1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 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>
32 #include <asm/div64.h>
33 #include <linux/seq_file.h>
34 #include <linux/namei.h>
36 #include <liblustre.h>
38 #include <linux/ext2_fs.h>
40 #include <linux/obd_support.h>
41 #include <linux/lustre_lib.h>
42 #include <linux/lustre_net.h>
43 #include <linux/lustre_idl.h>
44 #include <linux/lustre_dlm.h>
45 #include <linux/lustre_mds.h>
46 #include <linux/obd_class.h>
47 #include <linux/obd_ost.h>
48 #include <linux/lprocfs_status.h>
49 #include <linux/lustre_fsfilt.h>
50 #include <linux/obd_lmv.h>
51 #include <linux/lustre_lite.h>
52 #include <linux/lustre_audit.h>
53 #include "lmv_internal.h"
55 /* not defined for liblustre building */
56 #if !defined(ATOMIC_INIT)
57 #define ATOMIC_INIT(val) { (val) }
61 kmem_cache_t *obj_cache;
62 atomic_t obj_cache_count = ATOMIC_INIT(0);
64 static void lmv_activate_target(struct lmv_obd *lmv,
65 struct lmv_tgt_desc *tgt,
68 if (tgt->active == activate)
71 tgt->active = activate;
72 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
77 * -EINVAL : UUID can't be found in the LMV's target list
78 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
79 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
81 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
84 struct lmv_tgt_desc *tgt;
85 struct obd_device *obd;
89 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
90 lmv, uuid->uuid, activate);
92 spin_lock(&lmv->lmv_lock);
93 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
94 if (tgt->ltd_exp == NULL)
97 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
98 i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
100 if (obd_uuid_equals(uuid, &tgt->uuid))
104 if (i == lmv->desc.ld_tgt_count)
105 GOTO(out_lmv_lock, rc = -EINVAL);
107 obd = class_exp2obd(tgt->ltd_exp);
109 GOTO(out_lmv_lock, rc = -ENOTCONN);
111 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
112 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
113 obd->obd_type->typ_name, i);
114 LASSERT(strcmp(obd->obd_type->typ_name, OBD_MDC_DEVICENAME) == 0);
116 if (tgt->active == activate) {
117 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
118 activate ? "" : "in");
119 GOTO(out_lmv_lock, rc);
122 CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
123 obd, activate ? "" : "in");
125 lmv_activate_target(lmv, tgt, activate);
130 spin_unlock(&lmv->lmv_lock);
134 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
135 int active, void *data)
137 struct obd_uuid *uuid;
141 if (strcmp(watched->obd_type->typ_name, OBD_MDC_DEVICENAME)) {
142 CERROR("unexpected notification of %s %s!\n",
143 watched->obd_type->typ_name,
147 uuid = &watched->u.cli.cl_import->imp_target_uuid;
149 /* Set MDC as active before notifying the observer, so the observer can
150 * use the MDC normally.
152 rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
154 CERROR("%sactivation of %s failed: %d\n",
155 active ? "" : "de", uuid->uuid, rc);
159 if (obd->obd_observer)
160 /* Pass the notification up the chain. */
161 rc = obd_notify(obd->obd_observer, watched, active, data);
166 static int lmv_attach(struct obd_device *dev, obd_count len, void *data)
168 struct lprocfs_static_vars lvars;
172 lprocfs_init_vars(lmv, &lvars);
173 rc = lprocfs_obd_attach(dev, lvars.obd_vars);
176 struct proc_dir_entry *entry;
178 entry = create_proc_entry("target_obd_status", 0444,
179 dev->obd_proc_entry);
182 entry->proc_fops = &lmv_proc_target_fops;
189 static int lmv_detach(struct obd_device *dev)
191 return lprocfs_obd_detach(dev);
194 /* this is fake connect function. Its purpose is to initialize lmv and say
195 * caller that everything is okay. Real connection will be performed later. */
196 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
197 struct obd_uuid *cluuid, struct obd_connect_data *data,
201 struct proc_dir_entry *lmv_proc_dir;
203 struct lmv_obd *lmv = &obd->u.lmv;
204 struct obd_export *exp;
208 rc = class_connect(conn, obd, cluuid);
210 CERROR("class_connection() returned %d\n", rc);
214 exp = class_conn2export(conn);
216 /* we don't want to actually do the underlying connections more than
217 * once, so keep track. */
219 if (lmv->refcount > 1) {
220 class_export_put(exp);
226 lmv->cluuid = *cluuid;
227 lmv->connect_flags = flags;
229 memcpy(&lmv->conn_data, data, sizeof(*data));
232 lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
234 if (IS_ERR(lmv_proc_dir)) {
235 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
236 obd->obd_type->typ_name, obd->obd_name);
242 * all real clients should perform actual connection rightaway, because
243 * it is possible, that LMV will not have opportunity to connect
244 * targets, as MDC stuff will bit called directly, for instance while
245 * reading ../mdc/../kbytesfree procfs file, etc.
247 if (flags & OBD_OPT_REAL_CLIENT)
248 rc = lmv_check_connect(obd);
253 lprocfs_remove(lmv_proc_dir);
260 static void lmv_set_timeouts(struct obd_device *obd)
262 struct lmv_tgt_desc *tgts;
267 if (lmv->server_timeout == 0)
270 if (lmv->connected == 0)
273 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
274 if (tgts->ltd_exp == NULL)
277 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
278 "inter_mds", 0, NULL);
282 static int lmv_init_ea_size(struct obd_export *exp, int easize,
285 struct obd_device *obd = exp->exp_obd;
286 struct lmv_obd *lmv = &obd->u.lmv;
287 int i, rc = 0, change = 0;
290 if (lmv->max_easize < easize) {
291 lmv->max_easize = easize;
294 if (lmv->max_cookiesize < cookiesize) {
295 lmv->max_cookiesize = cookiesize;
301 if (lmv->connected == 0)
304 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
305 if (lmv->tgts[i].ltd_exp == NULL) {
306 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
310 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
312 CERROR("obd_init_ea_size() failed on MDT target %d, "
313 "error %d.\n", i, rc);
320 #define MAX_STRING_SIZE 128
322 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
324 struct lmv_obd *lmv = &obd->u.lmv;
325 struct obd_uuid *cluuid = &lmv->cluuid;
326 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
327 struct lustre_handle conn = {0, };
328 struct obd_device *mdc_obd;
329 struct obd_export *mdc_exp;
332 struct proc_dir_entry *lmv_proc_dir;
336 /* for MDS: don't connect to yourself */
337 if (obd_uuid_equals(&tgt->uuid, cluuid)) {
338 CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
339 /* XXX - the old code didn't increment active tgt count.
