1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003, 2004, 2005, 2006 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 # define EXPORT_SYMTAB
25 #define DEBUG_SUBSYSTEM S_LMV
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/pagemap.h>
33 #include <asm/div64.h>
34 #include <linux/seq_file.h>
35 #include <linux/namei.h>
37 #include <liblustre.h>
40 #include <linux/ext2_fs.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_log.h>
44 #include <obd_support.h>
45 #include <lustre_lib.h>
46 #include <lustre_net.h>
47 #include <obd_class.h>
48 #include <lprocfs_status.h>
49 #include <lustre_lite.h>
50 #include <lustre_fid.h>
51 #include "lmv_internal.h"
53 /* not defined for liblustre building */
54 #if !defined(ATOMIC_INIT)
55 #define ATOMIC_INIT(val) { (val) }
59 cfs_mem_cache_t *obj_cache;
60 atomic_t obj_cache_count = ATOMIC_INIT(0);
62 static void lmv_activate_target(struct lmv_obd *lmv,
63 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);
75 * -EINVAL : UUID can't be found in the LMV's target list
76 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
77 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
79 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
82 struct lmv_tgt_desc *tgt;
83 struct obd_device *obd;
87 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
88 lmv, uuid->uuid, activate);
90 spin_lock(&lmv->lmv_lock);
91 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
92 if (tgt->ltd_exp == NULL)
95 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
96 i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
98 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
102 if (i == lmv->desc.ld_tgt_count)
103 GOTO(out_lmv_lock, rc = -EINVAL);
105 obd = class_exp2obd(tgt->ltd_exp);
107 GOTO(out_lmv_lock, rc = -ENOTCONN);
109 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
110 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
111 obd->obd_type->typ_name, i);
112 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
114 if (tgt->ltd_active == activate) {
115 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
116 activate ? "" : "in");
117 GOTO(out_lmv_lock, rc);
120 CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
121 obd, activate ? "" : "in");
123 lmv_activate_target(lmv, tgt, activate);
128 spin_unlock(&lmv->lmv_lock);
132 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
133 struct obd_connect_data *data)
135 struct lmv_tgt_desc *tgt;
139 LASSERT(data != NULL);
141 spin_lock(&lmv->lmv_lock);
142 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
143 if (tgt->ltd_exp == NULL)
146 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
147 lmv->datas[tgt->ltd_idx] = *data;
151 spin_unlock(&lmv->lmv_lock);
155 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
156 enum obd_notify_event ev, void *data)
158 struct lmv_obd *lmv = &obd->u.lmv;
159 struct obd_uuid *uuid;
163 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
164 CERROR("unexpected notification of %s %s!\n",
165 watched->obd_type->typ_name,
170 uuid = &watched->u.cli.cl_target_uuid;
171 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
173 * Set MDC as active before notifying the observer, so the
174 * observer can use the MDC normally.
176 rc = lmv_set_mdc_active(lmv, uuid,
177 ev == OBD_NOTIFY_ACTIVE);
179 CERROR("%sactivation of %s failed: %d\n",
180 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
184 } else if (ev == OBD_NOTIFY_OCD) {
185 struct obd_connect_data *conn_data =
186 &watched->u.cli.cl_import->imp_connect_data;
188 /* Set connect data to desired target, update
189 * exp_connect_flags. */
190 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
192 CERROR("can't set connect data to target %s, rc %d\n",
198 * XXX: Make sure that ocd_connect_flags from all targets are
199 * the same. Otherwise one of MDTs runs wrong version or
200 * something like this. --umka
202 obd->obd_self_export->exp_connect_flags =
203 conn_data->ocd_connect_flags;
206 else if (ev == OBD_NOTIFY_DISCON) {
207 /* For disconnect event, flush fld cache for failout MDS case. */
208 fld_client_flush(&lmv->lmv_fld);
211 /* Pass the notification up the chain. */
212 if (obd->obd_observer)
213 rc = obd_notify(obd->obd_observer, watched, ev, data);
218 /* this is fake connect function. Its purpose is to initialize lmv and say
219 * caller that everything is okay. Real connection will be performed later. */
220 static int lmv_connect(const struct lu_env *env,
221 struct lustre_handle *conn, struct obd_device *obd,
222 struct obd_uuid *cluuid, struct obd_connect_data *data)
225 struct proc_dir_entry *lmv_proc_dir;
227 struct lmv_obd *lmv = &obd->u.lmv;
228 struct obd_export *exp;
232 rc = class_connect(conn, obd, cluuid);
234 CERROR("class_connection() returned %d\n", rc);
238 exp = class_conn2export(conn);
240 /* we don't want to actually do the underlying connections more than
241 * once, so keep track. */
243 if (lmv->refcount > 1) {
244 class_export_put(exp);
250 lmv->cluuid = *cluuid;
253 lmv->conn_data = *data;
256 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
258 if (IS_ERR(lmv_proc_dir)) {
259 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
260 obd->obd_type->typ_name, obd->obd_name);
265 /* all real clients should perform actual connection right away, because
266 * it is possible, that LMV will not have opportunity to connect targets
267 * and MDC stuff will be called directly, for instance while reading
268 * ../mdc/../kbytesfree procfs file, etc. */
269 if (data->ocd_connect_flags & OBD_CONNECT_REAL)
270 rc = lmv_check_connect(obd);
275 lprocfs_remove(&lmv_proc_dir);
282 static void lmv_set_timeouts(struct obd_device *obd)
284 struct lmv_tgt_desc *tgts;
289 if (lmv->server_timeout == 0)
292 if (lmv->connected == 0)
295 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
296 if (tgts->ltd_exp == NULL)
299 obd_set_info_async(tgts->ltd_exp, strlen("inter_mds"),
300 "inter_mds", 0, NULL, NULL);
304 static int lmv_init_ea_size(struct obd_export *exp, int easize,
305 int def_easize, int cookiesize)
307 struct obd_device *obd = exp->exp_obd;
308 struct lmv_obd *lmv = &obd->u.lmv;
309 int i, rc = 0, change = 0;
312 if (lmv->max_easize < easize) {
313 lmv->max_easize = easize;
316 if (lmv->max_def_easize < def_easize) {
317 lmv->max_def_easize = def_easize;
320 if (lmv->max_cookiesize < cookiesize) {
321 lmv->max_cookiesize = cookiesize;
327 if (lmv->connected == 0)
330 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
331 if (lmv->tgts[i].ltd_exp == NULL) {
332 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
336 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
339 CERROR("obd_init_ea_size() failed on MDT target %d, "
340 "error %d.\n", i, rc);
347 #define MAX_STRING_SIZE 128
349 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
351 struct lmv_obd *lmv = &obd->u.lmv;
352 struct obd_uuid *cluuid = &lmv->cluuid;
353 struct obd_connect_data *mdc_data = NULL;
354 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
355 struct lustre_handle conn = {0, };
356 struct obd_device *mdc_obd;
357 struct obd_export *mdc_exp;
358 struct lu_fld_target target;
361 struct proc_dir_entry *lmv_proc_dir;
365 /* for MDS: don't connect to yourself */
366 if (obd_uuid_equals(&tgt->ltd_uuid, cluuid)) {
367 CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
368 /* XXX - the old code didn't increment active tgt count.
