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_IO);
78 l = &o->opo_obj.do_lu;
80 /* For data object, OSP obj would always be the top
81 * object, i.e. hdr is always NULL, see lu_object_alloc.
82 * But for metadata object, we always build the object
83 * stack from MDT. i.e. mdt_object will be the top object
88 lu_object_header_init(h);
89 dt_object_init(&o->opo_obj, h, d);
90 lu_object_add_top(h, l);
92 dt_object_init(&o->opo_obj, h, d);
95 l->lo_ops = &osp_lu_obj_ops;
103 static struct dt_object
104 *osp_find_or_create(const struct lu_env *env, struct osp_device *osp,
105 struct lu_attr *attr, __u32 reg_id)
107 struct osp_thread_info *osi = osp_env_info(env);
108 struct dt_object_format dof = { 0 };
109 struct dt_object *dto;
113 lu_local_obj_fid(&osi->osi_fid, reg_id);
114 attr->la_valid = LA_MODE;
115 attr->la_mode = S_IFREG | 0644;
116 dof.dof_type = DFT_REGULAR;
117 dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
122 rc = dt_attr_get(env, dto, attr, NULL);
124 CERROR("%s: can't be initialized: rc = %d\n",
125 osp->opd_obd->obd_name, rc);
126 lu_object_put(env, &dto->do_lu);
132 static int osp_write_local_file(const struct lu_env *env,
133 struct osp_device *osp,
134 struct dt_object *dt_obj,
141 th = dt_trans_create(env, osp->opd_storage);
145 rc = dt_declare_record_write(env, dt_obj, buf->lb_len, offset, th);
148 rc = dt_trans_start_local(env, osp->opd_storage, th);
152 rc = dt_record_write(env, dt_obj, buf, &offset, th);
154 dt_trans_stop(env, osp->opd_storage, th);
158 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
160 struct osp_thread_info *osi = osp_env_info(env);
161 struct lu_fid *fid = &osp->opd_last_used_fid;
162 struct dt_object *dto;
166 dto = osp_find_or_create(env, osp, &osi->osi_attr, MDD_LOV_OBJ_OID);
168 RETURN(PTR_ERR(dto));
169 /* object will be released in device cleanup path */
170 if (osi->osi_attr.la_size >=
171 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
172 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
174 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
179 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
181 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
184 osp->opd_last_used_oid_file = dto;
187 /* object will be released in device cleanup path */
188 CERROR("%s: can't initialize lov_objid: rc = %d\n",
189 osp->opd_obd->obd_name, rc);
190 lu_object_put(env, &dto->do_lu);
191 osp->opd_last_used_oid_file = NULL;
195 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
197 struct osp_thread_info *osi = osp_env_info(env);
198 struct lu_fid *fid = &osp->opd_last_used_fid;
199 struct dt_object *dto;
203 dto = osp_find_or_create(env, osp, &osi->osi_attr, MDD_LOV_OBJ_OSEQ);
205 RETURN(PTR_ERR(dto));
207 /* object will be released in device cleanup path */
208 if (osi->osi_attr.la_size >=
209 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
210 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
212 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
217 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
219 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
222 osp->opd_last_used_seq_file = dto;
225 /* object will be released in device cleanup path */
226 CERROR("%s: can't initialize lov_seq: rc = %d\n",
227 osp->opd_obd->obd_name, rc);
228 lu_object_put(env, &dto->do_lu);
229 osp->opd_last_used_seq_file = NULL;
233 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
235 struct osp_thread_info *osi = osp_env_info(env);
239 fid_zero(&osp->opd_last_used_fid);
240 rc = osp_init_last_objid(env, osp);
242 CERROR("%s: Can not get ids %d from old objid!\n",
243 osp->opd_obd->obd_name, rc);
247 rc = osp_init_last_seq(env, osp);
249 CERROR("%s: Can not get ids %d from old objid!\n",
250 osp->opd_obd->obd_name, rc);
254 if (fid_oid(&osp->opd_last_used_fid) != 0 &&
255 fid_seq(&osp->opd_last_used_fid) == 0) {
256 /* Just upgrade from the old version,
257 * set the seq to be IDIF */
258 osp->opd_last_used_fid.f_seq =
259 fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
261 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
262 &osp->opd_last_used_fid.f_seq,
264 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
265 &osi->osi_lb, osi->osi_off);
267 CERROR("%s : Can not write seq file: rc = %d\n",
268 osp->opd_obd->obd_name, rc);
273 if (!fid_is_zero(&osp->opd_last_used_fid) &&
274 !fid_is_sane(&osp->opd_last_used_fid)) {
275 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
276 PFID(&osp->opd_last_used_fid));
277 GOTO(out, rc = -EINVAL);
