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 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
39 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 * Author: Di Wang <di.wang@intel.com>
43 * The Object Storage Proxy (OSP) module provides an implementation of
44 * the DT API for remote MDTs and OSTs. Every local OSP device (or
45 * object) is a proxy for a remote OSD device (or object). Thus OSP
46 * converts DT operations into RPCs, which are sent to the OUT service
47 * on a remote target, converted back to DT operations, and
48 * executed. Of course there are many ways in which this description
49 * is inaccurate but it's a good enough mental model. OSP is used by
50 * the MDT stack in several ways:
52 * - OSP devices allocate FIDs for the stripe sub-objects of a striped
55 * - OSP objects represent the remote MDT and OST objects that are
56 * the stripes of a striped object.
58 * - OSP devices log, send, and track synchronous operations (setattr
59 * and unlink) to remote targets.
61 * - OSP objects are the bottom slice of the compound LU object
62 * representing a remote MDT object: MDT/MDD/LOD/OSP.
64 * - OSP objects are used by LFSCK to represent remote OST objects
65 * during the verification of MDT-OST consistency.
67 * - OSP devices batch idempotent requests (declare_attr_get() and
68 * declare_xattr_get()) to the remote target and cache their results.
70 * In addition the OSP layer implements a subset of the OBD device API
71 * to support being a client of a remote target, connecting to other
72 * layers, and FID allocation.
75 #define DEBUG_SUBSYSTEM S_MDS
77 #include <obd_class.h>
78 #include <lustre_ioctl.h>
79 #include <lustre_param.h>
80 #include <lustre_log.h>
81 #include <lustre_mdc.h>
83 #include "osp_internal.h"
85 /* Slab for OSP object allocation */
86 struct kmem_cache *osp_object_kmem;
88 static struct lu_kmem_descr osp_caches[] = {
90 .ckd_cache = &osp_object_kmem,
91 .ckd_name = "osp_obj",
92 .ckd_size = sizeof(struct osp_object)
100 * Implementation of lu_device_operations::ldo_object_alloc
102 * Allocates an OSP object in memory, whose FID is on the remote target.
104 * \param[in] env execution environment
105 * \param[in] hdr The header of the object stack. If it is NULL, it
106 * means the object is not built from top device, i.e.
107 * it is a sub-stripe object of striped directory or
109 * \param[in] d OSP device
111 * \retval object object being created if the creation succeed.
112 * \retval NULL NULL if the creation failed.
114 struct lu_object *osp_object_alloc(const struct lu_env *env,
115 const struct lu_object_header *hdr,
118 struct lu_object_header *h = NULL;
119 struct osp_object *o;
122 OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
124 l = &o->opo_obj.do_lu;
126 /* If hdr is NULL, it means the object is not built
127 * from the top dev(MDT/OST), usually it happens when
128 * building striped object, like data object on MDT or
129 * striped object for directory */
132 lu_object_header_init(h);
133 dt_object_init(&o->opo_obj, h, d);
134 lu_object_add_top(h, l);
136 dt_object_init(&o->opo_obj, h, d);
139 l->lo_ops = &osp_lu_obj_ops;
148 * Find or create the local object
150 * Finds or creates the local file referenced by \a reg_id and return the
151 * attributes of the local file.
153 * \param[in] env execution environment
154 * \param[in] osp OSP device
155 * \param[out] attr attributes of the object
156 * \param[in] reg_id the local object ID of the file. It will be used
157 * to compose a local FID{FID_SEQ_LOCAL_FILE, reg_id, 0}
158 * to identify the object.
160 * \retval object object(dt_object) found or created
161 * \retval ERR_PTR(errno) ERR_PTR(errno) if not get the object.
163 static struct dt_object
164 *osp_find_or_create_local_file(const struct lu_env *env, struct osp_device *osp,
165 struct lu_attr *attr, __u32 reg_id)
167 struct osp_thread_info *osi = osp_env_info(env);
168 struct dt_object_format dof = { 0 };
169 struct dt_object *dto;
173 lu_local_obj_fid(&osi->osi_fid, reg_id);
174 attr->la_valid = LA_MODE;
175 attr->la_mode = S_IFREG | 0644;
176 dof.dof_type = DFT_REGULAR;
177 /* Find or create the local object by osi_fid. */
178 dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
183 /* Get attributes of the local object. */
184 rc = dt_attr_get(env, dto, attr, NULL);
186 CERROR("%s: can't be initialized: rc = %d\n",
187 osp->opd_obd->obd_name, rc);
188 lu_object_put(env, &dto->do_lu);
195 * Write data buffer to a local file object.
197 * \param[in] env execution environment
198 * \param[in] osp OSP device
199 * \param[in] dt_obj object written to
200 * \param[in] buf buffer containing byte array and length
201 * \param[in] offset write offset in the object in bytes
203 * \retval 0 0 if write succeed
204 * \retval -EFAULT -EFAULT if only part of buffer is written.
205 * \retval negative other negative errno if write failed.
207 static int osp_write_local_file(const struct lu_env *env,
208 struct osp_device *osp,
209 struct dt_object *dt_obj,
216 th = dt_trans_create(env, osp->opd_storage);
220 rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
223 rc = dt_trans_start_local(env, osp->opd_storage, th);
227 rc = dt_record_write(env, dt_obj, buf, &offset, th);
229 dt_trans_stop(env, osp->opd_storage, th);
234 * Initialize last ID object.