373 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
376 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
380 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
381 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
382 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
385 if (!mdc_obd->obd_set_up) {
386 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
390 rc = obd_connect(NULL, &conn, mdc_obd, &lmv_mdc_uuid,
393 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
397 mdc_exp = class_conn2export(&conn);
399 /* Init fid sequence client for this mdc. */
400 rc = obd_fid_init(mdc_exp);
404 /* Add new FLD target. */
405 target.ft_srv = NULL;
406 target.ft_exp = mdc_exp;
407 target.ft_idx = tgt->ltd_idx;
409 fld_client_add_target(&lmv->lmv_fld, &target);
411 mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
413 rc = obd_register_observer(mdc_obd, obd);
415 obd_disconnect(mdc_exp);
416 CERROR("target %s register_observer error %d\n",
417 tgt->ltd_uuid.uuid, rc);
421 if (obd->obd_observer) {
422 /* tell the mds_lmv about the new target */
423 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
424 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
426 obd_disconnect(mdc_exp);
432 tgt->ltd_exp = mdc_exp;
433 lmv->desc.ld_active_tgt_count++;
435 /* copy connect data, it may be used later */
436 lmv->datas[tgt->ltd_idx] = *mdc_data;
438 md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
439 lmv->max_def_easize, lmv->max_cookiesize);
441 CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
442 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
443 atomic_read(&obd->obd_refcount));
446 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
448 struct proc_dir_entry *mdc_symlink;
449 char name[MAX_STRING_SIZE + 1];
451 LASSERT(mdc_obd->obd_type != NULL);
452 LASSERT(mdc_obd->obd_type->typ_name != NULL);
453 name[MAX_STRING_SIZE] = '\0';
454 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
455 mdc_obd->obd_type->typ_name,
457 mdc_symlink = proc_symlink(mdc_obd->obd_name,
459 if (mdc_symlink == NULL) {
460 CERROR("could not register LMV target "
461 "/proc/fs/lustre/%s/%s/target_obds/%s.",
462 obd->obd_type->typ_name, obd->obd_name,
464 lprocfs_remove(&lmv_proc_dir);
472 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
474 struct lmv_obd *lmv = &obd->u.lmv;
475 struct lmv_tgt_desc *tgt;
479 CDEBUG(D_CONFIG, "tgt_uuid: %s.\n", tgt_uuid->uuid);
483 if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
484 lmv_init_unlock(lmv);
485 CERROR("can't add %s, LMV module compiled for %d MDCs. "
486 "That many MDCs already configured.\n",
487 tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
490 if (lmv->desc.ld_tgt_count == 0) {
491 struct obd_device *mdc_obd;
493 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
496 lmv_init_unlock(lmv);
497 CERROR("Target %s not attached\n", tgt_uuid->uuid);
501 rc = obd_llog_init(obd, NULL, mdc_obd, 0, NULL, tgt_uuid);
503 lmv_init_unlock(lmv);
504 CERROR("lmv failed to setup llogging subsystems\n");
507 spin_lock(&lmv->lmv_lock);
508 tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
509 tgt->ltd_uuid = *tgt_uuid;
510 spin_unlock(&lmv->lmv_lock);
512 if (lmv->connected) {
513 rc = lmv_connect_mdc(obd, tgt);
515 spin_lock(&lmv->lmv_lock);
516 lmv->desc.ld_tgt_count--;
517 memset(tgt, 0, sizeof(*tgt));
518 spin_unlock(&lmv->lmv_lock);
520 int easize = sizeof(struct lmv_stripe_md) +
521 lmv->desc.ld_tgt_count *
522 sizeof(struct lu_fid);
523 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
527 lmv_init_unlock(lmv);
531 /* performs a check if passed obd is connected. If no - connect it. */
532 int lmv_check_connect(struct obd_device *obd)
534 struct lmv_obd *lmv = &obd->u.lmv;
535 struct lmv_tgt_desc *tgt;
543 if (lmv->connected) {
544 lmv_init_unlock(lmv);
548 if (lmv->desc.ld_tgt_count == 0) {
549 CERROR("%s: no targets configured.\n", obd->obd_name);
553 CDEBUG(D_CONFIG, "time to connect %s to %s\n",
554 lmv->cluuid.uuid, obd->obd_name);
556 LASSERT(lmv->tgts != NULL);
558 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
559 rc = lmv_connect_mdc(obd, tgt);
564 lmv_set_timeouts(obd);
565 class_export_put(lmv->exp);
567 easize = lmv_get_easize(lmv);
568 lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
569 lmv_init_unlock(lmv);
578 --lmv->desc.ld_active_tgt_count;
579 rc2 = obd_disconnect(tgt->ltd_exp);
581 CERROR("error: LMV target %s disconnect on "
582 "MDC idx %d: error %d\n",
583 tgt->ltd_uuid.uuid, i, rc2);
587 class_disconnect(lmv->exp);
588 lmv_init_unlock(lmv);
592 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
595 struct proc_dir_entry *lmv_proc_dir;
597 struct lmv_obd *lmv = &obd->u.lmv;
598 struct obd_device *mdc_obd;
602 LASSERT(tgt != NULL);
603 LASSERT(obd != NULL);
605 mdc_obd = class_exp2obd(tgt->ltd_exp);
608 mdc_obd->obd_no_recov = obd->obd_no_recov;
611 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
613 struct proc_dir_entry *mdc_symlink;
615 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
617 lprocfs_remove(&mdc_symlink);
619 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
620 obd->obd_type->typ_name, obd->obd_name,
625 rc = obd_fid_fini(tgt->ltd_exp);
627 CERROR("Can't finanize fids factory\n");
629 CDEBUG(D_OTHER, "Disconnected from %s(%s) successfully\n",
630 tgt->ltd_exp->exp_obd->obd_name,
631 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
633 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
634 rc = obd_disconnect(tgt->ltd_exp);
636 if (tgt->ltd_active) {
637 CERROR("Target %s disconnect error %d\n",
638 tgt->ltd_uuid.uuid, rc);
642 lmv_activate_target(lmv, tgt, 0);
647 static int lmv_disconnect(struct obd_export *exp)
649 struct obd_device *obd = class_exp2obd(exp);
651 struct proc_dir_entry *lmv_proc_dir;
653 struct lmv_obd *lmv = &obd->u.lmv;
660 /* Only disconnect the underlying layers on the final disconnect. */
662 if (lmv->refcount != 0)
665 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
666 if (lmv->tgts[i].ltd_exp == NULL)
668 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
672 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
674 lprocfs_remove(&lmv_proc_dir);
676 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
677 obd->obd_type->typ_name, obd->obd_name);
683 * This is the case when no real connection is established by
684 * lmv_check_connect().
687 class_export_put(exp);
688 rc = class_disconnect(exp);
689 if (lmv->refcount == 0)
694 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
695 int len, void *karg, void *uarg)
697 struct obd_device *obddev = class_exp2obd(exp);
698 struct lmv_obd *lmv = &obddev->u.lmv;
699 int i, rc = 0, set = 0;
702 if (lmv->desc.ld_tgt_count == 0)
705 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
708 if (lmv->tgts[i].ltd_exp == NULL)
711 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
713 if (lmv->tgts[i].ltd_active) {
714 CERROR("error: iocontrol MDC %s on MDT"
715 "idx %d: err = %d\n",
716 lmv->tgts[i].ltd_uuid.uuid, i, err);
734 static int lmv_all_chars_policy(int count, const char *name,
745 static int lmv_nid_policy(struct lmv_obd *lmv)
747 struct obd_import *imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
750 * XXX Hack: to get nid we assume that underlying obd device is mdc.
752 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
753 return id % lmv->desc.ld_tgt_count;
756 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
761 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
763 op_data->op_namelen);
765 return lmv_nid_policy(lmv);
771 CERROR("unsupport type %d \n", type);
775 /* This is _inode_ placement policy function (not name). */
776 static int lmv_placement_policy(struct obd_device *obd,
777 struct md_op_data *op_data,
780 struct lmv_obd *lmv = &obd->u.lmv;
785 LASSERT(mds != NULL);
788 * Allocate new fid on target according to operation type and parent
791 obj = lmv_obj_grab(obd, &op_data->op_fid1);
792 if (obj != NULL || op_data->op_name == NULL ||
793 op_data->op_opc != LUSTRE_OPC_MKDIR) {
795 * Allocate fid for non-dir or for null name or for case parent
802 * If we have this flag turned on, and we see that
803 * parent dir is split, this means, that caller did not
804 * notice split yet. This is race and we would like to
805 * let caller know that.
807 if (op_data->op_bias & MDS_CHECK_SPLIT)
812 * Allocate new fid on same mds where parent fid is located and
813 * where operation will be sent. In case of split dir, ->op_fid1
814 * and ->op_mds here will contain fid and mds of slave directory
815 * object (assigned by caller).