344 mdc_obd = class_find_client_obd(&tgt->uuid, OBD_MDC_DEVICENAME,
347 CERROR("target %s not attached\n", tgt->uuid.uuid);
351 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
352 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
353 tgt->uuid.uuid, obd->obd_uuid.uuid,
356 if (!mdc_obd->obd_set_up) {
357 CERROR("target %s not set up\n", tgt->uuid.uuid);
361 rc = obd_connect(&conn, mdc_obd, &lmv_mdc_uuid, &lmv->conn_data,
364 CERROR("target %s connect error %d\n", tgt->uuid.uuid, rc);
368 mdc_exp = class_conn2export(&conn);
370 rc = obd_register_observer(mdc_obd, obd);
372 obd_disconnect(mdc_exp, 0);
373 CERROR("target %s register_observer error %d\n",
378 if (obd->obd_observer) {
379 /* tell the mds_lmv about the new target */
380 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd, 1,
381 (void *)(tgt - lmv->tgts));
383 obd_disconnect(mdc_exp, 0);
388 tgt->ltd_exp = mdc_exp;
390 lmv->desc.ld_active_tgt_count++;
392 obd_init_ea_size(tgt->ltd_exp, lmv->max_easize,
393 lmv->max_cookiesize);
394 CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
395 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
396 atomic_read(&obd->obd_refcount));
399 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
401 struct proc_dir_entry *mdc_symlink;
402 char name[MAX_STRING_SIZE + 1];
404 LASSERT(mdc_obd->obd_type != NULL);
405 LASSERT(mdc_obd->obd_type->typ_name != NULL);
406 name[MAX_STRING_SIZE] = '\0';
407 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
408 mdc_obd->obd_type->typ_name,
410 mdc_symlink = proc_symlink(mdc_obd->obd_name,
412 if (mdc_symlink == NULL) {
413 CERROR("could not register LMV target "
414 "/proc/fs/lustre/%s/%s/target_obds/%s.",
415 obd->obd_type->typ_name, obd->obd_name,
417 lprocfs_remove(lmv_proc_dir);
425 int lmv_add_mdc(struct obd_device *obd, struct obd_uuid *mdc_uuid)
427 struct lmv_obd *lmv = &obd->u.lmv;
428 struct lmv_tgt_desc *tgt;
432 CDEBUG(D_CONFIG, "mdc_uuid: %s.\n", mdc_uuid->uuid);
436 if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
437 lmv_init_unlock(lmv);
438 CERROR("can't add %s, LMV module compiled for %d MDCs. "
439 "That many MDCs already configured.\n",
440 mdc_uuid->uuid, LMV_MAX_TGT_COUNT);
444 if (lmv->desc.ld_tgt_count == 0) {
445 struct obd_device *mdc_obd;
447 mdc_obd = class_find_client_obd(mdc_uuid, OBD_MDC_DEVICENAME,
450 lmv_init_unlock(lmv);
451 CERROR("Target %s not attached\n", mdc_uuid->uuid);
455 rc = obd_llog_init(obd, &obd->obd_llogs, mdc_obd, 0, NULL);
457 lmv_init_unlock(lmv);
458 CERROR("lmv failed to setup llogging subsystems\n");
462 spin_lock(&lmv->lmv_lock);
463 tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
464 tgt->uuid = *mdc_uuid;
465 spin_unlock(&lmv->lmv_lock);
467 if (lmv->connected) {
468 rc = lmv_connect_mdc(obd, tgt);
470 spin_lock(&lmv->lmv_lock);
471 lmv->desc.ld_tgt_count--;
472 memset(tgt, 0, sizeof(*tgt));
473 spin_unlock(&lmv->lmv_lock);
475 int easize = sizeof(struct mea) +
476 lmv->desc.ld_tgt_count *
477 sizeof(struct lustre_id);
478 lmv_init_ea_size(obd->obd_self_export, easize, 0);
482 lmv_init_unlock(lmv);
486 /* performs a check if passed obd is connected. If no - connect it. */
487 int lmv_check_connect(struct obd_device *obd)
489 struct lmv_obd *lmv = &obd->u.lmv;
490 struct lmv_tgt_desc *tgt;
498 if (lmv->connected) {
499 lmv_init_unlock(lmv);
503 if (lmv->desc.ld_tgt_count == 0) {
504 CERROR("%s: no targets configured.\n", obd->obd_name);
508 CDEBUG(D_CONFIG, "time to connect %s to %s\n",
509 lmv->cluuid.uuid, obd->obd_name);
511 LASSERT(lmv->tgts != NULL);
513 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
514 rc = lmv_connect_mdc(obd, tgt);
519 lmv_set_timeouts(obd);
520 class_export_put(lmv->exp);
522 easize = lmv->desc.ld_tgt_count * sizeof(struct lustre_id) +
524 lmv_init_ea_size(obd->obd_self_export, easize, 0);
525 lmv_init_unlock(lmv);
534 --lmv->desc.ld_active_tgt_count;
535 rc2 = obd_disconnect(tgt->ltd_exp, 0);
537 CERROR("error: LMV target %s disconnect on "
538 "MDC idx %d: error %d\n",
539 tgt->uuid.uuid, i, rc2);
543 class_disconnect(lmv->exp, 0);
544 lmv_init_unlock(lmv);
548 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
550 struct obd_device *obd = class_exp2obd(exp);
551 struct lmv_obd *lmv = &obd->u.lmv;
554 struct proc_dir_entry *lmv_proc_dir;
562 /* Only disconnect the underlying layers on the final disconnect. */
564 if (lmv->refcount != 0)
568 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
571 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
572 struct obd_device *mdc_obd;
574 if (lmv->tgts[i].ltd_exp == NULL)
577 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
580 mdc_obd->obd_no_recov = obd->obd_no_recov;
584 struct proc_dir_entry *mdc_symlink;
586 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
588 lprocfs_remove(mdc_symlink);
590 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
591 obd->obd_type->typ_name, obd->obd_name,
596 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
597 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
598 lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
600 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
601 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
603 if (lmv->tgts[i].active) {
604 CERROR("Target %s disconnect error %d\n",
605 lmv->tgts[i].uuid.uuid, rc);
610 lmv_activate_target(lmv, &lmv->tgts[i], 0);
611 lmv->tgts[i].ltd_exp = NULL;
616 lprocfs_remove(lmv_proc_dir);
618 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
619 obd->obd_type->typ_name, obd->obd_name);
624 /* this is the case when no real connection is established by
625 * lmv_check_connect(). */
627 class_export_put(exp);
628 rc = class_disconnect(exp, 0);
629 if (lmv->refcount == 0)
634 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
635 int len, void *karg, void *uarg)
637 struct obd_device *obddev = class_exp2obd(exp);
638 struct lmv_obd *lmv = &obddev->u.lmv;
639 int i, rc = 0, set = 0;
642 if (lmv->desc.ld_tgt_count == 0)
645 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
648 if (lmv->tgts[i].ltd_exp == NULL)
651 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
653 if (lmv->tgts[i].active) {
