp;
354 lmv->desc.ld_active_tgt_count++;
356 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
358 rc = lqos_add_tgt(&lmv->lmv_qos, tgt);
360 obd_disconnect(mdc_exp);
364 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
365 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
366 atomic_read(&obd->obd_refcount));
368 lmv_statfs_check_update(obd, tgt);
370 if (lmv->lmv_tgts_kobj)
371 /* Even if we failed to create the link, that's fine */
372 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
373 &mdc_obd->obd_kset.kobj,
378 static void lmv_del_target(struct lmv_obd *lmv, int index)
380 if (lmv->tgts[index] == NULL)
383 lqos_del_tgt(&lmv->lmv_qos, lmv->tgts[index]);
385 OBD_FREE_PTR(lmv->tgts[index]);
386 lmv->tgts[index] = NULL;
390 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
391 __u32 index, int gen)
393 struct obd_device *mdc_obd;
394 struct lmv_obd *lmv = &obd->u.lmv;
395 struct lmv_tgt_desc *tgt;
396 int orig_tgt_count = 0;
401 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
402 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
405 CERROR("%s: Target %s not attached: rc = %d\n",
406 obd->obd_name, uuidp->uuid, -EINVAL);
410 mutex_lock(&lmv->lmv_init_mutex);
411 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
412 tgt = lmv->tgts[index];
413 CERROR("%s: UUID %s already assigned at LMV target index %d:"
414 " rc = %d\n", obd->obd_name,
415 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
416 mutex_unlock(&lmv->lmv_init_mutex);
420 if (index >= lmv->tgts_size) {
421 /* We need to reallocate the lmv target array. */
422 struct lmv_tgt_desc **newtgts, **old = NULL;
426 while (newsize < index + 1)
427 newsize = newsize << 1;
428 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
429 if (newtgts == NULL) {
430 mutex_unlock(&lmv->lmv_init_mutex);
434 if (lmv->tgts_size) {
435 memcpy(newtgts, lmv->tgts,
436 sizeof(*newtgts) * lmv->tgts_size);
438 oldsize = lmv->tgts_size;
442 lmv->tgts_size = newsize;
445 OBD_FREE(old, sizeof(*old) * oldsize);
447 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
453 mutex_unlock(&lmv->lmv_init_mutex);
457 mutex_init(&tgt->ltd_fid_mutex);
458 tgt->ltd_index = index;
459 tgt->ltd_uuid = *uuidp;
461 lmv->tgts[index] = tgt;
462 if (index >= lmv->desc.ld_tgt_count) {
463 orig_tgt_count = lmv->desc.ld_tgt_count;
464 lmv->desc.ld_tgt_count = index + 1;
467 if (lmv->connected == 0) {
468 /* lmv_check_connect() will connect this target. */
469 mutex_unlock(&lmv->lmv_init_mutex);
473 /* Otherwise let's connect it ourselves */
474 mutex_unlock(&lmv->lmv_init_mutex);
475 rc = lmv_connect_mdc(obd, tgt);
477 spin_lock(&lmv->lmv_lock);
478 if (lmv->desc.ld_tgt_count == index + 1)
479 lmv->desc.ld_tgt_count = orig_tgt_count;
480 memset(tgt, 0, sizeof(*tgt));
481 spin_unlock(&lmv->lmv_lock);
483 int easize = sizeof(struct lmv_stripe_md) +
484 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
485 lmv_init_ea_size(obd->obd_self_export, easize, 0);
491 static int lmv_check_connect(struct obd_device *obd)
493 struct lmv_obd *lmv = &obd->u.lmv;
494 struct lmv_tgt_desc *tgt;
503 mutex_lock(&lmv->lmv_init_mutex);
504 if (lmv->connected) {
505 mutex_unlock(&lmv->lmv_init_mutex);
509 if (lmv->desc.ld_tgt_count == 0) {
510 mutex_unlock(&lmv->lmv_init_mutex);
511 CERROR("%s: no targets configured.\n", obd->obd_name);
515 LASSERT(lmv->tgts != NULL);
517 if (lmv->tgts[0] == NULL) {
518 mutex_unlock(&lmv->lmv_init_mutex);
519 CERROR("%s: no target configured for index 0.\n",
524 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
525 obd->obd_uuid.uuid, obd->obd_name);
527 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
531 rc = lmv_connect_mdc(obd, tgt);
537 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
538 lmv_init_ea_size(obd->obd_self_export, easize, 0);
539 mutex_unlock(&lmv->lmv_init_mutex);
550 --lmv->desc.ld_active_tgt_count;
551 rc2 = obd_disconnect(tgt->ltd_exp);
553 CERROR("LMV target %s disconnect on "
554 "MDC idx %d: error %d\n",
555 tgt->ltd_uuid.uuid, i, rc2);
560 mutex_unlock(&lmv->lmv_init_mutex);
565 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
567 struct lmv_obd *lmv = &obd->u.lmv;
568 struct obd_device *mdc_obd;
572 LASSERT(tgt != NULL);
573 LASSERT(obd != NULL);
575 mdc_obd = class_exp2obd(tgt->ltd_exp);
578 mdc_obd->obd_force = obd->obd_force;
579 mdc_obd->obd_fail = obd->obd_fail;
580 mdc_obd->obd_no_recov = obd->obd_no_recov;
582 if (lmv->lmv_tgts_kobj)
583 sysfs_remove_link(lmv->lmv_tgts_kobj,
587 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
589 CERROR("Can't finanize fids factory\n");
591 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
592 tgt->ltd_exp->exp_obd->obd_name,
593 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
595 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
596 rc = obd_disconnect(tgt->ltd_exp);
598 if (tgt->ltd_active) {
599 CERROR("Target %s disconnect error %d\n",
600 tgt->ltd_uuid.uuid, rc);
604 lmv_activate_target(lmv, tgt, 0);
609 static int lmv_disconnect(struct obd_export *exp)
611 struct obd_device *obd = class_exp2obd(exp);
612 struct lmv_obd *lmv = &obd->u.lmv;
620 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
621 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
624 lmv_disconnect_mdc(obd, lmv->tgts[i]);
627 if (lmv->lmv_tgts_kobj)
628 kobject_put(lmv->lmv_tgts_kobj);
632 * This is the case when no real connection is established by
633 * lmv_check_connect().
636 class_export_put(exp);
637 rc = class_disconnect(exp);
643 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
646 struct obd_device *obddev = class_exp2obd(exp);
647 struct lmv_obd *lmv = &obddev->u.lmv;
648 struct getinfo_fid2path *gf;
649 struct lmv_tgt_desc *tgt;
650 struct getinfo_fid2path *remote_gf = NULL;
651 struct lu_fid root_fid;
652 int remote_gf_size = 0;
656 tgt = lmv_find_target(lmv, &gf->gf_fid);
658 RETURN(PTR_ERR(tgt));
660 root_fid = *gf->gf_u.gf_root_fid;
661 LASSERT(fid_is_sane(&root_fid));
664 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
665 if (rc != 0 && rc != -EREMOTE)
666 GOTO(out_fid2path, rc);
668 /* If remote_gf != NULL, it means just building the
669 * path on the remote MDT, copy this path segement to gf */
670 if (remote_gf != NULL) {
671 struct getinfo_fid2path *ori_gf;
675 ori_gf = (struct getinfo_fid2path *)karg;
676 if (strlen(ori_gf->gf_u.gf_path) + 1 +
677 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
678 GOTO(out_fid2path, rc = -EOVERFLOW);
680 ptr = ori_gf->gf_u.gf_path;
682 len = strlen(gf->gf_u.gf_path);
683 /* move the current path to the right to release space
684 * for closer-to-root part */
685 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
686 memcpy(ptr, gf->gf_u.gf_path, len);
690 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
691 tgt->ltd_exp->exp_obd->obd_name,
692 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
696 GOTO(out_fid2path, rc);
698 /* sigh, has to go to another MDT to do path building further */
699 if (remote_gf == NULL) {
700 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
701 OBD_ALLOC(remote_gf, remote_gf_size);
702 if (remote_gf == NULL)
703 GOTO(out_fid2path, rc = -ENOMEM);
704 remote_gf->gf_pathlen = PATH_MAX;
707 if (!fid_is_sane(&gf->gf_fid)) {
708 CERROR("%s: invalid FID "DFID": rc = %d\n",
709 tgt->ltd_exp->exp_obd->obd_name,
710 PFID(&gf->gf_fid), -EINVAL);
711 GOTO(out_fid2path, rc = -EINVAL);
714 tgt = lmv_find_target(lmv, &gf->gf_fid);
716 GOTO(out_fid2path, rc = -EINVAL);
718 remote_gf->gf_fid = gf->gf_fid;
719 remote_gf->gf_recno = -1;
720 remote_gf->gf_linkno = -1;
721 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
722 *remote_gf->gf_u.gf_root_fid = root_fid;
724 goto repeat_fid2path;
727 if (remote_gf != NULL)
728 OBD_FREE(remote_gf, remote_gf_size);
732 static int lmv_hsm_req_count(struct lmv_obd *lmv,
733 const struct hsm_user_request *hur,
734 const struct lmv_tgt_desc *tgt_mds)
738 struct lmv_tgt_desc *curr_tgt;
740 /* count how many requests must be sent to the given target */
741 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
742 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
743 if (IS_ERR(curr_tgt))
744 RETURN(PTR_ERR(curr_tgt));
745 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
751 static int lmv_hsm_req_build(struct lmv_obd *lmv,
752 struct hsm_user_request *hur_in,
753 const struct lmv_tgt_desc *tgt_mds,
754 struct hsm_user_request *hur_out)
757 struct lmv_tgt_desc *curr_tgt;
759 /* build the hsm_user_request for the given target */
760 hur_out->hur_request = hur_in->hur_request;
762 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
763 curr_tgt = lmv_find_target(lmv,
764 &hur_in->hur_user_item[i].hui_fid);
765 if (IS_ERR(curr_tgt))
766 RETURN(PTR_ERR(curr_tgt));
767 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
768 hur_out->hur_user_item[nr_out] =
769 hur_in->hur_user_item[i];
773 hur_out->hur_request.hr_itemcount = nr_out;
774 memcpy(hur_data(hur_out), hur_data(hur_in),
775 hur_in->hur_request.hr_data_len);
780 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
781 int len, struct lustre_kernelcomm *lk,
784 struct lmv_obd *lmv = &obd->u.lmv;
789 /* unregister request (call from llapi_hsm_copytool_fini) */
790 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
791 struct lmv_tgt_desc *tgt = lmv->tgts[i];
793 if (tgt == NULL || tgt->ltd_exp == NULL)
795 /* best effort: try to clean as much as possible
796 * (continue on error) */
797 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
800 /* Whatever the result, remove copytool from kuc groups.