280 CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
281 osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
284 if (osp->opd_last_used_oid_file != NULL) {
285 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
286 osp->opd_last_used_oid_file = NULL;
288 if (osp->opd_last_used_seq_file != NULL) {
289 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
290 osp->opd_last_used_seq_file = NULL;
297 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *d)
299 /* release last_used file */
300 if (d->opd_last_used_oid_file != NULL) {
301 lu_object_put(env, &d->opd_last_used_oid_file->do_lu);
302 d->opd_last_used_oid_file = NULL;
305 if (d->opd_last_used_seq_file != NULL) {
306 lu_object_put(env, &d->opd_last_used_seq_file->do_lu);
307 d->opd_last_used_seq_file = NULL;
311 static int osp_disconnect(struct osp_device *d)
313 struct obd_import *imp;
316 imp = d->opd_obd->u.cli.cl_import;
318 /* Mark import deactivated now, so we don't try to reconnect if any
319 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
320 * fully deactivate the import, or that would drop all requests. */
321 LASSERT(imp != NULL);
322 spin_lock(&imp->imp_lock);
323 imp->imp_deactive = 1;
324 spin_unlock(&imp->imp_lock);
326 ptlrpc_deactivate_import(imp);
328 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
329 * delete it regardless. (It's safe to delete an import that was
331 (void)ptlrpc_pinger_del_import(imp);
333 rc = ptlrpc_disconnect_import(imp, 0);
334 if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ESHUTDOWN)
337 CERROR("%s: can't disconnect: rc = %d\n",
338 d->opd_obd->obd_name, rc);
340 ptlrpc_invalidate_import(imp);
345 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
352 rc = osp_disconnect(d);
354 if (!d->opd_connect_mdt) {
355 /* stop sync thread */
358 /* stop precreate thread */
359 osp_precreate_fini(d);
361 /* release last_used file */
362 osp_last_used_fini(env, d);
365 obd_fid_fini(d->opd_obd);
370 static int osp_process_config(const struct lu_env *env,
371 struct lu_device *dev, struct lustre_cfg *lcfg)
373 struct osp_device *d = lu2osp_dev(dev);
374 struct lprocfs_static_vars lvars = { 0 };
379 switch (lcfg->lcfg_command) {
380 case LCFG_PRE_CLEANUP:
381 rc = osp_disconnect(d);
384 lu_dev_del_linkage(dev->ld_site, dev);
385 rc = osp_shutdown(env, d);
388 lprocfs_osp_init_vars(&lvars);
391 rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
396 /* class_process_proc_param() haven't found matching
397 * parameter and returned ENOSYS so that layer(s)
398 * below could use that. But OSP is the bottom, so
400 CERROR("%s: unknown param %s\n",
401 (char *)lustre_cfg_string(lcfg, 0),
402 (char *)lustre_cfg_string(lcfg, 1));
407 CERROR("%s: unknown command %u\n",
408 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
416 static int osp_recovery_complete(const struct lu_env *env,
417 struct lu_device *dev)
419 struct osp_device *osp = lu2osp_dev(dev);
423 osp->opd_recovery_completed = 1;
425 if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
426 wake_up(&osp->opd_pre_waitq);
430 const struct lu_device_operations osp_lu_ops = {
431 .ldo_object_alloc = osp_object_alloc,
432 .ldo_process_config = osp_process_config,
433 .ldo_recovery_complete = osp_recovery_complete,
437 * provides with statfs from corresponded OST
440 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
441 struct obd_statfs *sfs)
443 struct osp_device *d = dt2osp_dev(dev);
447 if (unlikely(d->opd_imp_active == 0))
450 if (d->opd_pre == NULL)
453 /* return recently updated data */
454 *sfs = d->opd_statfs;
457 * layer above osp (usually lod) can use ffree to estimate
458 * how many objects are available for immediate creation
460 spin_lock(&d->opd_pre_lock);
461 LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
462 fid_seq(&d->opd_pre_used_fid),
463 "last_created "DFID", next_fid "DFID"\n",
464 PFID(&d->opd_pre_last_created_fid),
465 PFID(&d->opd_pre_used_fid));
466 sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
467 fid_oid(&d->opd_pre_used_fid);
468 sfs->os_fprecreated -= d->opd_pre_reserved;
469 spin_unlock(&d->opd_pre_lock);
471 LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
473 CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
474 LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
475 sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
476 sfs->os_files, sfs->os_ffree);
480 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
485 * XXX: wake up sync thread, command it to start flushing asap?