654 CERROR("error: iocontrol MDC %s on MDT"
655 "idx %d: err = %d\n",
656 lmv->tgts[i].uuid.uuid, i, err);
669 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
671 struct lmv_obd *lmv = &obd->u.lmv;
672 struct lustre_cfg *lcfg = buf;
673 struct lmv_desc *desc;
677 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
678 CERROR("LMV setup requires a descriptor\n");
682 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
683 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
684 CERROR("descriptor size wrong: %d > %d\n",
685 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
689 lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
691 OBD_ALLOC(lmv->tgts, lmv->tgts_size);
692 if (lmv->tgts == NULL) {
693 CERROR("Out of memory\n");
697 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
698 lmv->desc.ld_tgt_count = 0;
699 lmv->desc.ld_active_tgt_count = 0;
700 lmv->max_cookiesize = 0;
703 spin_lock_init(&lmv->lmv_lock);
704 sema_init(&lmv->init_sem, 1);
706 rc = lmv_setup_mgr(obd);
708 CERROR("Can't setup LMV object manager, "
710 OBD_FREE(lmv->tgts, lmv->tgts_size);
717 static int lmv_cleanup(struct obd_device *obd, int flags)
719 struct lmv_obd *lmv = &obd->u.lmv;
722 lmv_cleanup_mgr(obd);
723 OBD_FREE(lmv->tgts, lmv->tgts_size);
728 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
730 struct lustre_cfg *lcfg = buf;
731 struct obd_uuid mdc_uuid;
735 switch(lcfg->lcfg_command) {
736 case LCFG_LMV_ADD_MDC:
737 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(mdc_uuid.uuid))
738 GOTO(out, rc = -EINVAL);
740 obd_str2uuid(&mdc_uuid, lustre_cfg_string(lcfg, 1));
741 rc = lmv_add_mdc(obd, &mdc_uuid);
744 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
745 GOTO(out, rc = -EINVAL);
752 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
753 unsigned long max_age)
755 struct lmv_obd *lmv = &obd->u.lmv;
756 struct obd_statfs *temp;
760 rc = lmv_check_connect(obd);
764 OBD_ALLOC(temp, sizeof(*temp));
768 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
769 if (lmv->tgts[i].ltd_exp == NULL)
772 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
774 CERROR("can't stat MDS #%d (%s), error %d\n", i,
775 lmv->tgts[i].ltd_exp->exp_obd->obd_name,
777 GOTO(out_free_temp, rc);
780 memcpy(osfs, temp, sizeof(*temp));
782 osfs->os_bavail += temp->os_bavail;
783 osfs->os_blocks += temp->os_blocks;
784 osfs->os_ffree += temp->os_ffree;
785 osfs->os_files += temp->os_files;
791 OBD_FREE(temp, sizeof(*temp));
795 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
797 struct obd_device *obd = exp->exp_obd;
798 struct lmv_obd *lmv = &obd->u.lmv;
802 rc = lmv_check_connect(obd);
806 rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
812 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
813 __u64 valid, const char *xattr_name,
814 const void *xattr_data, unsigned int xattr_datalen,
815 unsigned int ea_size, struct obd_capa *ocapa,
816 struct ptlrpc_request **request)
818 struct obd_device *obd = exp->exp_obd;
819 struct lmv_obd *lmv = &obd->u.lmv;
820 int rc, i = id_group(id);
824 rc = lmv_check_connect(obd);
828 LASSERT(i < lmv->desc.ld_tgt_count);
831 rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
832 xattr_name, xattr_data, xattr_datalen,
833 ea_size, ocapa, request);
837 obj = lmv_grab_obj(obd, id);
839 CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
840 OLID4(id), obj ? "(splitted)" : "");
843 * if object is splitted, then we loop over all the slaves and gather
844 * size attribute. In ideal world we would have to gather also mds field
845 * from all slaves, as object is spread over the cluster and this is
846 * definitely interesting information and it is not good to loss it,
850 struct mds_body *body;
852 if (*request == NULL) {
857 body = lustre_msg_buf((*request)->rq_repmsg, 0,
859 LASSERT(body != NULL);
863 for (i = 0; i < obj->objcount; i++) {
865 if (lmv->tgts[i].ltd_exp == NULL) {
866 CWARN("%s: NULL export for %d\n",
871 /* skip master obj. */
872 if (id_equal_fid(&obj->id, &obj->objs[i].id))
875 body->size += obj->objs[i].size;
885 static int lmv_access_check(struct obd_export *exp,
886 struct lustre_id *id,
887 struct ptlrpc_request **request)
889 struct obd_device *obd = exp->exp_obd;
890 struct lmv_obd *lmv = &obd->u.lmv;
891 int rc, i = id_group(id);
894 rc = lmv_check_connect(obd);
898 LASSERT(i < lmv->desc.ld_tgt_count);
899 rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
903 static int lmv_change_cbdata(struct obd_export *exp,
904 struct lustre_id *id,
908 struct obd_device *obd = exp->exp_obd;
909 struct lmv_obd *lmv = &obd->u.lmv;
913 rc = lmv_check_connect(obd);
917 CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
918 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
920 /* with CMD every object can have two locks in different
921 * namespaces: lookup lock in space of mds storing direntry
922 * and update/open lock in space of mds storing inode */
923 for (i = 0; i < lmv->desc.ld_tgt_count; i++)
924 md_change_cbdata(lmv->tgts[i].ltd_exp, id, it, data);
929 static int lmv_change_cbdata_name(struct obd_export *exp,
930 struct lustre_id *pid,
932 struct lustre_id *cid,
936 struct obd_device *obd = exp->exp_obd;
937 struct lmv_obd *lmv = &obd->u.lmv;
938 struct lustre_id rcid = *cid;
943 rc = lmv_check_connect(obd);
947 LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
948 LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
950 CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
951 OLID4(pid), len, name, OLID4(cid));
953 /* this is default mds for directory name belongs to. */
955 obj = lmv_grab_obj(obd, pid);
957 /* directory is splitted. look for right mds for this name. */
958 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
959 rcid = obj->objs[mds].id;
960 mds = id_group(&rcid);
963 rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
967 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
969 struct obd_device *obd = exp->exp_obd;
970 struct lmv_obd *lmv = &obd->u.lmv;
974 rc = lmv_check_connect(obd);
978 CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
979 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
980 rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
