4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osp/osp_dev.c
38 * Lustre OST Proxy Device
40 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 * Author: Mikhail Pershin <mike.pershin@intel.com>
42 * Author: Di Wang <di.wang@intel.com>
45 #define DEBUG_SUBSYSTEM S_MDS
47 #include <obd_class.h>
48 #include <lustre_param.h>
49 #include <lustre_log.h>
50 #include <lustre_mdc.h>
52 #include "osp_internal.h"
54 /* Slab for OSP object allocation */
55 struct kmem_cache *osp_object_kmem;
57 static struct lu_kmem_descr osp_caches[] = {
59 .ckd_cache = &osp_object_kmem,
60 .ckd_name = "osp_obj",
61 .ckd_size = sizeof(struct osp_object)
68 struct lu_object *osp_object_alloc(const struct lu_env *env,
69 const struct lu_object_header *hdr,
72 struct lu_object_header *h = NULL;
76 OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
78 l = &o->opo_obj.do_lu;
80 /* If hdr is NULL, it means the object is not built
81 * from the top dev(MDT/OST), usually it happens when
82 * building striped object, like data object on MDT or
83 * striped object for directory */
86 lu_object_header_init(h);
87 dt_object_init(&o->opo_obj, h, d);
88 lu_object_add_top(h, l);
90 dt_object_init(&o->opo_obj, h, d);
93 l->lo_ops = &osp_lu_obj_ops;
101 static struct dt_object
102 *osp_find_or_create(const struct lu_env *env, struct osp_device *osp,
103 struct lu_attr *attr, __u32 reg_id)
105 struct osp_thread_info *osi = osp_env_info(env);
106 struct dt_object_format dof = { 0 };
107 struct dt_object *dto;
111 lu_local_obj_fid(&osi->osi_fid, reg_id);
112 attr->la_valid = LA_MODE;
113 attr->la_mode = S_IFREG | 0644;
114 dof.dof_type = DFT_REGULAR;
115 dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
120 rc = dt_attr_get(env, dto, attr, NULL);
122 CERROR("%s: can't be initialized: rc = %d\n",
123 osp->opd_obd->obd_name, rc);
124 lu_object_put(env, &dto->do_lu);
130 static int osp_write_local_file(const struct lu_env *env,
131 struct osp_device *osp,
132 struct dt_object *dt_obj,
139 th = dt_trans_create(env, osp->opd_storage);
143 rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
146 rc = dt_trans_start_local(env, osp->opd_storage, th);
150 rc = dt_record_write(env, dt_obj, buf, &offset, th);
152 dt_trans_stop(env, osp->opd_storage, th);
156 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
158 struct osp_thread_info *osi = osp_env_info(env);
159 struct lu_fid *fid = &osp->opd_last_used_fid;
160 struct dt_object *dto;
164 dto = osp_find_or_create(env, osp, &osi->osi_attr, MDD_LOV_OBJ_OID);
166 RETURN(PTR_ERR(dto));
167 /* object will be released in device cleanup path */
168 if (osi->osi_attr.la_size >=
169 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
170 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
172 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
177 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
179 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
182 osp->opd_last_used_oid_file = dto;
185 /* object will be released in device cleanup path */
186 CERROR("%s: can't initialize lov_objid: rc = %d\n",
187 osp->opd_obd->obd_name, rc);
188 lu_object_put(env, &dto->do_lu);
189 osp->opd_last_used_oid_file = NULL;
193 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
195 struct osp_thread_info *osi = osp_env_info(env);
196 struct lu_fid *fid = &osp->opd_last_used_fid;
197 struct dt_object *dto;
201 dto = osp_find_or_create(env, osp, &osi->osi_attr, MDD_LOV_OBJ_OSEQ);
203 RETURN(PTR_ERR(dto));
205 /* object will be released in device cleanup path */
206 if (osi->osi_attr.la_size >=
207 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
208 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
210 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
215 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
217 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
220 osp->opd_last_used_seq_file = dto;
223 /* object will be released in device cleanup path */
224 CERROR("%s: can't initialize lov_seq: rc = %d\n",
225 osp->opd_obd->obd_name, rc);
226 lu_object_put(env, &dto->do_lu);
227 osp->opd_last_used_seq_file = NULL;
231 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
233 struct osp_thread_info *osi = osp_env_info(env);
237 fid_zero(&osp->opd_last_used_fid);
238 rc = osp_init_last_objid(env, osp);
240 CERROR("%s: Can not get ids %d from old objid!\n",
241 osp->opd_obd->obd_name, rc);
245 rc = osp_init_last_seq(env, osp);
247 CERROR("%s: Can not get ids %d from old objid!\n",
248 osp->opd_obd->obd_name, rc);
252 if (fid_oid(&osp->opd_last_used_fid) != 0 &&
253 fid_seq(&osp->opd_last_used_fid) == 0) {
254 /* Just upgrade from the old version,
255 * set the seq to be IDIF */
256 osp->opd_last_used_fid.f_seq =
257 fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
259 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