801 * Unreached coordinators will get EPIPE on next requests
802 * and will unregister automatically.
804 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
809 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
810 int len, struct lustre_kernelcomm *lk,
813 struct lmv_obd *lmv = &obd->u.lmv;
817 bool any_set = false;
818 struct kkuc_ct_data *kcd;
823 filp = fget(lk->lk_wfd);
827 if (lk->lk_flags & LK_FLG_DATANR)
828 kcd_size = offsetof(struct kkuc_ct_data,
829 kcd_archives[lk->lk_data_count]);
831 kcd_size = sizeof(*kcd);
833 OBD_ALLOC(kcd, kcd_size);
835 GOTO(err_fput, rc = -ENOMEM);
837 kcd->kcd_nr_archives = lk->lk_data_count;
838 if (lk->lk_flags & LK_FLG_DATANR) {
839 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
840 if (lk->lk_data_count > 0)
841 memcpy(kcd->kcd_archives, lk->lk_data,
842 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
844 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
847 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
848 lk->lk_group, kcd, kcd_size);
849 OBD_FREE(kcd, kcd_size);
853 /* All or nothing: try to register to all MDS.
854 * In case of failure, unregister from previous MDS,
855 * except if it because of inactive target. */
856 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
857 struct lmv_tgt_desc *tgt = lmv->tgts[i];
859 if (tgt == NULL || tgt->ltd_exp == NULL)
862 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
864 if (tgt->ltd_active) {
865 /* permanent error */
866 CERROR("%s: iocontrol MDC %s on MDT"
867 " idx %d cmd %x: err = %d\n",
868 lmv2obd_dev(lmv)->obd_name,
869 tgt->ltd_uuid.uuid, i, cmd, err);
871 lk->lk_flags |= LK_FLG_STOP;
872 /* unregister from previous MDS */
873 for (j = 0; j < i; j++) {
875 if (tgt == NULL || tgt->ltd_exp == NULL)
877 obd_iocontrol(cmd, tgt->ltd_exp, len,
880 GOTO(err_kkuc_rem, rc);
882 /* else: transient error.
883 * kuc will register to the missing MDT
891 /* no registration done: return error */
892 GOTO(err_kkuc_rem, rc = -ENOTCONN);
897 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
907 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
908 int len, void *karg, void __user *uarg)
910 struct obd_device *obddev = class_exp2obd(exp);
911 struct lmv_obd *lmv = &obddev->u.lmv;
912 struct lmv_tgt_desc *tgt = NULL;
916 __u32 count = lmv->desc.ld_tgt_count;
923 case IOC_OBD_STATFS: {
924 struct obd_ioctl_data *data = karg;
925 struct obd_device *mdc_obd;
926 struct obd_statfs stat_buf = {0};
929 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
930 if ((index >= count))
933 tgt = lmv->tgts[index];
934 if (tgt == NULL || !tgt->ltd_active)
937 mdc_obd = class_exp2obd(tgt->ltd_exp);
942 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
943 min((int) data->ioc_plen2,
944 (int) sizeof(struct obd_uuid))))
947 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
948 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
952 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
953 min((int) data->ioc_plen1,
954 (int) sizeof(stat_buf))))
958 case OBD_IOC_QUOTACTL: {
959 struct if_quotactl *qctl = karg;
960 struct obd_quotactl *oqctl;
962 if (qctl->qc_valid == QC_MDTIDX) {
963 if (count <= qctl->qc_idx)
966 tgt = lmv->tgts[qctl->qc_idx];
967 if (tgt == NULL || tgt->ltd_exp == NULL)
969 } else if (qctl->qc_valid == QC_UUID) {
970 for (i = 0; i < count; i++) {
974 if (!obd_uuid_equals(&tgt->ltd_uuid,
978 if (tgt->ltd_exp == NULL)
990 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
991 OBD_ALLOC_PTR(oqctl);
995 QCTL_COPY(oqctl, qctl);
996 rc = obd_quotactl(tgt->ltd_exp, oqctl);
998 QCTL_COPY(qctl, oqctl);
999 qctl->qc_valid = QC_MDTIDX;
1000 qctl->obd_uuid = tgt->ltd_uuid;
1002 OBD_FREE_PTR(oqctl);
1005 case LL_IOC_GET_CONNECT_FLAGS: {
1007 if (tgt == NULL || tgt->ltd_exp == NULL)
1009 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1012 case LL_IOC_FID2MDTIDX: {
1013 struct lu_fid *fid = karg;
1016 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1020 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1021 * point to user space memory for FID2MDTIDX. */
1022 *(__u32 *)uarg = mdt_index;
1025 case OBD_IOC_FID2PATH: {
1026 rc = lmv_fid2path(exp, len, karg, uarg);
1029 case LL_IOC_HSM_STATE_GET:
1030 case LL_IOC_HSM_STATE_SET:
1031 case LL_IOC_HSM_ACTION: {
1032 struct md_op_data *op_data = karg;
1034 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1036 RETURN(PTR_ERR(tgt));
1038 if (tgt->ltd_exp == NULL)
1041 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1044 case LL_IOC_HSM_PROGRESS: {
1045 const struct hsm_progress_kernel *hpk = karg;
1047 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1049 RETURN(PTR_ERR(tgt));
1050 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1053 case LL_IOC_HSM_REQUEST: {
1054 struct hsm_user_request *hur = karg;
1055 unsigned int reqcount = hur->hur_request.hr_itemcount;
1060 /* if the request is about a single fid
1061 * or if there is a single MDS, no need to split
1063 if (reqcount == 1 || count == 1) {
1064 tgt = lmv_find_target(lmv,
1065 &hur->hur_user_item[0].hui_fid);
1067 RETURN(PTR_ERR(tgt));
1068 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1070 /* split fid list to their respective MDS */
1071 for (i = 0; i < count; i++) {
1074 struct hsm_user_request *req;
1077 if (tgt == NULL || tgt->ltd_exp == NULL)
1080 nr = lmv_hsm_req_count(lmv, hur, tgt);
1083 if (nr == 0) /* nothing for this MDS */
1086 /* build a request with fids for this MDS */
1087 reqlen = offsetof(typeof(*hur),
1089 + hur->hur_request.hr_data_len;
1090 OBD_ALLOC_LARGE(req, reqlen);
1093 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1095 GOTO(hsm_req_err, rc1);
1096 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1099 if (rc1 != 0 && rc == 0)
1101 OBD_FREE_LARGE(req, reqlen);
1106 case LL_IOC_LOV_SWAP_LAYOUTS: {
1107 struct md_op_data *op_data = karg;
1108 struct lmv_tgt_desc *tgt1, *tgt2;
1110 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1112 RETURN(PTR_ERR(tgt1));
1114 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1116 RETURN(PTR_ERR(tgt2));
1118 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1121 /* only files on same MDT can have their layouts swapped */
1122 if (tgt1->ltd_index != tgt2->ltd_index)
1125 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1128 case LL_IOC_HSM_CT_START: {
1129 struct lustre_kernelcomm *lk = karg;
1130 if (lk->lk_flags & LK_FLG_STOP)
1131 rc = lmv_hsm_ct_unregister(obddev, cmd, len, lk, uarg);
1133 rc = lmv_hsm_ct_register(obddev, cmd, len, lk, uarg);
1137 for (i = 0; i < count; i++) {
1138 struct obd_device *mdc_obd;
1142 if (tgt == NULL || tgt->ltd_exp == NULL)
1144 /* ll_umount_begin() sets force flag but for lmv, not
1145 * mdc. Let's pass it through */
1146 mdc_obd = class_exp2obd(tgt->ltd_exp);
1147 mdc_obd->obd_force = obddev->obd_force;
1148 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1150 if (tgt->ltd_active) {
1151 CERROR("error: iocontrol MDC %s on MDT"
1152 " idx %d cmd %x: err = %d\n",
1153 tgt->ltd_uuid.uuid, i, cmd, err);
1167 * This is _inode_ placement policy function (not name).