984 static int lmv_close(struct obd_export *exp, struct mdc_op_data *op_data,
985 struct obd_client_handle *och,
986 struct ptlrpc_request **request)
988 struct obd_device *obd = exp->exp_obd;
989 struct lmv_obd *lmv = &obd->u.lmv;
990 int rc, i = id_group(&op_data->id1);
993 rc = lmv_check_connect(obd);
997 LASSERT(i < lmv->desc.ld_tgt_count);
998 CDEBUG(D_OTHER, "CLOSE "DLID4"\n", OLID4(&op_data->id1));
999 rc = md_close(lmv->tgts[i].ltd_exp, op_data, och, request);
1003 int lmv_get_mea_and_update_object(struct obd_export *exp,
1004 struct lustre_id *id)
1006 struct obd_device *obd = exp->exp_obd;
1007 struct lmv_obd *lmv = &obd->u.lmv;
1008 struct ptlrpc_request *req = NULL;
1009 struct lmv_obj *obj;
1010 struct lustre_md md;
1016 mealen = MEA_SIZE_LMV(lmv);
1018 valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1020 /* time to update mea of parent id */
1021 rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
1022 id, valid, NULL, NULL, 0, mealen, NULL, &req);
1024 CERROR("md_getattr() failed, error %d\n", rc);
1028 rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
1030 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
1035 GOTO(cleanup, rc = -ENODATA);
1037 obj = lmv_create_obj(exp, id, md.mea);
1043 obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
1048 ptlrpc_req_finished(req);
1052 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
1053 const void *data, int datalen, int mode, __u32 uid,
1054 __u32 gid, __u64 rdev, struct ptlrpc_request **request)
1056 struct obd_device *obd = exp->exp_obd;
1057 struct lmv_obd *lmv = &obd->u.lmv;
1058 struct mds_body *body;
1059 struct lmv_obj *obj;
1060 int rc, mds, loop = 0;
1063 rc = lmv_check_connect(obd);
1067 if (!lmv->desc.ld_active_tgt_count)
1070 LASSERT(++loop <= 2);
1071 obj = lmv_grab_obj(obd, &op_data->id1);
1073 mds = raw_name2idx(obj->hashtype, obj->objcount,
1074 op_data->name, op_data->namelen);
1075 op_data->id1 = obj->objs[mds].id;
1079 CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
1080 op_data->name, OLID4(&op_data->id1));
1082 rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
1083 op_data, data, datalen, mode, uid, gid, rdev,
1086 if (*request == NULL)
1089 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1094 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
1095 } else if (rc == -ERESTART) {
1097 * directory got splitted. time to update local object and
1098 * repeat the request with proper MDS.
1100 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
1102 ptlrpc_req_finished(*request);
1109 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
1111 struct obd_device *obd = exp->exp_obd;
1112 struct lmv_obd *lmv = &obd->u.lmv;
1116 rc = lmv_check_connect(obd);
1120 /* FIXME: choose right MDC here */
1121 CWARN("this method isn't implemented yet\n");
1122 rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
1127 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1128 struct lookup_intent *it, int lockmode,
1129 struct mdc_op_data *data, struct lustre_handle *lockh,
1130 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1131 ldlm_blocking_callback cb_blocking, void *cb_data)
1133 struct obd_device *obd = exp->exp_obd;
1134 struct lmv_obd *lmv = &obd->u.lmv;
1135 struct mea *mea = data->mea1;
1136 struct mdc_op_data *data2;
1140 OBD_ALLOC(data2, sizeof(*data2));
1144 LASSERT(mea != NULL);
1145 for (i = 0; i < mea->mea_count; i++) {
1146 memset(data2, 0, sizeof(*data2));
1147 data2->id1 = mea->mea_ids[i];
1148 mds = id_group(&data2->id1);
1150 if (lmv->tgts[mds].ltd_exp == NULL)
1153 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1154 lockmode, data2, lockh + i, lmm, lmmsize,
1155 cb_compl, cb_blocking, cb_data);
1157 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1158 OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1161 if (LUSTRE_IT(it)->it_data) {
1162 struct ptlrpc_request *req;
1163 req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1164 ptlrpc_req_finished(req);
1167 if (LUSTRE_IT(it)->it_status)
1168 GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1171 OBD_FREE(data2, sizeof(*data2));
1174 OBD_FREE(data2, sizeof(*data2));
1176 /* drop all taken locks */
1178 if (lockh[i].cookie)
1179 ldlm_lock_decref(lockh + i, lockmode);
1180 lockh[i].cookie = 0;
1186 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1187 struct lookup_intent *it, int lock_mode,
1188 struct mdc_op_data *data, struct lustre_handle *lockh,
1189 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1190 ldlm_blocking_callback cb_blocking, void *cb_data)
1192 struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1193 struct obd_device *obd = exp->exp_obd;
1194 struct lmv_obd *lmv = &obd->u.lmv;
1195 struct lustre_handle plock;
1196 struct mdc_op_data rdata;
1197 struct mds_body *body = NULL;
1201 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1202 LASSERT(body != NULL);
1204 if (!(body->valid & OBD_MD_MDS))
1207 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1208 LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1210 /* we got LOOKUP lock, but we really need attrs */
1211 pmode = LUSTRE_IT(it)->it_lock_mode;
1212 LASSERT(pmode != 0);
1213 memcpy(&plock, lockh, sizeof(plock));
1214 LUSTRE_IT(it)->it_lock_mode = 0;
1215 LUSTRE_IT(it)->it_data = NULL;
1216 LASSERT((body->valid & OBD_MD_FID) != 0);
1218 memcpy(&rdata, data, sizeof(rdata));
1219 rdata.id1 = body->id1;
1223 LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1224 ptlrpc_req_finished(req);
1226 rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1227 lock_type, it, lock_mode, &rdata, lockh, lmm,
1228 lmmsize, cb_compl, cb_blocking, cb_data);
1229 ldlm_lock_decref(&plock, pmode);
1234 lmv_enqueue(struct obd_export *exp, int lock_type,
1235 struct lookup_intent *it, int lock_mode,
1236 struct mdc_op_data *data, struct lustre_handle *lockh,
1237 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1238 ldlm_blocking_callback cb_blocking, void *cb_data)
1240 struct obd_device *obd = exp->exp_obd;
1241 struct lmv_obd *lmv = &obd->u.lmv;
1242 struct lmv_obj *obj;
1246 rc = lmv_check_connect(obd);
1250 if (data->mea1 && it->it_op == IT_UNLINK) {