260 &osp->opd_last_used_fid.f_seq,
262 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
263 &osi->osi_lb, osi->osi_off);
265 CERROR("%s : Can not write seq file: rc = %d\n",
266 osp->opd_obd->obd_name, rc);
271 if (!fid_is_zero(&osp->opd_last_used_fid) &&
272 !fid_is_sane(&osp->opd_last_used_fid)) {
273 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
274 PFID(&osp->opd_last_used_fid));
275 GOTO(out, rc = -EINVAL);
278 CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
279 osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
282 if (osp->opd_last_used_oid_file != NULL) {
283 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
284 osp->opd_last_used_oid_file = NULL;
286 if (osp->opd_last_used_seq_file != NULL) {
287 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
288 osp->opd_last_used_seq_file = NULL;
295 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *d)
297 /* release last_used file */
298 if (d->opd_last_used_oid_file != NULL) {
299 lu_object_put(env, &d->opd_last_used_oid_file->do_lu);
300 d->opd_last_used_oid_file = NULL;
303 if (d->opd_last_used_seq_file != NULL) {
304 lu_object_put(env, &d->opd_last_used_seq_file->do_lu);
305 d->opd_last_used_seq_file = NULL;
309 static int osp_disconnect(struct osp_device *d)
311 struct obd_import *imp;
314 imp = d->opd_obd->u.cli.cl_import;
316 /* Mark import deactivated now, so we don't try to reconnect if any
317 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
318 * fully deactivate the import, or that would drop all requests. */
319 LASSERT(imp != NULL);
320 spin_lock(&imp->imp_lock);
321 imp->imp_deactive = 1;
322 spin_unlock(&imp->imp_lock);
324 ptlrpc_deactivate_import(imp);
326 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
327 * delete it regardless. (It's safe to delete an import that was
329 (void)ptlrpc_pinger_del_import(imp);
331 rc = ptlrpc_disconnect_import(imp, 0);
333 CERROR("%s: can't disconnect: rc = %d\n",
334 d->opd_obd->obd_name, rc);
336 ptlrpc_invalidate_import(imp);
341 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
348 rc = osp_disconnect(d);
352 if (!d->opd_connect_mdt) {
353 /* stop precreate thread */
354 osp_precreate_fini(d);
356 /* release last_used file */
357 osp_last_used_fini(env, d);
360 obd_fid_fini(d->opd_obd);
365 static int osp_process_config(const struct lu_env *env,
366 struct lu_device *dev, struct lustre_cfg *lcfg)
368 struct osp_device *d = lu2osp_dev(dev);
369 struct lprocfs_static_vars lvars = { 0 };
374 switch (lcfg->lcfg_command) {
375 case LCFG_PRE_CLEANUP:
376 rc = osp_disconnect(d);
379 lu_dev_del_linkage(dev->ld_site, dev);
380 rc = osp_shutdown(env, d);
383 lprocfs_osp_init_vars(&lvars);
386 rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
391 /* class_process_proc_param() haven't found matching
392 * parameter and returned ENOSYS so that layer(s)
393 * below could use that. But OSP is the bottom, so
395 CERROR("%s: unknown param %s\n",
396 (char *)lustre_cfg_string(lcfg, 0),
397 (char *)lustre_cfg_string(lcfg, 1));
402 CERROR("%s: unknown command %u\n",
403 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
411 static int osp_recovery_complete(const struct lu_env *env,
412 struct lu_device *dev)
414 struct osp_device *osp = lu2osp_dev(dev);
418 osp->opd_recovery_completed = 1;
420 if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
421 wake_up(&osp->opd_pre_waitq);
425 const struct lu_device_operations osp_lu_ops = {
426 .ldo_object_alloc = osp_object_alloc,
427 .ldo_process_config = osp_process_config,
428 .ldo_recovery_complete = osp_recovery_complete,
432 * provides with statfs from corresponded OST
435 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
436 struct obd_statfs *sfs)
438 struct osp_device *d = dt2osp_dev(dev);
442 if (unlikely(d->opd_imp_active == 0))
445 if (d->opd_pre == NULL)
448 /* return recently updated data */
449 *sfs = d->opd_statfs;
452 * layer above osp (usually lod) can use ffree to estimate
453 * how many objects are available for immediate creation
455 spin_lock(&d->opd_pre_lock);
456 LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
457 fid_seq(&d->opd_pre_used_fid),
458 "last_created "DFID", next_fid "DFID"\n",
459 PFID(&d->opd_pre_last_created_fid),
460 PFID(&d->opd_pre_used_fid));
461 sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
462 fid_oid(&d->opd_pre_used_fid);
463 sfs->os_fprecreated -= d->opd_pre_reserved;
464 spin_unlock(&d->opd_pre_lock);
466 LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
468 CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
469 LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
470 sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
471 sfs->os_files, sfs->os_ffree);
475 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
480 * XXX: wake up sync thread, command it to start flushing asap?