1169 static u32 lmv_placement_policy(struct obd_device *obd,
1170 struct md_op_data *op_data)
1172 struct lmv_obd *lmv = &obd->u.lmv;
1173 struct lmv_user_md *lum;
1178 if (lmv->desc.ld_tgt_count == 1)
1181 lum = op_data->op_data;
1184 * 1. See if the stripe offset is specified by lum.
1185 * 2. If parent has default LMV, and its hash type is "space", choose
1186 * MDT with QoS. (see lmv_locate_tgt_qos()).
1187 * 3. Then check if default LMV stripe offset is not -1.
1188 * 4. Finally choose MDS by name hash if the parent
1189 * is striped directory. (see lmv_locate_tgt()).
1191 * presently explicit MDT location is not supported
1192 * for foreign dirs (as it can't be embedded into free
1193 * format LMV, like with lum_stripe_offset), so we only
1194 * rely on default stripe offset or then name hashing.
1196 if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
1197 le32_to_cpu(lum->lum_magic != LMV_MAGIC_FOREIGN) &&
1198 le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1199 mdt = le32_to_cpu(lum->lum_stripe_offset);
1200 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1201 !lmv_dir_striped(op_data->op_mea1) &&
1202 lmv_dir_qos_mkdir(op_data->op_default_mea1)) {
1203 mdt = op_data->op_mds;
1204 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1205 op_data->op_default_mea1 &&
1206 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1208 mdt = op_data->op_default_mea1->lsm_md_master_mdt_index;
1209 op_data->op_mds = mdt;
1211 mdt = op_data->op_mds;
1217 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1219 struct lmv_tgt_desc *tgt;
1223 tgt = lmv_get_target(lmv, mds, NULL);
1225 RETURN(PTR_ERR(tgt));
1228 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1229 * on server that seq in new allocated fid is not yet known.
1231 mutex_lock(&tgt->ltd_fid_mutex);
1233 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1234 GOTO(out, rc = -ENODEV);
1237 * Asking underlying tgt layer to allocate new fid.
1239 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1241 LASSERT(fid_is_sane(fid));
1247 mutex_unlock(&tgt->ltd_fid_mutex);
1251 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1252 struct lu_fid *fid, struct md_op_data *op_data)
1254 struct obd_device *obd = class_exp2obd(exp);
1255 struct lmv_obd *lmv = &obd->u.lmv;
1261 LASSERT(op_data != NULL);
1262 LASSERT(fid != NULL);
1264 mds = lmv_placement_policy(obd, op_data);
1266 rc = __lmv_fid_alloc(lmv, fid, mds);
1268 CERROR("Can't alloc new fid, rc %d\n", rc);
1273 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1275 struct lmv_obd *lmv = &obd->u.lmv;
1276 struct lmv_desc *desc;
1277 struct lnet_process_id lnet_id;
1283 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1284 CERROR("LMV setup requires a descriptor\n");
1288 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1289 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1290 CERROR("Lmv descriptor size wrong: %d > %d\n",
1291 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1295 lmv->tgts_size = 32U;
1296 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1297 if (lmv->tgts == NULL)
1300 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1301 lmv->desc.ld_tgt_count = 0;
1302 lmv->desc.ld_active_tgt_count = 0;
1303 lmv->desc.ld_qos_maxage = LMV_DESC_QOS_MAXAGE_DEFAULT;
1304 lmv->max_def_easize = 0;
1305 lmv->max_easize = 0;
1307 spin_lock_init(&lmv->lmv_lock);
1308 mutex_init(&lmv->lmv_init_mutex);
1310 /* Set up allocation policy (QoS and RR) */
1311 INIT_LIST_HEAD(&lmv->lmv_qos.lq_svr_list);
1312 init_rwsem(&lmv->lmv_qos.lq_rw_sem);
1313 lmv->lmv_qos.lq_dirty = 1;
1314 lmv->lmv_qos.lq_reset = 1;
1315 /* Default priority is toward free space balance */
1316 lmv->lmv_qos.lq_prio_free = 232;
1317 /* Default threshold for rr (roughly 17%) */
1318 lmv->lmv_qos.lq_threshold_rr = 43;
1320 lu_qos_rr_init(&lmv->lmv_qos.lq_rr);
1323 * initialize rr_index to lower 32bit of netid, so that client
1324 * can distribute subdirs evenly from the beginning.
1326 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1327 if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
1328 lmv->lmv_qos_rr_index = (u32)lnet_id.nid;
1333 rc = lmv_tunables_init(obd);
1335 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1338 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1339 LUSTRE_CLI_FLD_HASH_DHT);
1341 CERROR("Can't init FLD, err %d\n", rc);
1351 static int lmv_cleanup(struct obd_device *obd)
1353 struct lmv_obd *lmv = &obd->u.lmv;
1356 fld_client_fini(&lmv->lmv_fld);
1357 if (lmv->tgts != NULL) {
1359 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1360 if (lmv->tgts[i] == NULL)
1362 lmv_del_target(lmv, i);
1364 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1370 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1372 struct lustre_cfg *lcfg = buf;
1373 struct obd_uuid obd_uuid;
1379 switch (lcfg->lcfg_command) {
1381 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1382 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1383 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1384 GOTO(out, rc = -EINVAL);
1386 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1388 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1389 GOTO(out, rc = -EINVAL);
1390 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1391 GOTO(out, rc = -EINVAL);
1392 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1395 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1396 GOTO(out, rc = -EINVAL);
1402 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1406 if (flags & OBD_STATFS_FOR_MDT0)
1409 if (lmv->lmv_statfs_start || lmv->desc.ld_tgt_count == 1)
1410 return lmv->lmv_statfs_start;
1412 /* choose initial MDT for this client */
1414 struct lnet_process_id lnet_id;
1415 if (LNetGetId(i, &lnet_id) == -ENOENT)
1418 if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
1419 /* We dont need a full 64-bit modulus, just enough
1420 * to distribute the requests across MDTs evenly.
1422 lmv->lmv_statfs_start =
1423 (u32)lnet_id.nid % lmv->desc.ld_tgt_count;
1428 return lmv->lmv_statfs_start;
1431 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1432 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1434 struct obd_device *obd = class_exp2obd(exp);
1435 struct lmv_obd *lmv = &obd->u.lmv;
1436 struct obd_statfs *temp;
1441 OBD_ALLOC(temp, sizeof(*temp));
1445 /* distribute statfs among MDTs */
1446 idx = lmv_select_statfs_mdt(lmv, flags);
1448 for (i = 0; i < lmv->desc.ld_tgt_count; i++, idx++) {
1449 idx = idx % lmv->desc.ld_tgt_count;
1450 if (lmv->tgts[idx] == NULL || lmv->tgts[idx]->ltd_exp == NULL)
1453 rc = obd_statfs(env, lmv->tgts[idx]->ltd_exp, temp,
1456 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1457 lmv->tgts[idx]->ltd_exp->exp_obd->obd_name, i,
1459 GOTO(out_free_temp, rc);
1462 if (temp->os_state & OS_STATE_SUM ||
1463 flags == OBD_STATFS_FOR_MDT0) {
1464 /* reset to the last aggregated values
1465 * and don't sum with non-aggrated data */
1466 /* If the statfs is from mount, it needs to retrieve
1467 * necessary information from MDT0. i.e. mount does
1468 * not need the merged osfs from all of MDT. Also
1469 * clients can be mounted as long as MDT0 is in
1478 osfs->os_bavail += temp->os_bavail;
1479 osfs->os_blocks += temp->os_blocks;
1480 osfs->os_ffree += temp->os_ffree;
1481 osfs->os_files += temp->os_files;
1482 osfs->os_granted += temp->os_granted;
1488 OBD_FREE(temp, sizeof(*temp));
1492 static int lmv_statfs_update(void *cookie, int rc)
1494 struct obd_info *oinfo = cookie;
1495 struct obd_device *obd = oinfo->oi_obd;
1496 struct lmv_obd *lmv = &obd->u.lmv;
1497 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1498 struct obd_statfs *osfs = oinfo->oi_osfs;
1501 * NB: don't deactivate TGT upon error, because we may not trigger async
1502 * statfs any longer, then there is no chance to activate TGT.