1251 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1252 data, lockh, lmm, lmmsize,
1253 cb_compl, cb_blocking, cb_data);
1257 if (data->namelen) {
1258 obj = lmv_grab_obj(obd, &data->id1);
1260 /* directory is splitted. look for right mds for this
1262 mds = raw_name2idx(obj->hashtype, obj->objcount,
1263 (char *)data->name, data->namelen);
1264 data->id1 = obj->objs[mds].id;
1268 CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1271 rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1272 lock_type, it, lock_mode, data, lockh, lmm,
1273 lmmsize, cb_compl, cb_blocking, cb_data);
1274 if (rc == 0 && it->it_op == IT_OPEN)
1275 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1276 data, lockh, lmm, lmmsize,
1277 cb_compl, cb_blocking, cb_data);
1282 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1283 char *filename, int namelen, __u64 valid,
1284 unsigned int ea_size, struct ptlrpc_request **request)
1286 int rc, mds = id_group(id), loop = 0;
1287 struct obd_device *obd = exp->exp_obd;
1288 struct lmv_obd *lmv = &obd->u.lmv;
1289 struct lustre_id rid = *id;
1290 struct mds_body *body;
1291 struct lmv_obj *obj;
1294 rc = lmv_check_connect(obd);
1298 LASSERT(++loop <= 2);
1299 obj = lmv_grab_obj(obd, id);
1301 /* directory is splitted. look for right mds for this name */
1302 mds = raw_name2idx(obj->hashtype, obj->objcount,
1303 filename, namelen - 1);
1304 rid = obj->objs[mds].id;
1308 CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1309 namelen, filename, OLID4(id), OLID4(&rid));
1311 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1312 &rid, filename, namelen,
1313 valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1317 * this could be cross-node reference. in this case all we have
1318 * right now is lustre_id triple. we'd like to find other
1321 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1322 LASSERT(body != NULL);
1323 LASSERT((body->valid & OBD_MD_FID) != 0
1324 || body->valid == OBD_MD_FLID);
1326 if (body->valid & OBD_MD_MDS) {
1327 struct ptlrpc_request *req = NULL;
1330 CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1332 rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1333 &rid, NULL, 1, valid, ea_size, &req);
1334 ptlrpc_req_finished(*request);
1337 } else if (rc == -ERESTART) {
1338 /* directory got splitted. time to update local object and
1339 * repeat the request with proper MDS */
1340 rc = lmv_get_mea_and_update_object(exp, &rid);
1342 ptlrpc_req_finished(*request);
1350 * llite passes id of an target inode in data->id1 and id of directory in
1353 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1354 struct ptlrpc_request **request)
1356 struct obd_device *obd = exp->exp_obd;
1357 struct lmv_obd *lmv = &obd->u.lmv;
1358 struct lmv_obj *obj;
1362 rc = lmv_check_connect(obd);
1366 if (data->namelen != 0) {
1367 /* usual link request */
1368 obj = lmv_grab_obj(obd, &data->id2);
1370 rc = raw_name2idx(obj->hashtype, obj->objcount,
1371 data->name, data->namelen);
1372 data->id2 = obj->objs[rc].id;
1376 mds = id_group(&data->id2);
1378 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1379 OLID4(&data->id2), data->namelen, data->name,
1382 mds = id_group(&data->id1);
1384 /* request from MDS to acquire i_links for inode by id1 */
1385 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1389 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1390 mds, OLID4(&data->id1));
1391 rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1396 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1397 const char *old, int oldlen, const char *new, int newlen,
1398 struct ptlrpc_request **request)
1400 struct obd_device *obd = exp->exp_obd;
1401 struct lmv_obd *lmv = &obd->u.lmv;
1402 struct lmv_obj *obj;
1406 CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1407 oldlen, old, OLID4(&data->id1), newlen, new,
1410 rc = lmv_check_connect(obd);
1416 * MDS with old dir entry is asking another MDS to create name
1420 "create %*s(%d/%d) in "DLID4" pointing "
1421 "to "DLID4"\n", newlen, new, oldlen, newlen,
1422 OLID4(&data->id2), OLID4(&data->id1));
1424 mds = id_group(&data->id2);
1427 * target directory can be splitted, sowe should forward request
1430 obj = lmv_grab_obj(obd, &data->id2);
1432 mds = raw_name2idx(obj->hashtype, obj->objcount,
1433 (char *)new, newlen);
1434 data->id2 = obj->objs[mds].id;
1435 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1442 obj = lmv_grab_obj(obd, &data->id1);
1445 * directory is already splitted, so we have to forward request
1448 mds = raw_name2idx(obj->hashtype, obj->objcount,
1449 (char *)old, oldlen);
1450 data->id1 = obj->objs[mds].id;
1451 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1456 obj = lmv_grab_obj(obd, &data->id2);
1459 * directory is already splitted, so we have to forward request
1462 mds = raw_name2idx(obj->hashtype, obj->objcount,
1463 (char *)new, newlen);
1465 data->id2 = obj->objs[mds].id;
1466 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1471 mds = id_group(&data->id1);
1474 if (id_group(&data->id1) != id_group(&data->id2)) {
1475 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1476 OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1480 rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1481 new, newlen, request);
1485 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1486 struct iattr *iattr, void *ea, int ealen, void *ea2,
1487 int ea2len, void *ea3, int ea3len,
1488 struct ptlrpc_request **request)
1490 struct obd_device *obd = exp->exp_obd;
1491 struct lmv_obd *lmv = &obd->u.lmv;
1492 struct ptlrpc_request *req;
1493 struct mds_body *body;
1494 struct lmv_obj *obj;
1498 rc = lmv_check_connect(obd);
1502 obj = lmv_grab_obj(obd, &data->id1);
1504 CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1505 OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1508 for (i = 0; i < obj->objcount; i++) {
1509 data->id1 = obj->objs[i].id;
1511 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1512 data, iattr, ea, ealen, ea2, ea2len,
1515 if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1517 * this is master object and this request should
1518 * be returned back to llite.
