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) 2011, 2012, Intel, Inc.
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>
45 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_MDS
49 #include <obd_class.h>
50 #include <lustre_param.h>
52 #include "osp_internal.h"
54 /* Slab for OSP object allocation */
55 cfs_mem_cache_t *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;
78 OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, CFS_ALLOC_IO);
80 l = &o->opo_obj.do_lu;
83 lu_object_header_init(h);
84 dt_object_init(&o->opo_obj, h, d);
85 lu_object_add_top(h, l);
87 l->lo_ops = &osp_lu_obj_ops;
95 /* Update opd_last_used_id along with checking for gap in objid sequence */
96 void osp_update_last_id(struct osp_device *d, obd_id objid)
99 * we might have lost precreated objects due to VBR and precreate
100 * orphans, the gap in objid can be calculated properly only here
102 if (objid > le64_to_cpu(d->opd_last_used_id)) {
103 if (objid - le64_to_cpu(d->opd_last_used_id) > 1) {
104 d->opd_gap_start = le64_to_cpu(d->opd_last_used_id) + 1;
105 d->opd_gap_count = objid - d->opd_gap_start;
106 CDEBUG(D_HA, "Gap in objids: %d, start = %llu\n",
107 d->opd_gap_count, d->opd_gap_start);
109 d->opd_last_used_id = cpu_to_le64(objid);
113 static int osp_last_used_init(const struct lu_env *env, struct osp_device *m)
115 struct osp_thread_info *osi = osp_env_info(env);
116 struct dt_object_format dof = { 0 };
122 osi->osi_attr.la_valid = LA_MODE;
123 osi->osi_attr.la_mode = S_IFREG | 0644;
124 lu_local_obj_fid(&osi->osi_fid, MDD_LOV_OBJ_OID);
125 dof.dof_type = DFT_REGULAR;
126 o = dt_find_or_create(env, m->opd_storage, &osi->osi_fid, &dof,
131 rc = dt_attr_get(env, o, &osi->osi_attr, NULL);
135 m->opd_last_used_file = o;
137 if (osi->osi_attr.la_size >= sizeof(osi->osi_id) *
138 (m->opd_index + 1)) {
139 osp_objid_buf_prep(osi, m, m->opd_index);
140 rc = dt_record_read(env, o, &osi->osi_lb, &osi->osi_off);
144 /* reset value to 0, just to make sure and change file's size */
147 m->opd_last_used_id = 0;
148 osp_objid_buf_prep(osi, m, m->opd_index);
150 th = dt_trans_create(env, m->opd_storage);
152 GOTO(out, rc = PTR_ERR(th));
154 rc = dt_declare_record_write(env, m->opd_last_used_file,
155 osi->osi_lb.lb_len, osi->osi_off,
158 dt_trans_stop(env, m->opd_storage, th);
162 rc = dt_trans_start_local(env, m->opd_storage, th);
164 dt_trans_stop(env, m->opd_storage, th);
168 rc = dt_record_write(env, m->opd_last_used_file, &osi->osi_lb,
170 dt_trans_stop(env, m->opd_storage, th);
176 /* object will be released in device cleanup path */
177 CERROR("%s: can't initialize lov_objid: %d\n",
178 m->opd_obd->obd_name, rc);
179 lu_object_put(env, &o->do_lu);
180 m->opd_last_used_file = NULL;
184 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *d)
186 lu_object_put(env, &d->opd_last_used_file->do_lu);
187 d->opd_last_used_file = NULL;
190 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
192 struct obd_import *imp;
196 /* release last_used file */
197 osp_last_used_fini(env, d);
199 imp = d->opd_obd->u.cli.cl_import;
201 /* Mark import deactivated now, so we don't try to reconnect if any
202 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
203 * fully deactivate the import, or that would drop all requests. */