817 *mds = op_data->op_mds;
821 /* XXX: This should be removed later wehn we sure it is not
823 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, mds);
829 * Parent directory is not split and we want to create a
830 * directory in it. Let's calculate where to place it according
833 *mds = lmv_choose_mds(lmv, op_data, NID_TYPE);
841 CERROR("Can't choose MDS, err = %d\n", rc);
843 LASSERT(*mds < lmv->desc.ld_tgt_count);
849 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
852 struct lmv_tgt_desc *tgt = &lmv->tgts[mds];
856 /* New seq alloc and FLD setup should be atomic. */
857 down(&tgt->ltd_fid_sem);
859 /* Asking underlaying tgt layer to allocate new fid. */
860 rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
862 LASSERT(fid_is_sane(fid));
864 /* Client switches to new sequence, setup FLD. */
865 rc = fld_client_create(&lmv->lmv_fld, fid_seq(fid),
868 CERROR("Can't create fld entry, rc %d\n", rc);
869 /* Delete just allocated fid sequence */
870 obd_fid_delete(tgt->ltd_exp, NULL);
873 up(&tgt->ltd_fid_sem);
877 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
878 struct md_op_data *op_data)
880 struct obd_device *obd = class_exp2obd(exp);
881 struct lmv_obd *lmv = &obd->u.lmv;
886 LASSERT(op_data != NULL);
887 LASSERT(fid != NULL);
889 rc = lmv_placement_policy(obd, op_data, &mds);
891 CERROR("Can't get target for allocating fid, "
896 rc = __lmv_fid_alloc(lmv, fid, mds);
898 CERROR("Can't alloc new fid, rc %d\n", rc);
905 static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
910 if (lmv_obj_delete(exp, fid)) {
911 CDEBUG(D_OTHER, "lmv object "DFID" is destroyed.\n",
917 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
919 struct lmv_obd *lmv = &obd->u.lmv;
920 struct lprocfs_static_vars lvars;
921 struct lmv_desc *desc;
925 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
926 CERROR("LMV setup requires a descriptor\n");
930 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
931 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
932 CERROR("descriptor size wrong: %d > %d\n",
933 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
937 lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
939 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
940 if (lmv->tgts == NULL)
943 for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
944 sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
945 lmv->tgts[i].ltd_idx = i;
948 lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
950 OBD_ALLOC(lmv->datas, lmv->datas_size);
951 if (lmv->datas == NULL)
952 GOTO(out_free_tgts, rc = -ENOMEM);
954 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
955 lmv->desc.ld_tgt_count = 0;
956 lmv->desc.ld_active_tgt_count = 0;
957 lmv->max_cookiesize = 0;
958 lmv->max_def_easize = 0;
961 spin_lock_init(&lmv->lmv_lock);
962 sema_init(&lmv->init_sem, 1);
964 rc = lmv_obj_setup(obd);
966 CERROR("Can't setup LMV object manager, "
968 GOTO(out_free_datas, rc);
971 lprocfs_lmv_init_vars(&lvars);
972 lprocfs_obd_setup(obd, lvars.obd_vars);
975 struct proc_dir_entry *entry;
977 entry = create_proc_entry("target_obd_status", 0444,
978 obd->obd_proc_entry);
980 entry->proc_fops = &lmv_proc_target_fops;
985 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
986 LUSTRE_CLI_FLD_HASH_DHT);
988 CERROR("can't init FLD, err %d\n",
990 GOTO(out_free_datas, rc);
996 OBD_FREE(lmv->datas, lmv->datas_size);
999 OBD_FREE(lmv->tgts, lmv->tgts_size);
1004 static int lmv_cleanup(struct obd_device *obd)
1006 struct lmv_obd *lmv = &obd->u.lmv;
1009 fld_client_fini(&lmv->lmv_fld);
1010 lprocfs_obd_cleanup(obd);
1011 lmv_obj_cleanup(obd);
1012 OBD_FREE(lmv->datas, lmv->datas_size);
1013 OBD_FREE(lmv->tgts, lmv->tgts_size);
1018 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1020 struct lustre_cfg *lcfg = buf;
1021 struct obd_uuid tgt_uuid;
1025 switch(lcfg->lcfg_command) {
1027 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1028 GOTO(out, rc = -EINVAL);
1030 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1031 rc = lmv_add_target(obd, &tgt_uuid);
1034 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1035 GOTO(out, rc = -EINVAL);
1042 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1045 struct lmv_obd *lmv = &obd->u.lmv;
1046 struct obd_statfs *temp;
1050 rc = lmv_check_connect(obd);
1054 OBD_ALLOC(temp, sizeof(*temp));
1058 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1059 if (lmv->tgts[i].ltd_exp == NULL)
1062 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
1064 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1065 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1067 GOTO(out_free_temp, rc);
1072 osfs->os_bavail += temp->os_bavail;
1073 osfs->os_blocks += temp->os_blocks;
1074 osfs->os_ffree += temp->os_ffree;
1075 osfs->os_files += temp->os_files;
1081 OBD_FREE(temp, sizeof(*temp));
1085 static int lmv_getstatus(struct obd_export *exp,
1087 struct obd_capa **pc)
1089 struct obd_device *obd = exp->exp_obd;
1090 struct lmv_obd *lmv = &obd->u.lmv;
1094 rc = lmv_check_connect(obd);
1098 rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1103 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1104 struct obd_capa *oc, obd_valid valid, const char *name,
1105 const char *input, int input_size, int output_size,
1106 int flags, struct ptlrpc_request **request)
1108 struct obd_device *obd = exp->exp_obd;
1109 struct lmv_obd *lmv = &obd->u.lmv;
1110 struct obd_export *tgt_exp;
1114 rc = lmv_check_connect(obd);
1118 tgt_exp = lmv_find_export(lmv, fid);
1119 if (IS_ERR(tgt_exp))
1120 RETURN(PTR_ERR(tgt_exp));
1122 rc = md_getxattr(tgt_exp, fid, oc, valid, name, input, input_size,
1123 output_size, flags, request);
1128 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1129 struct obd_capa *oc, obd_valid valid, const char *name,
1130 const char *input, int input_size, int output_size,
1131 int flags, __u32 suppgid,
1132 struct ptlrpc_request **request)
1134 struct obd_device *obd = exp->exp_obd;
1135 struct lmv_obd *lmv = &obd->u.lmv;
1136 struct obd_export *tgt_exp;
1140 rc = lmv_check_connect(obd);
1144 tgt_exp = lmv_find_export(lmv, fid);
1145 if (IS_ERR(tgt_exp))
1146 RETURN(PTR_ERR(tgt_exp));
1148 rc = md_setxattr(tgt_exp, fid, oc, valid, name,
1149 input, input_size, output_size, flags, suppgid,
1155 static int lmv_getattr(struct obd_export *exp, const struct lu_fid *fid,
1156 struct obd_capa *oc, obd_valid valid, int ea_size,
1157 struct ptlrpc_request **request)
1159 struct obd_device *obd = exp->exp_obd;
1160 struct lmv_obd *lmv = &obd->u.lmv;
1161 struct obd_export *tgt_exp;
1162 struct lmv_obj *obj;
1166 rc = lmv_check_connect(obd);
1170 tgt_exp = lmv_find_export(lmv, fid);
1171 if (IS_ERR(tgt_exp))
1172 RETURN(PTR_ERR(tgt_exp));
1174 rc = md_getattr(tgt_exp, fid, oc, valid, ea_size, request);
1178 obj = lmv_obj_grab(obd, fid);
1180 CDEBUG(D_OTHER, "GETATTR for "DFID" %s\n", PFID(fid),
1181 obj ? "(split)" : "");
1184 * If object is split, then we loop over all the slaves and gather size
1185 * attribute. In ideal world we would have to gather also mds field from
1186 * all slaves, as object is spread over the cluster and this is
1187 * definitely interesting information and it is not good to loss it,
1191 struct mdt_body *body;
1193 if (*request == NULL) {
1198 body = req_capsule_server_get(&(*request)->rq_pill,
1200 LASSERT(body != NULL);
1204 for (i = 0; i < obj->lo_objcount; i++) {
1205 if (lmv->tgts[i].ltd_exp == NULL) {
1206 CWARN("%s: NULL export for %d\n",
1211 /* skip master obj. */
1212 if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid))
1215 lmv_update_body(body, &obj->lo_inodes[i]);
1218 lmv_obj_unlock(obj);
1225 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1226 ldlm_iterator_t it, void *data)
1228 struct obd_device *obd = exp->exp_obd;
1229 struct lmv_obd *lmv = &obd->u.lmv;
1233 rc = lmv_check_connect(obd);
1237 CDEBUG(D_OTHER, "CBDATA for "DFID"\n", PFID(fid));
1240 * With CMD every object can have two locks in different namespaces:
1241 * lookup lock in space of mds storing direntry and update/open lock in
1242 * space of mds storing inode.
1244 for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1245 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1250 static int lmv_close(struct obd_export *exp,
1251 struct md_op_data *op_data,
1252 struct md_open_data *mod,
1253 struct ptlrpc_request **request)
1255 struct obd_device *obd = exp->exp_obd;
1256 struct lmv_obd *lmv = &obd->u.lmv;
1257 struct obd_export *tgt_exp;
1261 rc = lmv_check_connect(obd);
1265 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1266 if (IS_ERR(tgt_exp))
1267 RETURN(PTR_ERR(tgt_exp));
1269 CDEBUG(D_OTHER, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1270 rc = md_close(tgt_exp, op_data, mod, request);
1275 * Called in the case MDS returns -ERESTART on create on open, what means that
1276 * directory is split and its LMV presentation object has to be updated.