1522 ptlrpc_req_finished(req);
1530 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1531 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1532 data, iattr, ea, ealen, ea2, ea2len, ea3,
1535 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1537 LASSERT(body != NULL);
1538 LASSERT((body->valid & OBD_MD_FID) != 0);
1539 LASSERT(id_group(&body->id1) == id_group(&data->id1));
1545 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1546 struct ptlrpc_request **request)
1548 struct obd_device *obd = exp->exp_obd;
1549 struct lmv_obd *lmv = &obd->u.lmv;
1553 rc = lmv_check_connect(obd);
1557 rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1562 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1563 struct ldlm_lock_desc *desc,
1564 void *data, int flag)
1566 struct lustre_handle lockh;
1567 struct lmv_obj *obj;
1572 case LDLM_CB_BLOCKING:
1573 ldlm_lock2handle(lock, &lockh);
1574 rc = ldlm_cli_cancel(&lockh);
1576 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1580 case LDLM_CB_CANCELING:
1581 /* time to drop cached attrs for dirobj */
1582 obj = lock->l_ast_data;
1584 CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1585 ", master "DLID4"\n",
1586 lock->l_resource->lr_name.name[3] == 1 ?
1587 "LOOKUP" : "UPDATE",
1588 lock->l_resource->lr_name.name[0],
1589 lock->l_resource->lr_name.name[1],
1600 static void lmv_remove_dots(struct page *page)
1602 unsigned limit = PAGE_CACHE_SIZE;
1603 char *kaddr = page_address(page);
1604 struct ext2_dir_entry_2 *p;
1605 unsigned offs, rec_len;
1607 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1608 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1609 rec_len = le16_to_cpu(p->rec_len);
1611 if ((p->name_len == 1 && p->name[0] == '.') ||
1612 (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1617 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1618 __u64 offset, struct page *page,
1619 struct ptlrpc_request **request)
1621 struct obd_device *obd = exp->exp_obd;
1622 struct lmv_obd *lmv = &obd->u.lmv;
1623 struct lustre_id rid = *id;
1624 struct lmv_obj *obj;
1628 #warning "we need well-desgined readdir() implementation"
1629 rc = lmv_check_connect(obd);
1633 LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1634 CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1635 offset, OLID4(&rid));
1637 obj = lmv_grab_obj(obd, id);
1641 /* find dirobj containing page with requested offset. */
1642 for (i = 0; i < obj->objcount; i++) {
1643 if (offset < obj->objs[i].size)
1645 offset -= obj->objs[i].size;
1647 rid = obj->objs[i].id;
1649 lmv_unlock_obj(obj);
1652 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1653 OLID4(&rid), (unsigned long)offset);
1655 rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1656 offset, page, request);
1658 if (rc == 0 && !id_equal_fid(&rid, id))
1659 /* this page isn't from master object. To avoid "." and ".."
1660 * duplication in directory, we have to remove them from all
1662 lmv_remove_dots(page);
1667 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1668 struct ptlrpc_request **req)
1670 struct obd_device *obd = exp->exp_obd;
1671 struct lmv_obd *lmv = &obd->u.lmv;
1672 struct mea *mea = data->mea1;
1673 struct mdc_op_data *data2;
1677 OBD_ALLOC(data2, sizeof(*data2));
1681 LASSERT(mea != NULL);
1682 for (i = 0; i < mea->mea_count; i++) {
1683 memset(data2, 0, sizeof(*data2));
1684 data2->id1 = mea->mea_ids[i];
1685 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1687 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1690 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1693 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1694 OLID4(&mea->mea_ids[i]), rc);
1697 ptlrpc_req_finished(*req);
1703 OBD_FREE(data2, sizeof(*data2));
1707 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1712 if (lmv_delete_obj(exp, id)) {
1713 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1719 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1720 struct ptlrpc_request **request)
1722 struct obd_device *obd = exp->exp_obd;
1723 struct lmv_obd *lmv = &obd->u.lmv;
1727 rc = lmv_check_connect(obd);
1731 if (data->namelen == 0 && data->mea1 != NULL) {
1732 /* mds asks to remove slave objects */
1733 rc = lmv_unlink_slaves(exp, data, request);
1737 if (data->namelen != 0) {
1738 struct lmv_obj *obj;
1740 obj = lmv_grab_obj(obd, &data->id1);
1742 i = raw_name2idx(obj->hashtype, obj->objcount,
1743 data->name, data->namelen);
1744 data->id1 = obj->objs[i].id;
1747 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1748 data->namelen, data->name, OLID4(&data->id1),
1751 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1754 rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1759 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1760 struct lustre_id *id)
1762 struct obd_device *obd = exp->exp_obd;
1763 struct lmv_obd *lmv = &obd->u.lmv;
1767 rc = lmv_check_connect(obd);
1769 RETURN(ERR_PTR(rc));
1770 obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1776 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1777 void *acl, int acl_size,
1778 struct lov_stripe_md **ea,
1779 struct obd_trans_info *oti)
1781 struct obd_device *obd = exp->exp_obd;
1782 struct lmv_obd *lmv = &obd->u.lmv;
1783 struct lov_stripe_md obj_md;
1784 struct lov_stripe_md *obj_mdp = &obj_md;
1788 LASSERT(ea == NULL);
1789 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1791 rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1792 acl, acl_size, &obj_mdp, oti);
1798 * to be called from MDS only. @oa should have correct store cookie and o_fid
1799 * values for "master" object, as it will be used.
1801 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1802 void *acl, int acl_size,
1803 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1805 struct obd_device *obd = exp->exp_obd;
1806 struct lmv_obd *lmv = &obd->u.lmv;
1807 struct lustre_id mid;
1812 rc = lmv_check_connect(obd);
1816 LASSERT(oa != NULL);
1819 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1821 CERROR("Can't create object, rc = %d\n", rc);
1825 /* acl is only suppied when mds create single remote obj */
1826 LASSERT(acl == NULL && acl_size == 0);
1829 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1831 CERROR("obd_alloc_diskmd() failed, error %d\n",
1842 * here we should take care about splitted dir, so store cookie and fid
1843 * for "master" object should already be allocated and passed in @oa.
1845 LASSERT(oa->o_id != 0);
1846 LASSERT(oa->o_fid != 0);
1848 /* save "master" object id */
1851 mea = (struct mea *)*ea;
1852 mea->mea_master = -1;
1853 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1855 if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1856 mea->mea_count = lmv->desc.ld_tgt_count;
1858 for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1859 struct lov_stripe_md obj_md;
1860 struct lov_stripe_md *obj_mdp = &obj_md;
1862 if (lmv->tgts[i].ltd_exp == NULL) {
1863 /* this is "master" MDS */
1864 mea->mea_master = i;
1865 mea->mea_ids[c] = mid;
1871 * "master" MDS should always be part of stripped dir,
1874 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1877 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1878 OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1880 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1883 CERROR("obd_create() failed on MDT target %d, "
1884 "error %d\n", c, rc);
1888 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1889 i, oa->o_id, oa->o_generation);
1893 * here, when object is created (or it is master and was passed
1894 * from caller) on desired MDS we save its fid to local mea_ids.