1505 spin_lock(&lmv->lmv_lock);
1506 tgt->ltd_statfs = *osfs;
1507 tgt->ltd_statfs_age = ktime_get_seconds();
1508 spin_unlock(&lmv->lmv_lock);
1509 lmv->lmv_qos.lq_dirty = 1;
1515 /* update tgt statfs async if it's ld_qos_maxage old */
1516 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1518 struct obd_info oinfo = {
1521 .oi_cb_up = lmv_statfs_update,
1525 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1526 obd->u.lmv.desc.ld_qos_maxage)
1529 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1534 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1537 struct obd_device *obd = exp->exp_obd;
1538 struct lmv_obd *lmv = &obd->u.lmv;
1542 rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1546 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1547 u64 obd_md_valid, const char *name, size_t buf_size,
1548 struct ptlrpc_request **req)
1550 struct obd_device *obd = exp->exp_obd;
1551 struct lmv_obd *lmv = &obd->u.lmv;
1552 struct lmv_tgt_desc *tgt;
1556 tgt = lmv_find_target(lmv, fid);
1558 RETURN(PTR_ERR(tgt));
1560 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1565 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1566 u64 obd_md_valid, const char *name,
1567 const void *value, size_t value_size,
1568 unsigned int xattr_flags, u32 suppgid,
1569 struct ptlrpc_request **req)
1571 struct obd_device *obd = exp->exp_obd;
1572 struct lmv_obd *lmv = &obd->u.lmv;
1573 struct lmv_tgt_desc *tgt;
1577 tgt = lmv_find_target(lmv, fid);
1579 RETURN(PTR_ERR(tgt));
1581 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1582 value, value_size, xattr_flags, suppgid, req);
1587 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1588 struct ptlrpc_request **request)
1590 struct obd_device *obd = exp->exp_obd;
1591 struct lmv_obd *lmv = &obd->u.lmv;
1592 struct lmv_tgt_desc *tgt;
1596 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1598 RETURN(PTR_ERR(tgt));
1600 if (op_data->op_flags & MF_GET_MDT_IDX) {
1601 op_data->op_mds = tgt->ltd_index;
1605 rc = md_getattr(tgt->ltd_exp, op_data, request);
1610 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1612 struct obd_device *obd = exp->exp_obd;
1613 struct lmv_obd *lmv = &obd->u.lmv;
1617 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1620 * With DNE every object can have two locks in different namespaces:
1621 * lookup lock in space of MDT storing direntry and update/open lock in
1622 * space of MDT storing inode.
1624 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1625 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1627 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1633 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1634 struct md_open_data *mod, struct ptlrpc_request **request)
1636 struct obd_device *obd = exp->exp_obd;
1637 struct lmv_obd *lmv = &obd->u.lmv;
1638 struct lmv_tgt_desc *tgt;
1642 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1644 RETURN(PTR_ERR(tgt));
1646 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1647 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1651 static struct lmv_tgt_desc *
1652 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1653 const char *name, int namelen, struct lu_fid *fid,
1654 __u32 *mds, bool post_migrate)
1656 struct lmv_tgt_desc *tgt;
1657 const struct lmv_oinfo *oinfo;
1659 if (!lmv_dir_striped(lsm) || !namelen) {
1660 tgt = lmv_find_target(lmv, fid);
1664 *mds = tgt->ltd_index;
1668 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1669 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1670 return ERR_PTR(-EBADF);
1671 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1673 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
1676 return ERR_CAST(oinfo);
1679 *fid = oinfo->lmo_fid;
1680 *mds = oinfo->lmo_mds;
1681 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1683 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1689 * Locate MDT of op_data->op_fid1
1691 * For striped directory, it will locate the stripe by name hash, if hash_type
1692 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1693 * which is set outside, and if dir is migrating, 'op_data->op_post_migrate'
1694 * indicates whether old or new layout is used to locate.
1696 * For plain direcotry, normally it will locate MDT by FID, but if this
1697 * directory has default LMV, and its hash type is "space", locate MDT with QoS.
1699 * \param[in] lmv LMV device
1700 * \param[in] op_data client MD stack parameters, name, namelen
1703 * retval pointer to the lmv_tgt_desc if succeed.
1704 * ERR_PTR(errno) if failed.
1706 struct lmv_tgt_desc *
1707 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1709 struct lmv_stripe_md *lsm = op_data->op_mea1;
1710 struct lmv_oinfo *oinfo;
1711 struct lmv_tgt_desc *tgt;
1713 if (lmv_dir_foreign(lsm))
1714 return ERR_PTR(-ENODATA);
1716 /* During creating VOLATILE file, it should honor the mdt
1717 * index if the file under striped dir is being restored, see
1719 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1720 (int)op_data->op_mds != -1) {
1721 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1725 if (lmv_dir_striped(lsm)) {
1728 /* refill the right parent fid */
1729 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1730 oinfo = &lsm->lsm_md_oinfo[i];
1731 if (oinfo->lmo_mds == op_data->op_mds) {
1732 op_data->op_fid1 = oinfo->lmo_fid;
1737 if (i == lsm->lsm_md_stripe_count)
1738 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1740 } else if (lmv_dir_bad_hash(lsm)) {
1741 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1742 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1744 op_data->op_fid1 = oinfo->lmo_fid;
1745 op_data->op_mds = oinfo->lmo_mds;
1746 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1747 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1748 lmv_dir_qos_mkdir(op_data->op_default_mea1) &&
1749 !lmv_dir_striped(lsm)) {
1750 tgt = lmv_locate_tgt_qos(lmv, &op_data->op_mds);
1751 if (tgt == ERR_PTR(-EAGAIN))
1752 tgt = lmv_locate_tgt_rr(lmv, &op_data->op_mds);
1754 * only update statfs when mkdir under dir with "space" hash,
1755 * this means the cached statfs may be stale, and current mkdir
1756 * may not follow QoS accurately, but it's not serious, and it
1757 * avoids periodic statfs when client doesn't mkdir under
1758 * "space" hashed directories.
1760 * TODO: after MDT support QoS object allocation, also update
1761 * statfs for 'lfs mkdir -i -1 ...", currently it's done in user
1765 struct obd_device *obd;
1767 obd = container_of(lmv, struct obd_device, u.lmv);
1768 lmv_statfs_check_update(obd, tgt);
1771 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1772 op_data->op_name, op_data->op_namelen,
1773 &op_data->op_fid1, &op_data->op_mds,
1774 op_data->op_post_migrate);
1780 /* Locate MDT of op_data->op_fid2 for link/rename */
1781 static struct lmv_tgt_desc *
1782 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1784 struct lmv_tgt_desc *tgt;
1787 LASSERT(op_data->op_name);
1788 if (lmv_dir_migrating(op_data->op_mea2)) {
1789 struct lu_fid fid1 = op_data->op_fid1;
1790 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1791 struct ptlrpc_request *request = NULL;
1794 * avoid creating new file under old layout of migrating
1795 * directory, check it here.
1797 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1798 op_data->op_name, op_data->op_namelen,
1799 &op_data->op_fid2, &op_data->op_mds, false);
1803 op_data->op_fid1 = op_data->op_fid2;
1804 op_data->op_mea1 = op_data->op_mea2;
1805 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1806 op_data->op_fid1 = fid1;
1807 op_data->op_mea1 = lsm1;
1809 ptlrpc_req_finished(request);
1810 RETURN(ERR_PTR(-EEXIST));
1814 RETURN(ERR_PTR(rc));
1817 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1818 op_data->op_namelen, &op_data->op_fid2,
1819 &op_data->op_mds, true);
1822 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1823 const void *data, size_t datalen, umode_t mode, uid_t uid,
1824 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1825 struct ptlrpc_request **request)
1827 struct obd_device *obd = exp->exp_obd;
1828 struct lmv_obd *lmv = &obd->u.lmv;
1829 struct lmv_tgt_desc *tgt;
1834 if (!lmv->desc.ld_active_tgt_count)
1837 if (lmv_dir_bad_hash(op_data->op_mea1))
1840 if (lmv_dir_migrating(op_data->op_mea1)) {
1842 * if parent is migrating, create() needs to lookup existing
1843 * name, to avoid creating new file under old layout of
1844 * migrating directory, check old layout here.