486 const struct dt_device_operations osp_dt_ops = {
487 .dt_statfs = osp_statfs,
489 .dt_trans_create = osp_trans_create,
490 .dt_trans_start = osp_trans_start,
491 .dt_trans_stop = osp_trans_stop,
494 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *m,
497 struct obd_connect_data *data = NULL;
498 struct obd_device *obd;
503 LASSERT(m->opd_storage_exp == NULL);
509 obd = class_name2obd(nextdev);
511 CERROR("%s: can't locate next device: %s\n",
512 m->opd_obd->obd_name, nextdev);
513 GOTO(out, rc = -ENOTCONN);
516 rc = obd_connect(env, &m->opd_storage_exp, obd, &obd->obd_uuid, data,
519 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
520 m->opd_obd->obd_name, nextdev, rc);
524 m->opd_dt_dev.dd_lu_dev.ld_site =
525 m->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
526 LASSERT(m->opd_dt_dev.dd_lu_dev.ld_site);
527 m->opd_storage = lu2dt_dev(m->opd_storage_exp->exp_obd->obd_lu_dev);
534 #define OSP_MAX_AUIF_MAX 512
536 static int osp_init0(const struct lu_env *env, struct osp_device *m,
537 struct lu_device_type *ldt, struct lustre_cfg *cfg)
539 struct obd_device *obd;
540 struct obd_import *imp;
542 char *src, *tgt, *mdt, *osdname = NULL;
548 mutex_init(&m->opd_async_requests_mutex);
549 /* We allow OSP_MAX_AUIF_MAX async updates in flight at most. */
550 sema_init(&m->opd_async_fc_sem, OSP_MAX_AUIF_MAX);
552 obd = class_name2obd(lustre_cfg_string(cfg, 0));
554 CERROR("Cannot find obd with name %s\n",
555 lustre_cfg_string(cfg, 0));
560 /* There is no record in the MDT configuration for the local disk
561 * device, so we have to extract this from elsewhere in the profile.
562 * The only information we get at setup is from the OSC records:
563 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
564 * Note that 1.8 generated configs are missing the -MDTxxxx part.