491 const struct dt_device_operations osp_dt_ops = {
492 .dt_statfs = osp_statfs,
494 .dt_trans_start = osp_trans_start,
495 .dt_trans_stop = osp_trans_stop,
498 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *m,
501 struct obd_connect_data *data = NULL;
502 struct obd_device *obd;
507 LASSERT(m->opd_storage_exp == NULL);
513 obd = class_name2obd(nextdev);
515 CERROR("%s: can't locate next device: %s\n",
516 m->opd_obd->obd_name, nextdev);
517 GOTO(out, rc = -ENOTCONN);
520 rc = obd_connect(env, &m->opd_storage_exp, obd, &obd->obd_uuid, data,
523 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
524 m->opd_obd->obd_name, nextdev, rc);
528 m->opd_dt_dev.dd_lu_dev.ld_site =
529 m->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
530 LASSERT(m->opd_dt_dev.dd_lu_dev.ld_site);
531 m->opd_storage = lu2dt_dev(m->opd_storage_exp->exp_obd->obd_lu_dev);
538 static int osp_init0(const struct lu_env *env, struct osp_device *m,
539 struct lu_device_type *ldt, struct lustre_cfg *cfg)
541 struct obd_device *obd;
542 struct obd_import *imp;
544 char *src, *tgt, *mdt, *osdname = NULL;
550 obd = class_name2obd(lustre_cfg_string(cfg, 0));
552 CERROR("Cannot find obd with name %s\n",
553 lustre_cfg_string(cfg, 0));
558 /* There is no record in the MDT configuration for the local disk
559 * device, so we have to extract this from elsewhere in the profile.
560 * The only information we get at setup is from the OSC records:
561 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
562 * Note that 1.8 generated configs are missing the -MDTxxxx part.
563 * We need to reconstruct the name of the underlying OSD from this:
564 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd". We
565 * also need to determine the OST index from this - will be used
566 * to calculate the offset in shared lov_objids file later */
568 src = lustre_cfg_string(cfg, 0);
572 tgt = strrchr(src, '-');
574 CERROR("%s: invalid target name %s\n",
575 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
579 if (strncmp(tgt, "-osc", 4) == 0) {
580 /* Old OSC name fsname-OSTXXXX-osc */
581 for (tgt--; tgt > src && *tgt != '-'; tgt--)
584 CERROR("%s: invalid target name %s\n",
585 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
589 if (strncmp(tgt, "-OST", 4) != 0) {
590 CERROR("%s: invalid target name %s\n",
591 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
595 idx = simple_strtol(tgt + 4, &mdt, 16);
596 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
597 CERROR("%s: invalid OST index in '%s'\n",
598 m->opd_obd->obd_name, src);
605 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
606 if (strncmp(tgt, "-MDT", 4) != 0 &&
607 strncmp(tgt, "-OST", 4) != 0) {
608 CERROR("%s: invalid target name %s\n",
609 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
613 idx = simple_strtol(tgt + 4, &mdt, 16);
614 if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
615 CERROR("%s: invalid OST index in '%s'\n",
616 m->opd_obd->obd_name, src);
620 /* Get MDT index from the name and set it to opd_group,
621 * which will be used by OSP to connect with OST */
623 if (tgt - src <= 12) {
624 CERROR("%s: invalid mdt index retrieve from %s\n",
625 m->opd_obd->obd_name, lustre_cfg_string(cfg, 0));
629 if (strncmp(tgt - 12, "-MDT", 4) == 0)
630 m->opd_connect_mdt = 1;
632 idx = simple_strtol(tgt - 8, &mdt, 16);
633 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
634 CERROR("%s: invalid OST index in '%s'\n",