236 * This function initializes the LAST_ID file, which stores the current last
237 * used id of data objects. The MDT will use the last used id and the last_seq
238 * (\see osp_init_last_seq()) to synchronize the precreate object cache with
241 * \param[in] env execution environment
242 * \param[in] osp OSP device
244 * \retval 0 0 if initialization succeed
245 * \retval negative negative errno if initialization failed
247 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
249 struct osp_thread_info *osi = osp_env_info(env);
250 struct lu_fid *fid = &osp->opd_last_used_fid;
251 struct dt_object *dto;
255 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
258 RETURN(PTR_ERR(dto));
260 /* object will be released in device cleanup path */
261 if (osi->osi_attr.la_size >=
262 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
263 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
265 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
270 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
272 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
277 osp->opd_last_used_oid_file = dto;
280 /* object will be released in device cleanup path */
281 CERROR("%s: can't initialize lov_objid: rc = %d\n",
282 osp->opd_obd->obd_name, rc);
283 lu_object_put(env, &dto->do_lu);
284 osp->opd_last_used_oid_file = NULL;
289 * Initialize last sequence object.
291 * This function initializes the LAST_SEQ file in the local OSD, which stores
292 * the current last used sequence of data objects. The MDT will use the last
293 * sequence and last id (\see osp_init_last_objid()) to synchronize the
294 * precreate object cache with OSTs.
296 * \param[in] env execution environment
297 * \param[in] osp OSP device
299 * \retval 0 0 if initialization succeed
300 * \retval negative negative errno if initialization failed
302 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
304 struct osp_thread_info *osi = osp_env_info(env);
305 struct lu_fid *fid = &osp->opd_last_used_fid;
306 struct dt_object *dto;
310 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
313 RETURN(PTR_ERR(dto));
315 /* object will be released in device cleanup path */
316 if (osi->osi_attr.la_size >=
317 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
318 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
320 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
325 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
327 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
330 osp->opd_last_used_seq_file = dto;
333 /* object will be released in device cleanup path */
334 CERROR("%s: can't initialize lov_seq: rc = %d\n",
335 osp->opd_obd->obd_name, rc);
336 lu_object_put(env, &dto->do_lu);
337 osp->opd_last_used_seq_file = NULL;
342 * Initialize last OID and sequence object.
344 * If the MDT is just upgraded to 2.4 from the lower version, where the
345 * LAST_SEQ file does not exist, the file will be created and IDIF sequence
346 * will be written into the file.
348 * \param[in] env execution environment
349 * \param[in] osp OSP device
351 * \retval 0 0 if initialization succeed
352 * \retval negative negative error if initialization failed
354 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
356 struct osp_thread_info *osi = osp_env_info(env);
360 fid_zero(&osp->opd_last_used_fid);
361 rc = osp_init_last_objid(env, osp);
363 CERROR("%s: Can not get ids %d from old objid!\n",
364 osp->opd_obd->obd_name, rc);
368 rc = osp_init_last_seq(env, osp);
370 CERROR("%s: Can not get ids %d from old objid!\n",
371 osp->opd_obd->obd_name, rc);
375 if (fid_oid(&osp->opd_last_used_fid) != 0 &&
376 fid_seq(&osp->opd_last_used_fid) == 0) {
377 /* Just upgrade from the old version,
378 * set the seq to be IDIF */
379 osp->opd_last_used_fid.f_seq =
380 fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
382 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
383 &osp->opd_last_used_fid.f_seq,
385 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
386 &osi->osi_lb, osi->osi_off);
388 CERROR("%s : Can not write seq file: rc = %d\n",
389 osp->opd_obd->obd_name, rc);
394 if (!fid_is_zero(&osp->opd_last_used_fid) &&
395 !fid_is_sane(&osp->opd_last_used_fid)) {
396 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
397 PFID(&osp->opd_last_used_fid));
398 GOTO(out, rc = -EINVAL);
401 CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
402 osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
405 if (osp->opd_last_used_oid_file != NULL) {
406 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
407 osp->opd_last_used_oid_file = NULL;
409 if (osp->opd_last_used_seq_file != NULL) {
410 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
411 osp->opd_last_used_seq_file = NULL;
419 * Release the last sequence and OID file objects in OSP device.
421 * \param[in] env execution environment
422 * \param[in] osp OSP device
424 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp)
426 /* release last_used file */
427 if (osp->opd_last_used_oid_file != NULL) {
428 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
429 osp->opd_last_used_oid_file = NULL;
432 if (osp->opd_last_used_seq_file != NULL) {
433 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
434 osp->opd_last_used_seq_file = NULL;
439 * Disconnects the connection between OSP and its correspondent MDT or OST, and
440 * the import will be marked as inactive. It will only be called during OSP
443 * \param[in] d OSP device being disconnected
445 * \retval 0 0 if disconnection succeed
446 * \retval negative negative errno if disconnection failed
448 static int osp_disconnect(struct osp_device *d)
450 struct obd_import *imp;
453 imp = d->opd_obd->u.cli.cl_import;
455 /* Mark import deactivated now, so we don't try to reconnect if any
456 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
457 * fully deactivate the import, or that would drop all requests. */
458 LASSERT(imp != NULL);
459 spin_lock(&imp->imp_lock);
460 imp->imp_deactive = 1;
461 spin_unlock(&imp->imp_lock);
463 ptlrpc_deactivate_import(imp);
465 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
466 * delete it regardless. (It's safe to delete an import that was
468 (void)ptlrpc_pinger_del_import(imp);
470 rc = ptlrpc_disconnect_import(imp, 0);
472 CERROR("%s: can't disconnect: rc = %d\n",
473 d->opd_obd->obd_name, rc);
475 ptlrpc_invalidate_import(imp);
481 * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
482 * precreate threads etc.