1846 tgt = lmv_locate_tgt(lmv, op_data);
1848 RETURN(PTR_ERR(tgt));
1850 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1852 ptlrpc_req_finished(*request);
1860 op_data->op_post_migrate = true;
1863 tgt = lmv_locate_tgt(lmv, op_data);
1865 RETURN(PTR_ERR(tgt));
1867 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1868 (int)op_data->op_namelen, op_data->op_name,
1869 PFID(&op_data->op_fid1), op_data->op_mds);
1871 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1875 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1876 /* Send the create request to the MDT where the object
1877 * will be located */
1878 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1880 RETURN(PTR_ERR(tgt));
1882 op_data->op_mds = tgt->ltd_index;
1885 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1886 PFID(&op_data->op_fid2), op_data->op_mds);
1888 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1889 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1890 cap_effective, rdev, request);
1892 if (*request == NULL)
1894 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1900 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1901 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1902 struct lustre_handle *lockh, __u64 extra_lock_flags)
1904 struct obd_device *obd = exp->exp_obd;
1905 struct lmv_obd *lmv = &obd->u.lmv;
1906 struct lmv_tgt_desc *tgt;
1910 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1912 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1914 RETURN(PTR_ERR(tgt));
1916 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1917 PFID(&op_data->op_fid1), tgt->ltd_index);
1919 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1926 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1927 struct ptlrpc_request **preq)
1929 struct obd_device *obd = exp->exp_obd;
1930 struct lmv_obd *lmv = &obd->u.lmv;
1931 struct lmv_tgt_desc *tgt;
1932 struct mdt_body *body;
1938 tgt = lmv_locate_tgt(lmv, op_data);
1940 RETURN(PTR_ERR(tgt));
1942 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1943 (int)op_data->op_namelen, op_data->op_name,
1944 PFID(&op_data->op_fid1), tgt->ltd_index);
1946 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1947 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1948 ptlrpc_req_finished(*preq);
1956 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1957 LASSERT(body != NULL);
1959 if (body->mbo_valid & OBD_MD_MDS) {
1960 op_data->op_fid1 = body->mbo_fid1;
1961 op_data->op_valid |= OBD_MD_FLCROSSREF;
1962 op_data->op_namelen = 0;
1963 op_data->op_name = NULL;
1965 ptlrpc_req_finished(*preq);
1974 #define md_op_data_fid(op_data, fl) \
1975 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1976 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1977 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1978 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1981 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1982 struct md_op_data *op_data, __u32 op_tgt,
1983 enum ldlm_mode mode, int bits, int flag)
1985 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1986 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1987 union ldlm_policy_data policy = { { 0 } };
1991 if (!fid_is_sane(fid))
1995 tgt = lmv_find_target(lmv, fid);
1997 RETURN(PTR_ERR(tgt));
2000 if (tgt->ltd_index != op_tgt) {
2001 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2002 policy.l_inodebits.bits = bits;
2003 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2004 mode, LCF_ASYNC, NULL);
2007 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2009 op_data->op_flags |= flag;
2017 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2020 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2021 struct ptlrpc_request **request)
2023 struct obd_device *obd = exp->exp_obd;
2024 struct lmv_obd *lmv = &obd->u.lmv;
2025 struct lmv_tgt_desc *tgt;
2029 LASSERT(op_data->op_namelen != 0);
2031 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2032 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2033 op_data->op_name, PFID(&op_data->op_fid1));
2035 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2036 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2037 op_data->op_cap = cfs_curproc_cap_pack();
2039 tgt = lmv_locate_tgt2(lmv, op_data);
2041 RETURN(PTR_ERR(tgt));
2044 * Cancel UPDATE lock on child (fid1).
2046 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2047 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2048 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2052 rc = md_link(tgt->ltd_exp, op_data, request);
2057 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2058 const char *name, size_t namelen,
2059 struct ptlrpc_request **request)
2061 struct obd_device *obd = exp->exp_obd;
2062 struct lmv_obd *lmv = &obd->u.lmv;
2063 struct lmv_stripe_md *lsm = op_data->op_mea1;
2064 struct lmv_tgt_desc *parent_tgt;
2065 struct lmv_tgt_desc *sp_tgt;
2066 struct lmv_tgt_desc *tp_tgt = NULL;
2067 struct lmv_tgt_desc *child_tgt;
2068 struct lmv_tgt_desc *tgt;
2069 struct lu_fid target_fid;
2074 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2076 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2077 PFID(&op_data->op_fid1), (int)namelen, name);
2079 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2080 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2081 op_data->op_cap = cfs_curproc_cap_pack();
2083 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2084 if (IS_ERR(parent_tgt))
2085 RETURN(PTR_ERR(parent_tgt));
2087 if (lmv_dir_striped(lsm)) {
2088 __u32 hash_type = lsm->lsm_md_hash_type;
2089 __u32 stripe_count = lsm->lsm_md_stripe_count;
2092 * old stripes are appended after new stripes for migrating
2095 if (lmv_dir_migrating(lsm)) {
2096 hash_type = lsm->lsm_md_migrate_hash;
2097 stripe_count -= lsm->lsm_md_migrate_offset;
2100 rc = lmv_name_to_stripe_index(hash_type, stripe_count, name,
2105 if (lmv_dir_migrating(lsm))
2106 rc += lsm->lsm_md_migrate_offset;
2108 /* save it in fid4 temporarily for early cancel */
2109 op_data->op_fid4 = lsm->lsm_md_oinfo[rc].lmo_fid;
2110 sp_tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[rc].lmo_mds,
2113 RETURN(PTR_ERR(sp_tgt));
2116 * if parent is being migrated too, fill op_fid2 with target
2117 * stripe fid, otherwise the target stripe is not created yet.
2119 if (lmv_dir_migrating(lsm)) {
2120 hash_type = lsm->lsm_md_hash_type &
2121 ~LMV_HASH_FLAG_MIGRATION;
2122 stripe_count = lsm->lsm_md_migrate_offset;
2124 rc = lmv_name_to_stripe_index(hash_type, stripe_count,
2129 op_data->op_fid2 = lsm->lsm_md_oinfo[rc].lmo_fid;
2130 tp_tgt = lmv_get_target(lmv,
2131 lsm->lsm_md_oinfo[rc].lmo_mds,
2134 RETURN(PTR_ERR(tp_tgt));
2137 sp_tgt = parent_tgt;
2140 child_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2141 if (IS_ERR(child_tgt))
2142 RETURN(PTR_ERR(child_tgt));
2144 if (!S_ISDIR(op_data->op_mode) && tp_tgt)
2145 rc = __lmv_fid_alloc(lmv, &target_fid, tp_tgt->ltd_index);
2147 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2152 * for directory, send migrate request to the MDT where the object will
2153 * be migrated to, because we can't create a striped directory remotely.
2155 * otherwise, send to the MDT where source is located because regular
2156 * file may open lease.
2158 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2159 * backward compatibility.
2161 if (S_ISDIR(op_data->op_mode) &&
2162 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2163 tgt = lmv_find_target(lmv, &target_fid);
2165 RETURN(PTR_ERR(tgt));
2170 /* cancel UPDATE lock of parent master object */
2171 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2172 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2176 /* cancel UPDATE lock of source parent */
2177 if (sp_tgt != parent_tgt) {
2179 * migrate RPC packs master object FID, because we can only pack
2180 * two FIDs in reint RPC, but MDS needs to know both source
2181 * parent and target parent, and it will obtain them from master
2182 * FID and LMV, the other FID in RPC is kept for target.
2184 * since this FID is not passed to MDC, cancel it anyway.
2186 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2187 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2191 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2193 op_data->op_fid4 = target_fid;
2195 /* cancel UPDATE locks of target parent */
2196 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2197 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2201 /* cancel LOOKUP lock of source if source is remote object */
2202 if (child_tgt != sp_tgt) {
2203 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2204 LCK_EX, MDS_INODELOCK_LOOKUP,
2205 MF_MDC_CANCEL_FID3);
2210 /* cancel ELC locks of source */
2211 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2212 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2216 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2221 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2222 const char *old, size_t oldlen,
2223 const char *new, size_t newlen,
2224 struct ptlrpc_request **request)
2226 struct obd_device *obd = exp->exp_obd;
2227 struct lmv_obd *lmv = &obd->u.lmv;
2228 struct lmv_tgt_desc *sp_tgt;
2229 struct lmv_tgt_desc *tp_tgt = NULL;
2230 struct lmv_tgt_desc *src_tgt = NULL;
2231 struct lmv_tgt_desc *tgt;
2232 struct mdt_body *body;
2237 LASSERT(oldlen != 0);
2239 if (op_data->op_cli_flags & CLI_MIGRATE) {
2240 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2244 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2245 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2246 op_data->op_cap = cfs_curproc_cap_pack();
2248 op_data->op_name = new;
2249 op_data->op_namelen = newlen;
2251 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2253 RETURN(PTR_ERR(tp_tgt));
2255 /* Since the target child might be destroyed, and it might become
2256 * orphan, and we can only check orphan on the local MDT right now, so
2257 * we send rename request to the MDT where target child is located. If
2258 * target child does not exist, then it will send the request to the
2260 if (fid_is_sane(&op_data->op_fid4)) {
2261 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2263 RETURN(PTR_ERR(tgt));
2268 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2270 /* cancel UPDATE locks of target parent */
2271 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2272 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2276 if (fid_is_sane(&op_data->op_fid4)) {
2277 /* cancel LOOKUP lock of target on target parent */
2278 if (tgt != tp_tgt) {
2279 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2280 tgt->ltd_index, LCK_EX,
2281 MDS_INODELOCK_LOOKUP,
2282 MF_MDC_CANCEL_FID4);
2288 if (fid_is_sane(&op_data->op_fid3)) {
2289 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2290 if (IS_ERR(src_tgt))
2291 RETURN(PTR_ERR(src_tgt));
2293 /* cancel ELC locks of source */
2294 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2295 LCK_EX, MDS_INODELOCK_ELC,
2296 MF_MDC_CANCEL_FID3);
2301 op_data->op_name = old;
2302 op_data->op_namelen = oldlen;
2304 sp_tgt = lmv_locate_tgt(lmv, op_data);
2306 RETURN(PTR_ERR(sp_tgt));
2308 /* cancel UPDATE locks of source parent */