1899 * store cookie should be defined here for both cases (master
1900 * object and not master), because master is already created.
1904 /* fill mea by store cookie and fid */
1905 obdo2id(&mea->mea_ids[c], oa);
1908 LASSERT(c == mea->mea_count);
1910 CDEBUG(D_OTHER, "%d dirobjects created\n",
1911 (int)mea->mea_count);
1916 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1917 struct obd_device *tgt, int count,
1918 struct llog_catid *logid)
1920 struct llog_ctxt *ctxt;
1924 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1927 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1928 ctxt->loc_imp = tgt->u.cli.cl_import;
1934 static int lmv_llog_finish(struct obd_device *obd,
1935 struct obd_llogs *llogs, int count)
1940 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1944 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1945 void *key, __u32 *vallen, void *val)
1947 struct obd_device *obd;
1948 struct lmv_obd *lmv;
1952 obd = class_exp2obd(exp);
1954 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1955 exp->exp_handle.h_cookie);
1960 if (keylen == strlen("mdsize") && !strcmp(key, "mdsize")) {
1961 __u32 *mdsize = val;
1962 *vallen = sizeof(__u32);
1963 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1964 + sizeof(struct mea);
1966 } else if (keylen == strlen("mdsnum") && !strcmp(key, "mdsnum")) {
1967 struct obd_uuid *cluuid = &lmv->cluuid;
1968 struct lmv_tgt_desc *tgts;
1969 __u32 *mdsnum = val;
1973 for (i = 0; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1974 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1975 *vallen = sizeof(__u32);
1981 } else if (keylen == strlen("rootid") && !strcmp(key, "rootid")) {
1982 rc = lmv_check_connect(obd);
1986 /* getting rootid from first MDS. */
1987 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1990 } else if (keylen >= strlen("lmvdesc") && !strcmp(key, "lmvdesc")) {
1991 struct lmv_desc *desc_ret = val;
1992 *desc_ret = lmv->desc;
1994 } else if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1995 struct lmv_tgt_desc *tgts;
1998 rc = lmv_check_connect(obd);
2002 LASSERT(*vallen == sizeof(__u32));
2003 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2006 /* all tgts should be connected when this get called. */
2007 if (!tgts || !tgts->ltd_exp) {
2008 CERROR("target not setup?\n");
2012 if (!obd_get_info(tgts->ltd_exp, keylen, key,
2017 } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2018 rc = lmv_check_connect(obd);
2022 /* forwarding this request to first MDS, it should know LOV
2024 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2027 } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2028 struct lmv_tgt_desc *tgts;
2031 rc = lmv_check_connect(obd);
2035 LASSERT(*vallen == sizeof(struct fid_extent));
2036 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2039 /* all tgts should be connected when this get called. */
2040 if (!tgts || !tgts->ltd_exp) {
2041 CERROR("target not setup?\n");
2045 rc = obd_get_info(tgts->ltd_exp, keylen, key,
2053 CDEBUG(D_IOCTL, "invalid key\n");
2057 int lmv_set_info(struct obd_export *exp, obd_count keylen,
2058 void *key, obd_count vallen, void *val)
2060 struct lmv_tgt_desc *tgt;
2061 struct obd_device *obd;
2062 struct lmv_obd *lmv;
2066 obd = class_exp2obd(exp);
2068 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2069 exp->exp_handle.h_cookie);
2074 if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
2075 lmv->server_timeout = 1;
2076 lmv_set_timeouts(obd);
2080 /* maybe this could be default */
2081 if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
2082 (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
2083 (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
2084 struct obd_export *exp;
2087 spin_lock(&lmv->lmv_lock);
2088 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2091 /* during setup time the connections to mdc might
2092 * haven't been established.
2095 struct obd_device *tgt_obd;
2097 tgt_obd = class_find_client_obd(&tgt->uuid,
2101 CERROR("can't set info %s, "
2102 "device %s not attached?\n",
2103 (char *) key, tgt->uuid.uuid);
2107 exp = tgt_obd->obd_self_export;
2110 err = obd_set_info(exp, keylen, key, vallen, val);
2114 spin_unlock(&lmv->lmv_lock);
2118 if (keylen == 5 && strcmp(key, "audit") == 0) {
2119 struct audit_attr_msg * msg = val;
2120 int mds = id_group(&msg->id);
2122 LASSERT(mds < lmv->desc.ld_tgt_count);
2124 if (IS_AUDIT_OP(msg->attr, AUDIT_FS)) {
2125 //FS audit, send message to all mds
2126 for (i = 0; i < lmv->desc.ld_tgt_count;i++) {
2127 obd_set_info(lmv->tgts[i].ltd_exp,
2128 keylen, key, vallen, val);
2131 else if (IS_AUDIT_OP(msg->attr, AUDIT_DIR)) {
2133 //if dir is splitted, send RPC to all mds involved
2134 struct lmv_obj *obj;
2135 struct lustre_id rid;
2138 obj = lmv_grab_obj(obd, &msg->id);
2141 for (i = 0; i < obj->objcount; i++) {
2142 rid = obj->objs[i].id;
2143 mds = id_group(&rid);
2144 obd_set_info(lmv->tgts[mds].ltd_exp,
2148 lmv_unlock_obj(obj);
2152 rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2153 keylen, key, vallen, val);
2157 //set audit for file
2158 rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2159 keylen, key, vallen, val);
2163 if (((keylen == strlen("flush_cred") &&
2164 strcmp(key, "flush_cred") == 0)) ||
2165 ((keylen == strlen("crypto_type") &&
2166 strcmp(key, "crypto_type") == 0))) {
2169 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2173 rc = obd_set_info(tgt->ltd_exp,
2174 keylen, key, vallen, val);
2182 if (keylen == strlen("ids") && memcmp(key, "ids", keylen) == 0) {
2183 struct lustre_id *id = (struct lustre_id *)val;
2185 rc = lmv_check_connect(obd);
2189 rc = obd_set_info(lmv->tgts[id_group(id)].ltd_exp,
2190 keylen, key, vallen, val);
2194 if (keylen == strlen("chkconnect") &&