484 * \param[in] env execution environment.
485 * \param[in] d OSP device being disconnected.
487 * \retval 0 0 if cleanup succeed
488 * \retval negative negative errno if cleanup failed
490 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
497 rc = osp_disconnect(d);
501 if (!d->opd_connect_mdt) {
502 /* stop precreate thread */
503 osp_precreate_fini(d);
505 /* release last_used file */
506 osp_last_used_fini(env, d);
509 obd_fid_fini(d->opd_obd);
515 * Implementation of osp_lu_ops::ldo_process_config
517 * This function processes config log records in OSP layer. It is usually
518 * called from the top layer of MDT stack, and goes through the stack by calling
519 * ldo_process_config of next layer.
521 * \param[in] env execution environment
522 * \param[in] dev lu_device of OSP
523 * \param[in] lcfg config log
525 * \retval 0 0 if the config log record is executed correctly.
526 * \retval negative negative errno if the record execution fails.
528 static int osp_process_config(const struct lu_env *env,
529 struct lu_device *dev, struct lustre_cfg *lcfg)
531 struct osp_device *d = lu2osp_dev(dev);
532 struct obd_device *obd = d->opd_obd;
537 switch (lcfg->lcfg_command) {
538 case LCFG_PRE_CLEANUP:
539 rc = osp_disconnect(d);
542 lu_dev_del_linkage(dev->ld_site, dev);
543 rc = osp_shutdown(env, d);
547 rc = class_process_proc_param(PARAM_OSC, obd->obd_vars,
552 /* class_process_proc_param() haven't found matching
553 * parameter and returned ENOSYS so that layer(s)
554 * below could use that. But OSP is the bottom, so
556 CERROR("%s: unknown param %s\n",
557 (char *)lustre_cfg_string(lcfg, 0),
558 (char *)lustre_cfg_string(lcfg, 1));
563 CERROR("%s: unknown command %u\n",
564 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
573 * Implementation of osp_lu_ops::ldo_recovery_complete
575 * This function is called after recovery is finished, and OSP layer
576 * will wake up precreate thread here.
578 * \param[in] env execution environment
579 * \param[in] dev lu_device of OSP
581 * \retval 0 0 unconditionally
583 static int osp_recovery_complete(const struct lu_env *env,
584 struct lu_device *dev)
586 struct osp_device *osp = lu2osp_dev(dev);
589 osp->opd_recovery_completed = 1;
591 if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
592 wake_up(&osp->opd_pre_waitq);
597 const struct lu_device_operations osp_lu_ops = {
598 .ldo_object_alloc = osp_object_alloc,
599 .ldo_process_config = osp_process_config,
600 .ldo_recovery_complete = osp_recovery_complete,
604 * Implementation of dt_device_operations::dt_statfs
606 * This function provides statfs status (for precreation) from
607 * corresponding OST. Note: this function only retrieves the status
608 * from the OSP device, and the real statfs RPC happens inside
609 * precreate thread (\see osp_statfs_update). Note: OSP for MDT does
610 * not need to retrieve statfs data for now.
612 * \param[in] env execution environment.
613 * \param[in] dev dt_device of OSP.
614 * \param[out] sfs holds the retrieved statfs data.
616 * \retval 0 0 statfs data was retrieved successfully or
617 * retrieval was not needed
618 * \retval negative negative errno if get statfs failed.
620 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
621 struct obd_statfs *sfs)
623 struct osp_device *d = dt2osp_dev(dev);
627 if (unlikely(d->opd_imp_active == 0))
630 if (d->opd_pre == NULL)
633 /* return recently updated data */
634 *sfs = d->opd_statfs;
637 * layer above osp (usually lod) can use ffree to estimate
638 * how many objects are available for immediate creation
640 spin_lock(&d->opd_pre_lock);
641 LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
642 fid_seq(&d->opd_pre_used_fid),
643 "last_created "DFID", next_fid "DFID"\n",
644 PFID(&d->opd_pre_last_created_fid),
645 PFID(&d->opd_pre_used_fid));
646 sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
647 fid_oid(&d->opd_pre_used_fid);
648 sfs->os_fprecreated -= d->opd_pre_reserved;
649 spin_unlock(&d->opd_pre_lock);
651 LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
653 CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
654 LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
655 sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
656 sfs->os_files, sfs->os_ffree);
660 static int osp_sync_timeout(void *data)
666 * Implementation of dt_device_operations::dt_sync
668 * This function synchronizes the OSP cache to the remote target. It wakes
669 * up unlink log threads and sends out unlink records to the remote OST.