204 cfs_spin_lock(&imp->imp_lock);
205 imp->imp_deactive = 1;
206 cfs_spin_unlock(&imp->imp_lock);
208 ptlrpc_deactivate_import(imp);
210 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
211 * delete it regardless. (It's safe to delete an import that was
213 (void)ptlrpc_pinger_del_import(imp);
215 rc = ptlrpc_disconnect_import(imp, 0);
217 CERROR("%s: can't disconnect: rc = %d\n",
218 d->opd_obd->obd_name, rc);
220 ptlrpc_invalidate_import(imp);
225 static int osp_process_config(const struct lu_env *env,
226 struct lu_device *dev, struct lustre_cfg *lcfg)
228 struct osp_device *d = lu2osp_dev(dev);
229 struct lprocfs_static_vars lvars = { 0 };
234 switch (lcfg->lcfg_command) {
236 lu_dev_del_linkage(dev->ld_site, dev);
237 rc = osp_shutdown(env, d);
240 lprocfs_osp_init_vars(&lvars);
243 rc = class_process_proc_param(PARAM_OSP, lvars.obd_vars,
248 /* class_process_proc_param() haven't found matching
249 * parameter and returned ENOSYS so that layer(s)
250 * below could use that. But OSP is the bottom, so
252 CERROR("%s: unknown param %s\n",
253 (char *)lustre_cfg_string(lcfg, 0),
254 (char *)lustre_cfg_string(lcfg, 1));
259 CERROR("%s: unknown command %u\n",
260 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
268 static int osp_recovery_complete(const struct lu_env *env,
269 struct lu_device *dev)
271 struct osp_device *osp = lu2osp_dev(dev);
275 osp->opd_recovery_completed = 1;
279 const struct lu_device_operations osp_lu_ops = {
280 .ldo_object_alloc = osp_object_alloc,
281 .ldo_process_config = osp_process_config,
282 .ldo_recovery_complete = osp_recovery_complete,
285 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
290 * XXX: wake up sync thread, command it to start flushing asap?
296 static const struct dt_device_operations osp_dt_ops = {
300 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *m,
303 struct obd_connect_data *data = NULL;
304 struct obd_device *obd;
309 LASSERT(m->opd_storage_exp == NULL);
315 obd = class_name2obd(nextdev);
317 CERROR("%s: can't locate next device: %s\n",
318 m->opd_obd->obd_name, nextdev);
319 GOTO(out, rc = -ENOTCONN);
322 rc = obd_connect(env, &m->opd_storage_exp, obd, &obd->obd_uuid, data,
325 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
326 m->opd_obd->obd_name, nextdev, rc);
330 m->opd_dt_dev.dd_lu_dev.ld_site =
331 m->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
332 LASSERT(m->opd_dt_dev.dd_lu_dev.ld_site);
333 m->opd_storage = lu2dt_dev(m->opd_storage_exp->exp_obd->obd_lu_dev);
340 static int osp_init0(const struct lu_env *env, struct osp_device *m,
341 struct lu_device_type *ldt, struct lustre_cfg *cfg)
343 struct lprocfs_static_vars lvars = { 0 };
344 struct proc_dir_entry *osc_proc_dir;
345 struct obd_import *imp;
347 char *src, *ost, *mdt, *osdname = NULL;
352 m->opd_obd = class_name2obd(lustre_cfg_string(cfg, 0));
353 if (m->opd_obd == NULL) {
354 CERROR("Cannot find obd with name %s\n",
355 lustre_cfg_string(cfg, 0));
359 /* There is no record in the MDT configuration for the local disk
360 * device, so we have to extract this from elsewhere in the profile.
361 * The only information we get at setup is from the OSC records:
362 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
363 * Note that 1.8 generated configs are missing the -MDTxxxx part.