2195 memcmp(key, "chkconnect", keylen) == 0) {
2196 rc = lmv_check_connect(obd);
2203 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2204 struct lov_stripe_md *lsm)
2206 struct obd_device *obd = class_exp2obd(exp);
2207 struct lmv_obd *lmv = &obd->u.lmv;
2208 struct mea *meap, *lsmp;
2212 mea_size = (sizeof(struct lustre_id) *
2213 lmv->desc.ld_tgt_count) + sizeof(struct mea);
2217 if (*lmmp && !lsm) {
2218 OBD_FREE(*lmmp, mea_size);
2223 if (*lmmp == NULL) {
2224 OBD_ALLOC(*lmmp, mea_size);
2232 lsmp = (struct mea *)lsm;
2233 meap = (struct mea *)*lmmp;
2235 if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2236 lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2239 meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2240 meap->mea_count = cpu_to_le32(lsmp->mea_count);
2241 meap->mea_master = cpu_to_le32(lsmp->mea_master);
2243 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2244 meap->mea_ids[i] = meap->mea_ids[i];
2245 id_cpu_to_le(&meap->mea_ids[i]);
2251 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2252 struct lov_mds_md *lmm, int lmm_size)
2254 struct obd_device *obd = class_exp2obd(exp);
2255 struct mea **tmea = (struct mea **)lsmp;
2256 struct mea *mea = (struct mea *)lmm;
2257 struct lmv_obd *lmv = &obd->u.lmv;
2258 int mea_size, i, rc = 0;
2262 mea_size = sizeof(struct lustre_id) *
2263 lmv->desc.ld_tgt_count + sizeof(struct mea);
2268 if (*lsmp != NULL && lmm == NULL) {
2269 OBD_FREE(*tmea, mea_size);
2273 LASSERT(mea_size == lmm_size);
2275 OBD_ALLOC(*tmea, mea_size);
2282 if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2283 mea->mea_magic == MEA_MAGIC_ALL_CHARS)
2285 magic = le32_to_cpu(mea->mea_magic);
2287 struct mea_old *old = (struct mea_old *)lmm;
2289 mea_size = sizeof(struct lustre_id) * old->mea_count +
2290 sizeof(struct mea_old);
2292 if (old->mea_count > 256 || old->mea_master > 256 ||
2293 lmm_size < mea_size || old->mea_master > old->mea_count) {
2294 CWARN("bad MEA: count %u, master %u, size %u\n",
2295 old->mea_count, old->mea_master, mea_size);
2296 GOTO(out_free_mea, rc = -EINVAL);
2298 magic = MEA_MAGIC_LAST_CHAR;
2301 (*tmea)->mea_magic = magic;
2302 (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2303 (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2305 for (i = 0; i < (*tmea)->mea_count; i++) {
2306 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2307 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2312 OBD_FREE(*tmea, mea_size);
2316 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2317 struct lov_stripe_md *ea, obd_count oa_bufs,
2318 struct brw_page *pgarr, struct obd_trans_info *oti)
2320 struct obd_device *obd = exp->exp_obd;
2321 struct lmv_obd *lmv = &obd->u.lmv;
2322 struct mea *mea = (struct mea *) ea;
2325 LASSERT(oa != NULL);
2326 LASSERT(ea != NULL);
2327 LASSERT(pgarr != NULL);
2328 LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2330 oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2331 oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2332 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2334 err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2335 oa, NULL, oa_bufs, pgarr, oti);
2339 static int lmv_cancel_unused(struct obd_export *exp,
2340 struct lov_stripe_md *lsm,
2341 int flags, void *opaque)
2343 struct obd_device *obd = exp->exp_obd;
2344 struct lmv_obd *lmv = &obd->u.lmv;
2348 LASSERT(lsm == NULL);
2350 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2351 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active)
2354 err = obd_cancel_unused(lmv->tgts[i].ltd_exp,
2355 NULL, flags, opaque);
2362 struct obd_ops lmv_obd_ops = {
2363 .o_owner = THIS_MODULE,
2364 .o_attach = lmv_attach,
2365 .o_detach = lmv_detach,
2366 .o_setup = lmv_setup,
2367 .o_cleanup = lmv_cleanup,
2368 .o_process_config = lmv_process_config,
2369 .o_connect = lmv_connect,
2370 .o_disconnect = lmv_disconnect,
2371 .o_statfs = lmv_statfs,
2372 .o_llog_init = lmv_llog_init,
2373 .o_llog_finish = lmv_llog_finish,
2374 .o_get_info = lmv_get_info,
2375 .o_set_info = lmv_set_info,
2376 .o_create = lmv_obd_create,
2377 .o_packmd = lmv_packmd,
2378 .o_unpackmd = lmv_unpackmd,
2380 .o_init_ea_size = lmv_init_ea_size,
2381 .o_notify = lmv_notify,
2382 .o_iocontrol = lmv_iocontrol,
2383 .o_cancel_unused = lmv_cancel_unused,
2386 struct md_ops lmv_md_ops = {
2387 .m_getstatus = lmv_getstatus,
2388 .m_getattr = lmv_getattr,
2389 .m_change_cbdata = lmv_change_cbdata,
2390 .m_change_cbdata_name = lmv_change_cbdata_name,
2391 .m_close = lmv_close,
2392 .m_create = lmv_create,
2393 .m_done_writing = lmv_done_writing,
2394 .m_enqueue = lmv_enqueue,
2395 .m_getattr_lock = lmv_getattr_lock,
2396 .m_intent_lock = lmv_intent_lock,
2398 .m_rename = lmv_rename,
2399 .m_setattr = lmv_setattr,
2401 .m_readpage = lmv_readpage,
2402 .m_unlink = lmv_unlink,
2403 .m_get_real_obd = lmv_get_real_obd,
2404 .m_valid_attrs = lmv_valid_attrs,
2405 .m_delete_inode = lmv_delete_inode,
2406 .m_access_check = lmv_access_check,
2409 int __init lmv_init(void)
2411 struct lprocfs_static_vars lvars;
2414 obj_cache = kmem_cache_create("lmv_objects",
2415 sizeof(struct lmv_obj),
2418 CERROR("error allocating lmv objects cache\n");
2422 lprocfs_init_vars(lmv, &lvars);
2423 rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2425 OBD_LMV_DEVICENAME);
2427 kmem_cache_destroy(obj_cache);
2433 static void lmv_exit(void)
2435 class_unregister_type(OBD_LMV_DEVICENAME);
2437 LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2438 "can't free lmv objects cache, %d object(s)"
2439 "still in use\n", atomic_read(&obj_cache_count));
2442 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2443 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2444 MODULE_LICENSE("GPL");
2446 module_init(lmv_init);
2447 module_exit(lmv_exit);