1278 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1280 struct obd_device *obd = exp->exp_obd;
1281 struct lmv_obd *lmv = &obd->u.lmv;
1282 struct ptlrpc_request *req = NULL;
1283 struct obd_export *tgt_exp;
1284 struct lmv_obj *obj;
1285 struct lustre_md md;
1291 mealen = lmv_get_easize(lmv);
1293 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1295 tgt_exp = lmv_find_export(lmv, fid);
1296 if (IS_ERR(tgt_exp))
1297 RETURN(PTR_ERR(tgt_exp));
1299 /* time to update mea of parent fid */
1300 rc = md_getattr(tgt_exp, fid, NULL, valid, mealen, &req);
1302 CERROR("md_getattr() failed, error %d\n", rc);
1306 rc = md_get_lustre_md(tgt_exp, req, NULL, exp, &md);
1308 CERROR("mdc_get_lustre_md() failed, error %d\n", rc);
1313 GOTO(cleanup, rc = -ENODATA);
1315 obj = lmv_obj_create(exp, fid, md.mea);
1321 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
1326 ptlrpc_req_finished(req);
1330 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1331 const void *data, int datalen, int mode, __u32 uid,
1332 __u32 gid, __u32 cap_effective, __u64 rdev,
1333 struct ptlrpc_request **request)
1335 struct obd_device *obd = exp->exp_obd;
1336 struct lmv_obd *lmv = &obd->u.lmv;
1337 struct obd_export *tgt_exp;
1338 struct lmv_obj *obj;
1342 rc = lmv_check_connect(obd);
1346 if (!lmv->desc.ld_active_tgt_count)
1351 obj = lmv_obj_grab(obd, &op_data->op_fid1);
1355 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1356 op_data->op_name, op_data->op_namelen);
1357 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1358 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1359 op_data->op_mds = obj->lo_inodes[mea_idx].li_mds;
1360 tgt_exp = lmv_get_export(lmv, op_data->op_mds);
1363 struct lmv_tgt_desc *tgt;
1365 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1366 op_data->op_bias |= MDS_CHECK_SPLIT;
1367 op_data->op_mds = tgt->ltd_idx;
1368 tgt_exp = tgt->ltd_exp;
1371 if (IS_ERR(tgt_exp))
1372 RETURN(PTR_ERR(tgt_exp));
1374 rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1375 if (rc == -ERESTART)
1380 CDEBUG(D_OTHER, "CREATE '%*s' on "DFID"\n", op_data->op_namelen,
1381 op_data->op_name, PFID(&op_data->op_fid1));
1383 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1384 rc = md_create(tgt_exp, op_data, data, datalen, mode, uid, gid,
1385 cap_effective, rdev, request);
1387 if (*request == NULL)
1389 CDEBUG(D_OTHER, "created - "DFID"\n", PFID(&op_data->op_fid1));
1390 } else if (rc == -ERESTART) {
1391 LASSERT(*request != NULL);
1392 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1393 "Got -ERESTART during create!\n");
1394 ptlrpc_req_finished(*request);
1398 * Directory got split. Time to update local object and repeat
1399 * the request with proper MDS.
1401 rc = lmv_handle_split(exp, &op_data->op_fid1);
1403 rc = lmv_alloc_slave_fids(obd, &op_data->op_fid1,
1404 op_data, &op_data->op_fid2);
1413 static int lmv_done_writing(struct obd_export *exp,
1414 struct md_op_data *op_data,
1415 struct md_open_data *mod)
1417 struct obd_device *obd = exp->exp_obd;
1418 struct lmv_obd *lmv = &obd->u.lmv;
1419 struct obd_export *tgt_exp;
1423 rc = lmv_check_connect(obd);
1427 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1428 if (IS_ERR(tgt_exp))
1429 RETURN(PTR_ERR(tgt_exp));
1431 rc = md_done_writing(tgt_exp, op_data, mod);
1436 lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1437 struct lookup_intent *it, struct md_op_data *op_data,
1438 struct lustre_handle *lockh, void *lmm, int lmmsize)
1440 struct obd_device *obd = exp->exp_obd;
1441 struct lmv_obd *lmv = &obd->u.lmv;
1442 struct lmv_stripe_md *mea = op_data->op_mea1;
1443 struct md_op_data *op_data2;
1444 struct obd_export *tgt_exp;
1448 OBD_ALLOC_PTR(op_data2);
1449 if (op_data2 == NULL)
1452 LASSERT(mea != NULL);
1453 for (i = 0; i < mea->mea_count; i++) {
1454 memset(op_data2, 0, sizeof(*op_data2));
1455 op_data2->op_fid1 = mea->mea_ids[i];
1456 op_data2->op_bias = 0;
1458 tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
1459 if (IS_ERR(tgt_exp))
1460 GOTO(cleanup, rc = PTR_ERR(tgt_exp));
1462 if (tgt_exp == NULL)
1465 rc = md_enqueue(tgt_exp, einfo, it, op_data2,
1466 lockh + i, lmm, lmmsize, 0);
1468 CDEBUG(D_OTHER, "take lock on slave "DFID" -> %d/%d\n",
1469 PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1474 if (it->d.lustre.it_data) {
1475 struct ptlrpc_request *req;
1476 req = (struct ptlrpc_request *)it->d.lustre.it_data;
1477 ptlrpc_req_finished(req);
1480 if (it->d.lustre.it_status)
1481 GOTO(cleanup, rc = it->d.lustre.it_status);
1486 OBD_FREE_PTR(op_data2);
1489 /* drop all taken locks */
1491 if (lockh[i].cookie)
1492 ldlm_lock_decref(lockh + i, einfo->ei_mode);
1493 lockh[i].cookie = 0;
1500 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1501 struct lookup_intent *it, struct md_op_data *op_data,
1502 struct lustre_handle *lockh, void *lmm, int lmmsize,
1503 int extra_lock_flags)
1505 struct ptlrpc_request *req = it->d.lustre.it_data;
1506 struct obd_device *obd = exp->exp_obd;
1507 struct lmv_obd *lmv = &obd->u.lmv;
1508 struct lustre_handle plock;
1509 struct obd_export *tgt_exp;
1510 struct md_op_data *rdata;
1511 struct lu_fid fid_copy;
1512 struct mdt_body *body;
1516 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1517 LASSERT(body != NULL);
1519 if (!(body->valid & OBD_MD_MDS))
1522 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID" -> "DFID"\n",
1523 LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1525 /* We got LOOKUP lock, but we really need attrs */
1526 pmode = it->d.lustre.it_lock_mode;
1527 LASSERT(pmode != 0);
1528 memcpy(&plock, lockh, sizeof(plock));
1529 it->d.lustre.it_lock_mode = 0;
1530 it->d.lustre.it_data = NULL;
1531 fid_copy = body->fid1;
1533 it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1534 ptlrpc_req_finished(req);
1536 tgt_exp = lmv_find_export(lmv, &fid_copy);
1537 if (IS_ERR(tgt_exp))
1538 GOTO(out, rc = PTR_ERR(tgt_exp));
1540 OBD_ALLOC_PTR(rdata);
1542 GOTO(out, rc = -ENOMEM);
1544 rdata->op_fid1 = fid_copy;
1545 rdata->op_bias = MDS_CROSS_REF;
1547 rc = md_enqueue(tgt_exp, einfo, it, rdata, lockh,
1548 lmm, lmmsize, extra_lock_flags);
1549 OBD_FREE_PTR(rdata);
1552 ldlm_lock_decref(&plock, pmode);
1557 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1558 struct lookup_intent *it, struct md_op_data *op_data,
1559 struct lustre_handle *lockh, void *lmm, int lmmsize,
1560 int extra_lock_flags)
1562 struct obd_device *obd = exp->exp_obd;
1563 struct lmv_obd *lmv = &obd->u.lmv;
1564 struct obd_export *tgt_exp = NULL;
1565 struct lmv_obj *obj;
1569 rc = lmv_check_connect(obd);
1573 if (op_data->op_mea1 && it->it_op == IT_UNLINK) {
1574 rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
1575 lockh, lmm, lmmsize);
1579 if (op_data->op_namelen) {
1580 obj = lmv_obj_grab(obd, &op_data->op_fid1);
1584 /* directory is split. look for right mds for this
1586 mea_idx = raw_name2idx(obj->lo_hashtype,
1588 (char *)op_data->op_name,
1589 op_data->op_namelen);