635 m->opd_obd->obd_name, src);
640 idx = tgt - src - 12;
642 /* check the fsname length, and after this everything else will fit */
643 if (idx > MTI_NAME_MAXLEN) {
644 CERROR("%s: fsname too long in '%s'\n",
645 m->opd_obd->obd_name, src);
649 OBD_ALLOC(osdname, MAX_OBD_NAME);
653 memcpy(osdname, src, idx); /* copy just the fsname part */
656 mdt = strstr(mdt, "-MDT");
657 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
658 strcat(osdname, "-MDT0000");
660 strcat(osdname, mdt);
661 strcat(osdname, "-osd");
662 CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
664 if (m->opd_connect_mdt) {
665 struct client_obd *cli = &m->opd_obd->u.cli;
667 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
668 if (!cli->cl_rpc_lock)
669 GOTO(out_fini, rc = -ENOMEM);
670 osp_init_rpc_lock(cli->cl_rpc_lock);
673 m->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
674 m->opd_dt_dev.dd_ops = &osp_dt_ops;
675 obd->obd_lu_dev = &m->opd_dt_dev.dd_lu_dev;
677 rc = osp_connect_to_osd(env, m, osdname);
681 rc = ptlrpcd_addref();
683 GOTO(out_disconnect, rc);
685 rc = client_obd_setup(obd, cfg);
687 CERROR("%s: can't setup obd: %d\n", m->opd_obd->obd_name, rc);
693 rc = obd_fid_init(m->opd_obd, NULL, m->opd_connect_mdt ?
694 LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
696 CERROR("%s: fid init error: rc = %d\n",
697 m->opd_obd->obd_name, rc);
701 if (!m->opd_connect_mdt) {
702 /* Initialize last id from the storage - will be
703 * used in orphan cleanup. */
704 rc = osp_last_used_init(env, m);
709 /* Initialize precreation thread, it handles new
710 * connections as well. */
711 rc = osp_init_precreate(m);
713 GOTO(out_last_used, rc);
715 * Initialize synhronization mechanism taking
716 * care of propogating changes to OST in near
717 * transactional manner.
719 rc = osp_sync_init(env, m);
721 GOTO(out_precreat, rc);
725 * Initiate connect to OST
727 ll_generate_random_uuid(uuid);
728 class_uuid_unparse(uuid, &m->opd_cluuid);
730 imp = obd->u.cli.cl_import;
732 rc = ptlrpc_init_import(imp);
736 OBD_FREE(osdname, MAX_OBD_NAME);
740 if (!m->opd_connect_mdt)
741 /* stop sync thread */
744 /* stop precreate thread */
745 if (!m->opd_connect_mdt)
746 osp_precreate_fini(m);
748 if (!m->opd_connect_mdt)
749 osp_last_used_fini(env, m);
751 ptlrpc_lprocfs_unregister_obd(obd);
752 lprocfs_obd_cleanup(obd);
754 lprocfs_remove(&m->opd_symlink);
755 client_obd_cleanup(obd);
759 if (m->opd_connect_mdt) {
760 struct client_obd *cli = &m->opd_obd->u.cli;
761 if (cli->cl_rpc_lock != NULL) {
762 OBD_FREE_PTR(cli->cl_rpc_lock);
763 cli->cl_rpc_lock = NULL;
766 obd_disconnect(m->opd_storage_exp);
769 OBD_FREE(osdname, MAX_OBD_NAME);
773 static struct lu_device *osp_device_free(const struct lu_env *env,
774 struct lu_device *lu)
776 struct osp_device *m = lu2osp_dev(lu);
780 if (cfs_atomic_read(&lu->ld_ref) && lu->ld_site) {
781 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
782 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
784 dt_device_fini(&m->opd_dt_dev);
789 static struct lu_device *osp_device_alloc(const struct lu_env *env,
790 struct lu_device_type *t,
791 struct lustre_cfg *lcfg)
793 struct osp_device *m;
798 l = ERR_PTR(-ENOMEM);
803 dt_device_init(&m->opd_dt_dev, t);
804 rc = osp_init0(env, m, t, lcfg);
806 osp_device_free(env, l);
813 static struct lu_device *osp_device_fini(const struct lu_env *env,
816 struct osp_device *m = lu2osp_dev(d);
817 struct obd_import *imp;
822 if (m->opd_storage_exp)