565 * We need to reconstruct the name of the underlying OSD from this:
566 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd". We
567 * also need to determine the OST index from this - will be used
568 * to calculate the offset in shared lov_objids file later */
570 src = lustre_cfg_string(cfg, 0);
574 tgt = strrchr(src, '-');
576 CERROR("%s: invalid target name %s\n",
577 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
581 if (strncmp(tgt, "-osc", 4) == 0) {
582 /* Old OSC name fsname-OSTXXXX-osc */
583 for (tgt--; tgt > src && *tgt != '-'; tgt--)
586 CERROR("%s: invalid target name %s\n",
587 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
591 if (strncmp(tgt, "-OST", 4) != 0) {
592 CERROR("%s: invalid target name %s\n",
593 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
597 idx = simple_strtol(tgt + 4, &mdt, 16);
598 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
599 CERROR("%s: invalid OST index in '%s'\n",
600 m->opd_obd->obd_name, src);
607 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
608 if (strncmp(tgt, "-MDT", 4) != 0 &&
609 strncmp(tgt, "-OST", 4) != 0) {
610 CERROR("%s: invalid target name %s\n",
611 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
615 idx = simple_strtol(tgt + 4, &mdt, 16);
616 if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
617 CERROR("%s: invalid OST index in '%s'\n",
618 m->opd_obd->obd_name, src);
622 /* Get MDT index from the name and set it to opd_group,
623 * which will be used by OSP to connect with OST */
625 if (tgt - src <= 12) {
626 CERROR("%s: invalid mdt index retrieve from %s\n",
627 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
631 if (strncmp(tgt - 12, "-MDT", 4) == 0)
632 m->opd_connect_mdt = 1;
634 idx = simple_strtol(tgt - 8, &mdt, 16);
635 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
636 CERROR("%s: invalid OST index in '%s'\n",
637 m->opd_obd->obd_name, src);
642 idx = tgt - src - 12;
644 /* check the fsname length, and after this everything else will fit */
645 if (idx > MTI_NAME_MAXLEN) {
646 CERROR("%s: fsname too long in '%s'\n",
647 m->opd_obd->obd_name, src);
651 OBD_ALLOC(osdname, MAX_OBD_NAME);
655 memcpy(osdname, src, idx); /* copy just the fsname part */
658 mdt = strstr(mdt, "-MDT");
659 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
660 strcat(osdname, "-MDT0000");
662 strcat(osdname, mdt);
663 strcat(osdname, "-osd");
664 CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
666 if (m->opd_connect_mdt) {
667 struct client_obd *cli = &m->opd_obd->u.cli;
669 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
670 if (!cli->cl_rpc_lock)
671 GOTO(out_fini, rc = -ENOMEM);
672 osp_init_rpc_lock(cli->cl_rpc_lock);
675 m->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
676 m->opd_dt_dev.dd_ops = &osp_dt_ops;
677 obd->obd_lu_dev = &m->opd_dt_dev.dd_lu_dev;
679 rc = osp_connect_to_osd(env, m, osdname);
683 rc = ptlrpcd_addref();
685 GOTO(out_disconnect, rc);
687 rc = client_obd_setup(obd, cfg);
689 CERROR("%s: can't setup obd: %d\n", m->opd_obd->obd_name, rc);
695 rc = obd_fid_init(m->opd_obd, NULL, m->opd_connect_mdt ?
696 LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
698 CERROR("%s: fid init error: rc = %d\n",
699 m->opd_obd->obd_name, rc);
703 if (!m->opd_connect_mdt) {
704 /* Initialize last id from the storage - will be
705 * used in orphan cleanup. */
706 rc = osp_last_used_init(env, m);
711 /* Initialize precreation thread, it handles new
712 * connections as well. */
713 rc = osp_init_precreate(m);
715 GOTO(out_last_used, rc);
719 * Initialize synhronization mechanism taking
720 * care of propogating changes to OST in near
721 * transactional manner.
723 rc = osp_sync_init(env, m);
725 GOTO(out_precreat, rc);
728 * Initiate connect to OST
730 ll_generate_random_uuid(uuid);
731 class_uuid_unparse(uuid, &m->opd_cluuid);
733 imp = obd->u.cli.cl_import;
735 rc = ptlrpc_init_import(imp);
739 OBD_FREE(osdname, MAX_OBD_NAME);
743 /* stop sync thread */
746 /* stop precreate thread */
747 if (!m->opd_connect_mdt)
748 osp_precreate_fini(m);
750 if (!m->opd_connect_mdt)
751 osp_last_used_fini(env, m);
753 ptlrpc_lprocfs_unregister_obd(obd);
754 lprocfs_obd_cleanup(obd);
756 lprocfs_remove(&m->opd_symlink);
757 client_obd_cleanup(obd);
761 if (m->opd_connect_mdt) {
762 struct client_obd *cli = &m->opd_obd->u.cli;
763 if (cli->cl_rpc_lock != NULL) {
764 OBD_FREE_PTR(cli->cl_rpc_lock);
765 cli->cl_rpc_lock = NULL;