1590 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1591 tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
1596 if (tgt_exp == NULL)
1597 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1598 if (IS_ERR(tgt_exp))
1599 RETURN(PTR_ERR(tgt_exp));
1601 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID"\n", LL_IT2STR(it),
1602 PFID(&op_data->op_fid1));
1604 rc = md_enqueue(tgt_exp, einfo, it, op_data, lockh,
1605 lmm, lmmsize, extra_lock_flags);
1607 if (rc == 0 && it->it_op == IT_OPEN)
1608 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1609 lmm, lmmsize, extra_lock_flags);
1614 lmv_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
1615 struct obd_capa *oc, const char *filename, int namelen,
1616 obd_valid valid, int ea_size, __u32 suppgid,
1617 struct ptlrpc_request **request)
1619 struct obd_device *obd = exp->exp_obd;
1620 struct lmv_obd *lmv = &obd->u.lmv;
1621 struct lu_fid rid = *fid;
1622 struct obd_export *tgt_exp;
1623 struct mdt_body *body;
1624 struct lmv_obj *obj;
1628 rc = lmv_check_connect(obd);
1635 obj = lmv_obj_grab(obd, &rid);
1639 /* Directory is split. Look for right mds for this name */
1640 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1641 filename, namelen - 1);
1642 rid = obj->lo_inodes[mea_idx].li_fid;
1643 tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
1645 valid &= ~OBD_MD_FLCKSPLIT;
1647 tgt_exp = lmv_find_export(lmv, &rid);
1648 valid |= OBD_MD_FLCKSPLIT;
1650 if (IS_ERR(tgt_exp))
1651 RETURN(PTR_ERR(tgt_exp));
1653 CDEBUG(D_OTHER, "getattr_name for %*s on "DFID" -> "DFID"\n",
1654 namelen, filename, PFID(fid), PFID(&rid));
1656 rc = md_getattr_name(tgt_exp, &rid, oc, filename, namelen, valid,
1657 ea_size, suppgid, request);
1659 body = req_capsule_server_get(&(*request)->rq_pill,
1661 LASSERT(body != NULL);
1663 if (body->valid & OBD_MD_MDS) {
1664 struct ptlrpc_request *req = NULL;
1667 CDEBUG(D_OTHER, "request attrs for "DFID"\n",
1670 tgt_exp = lmv_find_export(lmv, &rid);
1671 if (IS_ERR(tgt_exp)) {
1672 ptlrpc_req_finished(*request);
1673 RETURN(PTR_ERR(tgt_exp));
1676 rc = md_getattr_name(tgt_exp, &rid, NULL, NULL, 1,
1677 valid | OBD_MD_FLCROSSREF,
1678 ea_size, suppgid, &req);
1679 ptlrpc_req_finished(*request);
1682 } else if (rc == -ERESTART) {
1683 LASSERT(*request != NULL);
1684 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1685 "Got -ERESTART during getattr!\n");
1686 ptlrpc_req_finished(*request);
1690 * Directory got split. Time to update local object and repeat
1691 * the request with proper MDS.
1693 rc = lmv_handle_split(exp, &rid);
1700 #define md_op_data_fid(op_data, fl) \
1701 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1702 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1703 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1704 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1707 /* @tgt_exp is the export the metadata request is sent.
1708 * @fid_exp is the export the cancel should be sent for the current fid.
1709 * if @fid_exp is NULL, the export is found for the current fid.
1710 * @op_data keeps the current fid, which is pointed through @flag.
1711 * @mode, @bits -- lock match parameters. */
1712 static int lmv_early_cancel(struct lmv_obd *lmv, struct obd_export *tgt_exp,
1713 struct obd_export *fid_exp,
1714 struct md_op_data *op_data,
1715 ldlm_mode_t mode, int bits, int flag)
1717 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1718 ldlm_policy_data_t policy = {{0}};
1722 if (!fid_is_sane(fid))
1725 if (fid_exp == NULL)
1726 fid_exp = lmv_find_export(lmv, fid);
1728 if (tgt_exp == fid_exp) {
1729 /* The export is the same as on the target server, cancel
1730 * will be sent along with the main metadata operation. */
1731 op_data->op_flags |= flag;
1735 policy.l_inodebits.bits = bits;
1736 rc = md_cancel_unused(fid_exp, fid, &policy, mode, LDLM_FL_ASYNC, NULL);
1740 #ifdef EARLY_CANCEL_FOR_STRIPED_DIR_IS_READY
1741 /* Check if the fid in @op_data pointed to by flag is of the same export(s)
1742 * as @tgt_exp. Early cancels will be sent later by mdc code, otherwise, call
1743 * md_cancel_unused for child export(s). */
1744 static int lmv_early_cancel_stripes(struct obd_export *exp,
1745 struct obd_export *tgt_exp,
1746 struct md_op_data *op_data,
1747 ldlm_mode_t mode, int bits, int flag)
1749 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1750 struct obd_device *obd = exp->exp_obd;
1751 struct lmv_obd *lmv = &obd->u.lmv;
1752 struct obd_export *st_exp;
1753 struct lmv_obj *obj;
1757 if (!fid_is_sane(fid))
1760 obj = lmv_obj_grab(obd, fid);
1762 ldlm_policy_data_t policy = {{0}};
1763 struct lu_fid *st_fid;
1766 policy.l_inodebits.bits = bits;
1767 for (i = 0; i < obj->lo_objcount; i++) {
1768 st_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
1769 st_fid = &obj->lo_inodes[i].li_fid;
1770 if (tgt_exp != st_exp) {
1771 rc = md_cancel_unused(st_exp, st_fid, &policy,
1772 mode, LDLM_FL_ASYNC,
1777 /* Some export matches to @tgt_exp, do cancel
1778 * for its fid in mdc */
1780 op_data->op_flags |= flag;
1785 rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data,
1793 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1796 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1797 struct ptlrpc_request **request)
1799 struct obd_device *obd = exp->exp_obd;
1800 struct lmv_obd *lmv = &obd->u.lmv;
1801 struct obd_export *tgt_exp;
1802 struct lmv_obj *obj;
1807 rc = lmv_check_connect(obd);
1814 if (op_data->op_namelen != 0) {
1817 /* Usual link request */
1818 obj = lmv_obj_grab(obd, &op_data->op_fid2);
1820 mea_idx = raw_name2idx(obj->lo_hashtype,
1823 op_data->op_namelen);
1824 op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1825 mds = obj->lo_inodes[mea_idx].li_mds;
1828 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
1833 CDEBUG(D_OTHER,"link "DFID":%*s to "DFID"\n",
1834 PFID(&op_data->op_fid2), op_data->op_namelen,
1835 op_data->op_name, PFID(&op_data->op_fid1));
1837 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
1841 /* request from MDS to acquire i_links for inode by fid1 */
1842 CDEBUG(D_OTHER, "inc i_nlinks for "DFID"\n",
1843 PFID(&op_data->op_fid1));
1846 CDEBUG(D_OTHER, "forward to MDS #"LPU64" ("DFID")\n",
1847 mds, PFID(&op_data->op_fid1));
1849 op_data->op_fsuid = current->fsuid;
1850 op_data->op_fsgid = current->fsgid;
1851 op_data->op_cap = current->cap_effective;
1853 tgt_exp = lmv->tgts[mds].ltd_exp;
1854 if (op_data->op_namelen) {
1855 op_data->op_flags |= MF_MDC_CANCEL_FID2;
1856 /* Cancel UPDATE lock on child (fid1). */
1857 rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data, LCK_EX,
1858 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1861 rc = md_link(tgt_exp, op_data, request);
1862 if (rc == -ERESTART) {
1863 LASSERT(*request != NULL);
1864 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1865 "Got -ERESTART during link!\n");
1866 ptlrpc_req_finished(*request);
1870 * Directory got split. Time to update local object and repeat
1871 * the request with proper MDS.