364 * We need to reconstruct the name of the underlying OSD from this:
365 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd". We
366 * also need to determine the OST index from this - will be used
367 * to calculate the offset in shared lov_objids file later */
369 src = lustre_cfg_string(cfg, 0);
373 ost = strstr(src, "-OST");
377 idx = simple_strtol(ost + 4, &mdt, 16);
378 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
379 CERROR("%s: invalid OST index in '%s'\n",
380 m->opd_obd->obd_name, src);
381 GOTO(out_fini, rc = -EINVAL);
386 /* check the fsname length, and after this everything else will fit */
387 if (idx > MTI_NAME_MAXLEN) {
388 CERROR("%s: fsname too long in '%s'\n",
389 m->opd_obd->obd_name, src);
390 GOTO(out_fini, rc = -EINVAL);
393 OBD_ALLOC(osdname, MAX_OBD_NAME);
395 GOTO(out_fini, rc = -ENOMEM);
397 memcpy(osdname, src, idx); /* copy just the fsname part */
400 mdt = strstr(mdt, "-MDT");
401 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
402 strcat(osdname, "-MDT0000");
404 strcat(osdname, mdt);
405 strcat(osdname, "-osd");
406 CDEBUG(D_HA, "%s: connect to %s (%s)\n",
407 m->opd_obd->obd_name, osdname, src);
409 m->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
410 m->opd_dt_dev.dd_ops = &osp_dt_ops;
411 m->opd_obd->obd_lu_dev = &m->opd_dt_dev.dd_lu_dev;
413 rc = osp_connect_to_osd(env, m, osdname);
417 rc = ptlrpcd_addref();
419 GOTO(out_disconnect, rc);
421 rc = client_obd_setup(m->opd_obd, cfg);
423 CERROR("%s: can't setup obd: %d\n", m->opd_obd->obd_name, rc);
427 lprocfs_osp_init_vars(&lvars);
428 if (lprocfs_obd_setup(m->opd_obd, lvars.obd_vars) == 0)
429 ptlrpc_lprocfs_register_obd(m->opd_obd);
431 /* for compatibility we link old procfs's OSC entries to osp ones */
432 osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
434 cfs_proc_dir_entry_t *symlink = NULL;
437 OBD_ALLOC(name, strlen(m->opd_obd->obd_name) + 1);
439 GOTO(out, rc = -ENOMEM);
441 strcpy(name, m->opd_obd->obd_name);
442 if (strstr(name, "osc"))
443 symlink = lprocfs_add_symlink(name, osc_proc_dir,
445 m->opd_obd->obd_name);
446 OBD_FREE(name, strlen(m->opd_obd->obd_name) + 1);
447 m->opd_symlink = symlink;
451 * Initialize last id from the storage - will be used in orphan cleanup
453 rc = osp_last_used_init(env, m);
458 * Initiate connect to OST
460 ll_generate_random_uuid(uuid);
461 class_uuid_unparse(uuid, &m->opd_cluuid);
463 imp = m->opd_obd->u.cli.cl_import;
465 rc = ptlrpc_init_import(imp);
469 OBD_FREE(osdname, MAX_OBD_NAME);
473 osp_last_used_fini(env, m);
475 ptlrpc_lprocfs_unregister_obd(m->opd_obd);
476 lprocfs_obd_cleanup(m->opd_obd);
477 class_destroy_import(m->opd_obd->u.cli.cl_import);
478 client_obd_cleanup(m->opd_obd);
482 obd_disconnect(m->opd_storage_exp);
485 OBD_FREE(osdname, MAX_OBD_NAME);
489 static struct lu_device *osp_device_free(const struct lu_env *env,
490 struct lu_device *lu)
492 struct osp_device *m = lu2osp_dev(lu);
496 dt_device_fini(&m->opd_dt_dev);
501 static struct lu_device *osp_device_alloc(const struct lu_env *env,
502 struct lu_device_type *t,
503 struct lustre_cfg *lcfg)
505 struct osp_device *m;
510 l = ERR_PTR(-ENOMEM);
515 dt_device_init(&m->opd_dt_dev, t);
516 rc = osp_init0(env, m, t, lcfg);
518 osp_device_free(env, l);
525 static struct lu_device *osp_device_fini(const struct lu_env *env,
528 struct osp_device *m = lu2osp_dev(d);
529 struct obd_import *imp;
534 LASSERT(m->opd_storage_exp);
535 obd_disconnect(m->opd_storage_exp);
537 imp = m->opd_obd->u.cli.cl_import;
539 if (imp->imp_rq_pool) {
540 ptlrpc_free_rq_pool(imp->imp_rq_pool);
541 imp->imp_rq_pool = NULL;
544 obd_cleanup_client_import(m->opd_obd);
547 lprocfs_remove(&m->opd_symlink);
550 ptlrpc_lprocfs_unregister_obd(m->opd_obd);
551 lprocfs_obd_cleanup(m->opd_obd);
553 rc = client_obd_cleanup(m->opd_obd);
554 LASSERTF(rc == 0, "error %d\n", rc);
561 static int osp_reconnect(const struct lu_env *env,
562 struct obd_export *exp, struct obd_device *obd,
563 struct obd_uuid *cluuid,
564 struct obd_connect_data *data,
571 * we use exports to track all LOD users
573 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
574 struct obd_device *obd, struct obd_uuid *cluuid,