768 obd_disconnect(m->opd_storage_exp);
771 OBD_FREE(osdname, MAX_OBD_NAME);
775 static struct lu_device *osp_device_free(const struct lu_env *env,
776 struct lu_device *lu)
778 struct osp_device *m = lu2osp_dev(lu);
782 if (atomic_read(&lu->ld_ref) && lu->ld_site) {
783 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
784 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
786 dt_device_fini(&m->opd_dt_dev);
791 static struct lu_device *osp_device_alloc(const struct lu_env *env,
792 struct lu_device_type *t,
793 struct lustre_cfg *lcfg)
795 struct osp_device *m;
800 l = ERR_PTR(-ENOMEM);
805 dt_device_init(&m->opd_dt_dev, t);
806 rc = osp_init0(env, m, t, lcfg);
808 osp_device_free(env, l);
815 static struct lu_device *osp_device_fini(const struct lu_env *env,
818 struct osp_device *m = lu2osp_dev(d);
819 struct obd_import *imp;
824 if (m->opd_async_requests != NULL) {
825 out_destroy_update_req(m->opd_async_requests);
826 m->opd_async_requests = NULL;
829 if (m->opd_storage_exp)
830 obd_disconnect(m->opd_storage_exp);
832 imp = m->opd_obd->u.cli.cl_import;
834 if (imp->imp_rq_pool) {
835 ptlrpc_free_rq_pool(imp->imp_rq_pool);
836 imp->imp_rq_pool = NULL;
840 lprocfs_remove(&m->opd_symlink);
843 ptlrpc_lprocfs_unregister_obd(m->opd_obd);
844 lprocfs_obd_cleanup(m->opd_obd);
846 if (m->opd_connect_mdt) {
847 struct client_obd *cli = &m->opd_obd->u.cli;
848 if (cli->cl_rpc_lock != NULL) {
849 OBD_FREE_PTR(cli->cl_rpc_lock);
850 cli->cl_rpc_lock = NULL;
854 rc = client_obd_cleanup(m->opd_obd);
855 LASSERTF(rc == 0, "error %d\n", rc);
862 static int osp_reconnect(const struct lu_env *env,
863 struct obd_export *exp, struct obd_device *obd,
864 struct obd_uuid *cluuid,
865 struct obd_connect_data *data,
871 static int osp_prepare_fid_client(struct osp_device *osp)
873 LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
874 if (osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL)
877 LASSERT(osp->opd_exp != NULL);
878 osp->opd_obd->u.cli.cl_seq->lcs_exp =
879 class_export_get(osp->opd_exp);
884 * we use exports to track all LOD users
886 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
887 struct obd_device *obd, struct obd_uuid *cluuid,
888 struct obd_connect_data *data, void *localdata)
890 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
891 struct obd_connect_data *ocd;
892 struct obd_import *imp;
893 struct lustre_handle conn;
898 CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
900 rc = class_connect(&conn, obd, cluuid);
904 *exp = class_conn2export(&conn);
905 /* Why should there ever be more than 1 connect? */
907 LASSERT(osp->opd_connects == 1);
911 imp = osp->opd_obd->u.cli.cl_import;
912 imp->imp_dlm_handle = conn;
914 LASSERT(data != NULL);
915 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
916 ocd = &imp->imp_connect_data;
919 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
921 ocd->ocd_version = LUSTRE_VERSION_CODE;
922 ocd->ocd_index = data->ocd_index;
923 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
925 rc = ptlrpc_connect_import(imp);
927 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
931 ptlrpc_pinger_add_import(imp);
937 * once last export (we don't count self-export) disappeared
938 * osp can be released
940 static int osp_obd_disconnect(struct obd_export *exp)
942 struct obd_device *obd = exp->exp_obd;
943 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
947 /* Only disconnect the underlying layers on the final disconnect. */
948 LASSERT(osp->opd_connects == 1);
951 rc = class_disconnect(exp);
953 CERROR("%s: class disconnect error: rc = %d\n",
958 /* destroy the device */
959 class_manual_cleanup(obd);
965 * lprocfs helpers still use OBD API, let's keep obd_statfs() support
967 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
968 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
970 struct obd_statfs *msfs;
971 struct ptlrpc_request *req;
972 struct obd_import *imp = NULL;
977 /* Since the request might also come from lprocfs, so we need
978 * sync this with client_disconnect_export Bug15684 */
979 down_read(&exp->exp_obd->u.cli.cl_sem);
980 if (exp->exp_obd->u.cli.cl_import)
981 imp = class_import_get(exp->exp_obd->u.cli.cl_import);
982 up_read(&exp->exp_obd->u.cli.cl_sem);
986 /* We could possibly pass max_age in the request (as an absolute
987 * timestamp or a "seconds.usec ago") so the target can avoid doing
988 * extra calls into the filesystem if that isn't necessary (e.g.