1873 rc = lmv_handle_split(exp, &op_data->op_fid2);
1881 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1882 const char *old, int oldlen, const char *new, int newlen,
1883 struct ptlrpc_request **request)
1885 struct obd_export *tgt_exp = NULL, *src_exp;
1886 struct obd_device *obd = exp->exp_obd;
1887 struct lmv_obd *lmv = &obd->u.lmv;
1888 int rc, mea_idx, loop = 0;
1889 struct lmv_obj *obj;
1893 CDEBUG(D_OTHER, "rename %*s in "DFID" to %*s in "DFID"\n",
1894 oldlen, old, PFID(&op_data->op_fid1),
1895 newlen, new, PFID(&op_data->op_fid2));
1897 rc = lmv_check_connect(obd);
1903 * MDS with old dir entry is asking another MDS to create name
1907 "create %*s(%d/%d) in "DFID" pointing "
1908 "to "DFID"\n", newlen, new, oldlen, newlen,
1909 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1));
1911 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds1);
1916 * Target directory can be split, sowe should forward request to
1919 obj = lmv_obj_grab(obd, &op_data->op_fid2);
1921 mea_idx = raw_name2idx(obj->lo_hashtype,
1923 (char *)new, newlen);
1924 op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1925 CDEBUG(D_OTHER, "Parent obj "DFID"\n",
1926 PFID(&op_data->op_fid2));
1935 obj = lmv_obj_grab(obd, &op_data->op_fid1);
1938 * directory is already split, so we have to forward request to
1941 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1942 (char *)old, oldlen);
1943 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1944 mds1 = obj->lo_inodes[mea_idx].li_mds;
1945 CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
1948 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
1953 obj = lmv_obj_grab(obd, &op_data->op_fid2);
1956 * Directory is already split, so we have to forward request to
1959 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1960 (char *)new, newlen);
1962 mds2 = obj->lo_inodes[mea_idx].li_mds;
1963 op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1964 CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
1967 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
1973 op_data->op_fsuid = current->fsuid;
1974 op_data->op_fsgid = current->fsgid;
1975 op_data->op_cap = current->cap_effective;
1977 src_exp = lmv_get_export(lmv, mds1);
1978 tgt_exp = lmv_get_export(lmv, mds2);
1980 /* LOOKUP lock on src child (fid3) should also be cancelled for
1981 * src_exp in mdc_rename. */
1982 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1984 /* Cancel UPDATE locks on tgt parent (fid2), tgt_exp is its
1986 rc = lmv_early_cancel(lmv, src_exp, tgt_exp, op_data, LCK_EX,
1987 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
1989 /* Cancel LOOKUP locks on tgt child (fid4) for parent tgt_exp.*/
1991 rc = lmv_early_cancel(lmv, src_exp, tgt_exp, op_data,
1992 LCK_EX, MDS_INODELOCK_LOOKUP,
1993 MF_MDC_CANCEL_FID4);
1995 /* XXX: the case when child is a striped dir is not supported.
1996 * Only the master stripe has all locks cancelled early. */
1997 /* Cancel all the locks on tgt child (fid4). */
1999 rc = lmv_early_cancel(lmv, src_exp, NULL, op_data,
2000 LCK_EX, MDS_INODELOCK_FULL,
2001 MF_MDC_CANCEL_FID4);
2005 rc = md_rename(src_exp, op_data, old, oldlen,
2006 new, newlen, request);
2007 if (rc == -ERESTART) {
2008 LASSERT(*request != NULL);
2009 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2010 "Got -ERESTART during rename!\n");
2011 ptlrpc_req_finished(*request);
2015 * Directory got split. Time to update local object and repeat
2016 * the request with proper MDS.
2018 rc = lmv_handle_split(exp, &op_data->op_fid1);
2025 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2026 void *ea, int ealen, void *ea2, int ea2len,
2027 struct ptlrpc_request **request,
2028 struct md_open_data **mod)
2030 struct obd_device *obd = exp->exp_obd;
2031 struct lmv_obd *lmv = &obd->u.lmv;
2032 struct ptlrpc_request *req;
2033 struct obd_export *tgt_exp;
2034 struct lmv_obj *obj;
2038 rc = lmv_check_connect(obd);
2042 obj = lmv_obj_grab(obd, &op_data->op_fid1);
2044 CDEBUG(D_OTHER, "SETATTR for "DFID", valid 0x%x%s\n",
2045 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2046 obj ? ", split" : "");
2048 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2050 for (i = 0; i < obj->lo_objcount; i++) {
2051 op_data->op_fid1 = obj->lo_inodes[i].li_fid;
2053 tgt_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
2054 if (IS_ERR(tgt_exp)) {
2055 rc = PTR_ERR(tgt_exp);
2059 rc = md_setattr(tgt_exp, op_data, ea, ealen,
2060 ea2, ea2len, &req, mod);
2062 if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid)) {
2064 * this is master object and this request should
2065 * be returned back to llite.
2069 ptlrpc_req_finished(req);
2077 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
2078 if (IS_ERR(tgt_exp))
2079 RETURN(PTR_ERR(tgt_exp));
2081 rc = md_setattr(tgt_exp, op_data, ea, ealen, ea2,
2082 ea2len, request, mod);
2087 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2088 struct obd_capa *oc, struct ptlrpc_request **request)
2090 struct obd_device *obd = exp->exp_obd;
2091 struct lmv_obd *lmv = &obd->u.lmv;
2092 struct obd_export *tgt_exp;
2096 rc = lmv_check_connect(obd);
2100 tgt_exp = lmv_find_export(lmv, fid);
2101 if (IS_ERR(tgt_exp))
2102 RETURN(PTR_ERR(tgt_exp));
2104 rc = md_sync(tgt_exp, fid, oc, request);
2108 /* main purpose of LMV blocking ast is to remove split directory LMV
2109 * presentation object (struct lmv_obj) attached to the lock being revoked. */
2110 int lmv_blocking_ast(struct ldlm_lock *lock,
2111 struct ldlm_lock_desc *desc,
2112 void *data, int flag)
2114 struct lustre_handle lockh;
2115 struct lmv_obj *obj;
2120 case LDLM_CB_BLOCKING:
2121 ldlm_lock2handle(lock, &lockh);
2122 rc = ldlm_cli_cancel(&lockh);
2124 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2128 case LDLM_CB_CANCELING:
2129 /* time to drop cached attrs for dirobj */
2130 obj = lock->l_ast_data;
2132 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
2133 ", master "DFID"\n",
2134 lock->l_resource->lr_name.name[3] == 1 ?
2135 "LOOKUP" : "UPDATE",
2136 lock->l_resource->lr_name.name[0],
2137 lock->l_resource->lr_name.name[1],
2138 PFID(&obj->lo_fid));
2148 static void lmv_hash_adjust(__u32 *hash, __u32 hash_adj)
2152 val = le32_to_cpu(*hash);
2154 val += MAX_HASH_SIZE;
2155 if (val != DIR_END_OFF)
2156 *hash = cpu_to_le32(val - hash_adj);
2159 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2162 struct obd_import *imp;
2165 * XXX Hack: to get nid we assume that underlying obd device is mdc.
2167 imp = class_exp2cliimp(exp);
2168 id = imp->imp_connection->c_self + fid_flatten(fid);
2170 CDEBUG(D_INFO, "node rank: %llx "DFID" %llx %llx\n",
2171 imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2173 return id ^ (id >> 32);
2176 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2177 struct obd_capa *oc, __u64 offset64, struct page *page,
2178 struct ptlrpc_request **request)
2180 struct obd_device *obd = exp->exp_obd;
2181 struct lmv_obd *lmv = &obd->u.lmv;
2182 struct obd_export *tgt_exp;
2183 struct lu_fid rid = *fid;
2184 struct lmv_obj *obj;
2196 offset0 = offset = offset64;
2198 * Check that offset is representable by 32bit number.