671 * \param[in] env execution environment
672 * \param[in] dev dt_device of OSP
674 * \retval 0 0 if synchronization succeeds
675 * \retval negative negative errno if synchronization fails
677 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
679 struct osp_device *d = dt2osp_dev(dev);
681 struct l_wait_info lwi = { 0 };
682 unsigned long id, old;
684 unsigned long start = cfs_time_current();
687 if (unlikely(d->opd_imp_active == 0))
690 id = d->opd_syn_last_used_id;
692 CDEBUG(D_OTHER, "%s: id: used %lu, processed %lu\n",
693 d->opd_obd->obd_name, id, d->opd_syn_last_processed_id);
695 /* wait till all-in-line are processed */
696 while (d->opd_syn_last_processed_id < id) {
698 old = d->opd_syn_last_processed_id;
700 /* make sure the connection is fine */
701 expire = cfs_time_shift(obd_timeout);
702 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
703 osp_sync_timeout, d);
704 l_wait_event(d->opd_syn_barrier_waitq,
705 d->opd_syn_last_processed_id >= id,
708 if (d->opd_syn_last_processed_id >= id)
711 if (d->opd_syn_last_processed_id != old) {
712 /* some progress have been made,
717 /* no changes and expired, something is wrong */
718 GOTO(out, rc = -ETIMEDOUT);
721 /* block new processing (barrier>0 - few callers are possible */
722 atomic_inc(&d->opd_syn_barrier);
724 CDEBUG(D_OTHER, "%s: %u in flight\n", d->opd_obd->obd_name,
725 d->opd_syn_rpc_in_flight);
727 /* wait till all-in-flight are replied, so executed by the target */
728 /* XXX: this is used by LFSCK at the moment, which doesn't require
729 * all the changes to be committed, but in general it'd be
730 * better to wait till commit */
731 while (d->opd_syn_rpc_in_flight > 0) {
733 old = d->opd_syn_rpc_in_flight;
735 expire = cfs_time_shift(obd_timeout);
736 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
737 osp_sync_timeout, d);
738 l_wait_event(d->opd_syn_barrier_waitq,
739 d->opd_syn_rpc_in_flight == 0, &lwi);
741 if (d->opd_syn_rpc_in_flight == 0)
744 if (d->opd_syn_rpc_in_flight != old) {
745 /* some progress have been made */
749 /* no changes and expired, something is wrong */
750 GOTO(out, rc = -ETIMEDOUT);
753 CDEBUG(D_OTHER, "%s: done in %lu\n", d->opd_obd->obd_name,
754 cfs_time_current() - start);
756 /* resume normal processing (barrier=0) */
757 atomic_dec(&d->opd_syn_barrier);
758 __osp_sync_check_for_work(d);
763 const struct dt_device_operations osp_dt_ops = {
764 .dt_statfs = osp_statfs,
766 .dt_trans_create = osp_trans_create,
767 .dt_trans_start = osp_trans_start,
768 .dt_trans_stop = osp_trans_stop,
772 * Connect OSP to local OSD.
774 * Locate the local OSD referenced by \a nextdev and connect to it. Sometimes,
775 * OSP needs to access the local OSD to store some information. For example,
776 * during precreate, it needs to update last used OID and sequence file
777 * (LAST_SEQ) in local OSD.
779 * \param[in] env execution environment
780 * \param[in] osp OSP device
781 * \param[in] nextdev the name of local OSD
783 * \retval 0 0 connection succeeded
784 * \retval negative negative errno connection failed
786 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *osp,
789 struct obd_connect_data *data = NULL;
790 struct obd_device *obd;
795 LASSERT(osp->opd_storage_exp == NULL);
801 obd = class_name2obd(nextdev);
803 CERROR("%s: can't locate next device: %s\n",
804 osp->opd_obd->obd_name, nextdev);
805 GOTO(out, rc = -ENOTCONN);
808 rc = obd_connect(env, &osp->opd_storage_exp, obd, &obd->obd_uuid, data,
811 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
812 osp->opd_obd->obd_name, nextdev, rc);
816 osp->opd_dt_dev.dd_lu_dev.ld_site =
817 osp->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
818 LASSERT(osp->opd_dt_dev.dd_lu_dev.ld_site);
819 osp->opd_storage = lu2dt_dev(osp->opd_storage_exp->exp_obd->obd_lu_dev);
827 * Initialize OSP device according to the parameters in the configuration
830 * Reconstruct the local device name from the configuration profile, and
831 * initialize necessary threads and structures according to the OSP type
834 * Since there is no record in the MDT configuration for the local disk
835 * device, we have to extract this from elsewhere in the profile.
836 * The only information we get at setup is from the OSC records:
837 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
839 * Note: configs generated by Lustre 1.8 are missing the -MDTxxxx part,
840 * so, we need to reconstruct the name of the underlying OSD from this:
841 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd".
843 * \param[in] env execution environment
844 * \param[in] osp OSP device
845 * \param[in] ldt lu device type of OSP
846 * \param[in] cfg configuration log
848 * \retval 0 0 if OSP initialization succeeded.
849 * \retval negative negative errno if OSP initialization failed.