989 * during mount that would help a bit). Having relative timestamps
990 * is not so great if request processing is slow, while absolute
991 * timestamps are not ideal because they need time synchronization. */
992 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
994 class_import_put(imp);
999 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
1001 ptlrpc_request_free(req);
1004 ptlrpc_request_set_replen(req);
1005 req->rq_request_portal = OST_CREATE_PORTAL;
1006 ptlrpc_at_set_req_timeout(req);
1008 if (flags & OBD_STATFS_NODELAY) {
1009 /* procfs requests not want stat in wait for avoid deadlock */
1010 req->rq_no_resend = 1;
1011 req->rq_no_delay = 1;
1014 rc = ptlrpc_queue_wait(req);
1018 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1020 GOTO(out, rc = -EPROTO);
1026 ptlrpc_req_finished(req);
1030 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1031 enum obd_import_event event)
1033 struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1036 case IMP_EVENT_DISCON:
1037 d->opd_got_disconnected = 1;
1038 d->opd_imp_connected = 0;
1039 if (d->opd_connect_mdt)
1042 if (d->opd_pre != NULL) {
1043 osp_pre_update_status(d, -ENODEV);
1044 wake_up(&d->opd_pre_waitq);
1047 CDEBUG(D_HA, "got disconnected\n");
1049 case IMP_EVENT_INACTIVE:
1050 d->opd_imp_active = 0;
1051 if (d->opd_connect_mdt)
1054 if (d->opd_pre != NULL) {
1055 osp_pre_update_status(d, -ENODEV);
1056 wake_up(&d->opd_pre_waitq);
1059 CDEBUG(D_HA, "got inactive\n");
1061 case IMP_EVENT_ACTIVE:
1062 d->opd_imp_active = 1;
1064 if (osp_prepare_fid_client(d) != 0)
1067 if (d->opd_got_disconnected)
1068 d->opd_new_connection = 1;
1069 d->opd_imp_connected = 1;
1070 d->opd_imp_seen_connected = 1;
1071 if (d->opd_connect_mdt)
1074 if (d->opd_pre != NULL)
1075 wake_up(&d->opd_pre_waitq);
1077 __osp_sync_check_for_work(d);
1078 CDEBUG(D_HA, "got connected\n");
1080 case IMP_EVENT_INVALIDATE:
1081 if (obd->obd_namespace == NULL)
1083 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1086 case IMP_EVENT_DEACTIVATE:
1087 case IMP_EVENT_ACTIVATE:
1090 CERROR("%s: unsupported import event: %#x\n",
1091 obd->obd_name, event);
1096 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1097 void *karg, void *uarg)
1099 struct obd_device *obd = exp->exp_obd;
1100 struct osp_device *d;
1101 struct obd_ioctl_data *data = karg;
1106 LASSERT(obd->obd_lu_dev);
1107 d = lu2osp_dev(obd->obd_lu_dev);
1108 LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1110 if (!try_module_get(THIS_MODULE)) {
1111 CERROR("%s: can't get module. Is it alive?", obd->obd_name);
1116 case OBD_IOC_CLIENT_RECOVER:
1117 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1118 data->ioc_inlbuf1, 0);
1122 case IOC_OSC_SET_ACTIVE:
1123 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1126 case OBD_IOC_PING_TARGET:
1127 rc = ptlrpc_obd_ping(obd);
1130 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1131 cmd, current_comm());
1134 module_put(THIS_MODULE);
1138 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1139 __u32 keylen, void *key, __u32 *vallen, void *val,
1140 struct lov_stripe_md *lsm)
1144 if (KEY_IS(KEY_OSP_CONNECTED)) {
1145 struct obd_device *obd = exp->exp_obd;
1146 struct osp_device *osp;
1148 if (!obd->obd_set_up || obd->obd_stopping)
1151 osp = lu2osp_dev(obd->obd_lu_dev);
1154 * 1.8/2.0 behaviour is that OST being connected once at least
1155 * is considered "healthy". and one "healthy" OST is enough to
1156 * allow lustre clients to connect to MDS
1158 RETURN(!osp->opd_imp_seen_connected);
1164 int osp_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1165 struct md_op_data *op_data)
1167 struct client_obd *cli = &exp->exp_obd->u.cli;