823 obd_disconnect(m->opd_storage_exp);
825 imp = m->opd_obd->u.cli.cl_import;
827 if (imp->imp_rq_pool) {
828 ptlrpc_free_rq_pool(imp->imp_rq_pool);
829 imp->imp_rq_pool = NULL;
833 lprocfs_remove(&m->opd_symlink);
836 ptlrpc_lprocfs_unregister_obd(m->opd_obd);
837 lprocfs_obd_cleanup(m->opd_obd);
839 if (m->opd_connect_mdt) {
840 struct client_obd *cli = &m->opd_obd->u.cli;
841 if (cli->cl_rpc_lock != NULL) {
842 OBD_FREE_PTR(cli->cl_rpc_lock);
843 cli->cl_rpc_lock = NULL;
847 rc = client_obd_cleanup(m->opd_obd);
848 LASSERTF(rc == 0, "error %d\n", rc);
855 static int osp_reconnect(const struct lu_env *env,
856 struct obd_export *exp, struct obd_device *obd,
857 struct obd_uuid *cluuid,
858 struct obd_connect_data *data,
864 static int osp_prepare_fid_client(struct osp_device *osp)
866 LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
867 if (osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL)
870 LASSERT(osp->opd_exp != NULL);
871 osp->opd_obd->u.cli.cl_seq->lcs_exp =
872 class_export_get(osp->opd_exp);
877 * we use exports to track all LOD users
879 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
880 struct obd_device *obd, struct obd_uuid *cluuid,
881 struct obd_connect_data *data, void *localdata)
883 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
884 struct obd_connect_data *ocd;
885 struct obd_import *imp;
886 struct lustre_handle conn;
891 CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
893 rc = class_connect(&conn, obd, cluuid);
897 *exp = class_conn2export(&conn);
898 /* Why should there ever be more than 1 connect? */
900 LASSERT(osp->opd_connects == 1);
904 imp = osp->opd_obd->u.cli.cl_import;
905 imp->imp_dlm_handle = conn;
907 LASSERT(data != NULL);
908 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
909 ocd = &imp->imp_connect_data;
912 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
914 ocd->ocd_version = LUSTRE_VERSION_CODE;
915 ocd->ocd_index = data->ocd_index;
916 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
918 rc = ptlrpc_connect_import(imp);
920 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
924 ptlrpc_pinger_add_import(imp);
930 * once last export (we don't count self-export) disappeared
931 * osp can be released
933 static int osp_obd_disconnect(struct obd_export *exp)
935 struct obd_device *obd = exp->exp_obd;
936 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
940 /* Only disconnect the underlying layers on the final disconnect. */
941 LASSERT(osp->opd_connects == 1);
944 rc = class_disconnect(exp);
946 CERROR("%s: class disconnect error: rc = %d\n",
951 /* destroy the device */
952 class_manual_cleanup(obd);
958 * lprocfs helpers still use OBD API, let's keep obd_statfs() support
960 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
961 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
963 struct obd_statfs *msfs;
964 struct ptlrpc_request *req;
965 struct obd_import *imp = NULL;
970 /* Since the request might also come from lprocfs, so we need
971 * sync this with client_disconnect_export Bug15684 */
972 down_read(&exp->exp_obd->u.cli.cl_sem);
973 if (exp->exp_obd->u.cli.cl_import)
974 imp = class_import_get(exp->exp_obd->u.cli.cl_import);
975 up_read(&exp->exp_obd->u.cli.cl_sem);
979 /* We could possibly pass max_age in the request (as an absolute
980 * timestamp or a "seconds.usec ago") so the target can avoid doing
981 * extra calls into the filesystem if that isn't necessary (e.g.