851 static int osp_init0(const struct lu_env *env, struct osp_device *osp,
852 struct lu_device_type *ldt, struct lustre_cfg *cfg)
854 struct obd_device *obd;
855 struct obd_import *imp;
857 char *src, *tgt, *mdt, *osdname = NULL;
863 mutex_init(&osp->opd_async_requests_mutex);
865 obd = class_name2obd(lustre_cfg_string(cfg, 0));
867 CERROR("Cannot find obd with name %s\n",
868 lustre_cfg_string(cfg, 0));
873 src = lustre_cfg_string(cfg, 0);
877 tgt = strrchr(src, '-');
879 CERROR("%s: invalid target name %s: rc = %d\n",
880 osp->opd_obd->obd_name, lustre_cfg_string(cfg, 0),
885 if (strncmp(tgt, "-osc", 4) == 0) {
886 /* Old OSC name fsname-OSTXXXX-osc */
887 for (tgt--; tgt > src && *tgt != '-'; tgt--)
890 CERROR("%s: invalid target name %s: rc = %d\n",
891 osp->opd_obd->obd_name,
892 lustre_cfg_string(cfg, 0), -EINVAL);
896 if (strncmp(tgt, "-OST", 4) != 0) {
897 CERROR("%s: invalid target name %s: rc = %d\n",
898 osp->opd_obd->obd_name,
899 lustre_cfg_string(cfg, 0), -EINVAL);
903 idx = simple_strtol(tgt + 4, &mdt, 16);
904 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
905 CERROR("%s: invalid OST index in '%s': rc = %d\n",
906 osp->opd_obd->obd_name, src, -EINVAL);
909 osp->opd_index = idx;
913 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
914 if (strncmp(tgt, "-MDT", 4) != 0 &&
915 strncmp(tgt, "-OST", 4) != 0) {
916 CERROR("%s: invalid target name %s: rc = %d\n",
917 osp->opd_obd->obd_name,
918 lustre_cfg_string(cfg, 0), -EINVAL);
922 idx = simple_strtol(tgt + 4, &mdt, 16);
923 if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
924 CERROR("%s: invalid OST index in '%s': rc = %d\n",
925 osp->opd_obd->obd_name, src, -EINVAL);
929 /* Get MDT index from the name and set it to opd_group,
930 * which will be used by OSP to connect with OST */
931 osp->opd_group = idx;
932 if (tgt - src <= 12) {
933 CERROR("%s: invalid mdt index from %s: rc =%d\n",
934 osp->opd_obd->obd_name,
935 lustre_cfg_string(cfg, 0), -EINVAL);
939 if (strncmp(tgt - 12, "-MDT", 4) == 0)
940 osp->opd_connect_mdt = 1;
942 idx = simple_strtol(tgt - 8, &mdt, 16);
943 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
944 CERROR("%s: invalid OST index in '%s': rc =%d\n",
945 osp->opd_obd->obd_name, src, -EINVAL);
949 osp->opd_index = idx;
950 idx = tgt - src - 12;
952 /* check the fsname length, and after this everything else will fit */
953 if (idx > MTI_NAME_MAXLEN) {
954 CERROR("%s: fsname too long in '%s': rc = %d\n",
955 osp->opd_obd->obd_name, src, -EINVAL);
959 OBD_ALLOC(osdname, MAX_OBD_NAME);
963 memcpy(osdname, src, idx); /* copy just the fsname part */
966 mdt = strstr(mdt, "-MDT");
967 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
968 strcat(osdname, "-MDT0000");
970 strcat(osdname, mdt);
971 strcat(osdname, "-osd");
972 CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
974 if (osp->opd_connect_mdt) {
975 struct client_obd *cli = &osp->opd_obd->u.cli;
977 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
978 if (!cli->cl_rpc_lock)
979 GOTO(out_fini, rc = -ENOMEM);
980 osp_init_rpc_lock(cli->cl_rpc_lock);
983 osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
984 osp->opd_dt_dev.dd_ops = &osp_dt_ops;
986 obd->obd_lu_dev = &osp->opd_dt_dev.dd_lu_dev;
988 rc = osp_connect_to_osd(env, osp, osdname);
992 rc = ptlrpcd_addref();
994 GOTO(out_disconnect, rc);
996 rc = client_obd_setup(obd, cfg);
998 CERROR("%s: can't setup obd: rc = %d\n", osp->opd_obd->obd_name,
1003 osp_lprocfs_init(osp);
1005 rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
1006 LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
1008 CERROR("%s: fid init error: rc = %d\n",
1009 osp->opd_obd->obd_name, rc);
1013 if (!osp->opd_connect_mdt) {
1014 /* Initialize last id from the storage - will be
1015 * used in orphan cleanup. */
1016 rc = osp_last_used_init(env, osp);
1021 /* Initialize precreation thread, it handles new
1022 * connections as well. */
1023 rc = osp_init_precreate(osp);
1025 GOTO(out_last_used, rc);
1029 * Initialize synhronization mechanism taking
1030 * care of propogating changes to OST in near
1031 * transactional manner.
1033 rc = osp_sync_init(env, osp);
1035 GOTO(out_precreat, rc);
1038 * Initiate connect to OST
1040 ll_generate_random_uuid(uuid);
1041 class_uuid_unparse(uuid, &osp->opd_cluuid);
1043 imp = obd->u.cli.cl_import;
1045 rc = ptlrpc_init_import(imp);
1049 OBD_FREE(osdname, MAX_OBD_NAME);
1053 /* stop sync thread */
1056 /* stop precreate thread */
1057 if (!osp->opd_connect_mdt)
1058 osp_precreate_fini(osp);
1060 if (!osp->opd_connect_mdt)
1061 osp_last_used_fini(env, osp);
1063 ptlrpc_lprocfs_unregister_obd(obd);
1064 lprocfs_obd_cleanup(obd);
1065 if (osp->opd_symlink)
1066 lprocfs_remove(&osp->opd_symlink);
1067 client_obd_cleanup(obd);
1071 if (osp->opd_connect_mdt) {
1072 struct client_obd *cli = &osp->opd_obd->u.cli;
1073 if (cli->cl_rpc_lock != NULL) {
1074 OBD_FREE_PTR(cli->cl_rpc_lock);
1075 cli->cl_rpc_lock = NULL;
1078 obd_disconnect(osp->opd_storage_exp);
1081 OBD_FREE(osdname, MAX_OBD_NAME);
1086 * Implementation of lu_device_type_operations::ldto_device_free
1088 * Free the OSP device in memory. No return value is needed for now,
1089 * so always return NULL to comply with the interface.