1168 struct lu_client_seq *seq = cli->cl_seq;
1171 RETURN(seq_client_alloc_fid(NULL, seq, fid));
1174 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1175 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1176 static void osp_key_exit(const struct lu_context *ctx,
1177 struct lu_context_key *key, void *data)
1179 struct osp_thread_info *info = data;
1181 info->osi_attr.la_valid = 0;
1184 struct lu_context_key osp_thread_key = {
1185 .lct_tags = LCT_MD_THREAD,
1186 .lct_init = osp_key_init,
1187 .lct_fini = osp_key_fini,
1188 .lct_exit = osp_key_exit
1191 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1192 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1194 struct lu_context_key osp_txn_key = {
1195 .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE,
1196 .lct_init = osp_txn_key_init,
1197 .lct_fini = osp_txn_key_fini
1199 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1201 static struct lu_device_type_operations osp_device_type_ops = {
1202 .ldto_init = osp_type_init,
1203 .ldto_fini = osp_type_fini,
1205 .ldto_start = osp_type_start,
1206 .ldto_stop = osp_type_stop,
1208 .ldto_device_alloc = osp_device_alloc,
1209 .ldto_device_free = osp_device_free,
1211 .ldto_device_fini = osp_device_fini
1214 static struct lu_device_type osp_device_type = {
1215 .ldt_tags = LU_DEVICE_DT,
1216 .ldt_name = LUSTRE_OSP_NAME,
1217 .ldt_ops = &osp_device_type_ops,
1218 .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
1221 static struct obd_ops osp_obd_device_ops = {
1222 .o_owner = THIS_MODULE,
1223 .o_add_conn = client_import_add_conn,
1224 .o_del_conn = client_import_del_conn,
1225 .o_reconnect = osp_reconnect,
1226 .o_connect = osp_obd_connect,
1227 .o_disconnect = osp_obd_disconnect,
1228 .o_get_info = osp_obd_get_info,
1229 .o_import_event = osp_import_event,
1230 .o_iocontrol = osp_iocontrol,
1231 .o_statfs = osp_obd_statfs,
1232 .o_fid_init = client_fid_init,
1233 .o_fid_fini = client_fid_fini,
1234 .o_fid_alloc = osp_fid_alloc,
1237 struct llog_operations osp_mds_ost_orig_logops;
1239 static int __init osp_mod_init(void)
1241 struct lprocfs_static_vars lvars;
1242 cfs_proc_dir_entry_t *osc_proc_dir;
1245 rc = lu_kmem_init(osp_caches);
1249 lprocfs_osp_init_vars(&lvars);
1251 rc = class_register_type(&osp_obd_device_ops, NULL, NULL,
1252 #ifndef HAVE_ONLY_PROCFS_SEQ
1255 LUSTRE_OSP_NAME, &osp_device_type);
1257 /* create "osc" entry in procfs for compatibility purposes */
1259 lu_kmem_fini(osp_caches);
1263 lprocfs_lwp_init_vars(&lvars);
1265 rc = class_register_type(&lwp_obd_device_ops, NULL, NULL,
1266 #ifndef HAVE_ONLY_PROCFS_SEQ
1269 LUSTRE_LWP_NAME, &lwp_device_type);
1271 class_unregister_type(LUSTRE_OSP_NAME);
1272 lu_kmem_fini(osp_caches);
1276 /* Note: add_rec/delcare_add_rec will be only used by catalogs */
1277 osp_mds_ost_orig_logops = llog_osd_ops;
1278 osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
1279 osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1281 osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
1282 if (osc_proc_dir == NULL) {
1283 osc_proc_dir = lprocfs_register("osc", proc_lustre_root, NULL,
1285 if (IS_ERR(osc_proc_dir))
1286 CERROR("osp: can't create compat entry \"osc\": %d\n",
1287 (int) PTR_ERR(osc_proc_dir));
1292 static void __exit osp_mod_exit(void)
1294 lprocfs_try_remove_proc_entry("osc", proc_lustre_root);
1296 class_unregister_type(LUSTRE_LWP_NAME);
1297 class_unregister_type(LUSTRE_OSP_NAME);
1298 lu_kmem_fini(osp_caches);
1301 MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
1302 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
1303 MODULE_LICENSE("GPL");
1305 cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);