2309 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2310 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2314 if (fid_is_sane(&op_data->op_fid3)) {
2315 /* cancel LOOKUP lock of source on source parent */
2316 if (src_tgt != sp_tgt) {
2317 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2318 tgt->ltd_index, LCK_EX,
2319 MDS_INODELOCK_LOOKUP,
2320 MF_MDC_CANCEL_FID3);
2327 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2328 PFID(&op_data->op_fid1), (int)oldlen, old,
2329 PFID(&op_data->op_fid2), (int)newlen, new);
2331 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2333 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2334 ptlrpc_req_finished(*request);
2339 if (rc && rc != -EXDEV)
2342 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2346 /* Not cross-ref case, just get out of here. */
2347 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2350 op_data->op_fid4 = body->mbo_fid1;
2352 ptlrpc_req_finished(*request);
2355 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2357 RETURN(PTR_ERR(tgt));
2359 if (fid_is_sane(&op_data->op_fid4)) {
2360 /* cancel LOOKUP lock of target on target parent */
2361 if (tgt != tp_tgt) {
2362 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2363 tgt->ltd_index, LCK_EX,
2364 MDS_INODELOCK_LOOKUP,
2365 MF_MDC_CANCEL_FID4);
2374 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2375 void *ea, size_t ealen, struct ptlrpc_request **request)
2377 struct obd_device *obd = exp->exp_obd;
2378 struct lmv_obd *lmv = &obd->u.lmv;
2379 struct lmv_tgt_desc *tgt;
2383 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2384 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2385 op_data->op_xvalid);
2387 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2388 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2390 RETURN(PTR_ERR(tgt));
2392 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2397 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2398 struct ptlrpc_request **request)
2400 struct obd_device *obd = exp->exp_obd;
2401 struct lmv_obd *lmv = &obd->u.lmv;
2402 struct lmv_tgt_desc *tgt;
2406 tgt = lmv_find_target(lmv, fid);
2408 RETURN(PTR_ERR(tgt));
2410 rc = md_fsync(tgt->ltd_exp, fid, request);
2414 struct stripe_dirent {
2415 struct page *sd_page;
2416 struct lu_dirpage *sd_dp;
2417 struct lu_dirent *sd_ent;
2421 struct lmv_dir_ctxt {
2422 struct lmv_obd *ldc_lmv;
2423 struct md_op_data *ldc_op_data;
2424 struct md_callback *ldc_cb_op;
2427 struct stripe_dirent ldc_stripes[0];
2430 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2432 if (stripe->sd_page) {
2433 kunmap(stripe->sd_page);
2434 put_page(stripe->sd_page);
2435 stripe->sd_page = NULL;
2436 stripe->sd_ent = NULL;
2440 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2444 for (i = 0; i < ctxt->ldc_count; i++)
2445 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2448 /* if @ent is dummy, or . .., get next */
2449 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2450 struct lu_dirent *ent,
2453 for (; ent; ent = lu_dirent_next(ent)) {
2454 /* Skip dummy entry */
2455 if (le16_to_cpu(ent->lde_namelen) == 0)
2458 /* skip . and .. for other stripes */
2460 (strncmp(ent->lde_name, ".",
2461 le16_to_cpu(ent->lde_namelen)) == 0 ||
2462 strncmp(ent->lde_name, "..",
2463 le16_to_cpu(ent->lde_namelen)) == 0))
2466 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2473 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2474 struct stripe_dirent *stripe,
2477 struct md_op_data *op_data = ctxt->ldc_op_data;
2478 struct lmv_oinfo *oinfo;
2479 struct lu_fid fid = op_data->op_fid1;
2480 struct inode *inode = op_data->op_data;
2481 struct lmv_tgt_desc *tgt;
2482 struct lu_dirent *ent = stripe->sd_ent;
2483 __u64 hash = ctxt->ldc_hash;
2488 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2492 if (stripe->sd_page) {
2493 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2495 /* @hash should be the last dirent hash */
2496 LASSERTF(hash <= end,
2497 "ctxt@%p stripe@%p hash %llx end %llx\n",
2498 ctxt, stripe, hash, end);
2499 /* unload last page */
2500 stripe_dirent_unload(stripe);
2502 if (end == MDS_DIR_END_OFF) {
2503 stripe->sd_eof = true;
2509 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2510 if (!oinfo->lmo_root) {
2515 tgt = lmv_get_target(ctxt->ldc_lmv, oinfo->lmo_mds, NULL);
2521 /* op_data is shared by stripes, reset after use */
2522 op_data->op_fid1 = oinfo->lmo_fid;
2523 op_data->op_fid2 = oinfo->lmo_fid;
2524 op_data->op_data = oinfo->lmo_root;
2526 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2529 op_data->op_fid1 = fid;
2530 op_data->op_fid2 = fid;
2531 op_data->op_data = inode;
2536 stripe->sd_dp = page_address(stripe->sd_page);
2537 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2539 /* in case a page filled with ., .. and dummy, read next */
2542 stripe->sd_ent = ent;
2545 /* treat error as eof, so dir can be partially accessed */
2546 stripe->sd_eof = true;
2547 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2548 "directory is partially accessed!\n",
2549 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2556 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2558 struct obd_device *obd = exp->exp_obd;
2559 struct lmv_obd *lmv = &obd->u.lmv;
2560 struct lmv_tgt_desc *tgt;
2564 rc = lmv_check_connect(obd);
2568 tgt = lmv_find_target(lmv, &data->op_fid1);
2570 RETURN(PTR_ERR(tgt));
2572 data->op_flags |= MF_MDC_CANCEL_FID1;
2573 rc = md_file_resync(tgt->ltd_exp, data);
2578 * Get dirent with the closest hash for striped directory
2580 * This function will search the dir entry, whose hash value is the
2581 * closest(>=) to hash from all of sub-stripes, and it is only being called
2582 * for striped directory.
2584 * \param[in] ctxt dir read context
2586 * \retval dirent get the entry successfully
2587 * NULL does not get the entry, normally it means
2588 * it reaches the end of the directory, while read
2589 * stripe dirent error is ignored to allow partial
2592 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2594 struct stripe_dirent *stripe;
2595 struct lu_dirent *ent = NULL;
2599 /* TODO: optimize with k-way merge sort */
2600 for (i = 0; i < ctxt->ldc_count; i++) {
2601 stripe = &ctxt->ldc_stripes[i];
2605 if (!stripe->sd_ent) {
2606 stripe_dirent_load(ctxt, stripe, i);
2607 if (!stripe->sd_ent) {
2608 LASSERT(stripe->sd_eof);
2614 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2615 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2617 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2624 stripe = &ctxt->ldc_stripes[min];
2625 ent = stripe->sd_ent;
2626 /* pop found dirent */
2627 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2635 * Build dir entry page for striped directory
2637 * This function gets one entry by @offset from a striped directory. It will
2638 * read entries from all of stripes, and choose one closest to the required
2639 * offset(&offset). A few notes
2640 * 1. skip . and .. for non-zero stripes, because there can only have one .
2641 * and .. in a directory.
2642 * 2. op_data will be shared by all of stripes, instead of allocating new
2643 * one, so need to restore before reusing.
2645 * \param[in] exp obd export refer to LMV
2646 * \param[in] op_data hold those MD parameters of read_entry
2647 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2648 * \param[in] offset starting hash offset
2649 * \param[out] ppage the page holding the entry. Note: because the entry
2650 * will be accessed in upper layer, so we need hold the
2651 * page until the usages of entry is finished, see
2652 * ll_dir_entry_next.
2654 * retval =0 if get entry successfully
2655 * <0 cannot get entry
2657 static int lmv_striped_read_page(struct obd_export *exp,
2658 struct md_op_data *op_data,
2659 struct md_callback *cb_op,
2660 __u64 offset, struct page **ppage)
2662 struct page *page = NULL;
2663 struct lu_dirpage *dp;
2665 struct lu_dirent *ent;
2666 struct lu_dirent *last_ent;
2668 struct lmv_dir_ctxt *ctxt;
2669 struct lu_dirent *next = NULL;
2675 /* Allocate a page and read entries from all of stripes and fill
2676 * the page by hash order */
2677 page = alloc_page(GFP_KERNEL);
2681 /* Initialize the entry page */
2683 memset(dp, 0, sizeof(*dp));
2684 dp->ldp_hash_start = cpu_to_le64(offset);
2687 left_bytes = PAGE_SIZE - sizeof(*dp);
2691 /* initalize dir read context */
2692 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2693 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2695 GOTO(free_page, rc = -ENOMEM);
2696 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2697 ctxt->ldc_op_data = op_data;
2698 ctxt->ldc_cb_op = cb_op;
2699 ctxt->ldc_hash = offset;
2700 ctxt->ldc_count = stripe_count;
2703 next = lmv_dirent_next(ctxt);
2705 /* end of directory */
2707 ctxt->ldc_hash = MDS_DIR_END_OFF;
2710 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2712 ent_size = le16_to_cpu(next->lde_reclen);
2714 /* the last entry lde_reclen is 0, but it might not be the last
2715 * one of this temporay dir page */
2717 ent_size = lu_dirent_calc_size(
2718 le16_to_cpu(next->lde_namelen),
2719 le32_to_cpu(next->lde_attrs));
2721 if (ent_size > left_bytes)
2724 memcpy(ent, next, ent_size);
2726 /* Replace . with master FID and Replace .. with the parent FID
2727 * of master object */
2728 if (strncmp(ent->lde_name, ".",
2729 le16_to_cpu(ent->lde_namelen)) == 0 &&
2730 le16_to_cpu(ent->lde_namelen) == 1)
2731 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2732 else if (strncmp(ent->lde_name, "..",
2733 le16_to_cpu(ent->lde_namelen)) == 0 &&
2734 le16_to_cpu(ent->lde_namelen) == 2)
2735 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2737 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2738 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2739 le64_to_cpu(ent->lde_hash));
2741 left_bytes -= ent_size;
2742 ent->lde_reclen = cpu_to_le16(ent_size);
2744 ent = (void *)ent + ent_size;
2747 last_ent->lde_reclen = 0;
2750 dp->ldp_flags |= LDF_EMPTY;
2751 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2752 dp->ldp_flags |= LDF_COLLIDE;
2753 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2754 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2756 put_lmv_dir_ctxt(ctxt);
2757 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2770 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2771 struct md_callback *cb_op, __u64 offset,
2772 struct page **ppage)
2774 struct obd_device *obd = exp->exp_obd;
2775 struct lmv_obd *lmv = &obd->u.lmv;
2776 struct lmv_tgt_desc *tgt;
2781 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2784 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2785 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2789 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2791 RETURN(PTR_ERR(tgt));
2793 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2799 * Unlink a file/directory
2801 * Unlink a file or directory under the parent dir. The unlink request
2802 * usually will be sent to the MDT where the child is located, but if
2803 * the client does not have the child FID then request will be sent to the
2804 * MDT where the parent is located.