1091 * \param[in] env execution environment
1092 * \param[in] lu lu_device of OSP
1094 * \retval NULL NULL unconditionally
1096 static struct lu_device *osp_device_free(const struct lu_env *env,
1097 struct lu_device *lu)
1099 struct osp_device *osp = lu2osp_dev(lu);
1101 if (atomic_read(&lu->ld_ref) && lu->ld_site) {
1102 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1103 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1105 dt_device_fini(&osp->opd_dt_dev);
1112 * Implementation of lu_device_type_operations::ldto_device_alloc
1114 * This function allocates and initializes OSP device in memory according to
1117 * \param[in] env execution environment
1118 * \param[in] type device type of OSP
1119 * \param[in] lcfg config log
1121 * \retval pointer the pointer of allocated OSP if succeed.
1122 * \retval ERR_PTR(errno) ERR_PTR(errno) if failed.
1124 static struct lu_device *osp_device_alloc(const struct lu_env *env,
1125 struct lu_device_type *type,
1126 struct lustre_cfg *lcfg)
1128 struct osp_device *osp;
1129 struct lu_device *ld;
1133 ld = ERR_PTR(-ENOMEM);
1137 ld = osp2lu_dev(osp);
1138 dt_device_init(&osp->opd_dt_dev, type);
1139 rc = osp_init0(env, osp, type, lcfg);
1141 osp_device_free(env, ld);
1149 * Implementation of lu_device_type_operations::ldto_device_fini
1151 * This function cleans up the OSP device, i.e. release and free those
1152 * attached items in osp_device.
1154 * \param[in] env execution environment
1155 * \param[in] ld lu_device of OSP
1157 * \retval NULL NULL if cleanup succeeded.
1158 * \retval ERR_PTR(errno) ERR_PTR(errno) if cleanup failed.
1160 static struct lu_device *osp_device_fini(const struct lu_env *env,
1161 struct lu_device *ld)
1163 struct osp_device *osp = lu2osp_dev(ld);
1164 struct obd_import *imp;
1169 if (osp->opd_async_requests != NULL) {
1170 dt_update_request_destroy(osp->opd_async_requests);
1171 osp->opd_async_requests = NULL;
1174 if (osp->opd_storage_exp)
1175 obd_disconnect(osp->opd_storage_exp);
1177 imp = osp->opd_obd->u.cli.cl_import;
1179 if (imp->imp_rq_pool) {
1180 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1181 imp->imp_rq_pool = NULL;
1184 if (osp->opd_symlink)
1185 lprocfs_remove(&osp->opd_symlink);
1187 LASSERT(osp->opd_obd);
1188 ptlrpc_lprocfs_unregister_obd(osp->opd_obd);
1189 lprocfs_obd_cleanup(osp->opd_obd);
1191 if (osp->opd_connect_mdt) {
1192 struct client_obd *cli = &osp->opd_obd->u.cli;
1193 if (cli->cl_rpc_lock != NULL) {
1194 OBD_FREE_PTR(cli->cl_rpc_lock);
1195 cli->cl_rpc_lock = NULL;
1199 rc = client_obd_cleanup(osp->opd_obd);
1202 RETURN(ERR_PTR(rc));
1211 * Implementation of obd_ops::o_reconnect
1213 * This function is empty and does not need to do anything for now.
1215 static int osp_reconnect(const struct lu_env *env,
1216 struct obd_export *exp, struct obd_device *obd,
1217 struct obd_uuid *cluuid,
1218 struct obd_connect_data *data,
1225 * Implementation of obd_ops::o_connect
1227 * Connect OSP to the remote target (MDT or OST). Allocate the
1228 * export and return it to the LOD, which calls this function
1229 * for each OSP to connect it to the remote target. This function
1230 * is currently only called once per OSP.
1232 * \param[in] env execution environment
1233 * \param[out] exp export connected to OSP
1234 * \param[in] obd OSP device
1235 * \param[in] cluuid OSP device client uuid
1236 * \param[in] data connect_data to be used to connect to the remote
1238 * \param[in] localdata necessary for the API interface, but not used in
1241 * \retval 0 0 if the connection succeeded.
1242 * \retval negative negative errno if the connection failed.
1244 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
1245 struct obd_device *obd, struct obd_uuid *cluuid,
1246 struct obd_connect_data *data, void *localdata)
1248 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1249 struct obd_connect_data *ocd;
1250 struct obd_import *imp;
1251 struct lustre_handle conn;
1256 CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
1258 rc = class_connect(&conn, obd, cluuid);
1262 *exp = class_conn2export(&conn);
1263 /* Why should there ever be more than 1 connect? */
1264 osp->opd_connects++;
1265 LASSERT(osp->opd_connects == 1);
1267 osp->opd_exp = *exp;
1269 imp = osp->opd_obd->u.cli.cl_import;
1270 imp->imp_dlm_handle = conn;
1272 LASSERT(data != NULL);
1273 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
1274 ocd = &imp->imp_connect_data;
1277 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1279 ocd->ocd_version = LUSTRE_VERSION_CODE;
1280 ocd->ocd_index = data->ocd_index;
1281 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1283 rc = ptlrpc_connect_import(imp);
1285 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
1289 ptlrpc_pinger_add_import(imp);
1295 * Implementation of obd_ops::o_disconnect
1297 * Disconnect the export for the OSP. This is called by LOD to release the
1298 * OSP during cleanup (\see lod_del_device()). The OSP will be released after
1299 * the export is released.