982 * during mount that would help a bit). Having relative timestamps
983 * is not so great if request processing is slow, while absolute
984 * timestamps are not ideal because they need time synchronization. */
985 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
987 class_import_put(imp);
992 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
994 ptlrpc_request_free(req);
997 ptlrpc_request_set_replen(req);
998 req->rq_request_portal = OST_CREATE_PORTAL;
999 ptlrpc_at_set_req_timeout(req);
1001 if (flags & OBD_STATFS_NODELAY) {
1002 /* procfs requests not want stat in wait for avoid deadlock */
1003 req->rq_no_resend = 1;
1004 req->rq_no_delay = 1;
1007 rc = ptlrpc_queue_wait(req);
1011 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1013 GOTO(out, rc = -EPROTO);
1019 ptlrpc_req_finished(req);
1023 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1024 enum obd_import_event event)
1026 struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1029 case IMP_EVENT_DISCON:
1030 d->opd_got_disconnected = 1;
1031 d->opd_imp_connected = 0;
1032 if (d->opd_connect_mdt)
1035 if (d->opd_pre != NULL) {
1036 osp_pre_update_status(d, -ENODEV);
1037 wake_up(&d->opd_pre_waitq);
1040 CDEBUG(D_HA, "got disconnected\n");
1042 case IMP_EVENT_INACTIVE:
1043 d->opd_imp_active = 0;
1044 if (d->opd_connect_mdt)
1047 if (d->opd_pre != NULL) {
1048 osp_pre_update_status(d, -ENODEV);
1049 wake_up(&d->opd_pre_waitq);
1052 CDEBUG(D_HA, "got inactive\n");
1054 case IMP_EVENT_ACTIVE:
1055 d->opd_imp_active = 1;
1057 if (osp_prepare_fid_client(d) != 0)
1060 if (d->opd_got_disconnected)
1061 d->opd_new_connection = 1;
1062 d->opd_imp_connected = 1;
1063 d->opd_imp_seen_connected = 1;
1064 if (d->opd_connect_mdt)
1067 if (d->opd_pre != NULL)
1068 wake_up(&d->opd_pre_waitq);
1070 __osp_sync_check_for_work(d);
1071 CDEBUG(D_HA, "got connected\n");
1073 case IMP_EVENT_INVALIDATE:
1074 if (obd->obd_namespace == NULL)
1076 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1079 case IMP_EVENT_DEACTIVATE:
1080 case IMP_EVENT_ACTIVATE:
1083 CERROR("%s: unsupported import event: %#x\n",
1084 obd->obd_name, event);
1089 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1090 void *karg, void *uarg)
1092 struct obd_device *obd = exp->exp_obd;
1093 struct osp_device *d;
1094 struct obd_ioctl_data *data = karg;
1099 LASSERT(obd->obd_lu_dev);
1100 d = lu2osp_dev(obd->obd_lu_dev);
1101 LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1103 if (!try_module_get(THIS_MODULE)) {
1104 CERROR("%s: can't get module. Is it alive?", obd->obd_name);
1109 case OBD_IOC_CLIENT_RECOVER:
1110 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1111 data->ioc_inlbuf1, 0);
1115 case IOC_OSC_SET_ACTIVE:
1116 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1119 case OBD_IOC_PING_TARGET:
1120 rc = ptlrpc_obd_ping(obd);
1123 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1124 cmd, current_comm());
1127 module_put(THIS_MODULE);
1131 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1132 __u32 keylen, void *key, __u32 *vallen, void *val,
1133 struct lov_stripe_md *lsm)
1137 if (KEY_IS(KEY_OSP_CONNECTED)) {
1138 struct obd_device *obd = exp->exp_obd;
1139 struct osp_device *osp;
1141 if (!obd->obd_set_up || obd->obd_stopping)
1144 osp = lu2osp_dev(obd->obd_lu_dev);
1147 * 1.8/2.0 behaviour is that OST being connected once at least
1148 * is considered "healthy". and one "healthy" OST is enough to
1149 * allow lustre clients to connect to MDS
1151 RETURN(!osp->opd_imp_seen_connected);