2806 * If the parent is a striped directory then it also needs to locate which
2807 * stripe the name of the child is located, and replace the parent FID
2808 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2809 * it will walk through all of sub-stripes until the child is being
2812 * \param[in] exp export refer to LMV
2813 * \param[in] op_data different parameters transferred beween client
2814 * MD stacks, name, namelen, FIDs etc.
2815 * op_fid1 is the parent FID, op_fid2 is the child
2817 * \param[out] request point to the request of unlink.
2819 * retval 0 if succeed
2820 * negative errno if failed.
2822 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2823 struct ptlrpc_request **request)
2825 struct obd_device *obd = exp->exp_obd;
2826 struct lmv_obd *lmv = &obd->u.lmv;
2827 struct lmv_tgt_desc *tgt;
2828 struct lmv_tgt_desc *parent_tgt;
2829 struct mdt_body *body;
2834 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2835 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2836 op_data->op_cap = cfs_curproc_cap_pack();
2839 parent_tgt = lmv_locate_tgt(lmv, op_data);
2840 if (IS_ERR(parent_tgt))
2841 RETURN(PTR_ERR(parent_tgt));
2843 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2844 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2846 RETURN(PTR_ERR(tgt));
2852 * If child's fid is given, cancel unused locks for it if it is from
2853 * another export than parent.
2855 * LOOKUP lock for child (fid3) should also be cancelled on parent
2856 * tgt_tgt in mdc_unlink().
2858 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2860 if (parent_tgt != tgt)
2861 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
2862 LCK_EX, MDS_INODELOCK_LOOKUP,
2863 MF_MDC_CANCEL_FID3);
2865 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2866 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2870 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2871 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
2874 rc = md_unlink(tgt->ltd_exp, op_data, request);
2875 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2876 ptlrpc_req_finished(*request);
2884 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2888 /* Not cross-ref case, just get out of here. */
2889 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2892 /* This is a remote object, try remote MDT. */
2893 op_data->op_fid2 = body->mbo_fid1;
2894 ptlrpc_req_finished(*request);
2897 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2899 RETURN(PTR_ERR(tgt));
2904 static int lmv_precleanup(struct obd_device *obd)
2907 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2908 fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2909 lprocfs_obd_cleanup(obd);
2910 lprocfs_free_md_stats(obd);
2915 * Get by key a value associated with a LMV device.
2917 * Dispatch request to lower-layer devices as needed.
2919 * \param[in] env execution environment for this thread
2920 * \param[in] exp export for the LMV device
2921 * \param[in] keylen length of key identifier
2922 * \param[in] key identifier of key to get value for
2923 * \param[in] vallen size of \a val
2924 * \param[out] val pointer to storage location for value
2925 * \param[in] lsm optional striping metadata of object
2927 * \retval 0 on success
2928 * \retval negative negated errno on failure
2930 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2931 __u32 keylen, void *key, __u32 *vallen, void *val)
2933 struct obd_device *obd;
2934 struct lmv_obd *lmv;
2938 obd = class_exp2obd(exp);
2940 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2941 exp->exp_handle.h_cookie);
2946 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2949 LASSERT(*vallen == sizeof(__u32));
2950 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2951 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2953 * All tgts should be connected when this gets called.
2955 if (tgt == NULL || tgt->ltd_exp == NULL)
2958 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2963 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2964 KEY_IS(KEY_DEFAULT_EASIZE) ||
2965 KEY_IS(KEY_CONN_DATA)) {
2967 * Forwarding this request to first MDS, it should know LOV
2970 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2972 if (!rc && KEY_IS(KEY_CONN_DATA))
2973 exp->exp_connect_data = *(struct obd_connect_data *)val;
2975 } else if (KEY_IS(KEY_TGT_COUNT)) {
2976 *((int *)val) = lmv->desc.ld_tgt_count;
2980 CDEBUG(D_IOCTL, "Invalid key\n");
2985 * Asynchronously set by key a value associated with a LMV device.
2987 * Dispatch request to lower-layer devices as needed.
2989 * \param[in] env execution environment for this thread
2990 * \param[in] exp export for the LMV device
2991 * \param[in] keylen length of key identifier
2992 * \param[in] key identifier of key to store value for
2993 * \param[in] vallen size of value to store
2994 * \param[in] val pointer to data to be stored
2995 * \param[in] set optional list of related ptlrpc requests
2997 * \retval 0 on success
2998 * \retval negative negated errno on failure
3000 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3001 __u32 keylen, void *key, __u32 vallen, void *val,
3002 struct ptlrpc_request_set *set)
3004 struct lmv_tgt_desc *tgt = NULL;
3005 struct obd_device *obd;
3006 struct lmv_obd *lmv;
3010 obd = class_exp2obd(exp);
3012 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3013 exp->exp_handle.h_cookie);
3018 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3019 KEY_IS(KEY_DEFAULT_EASIZE)) {
3022 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3025 if (tgt == NULL || tgt->ltd_exp == NULL)
3028 err = obd_set_info_async(env, tgt->ltd_exp,
3029 keylen, key, vallen, val, set);
3040 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3041 const struct lmv_mds_md_v1 *lmm1)
3043 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3050 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3051 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3052 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3053 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3054 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3056 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3057 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3058 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3059 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3060 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3061 sizeof(lsm->lsm_md_pool_name));
3063 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3066 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %#x "
3067 "layout_version %d\n", lsm->lsm_md_stripe_count,
3068 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3069 lsm->lsm_md_layout_version);
3071 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3072 for (i = 0; i < stripe_count; i++) {
3073 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3074 &lmm1->lmv_stripe_fids[i]);
3076 * set default value -1, so lmv_locate_tgt() knows this stripe
3077 * target is not initialized.
3079 lsm->lsm_md_oinfo[i].lmo_mds = (u32)-1;
3080 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3083 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3084 &lsm->lsm_md_oinfo[i].lmo_mds);
3091 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3092 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3098 static inline int lmv_unpack_user_md(struct obd_export *exp,
3099 struct lmv_stripe_md *lsm,
3100 const struct lmv_user_md *lmu)
3102 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3103 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3104 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3105 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3110 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3111 const union lmv_mds_md *lmm, size_t lmm_size)
3113 struct lmv_stripe_md *lsm;
3116 bool allocated = false;
3119 LASSERT(lsmp != NULL);
3123 if (lsm != NULL && lmm == NULL) {
3125 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3127 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3130 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3132 OBD_FREE_LARGE(lfm, lfm_size);
3136 if (lmv_dir_striped(lsm)) {
3137 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3138 if (lsm->lsm_md_oinfo[i].lmo_root)
3139 iput(lsm->lsm_md_oinfo[i].lmo_root);
3141 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3143 lsm_size = lmv_stripe_md_size(0);
3145 OBD_FREE(lsm, lsm_size);
3150 /* foreign lmv case */
3151 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3152 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3155 OBD_ALLOC_LARGE(lfm, lmm_size);
3158 *lsmp = (struct lmv_stripe_md *)lfm;
3160 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3161 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3162 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3163 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3164 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3169 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3173 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3174 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3175 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3176 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3181 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3182 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3185 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3186 * stripecount should be 0 then.