1301 * \param[in] exp export to be disconnected.
1303 * \retval 0 0 if disconnection succeed
1304 * \retval negative negative errno if disconnection failed
1306 static int osp_obd_disconnect(struct obd_export *exp)
1308 struct obd_device *obd = exp->exp_obd;
1309 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1313 /* Only disconnect the underlying layers on the final disconnect. */
1314 LASSERT(osp->opd_connects == 1);
1315 osp->opd_connects--;
1317 rc = class_disconnect(exp);
1319 CERROR("%s: class disconnect error: rc = %d\n",
1324 /* destroy the device */
1325 class_manual_cleanup(obd);
1331 * Implementation of obd_ops::o_statfs
1333 * Send a RPC to the remote target to get statfs status. This is only used
1334 * in lprocfs helpers by obd_statfs.
1336 * \param[in] env execution environment
1337 * \param[in] exp connection state from this OSP to the parent (LOD)
1339 * \param[out] osfs hold the statfs result
1340 * \param[in] unused Not used in this function for now
1341 * \param[in] flags flags to indicate how OSP will issue the RPC
1343 * \retval 0 0 if statfs succeeded.
1344 * \retval negative negative errno if statfs failed.
1346 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
1347 struct obd_statfs *osfs, __u64 unused, __u32 flags)
1349 struct obd_statfs *msfs;
1350 struct ptlrpc_request *req;
1351 struct obd_import *imp = NULL;
1356 /* Since the request might also come from lprocfs, so we need
1357 * sync this with client_disconnect_export Bug15684 */
1358 down_read(&exp->exp_obd->u.cli.cl_sem);
1359 if (exp->exp_obd->u.cli.cl_import)
1360 imp = class_import_get(exp->exp_obd->u.cli.cl_import);
1361 up_read(&exp->exp_obd->u.cli.cl_sem);
1365 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
1367 class_import_put(imp);
1372 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
1374 ptlrpc_request_free(req);
1377 ptlrpc_request_set_replen(req);
1378 req->rq_request_portal = OST_CREATE_PORTAL;
1379 ptlrpc_at_set_req_timeout(req);
1381 if (flags & OBD_STATFS_NODELAY) {
1382 /* procfs requests not want stat in wait for avoid deadlock */
1383 req->rq_no_resend = 1;
1384 req->rq_no_delay = 1;
1387 rc = ptlrpc_queue_wait(req);
1391 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1393 GOTO(out, rc = -EPROTO);
1399 ptlrpc_req_finished(req);
1404 * Prepare fid client.
1406 * This function prepares the FID client for the OSP. It will check and assign
1407 * the export (to MDT0) for its FID client, so OSP can allocate super sequence
1408 * or lookup sequence in FLDB of MDT0.
1410 * \param[in] osp OSP device
1412 static void osp_prepare_fid_client(struct osp_device *osp)
1414 LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
1415 if (osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL)
1418 LASSERT(osp->opd_exp != NULL);
1419 osp->opd_obd->u.cli.cl_seq->lcs_exp =
1420 class_export_get(osp->opd_exp);
1424 * Implementation of obd_ops::o_import_event
1426 * This function is called when some related import event happens. It will
1427 * mark the necessary flags according to the event and notify the necessary
1428 * threads (mainly precreate thread).
1430 * \param[in] obd OSP OBD device
1431 * \param[in] imp import attached from OSP to remote (OST/MDT) service
1432 * \param[in] event event related to remote service (IMP_EVENT_*)
1434 * \retval 0 0 if the event handling succeeded.
1435 * \retval negative negative errno if the event handling failed.
1437 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1438 enum obd_import_event event)
1440 struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1443 case IMP_EVENT_DISCON:
1444 d->opd_got_disconnected = 1;
1445 d->opd_imp_connected = 0;
1446 if (d->opd_connect_mdt)
1449 if (d->opd_pre != NULL) {
1450 osp_pre_update_status(d, -ENODEV);
1451 wake_up(&d->opd_pre_waitq);
1454 CDEBUG(D_HA, "got disconnected\n");
1456 case IMP_EVENT_INACTIVE:
1457 d->opd_imp_active = 0;
1458 if (d->opd_connect_mdt)
1461 if (d->opd_pre != NULL) {
1462 osp_pre_update_status(d, -ENODEV);
1463 wake_up(&d->opd_pre_waitq);
1466 CDEBUG(D_HA, "got inactive\n");
1468 case IMP_EVENT_ACTIVE:
1469 d->opd_imp_active = 1;
1471 osp_prepare_fid_client(d);
1472 if (d->opd_got_disconnected)
1473 d->opd_new_connection = 1;
1474 d->opd_imp_connected = 1;
1475 d->opd_imp_seen_connected = 1;
1476 if (d->opd_connect_mdt)
1479 if (d->opd_pre != NULL)
1480 wake_up(&d->opd_pre_waitq);
1482 __osp_sync_check_for_work(d);
1483 CDEBUG(D_HA, "got connected\n");
1485 case IMP_EVENT_INVALIDATE:
1486 if (obd->obd_namespace == NULL)
1488 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1491 case IMP_EVENT_DEACTIVATE:
1492 case IMP_EVENT_ACTIVATE:
1495 CERROR("%s: unsupported import event: %#x\n",
1496 obd->obd_name, event);
1502 * Implementation of obd_ops: o_iocontrol
1504 * This function is the ioctl handler for OSP. Note: lctl will access the OSP
1505 * directly by ioctl, instead of through the MDS stack.