1157 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1158 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1159 static void osp_key_exit(const struct lu_context *ctx,
1160 struct lu_context_key *key, void *data)
1162 struct osp_thread_info *info = data;
1164 info->osi_attr.la_valid = 0;
1167 struct lu_context_key osp_thread_key = {
1168 .lct_tags = LCT_MD_THREAD,
1169 .lct_init = osp_key_init,
1170 .lct_fini = osp_key_fini,
1171 .lct_exit = osp_key_exit
1174 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1175 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1177 struct lu_context_key osp_txn_key = {
1178 .lct_tags = LCT_OSP_THREAD,
1179 .lct_init = osp_txn_key_init,
1180 .lct_fini = osp_txn_key_fini
1182 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1184 static struct lu_device_type_operations osp_device_type_ops = {
1185 .ldto_init = osp_type_init,
1186 .ldto_fini = osp_type_fini,
1188 .ldto_start = osp_type_start,
1189 .ldto_stop = osp_type_stop,
1191 .ldto_device_alloc = osp_device_alloc,
1192 .ldto_device_free = osp_device_free,
1194 .ldto_device_fini = osp_device_fini
1197 static struct lu_device_type osp_device_type = {
1198 .ldt_tags = LU_DEVICE_DT,
1199 .ldt_name = LUSTRE_OSP_NAME,
1200 .ldt_ops = &osp_device_type_ops,
1201 .ldt_ctx_tags = LCT_MD_THREAD
1204 static struct obd_ops osp_obd_device_ops = {
1205 .o_owner = THIS_MODULE,
1206 .o_add_conn = client_import_add_conn,
1207 .o_del_conn = client_import_del_conn,
1208 .o_reconnect = osp_reconnect,
1209 .o_connect = osp_obd_connect,
1210 .o_disconnect = osp_obd_disconnect,
1211 .o_get_info = osp_obd_get_info,
1212 .o_import_event = osp_import_event,
1213 .o_iocontrol = osp_iocontrol,
1214 .o_statfs = osp_obd_statfs,
1215 .o_fid_init = client_fid_init,
1216 .o_fid_fini = client_fid_fini,
1219 struct llog_operations osp_mds_ost_orig_logops;
1221 static int __init osp_mod_init(void)
1223 struct lprocfs_static_vars lvars;
1224 cfs_proc_dir_entry_t *osc_proc_dir;
1227 rc = lu_kmem_init(osp_caches);
1231 lprocfs_osp_init_vars(&lvars);
1233 rc = class_register_type(&osp_obd_device_ops, NULL, NULL,
1234 #ifndef HAVE_ONLY_PROCFS_SEQ
1237 LUSTRE_OSP_NAME, &osp_device_type);
1239 /* create "osc" entry in procfs for compatibility purposes */
1241 lu_kmem_fini(osp_caches);
1245 lprocfs_lwp_init_vars(&lvars);
1247 rc = class_register_type(&lwp_obd_device_ops, NULL, NULL,
1248 #ifndef HAVE_ONLY_PROCFS_SEQ
1251 LUSTRE_LWP_NAME, &lwp_device_type);
1253 class_unregister_type(LUSTRE_OSP_NAME);
1254 lu_kmem_fini(osp_caches);
1258 /* Note: add_rec/delcare_add_rec will be only used by catalogs */
1259 osp_mds_ost_orig_logops = llog_osd_ops;
1260 osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
1261 osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1263 osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
1264 if (osc_proc_dir == NULL) {
1265 osc_proc_dir = lprocfs_register("osc", proc_lustre_root, NULL,
1267 if (IS_ERR(osc_proc_dir))
1268 CERROR("osp: can't create compat entry \"osc\": %d\n",
1269 (int) PTR_ERR(osc_proc_dir));
1274 static void __exit osp_mod_exit(void)
1276 lprocfs_try_remove_proc_entry("osc", proc_lustre_root);
1278 class_unregister_type(LUSTRE_LWP_NAME);
1279 class_unregister_type(LUSTRE_OSP_NAME);
1280 lu_kmem_fini(osp_caches);
1283 MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
1284 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
1285 MODULE_LICENSE("GPL");
1287 cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);