3188 lsm_size = lmv_stripe_md_size(0);
3191 OBD_ALLOC(lsm, lsm_size);
3198 switch (le32_to_cpu(lmm->lmv_magic)) {
3200 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3202 case LMV_USER_MAGIC:
3203 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3206 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3207 le32_to_cpu(lmm->lmv_magic));
3212 if (rc != 0 && allocated) {
3213 OBD_FREE(lsm, lsm_size);
3220 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3222 lmv_unpackmd(NULL, &lsm, NULL, 0);
3224 EXPORT_SYMBOL(lmv_free_memmd);
3226 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3227 union ldlm_policy_data *policy,
3228 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3231 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3236 LASSERT(fid != NULL);
3238 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3239 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3242 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3245 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3253 static int lmv_set_lock_data(struct obd_export *exp,
3254 const struct lustre_handle *lockh,
3255 void *data, __u64 *bits)
3257 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3258 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3262 if (tgt == NULL || tgt->ltd_exp == NULL)
3264 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3268 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3269 const struct lu_fid *fid, enum ldlm_type type,
3270 union ldlm_policy_data *policy,
3271 enum ldlm_mode mode, struct lustre_handle *lockh)
3273 struct obd_device *obd = exp->exp_obd;
3274 struct lmv_obd *lmv = &obd->u.lmv;
3280 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3283 * With DNE every object can have two locks in different namespaces:
3284 * lookup lock in space of MDT storing direntry and update/open lock in
3285 * space of MDT storing inode. Try the MDT that the FID maps to first,
3286 * since this can be easily found, and only try others if that fails.
3288 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3289 i < lmv->desc.ld_tgt_count;
3290 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3292 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3293 obd->obd_name, PFID(fid), tgt);
3297 if (lmv->tgts[tgt] == NULL ||
3298 lmv->tgts[tgt]->ltd_exp == NULL ||
3299 lmv->tgts[tgt]->ltd_active == 0)
3302 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3303 type, policy, mode, lockh);
3311 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3312 struct obd_export *dt_exp, struct obd_export *md_exp,
3313 struct lustre_md *md)
3315 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3316 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3318 if (tgt == NULL || tgt->ltd_exp == NULL)
3321 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3324 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3326 struct obd_device *obd = exp->exp_obd;
3327 struct lmv_obd *lmv = &obd->u.lmv;
3328 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3331 if (md->default_lmv) {
3332 lmv_free_memmd(md->default_lmv);
3333 md->default_lmv = NULL;
3335 if (md->lmv != NULL) {
3336 lmv_free_memmd(md->lmv);
3339 if (tgt == NULL || tgt->ltd_exp == NULL)
3341 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3344 int lmv_set_open_replay_data(struct obd_export *exp,
3345 struct obd_client_handle *och,
3346 struct lookup_intent *it)
3348 struct obd_device *obd = exp->exp_obd;
3349 struct lmv_obd *lmv = &obd->u.lmv;
3350 struct lmv_tgt_desc *tgt;
3353 tgt = lmv_find_target(lmv, &och->och_fid);
3355 RETURN(PTR_ERR(tgt));
3357 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3360 int lmv_clear_open_replay_data(struct obd_export *exp,
3361 struct obd_client_handle *och)
3363 struct obd_device *obd = exp->exp_obd;
3364 struct lmv_obd *lmv = &obd->u.lmv;
3365 struct lmv_tgt_desc *tgt;
3368 tgt = lmv_find_target(lmv, &och->och_fid);
3370 RETURN(PTR_ERR(tgt));
3372 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3375 int lmv_intent_getattr_async(struct obd_export *exp,
3376 struct md_enqueue_info *minfo)
3378 struct md_op_data *op_data = &minfo->mi_data;
3379 struct obd_device *obd = exp->exp_obd;
3380 struct lmv_obd *lmv = &obd->u.lmv;
3381 struct lmv_tgt_desc *tgt = NULL;
3385 if (!fid_is_sane(&op_data->op_fid2))
3388 tgt = lmv_find_target(lmv, &op_data->op_fid1);
3390 RETURN(PTR_ERR(tgt));
3393 * no special handle for remote dir, which needs to fetch both LOOKUP
3394 * lock on parent, and then UPDATE lock on child MDT, which makes all
3395 * complicated because this is done async. So only LOOKUP lock is
3396 * fetched for remote dir, but considering remote dir is rare case,
3397 * and not supporting it in statahead won't cause any issue, just leave
3401 rc = md_intent_getattr_async(tgt->ltd_exp, minfo);
3405 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3406 struct lu_fid *fid, __u64 *bits)
3408 struct obd_device *obd = exp->exp_obd;
3409 struct lmv_obd *lmv = &obd->u.lmv;
3410 struct lmv_tgt_desc *tgt;
3414 tgt = lmv_find_target(lmv, fid);
3416 RETURN(PTR_ERR(tgt));
3418 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3422 int lmv_get_fid_from_lsm(struct obd_export *exp,
3423 const struct lmv_stripe_md *lsm,
3424 const char *name, int namelen, struct lu_fid *fid)
3426 const struct lmv_oinfo *oinfo;
3428 LASSERT(lmv_dir_striped(lsm));
3430 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3432 return PTR_ERR(oinfo);
3434 *fid = oinfo->lmo_fid;
3440 * For lmv, only need to send request to master MDT, and the master MDT will
3441 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3442 * we directly fetch data from the slave MDTs.
3444 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3445 struct obd_quotactl *oqctl)
3447 struct obd_device *obd = class_exp2obd(exp);
3448 struct lmv_obd *lmv = &obd->u.lmv;
3449 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3452 __u64 curspace, curinodes;
3456 tgt->ltd_exp == NULL ||
3458 lmv->desc.ld_tgt_count == 0) {
3459 CERROR("master lmv inactive\n");
3463 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3464 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3468 curspace = curinodes = 0;
3469 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3473 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3476 err = obd_quotactl(tgt->ltd_exp, oqctl);
3478 CERROR("getquota on mdt %d failed. %d\n", i, err);
3482 curspace += oqctl->qc_dqblk.dqb_curspace;
3483 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3486 oqctl->qc_dqblk.dqb_curspace = curspace;
3487 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3492 static int lmv_merge_attr(struct obd_export *exp,
3493 const struct lmv_stripe_md *lsm,
3494 struct cl_attr *attr,
3495 ldlm_blocking_callback cb_blocking)
3500 if (!lmv_dir_striped(lsm))
3503 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3507 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3508 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3514 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3515 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3516 i_size_read(inode), (unsigned long long)inode->i_blocks,
3517 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3518 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3520 /* for slave stripe, it needs to subtract nlink for . and .. */
3522 attr->cat_nlink += inode->i_nlink - 2;
3524 attr->cat_nlink = inode->i_nlink;
3526 attr->cat_size += i_size_read(inode);
3527 attr->cat_blocks += inode->i_blocks;
3529 if (attr->cat_atime < inode->i_atime.tv_sec)
3530 attr->cat_atime = inode->i_atime.tv_sec;
3532 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3533 attr->cat_ctime = inode->i_ctime.tv_sec;
3535 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3536 attr->cat_mtime = inode->i_mtime.tv_sec;
3541 struct obd_ops lmv_obd_ops = {
3542 .o_owner = THIS_MODULE,
3543 .o_setup = lmv_setup,
3544 .o_cleanup = lmv_cleanup,
3545 .o_precleanup = lmv_precleanup,
3546 .o_process_config = lmv_process_config,
3547 .o_connect = lmv_connect,
3548 .o_disconnect = lmv_disconnect,
3549 .o_statfs = lmv_statfs,
3550 .o_get_info = lmv_get_info,
3551 .o_set_info_async = lmv_set_info_async,
3552 .o_notify = lmv_notify,
3553 .o_get_uuid = lmv_get_uuid,
3554 .o_fid_alloc = lmv_fid_alloc,
3555 .o_iocontrol = lmv_iocontrol,
3556 .o_quotactl = lmv_quotactl
3559 struct md_ops lmv_md_ops = {
3560 .m_get_root = lmv_get_root,
3561 .m_null_inode = lmv_null_inode,
3562 .m_close = lmv_close,
3563 .m_create = lmv_create,
3564 .m_enqueue = lmv_enqueue,
3565 .m_getattr = lmv_getattr,
3566 .m_getxattr = lmv_getxattr,
3567 .m_getattr_name = lmv_getattr_name,
3568 .m_intent_lock = lmv_intent_lock,
3570 .m_rename = lmv_rename,
3571 .m_setattr = lmv_setattr,
3572 .m_setxattr = lmv_setxattr,
3573 .m_fsync = lmv_fsync,
3574 .m_file_resync = lmv_file_resync,
3575 .m_read_page = lmv_read_page,
3576 .m_unlink = lmv_unlink,
3577 .m_init_ea_size = lmv_init_ea_size,
3578 .m_cancel_unused = lmv_cancel_unused,
3579 .m_set_lock_data = lmv_set_lock_data,
3580 .m_lock_match = lmv_lock_match,
3581 .m_get_lustre_md = lmv_get_lustre_md,
3582 .m_free_lustre_md = lmv_free_lustre_md,
3583 .m_merge_attr = lmv_merge_attr,
3584 .m_set_open_replay_data = lmv_set_open_replay_data,
3585 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3586 .m_intent_getattr_async = lmv_intent_getattr_async,
3587 .m_revalidate_lock = lmv_revalidate_lock,
3588 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3589 .m_unpackmd = lmv_unpackmd,
3592 static int __init lmv_init(void)
3594 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3595 LUSTRE_LMV_NAME, NULL);
3598 static void __exit lmv_exit(void)
3600 class_unregister_type(LUSTRE_LMV_NAME);
3603 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3604 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3605 MODULE_VERSION(LUSTRE_VERSION_STRING);
3606 MODULE_LICENSE("GPL");
3608 module_init(lmv_init);
3609 module_exit(lmv_exit);