575 struct obd_connect_data *data, void *localdata)
577 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
578 struct obd_connect_data *ocd;
579 struct obd_import *imp;
580 struct lustre_handle conn;
585 CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
587 rc = class_connect(&conn, obd, cluuid);
591 *exp = class_conn2export(&conn);
593 /* Why should there ever be more than 1 connect? */
595 LASSERT(osp->opd_connects == 1);
597 imp = osp->opd_obd->u.cli.cl_import;
598 imp->imp_dlm_handle = conn;
600 ocd = &imp->imp_connect_data;
601 ocd->ocd_connect_flags = OBD_CONNECT_AT |
604 #ifdef HAVE_LRU_RESIZE_SUPPORT
605 OBD_CONNECT_LRU_RESIZE |
608 OBD_CONNECT_OSS_CAPA |
609 OBD_CONNECT_REQPORTAL |
610 OBD_CONNECT_SKIP_ORPHAN |
612 ocd->ocd_version = LUSTRE_VERSION_CODE;
613 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
614 ocd->ocd_index = data->ocd_index;
615 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
617 rc = ptlrpc_connect_import(imp);
619 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
623 ptlrpc_pinger_add_import(imp);
630 * once last export (we don't count self-export) disappeared
631 * osp can be released
633 static int osp_obd_disconnect(struct obd_export *exp)
635 struct obd_device *obd = exp->exp_obd;
636 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
641 /* Only disconnect the underlying layers on the final disconnect. */
642 LASSERT(osp->opd_connects == 1);
645 rc = class_disconnect(exp);
647 /* destroy the device */
649 class_manual_cleanup(obd);
654 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
655 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
656 static void osp_key_exit(const struct lu_context *ctx,
657 struct lu_context_key *key, void *data)
659 struct osp_thread_info *info = data;
661 info->osi_attr.la_valid = 0;
664 struct lu_context_key osp_thread_key = {
665 .lct_tags = LCT_MD_THREAD,
666 .lct_init = osp_key_init,
667 .lct_fini = osp_key_fini,
668 .lct_exit = osp_key_exit
671 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
672 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
674 struct lu_context_key osp_txn_key = {
675 .lct_tags = LCT_OSP_THREAD,
676 .lct_init = osp_txn_key_init,
677 .lct_fini = osp_txn_key_fini
679 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
681 static struct lu_device_type_operations osp_device_type_ops = {
682 .ldto_init = osp_type_init,
683 .ldto_fini = osp_type_fini,
685 .ldto_start = osp_type_start,
686 .ldto_stop = osp_type_stop,
688 .ldto_device_alloc = osp_device_alloc,
689 .ldto_device_free = osp_device_free,
691 .ldto_device_fini = osp_device_fini
694 static struct lu_device_type osp_device_type = {
695 .ldt_tags = LU_DEVICE_DT,
696 .ldt_name = LUSTRE_OSP_NAME,
697 .ldt_ops = &osp_device_type_ops,
698 .ldt_ctx_tags = LCT_MD_THREAD
701 static struct obd_ops osp_obd_device_ops = {
702 .o_owner = THIS_MODULE,
703 .o_add_conn = client_import_add_conn,
704 .o_del_conn = client_import_del_conn,
705 .o_reconnect = osp_reconnect,
706 .o_connect = osp_obd_connect,
707 .o_disconnect = osp_obd_disconnect,
710 static int __init osp_mod_init(void)
712 struct lprocfs_static_vars lvars;
713 cfs_proc_dir_entry_t *osc_proc_dir;
716 rc = lu_kmem_init(osp_caches);
720 lprocfs_osp_init_vars(&lvars);
722 rc = class_register_type(&osp_obd_device_ops, NULL, lvars.module_vars,
723 LUSTRE_OSP_NAME, &osp_device_type);
725 /* create "osc" entry in procfs for compatibility purposes */
727 lu_kmem_fini(osp_caches);
731 osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
732 if (osc_proc_dir == NULL) {
733 osc_proc_dir = lprocfs_register("osc", proc_lustre_root, NULL,
735 if (IS_ERR(osc_proc_dir))
736 CERROR("osp: can't create compat entry \"osc\": %d\n",
737 (int) PTR_ERR(osc_proc_dir));
742 static void __exit osp_mod_exit(void)
744 lprocfs_try_remove_proc_entry("osc", proc_lustre_root);
746 class_unregister_type(LUSTRE_OSP_NAME);
747 lu_kmem_fini(osp_caches);
750 MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
751 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
752 MODULE_LICENSE("GPL");
754 cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);