2200 LASSERT((__u64)offset == offset64);
2202 rc = lmv_check_connect(obd);
2206 CDEBUG(D_INFO, "READPAGE at %x from "DFID"\n", offset, PFID(&rid));
2208 obj = lmv_obj_grab(obd, fid);
2210 struct lmv_inode *loi;
2214 nr = obj->lo_objcount;
2216 seg_size = MAX_HASH_SIZE / nr;
2217 loi = obj->lo_inodes;
2218 rank = lmv_node_rank(lmv_get_export(lmv, loi[0].li_mds),
2220 tgt0 = (offset / seg_size) % nr;
2221 tgt = (tgt0 + rank) % nr;
2227 * Last segment has unusual length due to division
2230 hash_adj = MAX_HASH_SIZE - seg_size * nr;
2234 hash_adj += rank * seg_size;
2236 CDEBUG(D_INFO, "hash_adj: %x %x %x/%x -> %x/%x\n",
2237 rank, hash_adj, offset, tgt0, offset + hash_adj, tgt);
2239 offset = (offset + hash_adj) % MAX_HASH_SIZE;
2240 rid = obj->lo_inodes[tgt].li_fid;
2241 tgt_exp = lmv_get_export(lmv, loi[tgt].li_mds);
2243 CDEBUG(D_INFO, "forward to "DFID" with offset %lu i %d\n",
2244 PFID(&rid), (unsigned long)offset, tgt);
2246 tgt_exp = lmv_find_export(lmv, &rid);
2248 if (IS_ERR(tgt_exp))
2249 GOTO(cleanup, rc = PTR_ERR(tgt_exp));
2251 rc = md_readpage(tgt_exp, &rid, oc, offset, page, request);
2255 struct lu_dirpage *dp;
2256 struct lu_dirent *ent;
2258 dp = cfs_kmap(page);
2260 lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2261 lmv_hash_adjust(&dp->ldp_hash_end, hash_adj);
2262 LASSERT(cpu_to_le32(dp->ldp_hash_start) <= offset0);
2264 for (ent = lu_dirent_start(dp); ent != NULL;
2265 ent = lu_dirent_next(ent))
2266 lmv_hash_adjust(&ent->lde_hash, hash_adj);
2268 if (tgt0 != nr - 1) {
2271 end = le32_to_cpu(dp->ldp_hash_end);
2272 if (end == DIR_END_OFF) {
2273 dp->ldp_hash_end = cpu_to_le32(seg_size *
2275 CDEBUG(D_INFO, ""DFID" reset end %x tgt %d\n",
2277 le32_to_cpu(dp->ldp_hash_end), tgt);
2283 * Here we could remove "." and ".." from all pages which at not from
2284 * master. But MDS has only "." and ".." for master dir.
2289 lmv_obj_unlock(obj);
2295 static int lmv_unlink_slaves(struct obd_export *exp,
2296 struct md_op_data *op_data,
2297 struct ptlrpc_request **req)
2299 struct obd_device *obd = exp->exp_obd;
2300 struct lmv_obd *lmv = &obd->u.lmv;
2301 struct lmv_stripe_md *mea = op_data->op_mea1;
2302 struct md_op_data *op_data2;
2303 struct obd_export *tgt_exp;
2307 OBD_ALLOC_PTR(op_data2);
2308 if (op_data2 == NULL)
2311 op_data2->op_mode = S_IFDIR;
2312 op_data2->op_fsuid = current->fsuid;
2313 op_data2->op_fsgid = current->fsgid;
2314 op_data2->op_bias = 0;
2316 LASSERT(mea != NULL);
2317 for (i = 0; i < mea->mea_count; i++) {
2318 memset(op_data2, 0, sizeof(*op_data2));
2319 op_data2->op_fid1 = mea->mea_ids[i];
2320 tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
2321 if (IS_ERR(tgt_exp))
2322 GOTO(out_free_op_data2, rc = PTR_ERR(tgt_exp));
2324 if (tgt_exp == NULL)
2327 rc = md_unlink(tgt_exp, op_data2, req);
2329 CDEBUG(D_OTHER, "unlink slave "DFID" -> %d\n",
2330 PFID(&mea->mea_ids[i]), rc);
2333 ptlrpc_req_finished(*req);
2337 GOTO(out_free_op_data2, rc);
2342 OBD_FREE_PTR(op_data2);
2346 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2347 struct ptlrpc_request **request)
2349 struct obd_device *obd = exp->exp_obd;
2350 struct lmv_obd *lmv = &obd->u.lmv;
2351 struct obd_export *tgt_exp = NULL;
2352 struct lmv_obj *obj;
2356 rc = lmv_check_connect(obd);
2360 if (op_data->op_namelen == 0 && op_data->op_mea1 != NULL) {
2361 /* mds asks to remove slave objects */
2362 rc = lmv_unlink_slaves(exp, op_data, request);
2369 if (op_data->op_namelen != 0) {
2372 obj = lmv_obj_grab(obd, &op_data->op_fid1);
2374 mea_idx = raw_name2idx(obj->lo_hashtype,
2377 op_data->op_namelen);
2378 op_data->op_bias &= ~MDS_CHECK_SPLIT;
2379 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
2380 tgt_exp = lmv_get_export(lmv,
2381 obj->lo_inodes[mea_idx].li_mds);
2383 CDEBUG(D_OTHER, "unlink '%*s' in "DFID" -> %u\n",
2384 op_data->op_namelen, op_data->op_name,
2385 PFID(&op_data->op_fid1), mea_idx);
2388 CDEBUG(D_OTHER, "drop i_nlink on "DFID"\n",
2389 PFID(&op_data->op_fid1));
2391 if (tgt_exp == NULL) {
2392 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
2393 if (IS_ERR(tgt_exp))
2394 RETURN(PTR_ERR(tgt_exp));
2395 op_data->op_bias |= MDS_CHECK_SPLIT;
2398 op_data->op_fsuid = current->fsuid;
2399 op_data->op_fsgid = current->fsgid;
2400 op_data->op_cap = current->cap_effective;
2402 /* If child's fid is given, cancel unused locks for it if it is from
2403 * another export than parent. */
2404 if (op_data->op_namelen) {
2405 /* LOOKUP lock for child (fid3) should also be cancelled on
2406 * parent tgt_exp in mdc_unlink(). */
2407 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2409 /* XXX: the case when child is a striped dir is not supported.
2410 * Only the master stripe has all locks cancelled early. */
2411 /* Cancel FULL locks on child (fid3). */
2412 rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data, LCK_EX,
2413 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2416 rc = md_unlink(tgt_exp, op_data, request);
2417 if (rc == -ERESTART) {
2418 LASSERT(*request != NULL);
2419 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2420 "Got -ERESTART during unlink!\n");
2421 ptlrpc_req_finished(*request);
2425 * Directory got split. Time to update local object and repeat
2426 * the request with proper MDS.
2428 rc = lmv_handle_split(exp, &op_data->op_fid1);
2435 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs* llogs,
2436 struct obd_device *tgt, int count,
2437 struct llog_catid *logid, struct obd_uuid *uuid)
2439 struct llog_ctxt *ctxt;
2443 rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
2446 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
2447 ctxt->loc_imp = tgt->u.cli.cl_import;
2453 static int lmv_llog_finish(struct obd_device *obd, int count)
2458 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
2462 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2467 case OBD_CLEANUP_EARLY:
2468 /* XXX: here should be calling obd_precleanup() down to
2471 case OBD_CLEANUP_SELF_EXP:
2472 rc = obd_llog_finish(obd, 0);
2474 CERROR("failed to cleanup llogging subsystems\n");
2482 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2483 void *key, __u32 *vallen, void *val)
2485 struct obd_device *obd;
2486 struct lmv_obd *lmv;
2490 obd = class_exp2obd(exp);
2492 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2493 exp->exp_handle.h_cookie);
2498 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2499 struct lmv_tgt_desc *tgts;
2502 rc = lmv_check_connect(obd);
2506 LASSERT(*vallen == sizeof(__u32));
2507 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2510 /* all tgts should be connected when this get called. */
2511 if (!tgts || !tgts->ltd_exp) {
2512 CERROR("target not setup?\n");
2516 if (!obd_get_info(tgts->ltd_exp, keylen, key,
2521 } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2522 rc = lmv_check_connect(obd);
2526 /* forwarding this request to first MDS, it should know LOV
2528 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2530 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2531 exp->exp_connect_flags =
2532 ((struct obd_connect_data *)val)->ocd_connect_flags;
2537 CDEBUG(D_IOCTL, "invalid key\n");
2541 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2542 void *key, obd_count vallen, void *val,
2543 struct ptlrpc_request_set *set)
2545 struct lmv_tgt_desc *tgt;
2546 struct obd_device *obd;
2547 struct lmv_obd *lmv;
2551 obd = class_exp2obd(exp);
2553 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2554 exp->exp_handle.h_cookie);
2559 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2560 KEY_IS(KEY_INIT_RECOV_BACKUP)) {
2563 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2564 tgt = &lmv->tgts[i];
2569 err = obd_set_info_async(tgt->ltd_exp,
2570 keylen, key, vallen, val, set);
2581 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2582 struct lov_stripe_md *lsm)
2584 struct obd_device *obd = class_exp2obd(exp);
2585 struct lmv_obd *lmv = &obd->u.lmv;
2586 struct lmv_stripe_md *meap, *lsmp;
2590 mea_size = lmv_get_easize(lmv);
2594 if (*lmmp && !lsm) {
2595 OBD_FREE(*lmmp, mea_size);
2600 if (*lmmp == NULL) {
2601 OBD_ALLOC(*lmmp, mea_size);
2609 lsmp = (struct lmv_stripe_md *)lsm;
2610 meap = (struct lmv_stripe_md *)*lmmp;
2612 if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2613 lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2616 meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2617 meap->mea_count = cpu_to_le32(lsmp->mea_count);
2618 meap->mea_master = cpu_to_le32(lsmp->mea_master);
2620 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2621 meap->mea_ids[i] = meap->mea_ids[i];
2622 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2628 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2629 struct lov_mds_md *lmm, int lmm_size)
2631 struct obd_device *obd = class_exp2obd(exp);
2632 struct lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp;
2633 struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm;
2634 struct lmv_obd *lmv = &obd->u.lmv;
2639 mea_size = lmv_get_easize(lmv);
2643 if (*lsmp != NULL && lmm == NULL) {
2644 OBD_FREE(*tmea, mea_size);
2649 LASSERT(mea_size == lmm_size);
2651 OBD_ALLOC(*tmea, mea_size);
2658 if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2659 mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2660 mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2662 magic = le32_to_cpu(mea->mea_magic);
2664 /* old mea is not handled here */
2668 (*tmea)->mea_magic = magic;
2669 (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2670 (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2672 for (i = 0; i < (*tmea)->mea_count; i++) {
2673 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2674 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2679 static int lmv_cancel_unused(struct obd_export *exp,
2680 const struct lu_fid *fid,
2681 ldlm_policy_data_t *policy,
2682 ldlm_mode_t mode, int flags, void *opaque)
2684 struct obd_device *obd = exp->exp_obd;
2685 struct lmv_obd *lmv = &obd->u.lmv;
2689 LASSERT(fid != NULL);
2691 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2692 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2695 err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
2696 policy, mode, flags, opaque);
2703 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
2705 struct obd_device *obd = exp->exp_obd;
2706 struct lmv_obd *lmv = &obd->u.lmv;
2709 RETURN(md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data));
2712 ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
2713 const struct lu_fid *fid, ldlm_type_t type,
2714 ldlm_policy_data_t *policy, ldlm_mode_t mode,
2715 struct lustre_handle *lockh)
2717 struct obd_device *obd = exp->exp_obd;
2718 struct lmv_obd *lmv = &obd->u.lmv;
2723 CDEBUG(D_OTHER, "lock match for "DFID"\n", PFID(fid));
2725 /* with CMD every object can have two locks in different namespaces:
2726 * lookup lock in space of mds storing direntry and update/open lock in
2727 * space of mds storing inode. Thus we check all targets, not only that
2728 * one fid was created in. */
2729 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2730 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2731 type, policy, mode, lockh);
2739 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2740 struct obd_export *dt_exp, struct obd_export *md_exp,
2741 struct lustre_md *md)
2743 struct obd_device *obd = exp->exp_obd;
2744 struct lmv_obd *lmv = &obd->u.lmv;
2748 rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
2752 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2754 struct obd_device *obd = exp->exp_obd;
2755 struct lmv_obd *lmv = &obd->u.lmv;
2759 obd_free_memmd(exp, (struct lov_stripe_md**)&md->mea);
2760 RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2763 int lmv_set_open_replay_data(struct obd_export *exp,
2764 struct obd_client_handle *och,
2765 struct ptlrpc_request *open_req)
2767 struct obd_device *obd = exp->exp_obd;
2768 struct lmv_obd *lmv = &obd->u.lmv;
2769 struct obd_export *tgt_exp;
2773 tgt_exp = lmv_find_export(lmv, &och->och_fid);
2774 if (IS_ERR(tgt_exp))
2775 RETURN(PTR_ERR(tgt_exp));
2777 RETURN(md_set_open_replay_data(tgt_exp, och, open_req));
2780 int lmv_clear_open_replay_data(struct obd_export *exp,
2781 struct obd_client_handle *och)
2783 struct obd_device *obd = exp->exp_obd;
2784 struct lmv_obd *lmv = &obd->u.lmv;
2785 struct obd_export *tgt_exp;
2788 tgt_exp = lmv_find_export(lmv, &och->och_fid);
2789 if (IS_ERR(tgt_exp))
2790 RETURN(PTR_ERR(tgt_exp));
2792 RETURN(md_clear_open_replay_data(tgt_exp, och));
2795 static int lmv_get_remote_perm(struct obd_export *exp,
2796 const struct lu_fid *fid,
2797 struct obd_capa *oc, __u32 suppgid,
2798 struct ptlrpc_request **request)
2800 struct obd_device *obd = exp->exp_obd;
2801 struct lmv_obd *lmv = &obd->u.lmv;
2802 struct obd_export *tgt_exp;
2807 rc = lmv_check_connect(obd);
2811 tgt_exp = lmv_find_export(lmv, fid);
2812 if (IS_ERR(tgt_exp))
2813 RETURN(PTR_ERR(tgt_exp));
2815 rc = md_get_remote_perm(tgt_exp, fid, oc, suppgid, request);
2820 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2823 struct obd_device *obd = exp->exp_obd;
2824 struct lmv_obd *lmv = &obd->u.lmv;
2825 struct obd_export *tgt_exp;
2829 rc = lmv_check_connect(obd);
2833 tgt_exp = lmv_find_export(lmv, &oc->c_capa.lc_fid);
2834 if (IS_ERR(tgt_exp))
2835 RETURN(PTR_ERR(tgt_exp));
2837 rc = md_renew_capa(tgt_exp, oc, cb);
2841 struct obd_ops lmv_obd_ops = {
2842 .o_owner = THIS_MODULE,
2843 .o_setup = lmv_setup,
2844 .o_cleanup = lmv_cleanup,
2845 .o_precleanup = lmv_precleanup,
2846 .o_process_config = lmv_process_config,
2847 .o_connect = lmv_connect,
2848 .o_disconnect = lmv_disconnect,
2849 .o_statfs = lmv_statfs,
2850 .o_llog_init = lmv_llog_init,
2851 .o_llog_finish = lmv_llog_finish,
2852 .o_get_info = lmv_get_info,
2853 .o_set_info_async = lmv_set_info_async,
2854 .o_packmd = lmv_packmd,
2855 .o_unpackmd = lmv_unpackmd,
2856 .o_notify = lmv_notify,
2857 .o_iocontrol = lmv_iocontrol,
2858 .o_fid_delete = lmv_fid_delete
2861 struct md_ops lmv_md_ops = {
2862 .m_getstatus = lmv_getstatus,
2863 .m_change_cbdata = lmv_change_cbdata,
2864 .m_close = lmv_close,
2865 .m_create = lmv_create,
2866 .m_done_writing = lmv_done_writing,
2867 .m_enqueue = lmv_enqueue,
2868 .m_getattr = lmv_getattr,
2869 .m_getxattr = lmv_getxattr,
2870 .m_getattr_name = lmv_getattr_name,
2871 .m_intent_lock = lmv_intent_lock,
2873 .m_rename = lmv_rename,
2874 .m_setattr = lmv_setattr,
2875 .m_setxattr = lmv_setxattr,
2877 .m_readpage = lmv_readpage,
2878 .m_unlink = lmv_unlink,
2879 .m_init_ea_size = lmv_init_ea_size,
2880 .m_cancel_unused = lmv_cancel_unused,
2881 .m_set_lock_data = lmv_set_lock_data,
2882 .m_lock_match = lmv_lock_match,
2883 .m_get_lustre_md = lmv_get_lustre_md,
2884 .m_free_lustre_md = lmv_free_lustre_md,
2885 .m_set_open_replay_data = lmv_set_open_replay_data,
2886 .m_clear_open_replay_data = lmv_clear_open_replay_data,
2887 .m_get_remote_perm = lmv_get_remote_perm,
2888 .m_renew_capa = lmv_renew_capa
2891 int __init lmv_init(void)
2893 struct lprocfs_static_vars lvars;
2896 obj_cache = cfs_mem_cache_create("lmv_objects",
2897 sizeof(struct lmv_obj),
2900 CERROR("error allocating lmv objects cache\n");
2904 lprocfs_lmv_init_vars(&lvars);
2905 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2906 lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2908 cfs_mem_cache_destroy(obj_cache);
2914 static void lmv_exit(void)
2918 class_unregister_type(LUSTRE_LMV_NAME);
2920 rc = cfs_mem_cache_destroy(obj_cache);
2922 "can't free lmv objects cache, %d object(s)"
2923 "still in use\n", atomic_read(&obj_cache_count));
2926 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2927 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2928 MODULE_LICENSE("GPL");
2930 module_init(lmv_init);
2931 module_exit(lmv_exit);