1507 * param[in] cmd ioctl command.
1508 * param[in] exp export of this OSP.
1509 * param[in] len data length of \a karg.
1510 * param[in] karg input argument which is packed as
1512 * param[out] uarg pointer to userspace buffer (must access by
1515 * \retval 0 0 if the ioctl handling succeeded.
1516 * \retval negative negative errno if the ioctl handling failed.
1518 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1519 void *karg, void *uarg)
1521 struct obd_device *obd = exp->exp_obd;
1522 struct osp_device *d;
1523 struct obd_ioctl_data *data = karg;
1528 LASSERT(obd->obd_lu_dev);
1529 d = lu2osp_dev(obd->obd_lu_dev);
1530 LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1532 if (!try_module_get(THIS_MODULE)) {
1533 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1534 module_name(THIS_MODULE));
1539 case OBD_IOC_CLIENT_RECOVER:
1540 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1541 data->ioc_inlbuf1, 0);
1545 case IOC_OSC_SET_ACTIVE:
1546 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1549 case OBD_IOC_PING_TARGET:
1550 rc = ptlrpc_obd_ping(obd);
1553 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1554 cmd, current_comm());
1557 module_put(THIS_MODULE);
1562 * Implementation of obd_ops::o_get_info
1564 * Retrieve information by key. Retrieval starts from the top layer
1565 * (MDT) of the MDS stack and traverses the stack by calling the
1566 * obd_get_info() method of the next sub-layer.
1568 * \param[in] env execution environment
1569 * \param[in] exp export of this OSP
1570 * \param[in] keylen length of \a key
1571 * \param[in] key the key
1572 * \param[out] vallen length of \a val
1573 * \param[out] val holds the value returned by the key
1574 * \param[in] unused necessary for the interface but unused
1576 * \retval 0 0 if getting information succeeded.
1577 * \retval negative negative errno if getting information failed.
1579 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1580 __u32 keylen, void *key, __u32 *vallen, void *val,
1581 struct lov_stripe_md *unused)
1585 if (KEY_IS(KEY_OSP_CONNECTED)) {
1586 struct obd_device *obd = exp->exp_obd;
1587 struct osp_device *osp;
1589 if (!obd->obd_set_up || obd->obd_stopping)
1592 osp = lu2osp_dev(obd->obd_lu_dev);
1595 * 1.8/2.0 behaviour is that OST being connected once at least
1596 * is considered "healthy". and one "healthy" OST is enough to
1597 * allow lustre clients to connect to MDS
1599 RETURN(!osp->opd_imp_seen_connected);
1606 * Implementation of obd_ops: o_fid_alloc
1608 * Allocate a FID. There are two cases in which OSP performs
1611 * 1. FID precreation for data objects, which is done in
1612 * osp_precreate_fids() instead of this function.
1613 * 2. FID allocation for each sub-stripe of a striped directory.
1614 * Similar to other FID clients, OSP requests the sequence
1615 * from its corresponding remote MDT, which in turn requests
1616 * sequences from the sequence controller (MDT0).
1618 * \param[in] env execution environment
1619 * \param[in] exp export of the OSP
1620 * \param[out] fid FID being allocated
1621 * \param[in] unused necessary for the interface but unused.
1623 * \retval 0 0 FID allocated successfully.
1624 * \retval 1 1 FID allocated successfully and new sequence
1625 * requested from seq meta server
1626 * \retval negative negative errno if FID allocation failed.
1628 int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1629 struct lu_fid *fid, struct md_op_data *unused)
1631 struct client_obd *cli = &exp->exp_obd->u.cli;
1632 struct osp_device *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev);
1633 struct lu_client_seq *seq = cli->cl_seq;
1636 LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
1637 /* Sigh, fid client is not ready yet */
1638 if (osp->opd_obd->u.cli.cl_seq->lcs_exp == NULL)
1641 RETURN(seq_client_alloc_fid(env, seq, fid));
1644 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1645 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1646 static void osp_key_exit(const struct lu_context *ctx,
1647 struct lu_context_key *key, void *data)
1649 struct osp_thread_info *info = data;
1651 info->osi_attr.la_valid = 0;
1654 struct lu_context_key osp_thread_key = {
1655 .lct_tags = LCT_MD_THREAD,
1656 .lct_init = osp_key_init,
1657 .lct_fini = osp_key_fini,
1658 .lct_exit = osp_key_exit
1661 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1662 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1664 struct lu_context_key osp_txn_key = {
1665 .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE,
1666 .lct_init = osp_txn_key_init,
1667 .lct_fini = osp_txn_key_fini
1669 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1671 static struct lu_device_type_operations osp_device_type_ops = {
1672 .ldto_init = osp_type_init,
1673 .ldto_fini = osp_type_fini,
1675 .ldto_start = osp_type_start,
1676 .ldto_stop = osp_type_stop,
1678 .ldto_device_alloc = osp_device_alloc,
1679 .ldto_device_free = osp_device_free,
1681 .ldto_device_fini = osp_device_fini
1684 static struct lu_device_type osp_device_type = {
1685 .ldt_tags = LU_DEVICE_DT,
1686 .ldt_name = LUSTRE_OSP_NAME,
1687 .ldt_ops = &osp_device_type_ops,
1688 .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
1691 static struct obd_ops osp_obd_device_ops = {
1692 .o_owner = THIS_MODULE,
1693 .o_add_conn = client_import_add_conn,
1694 .o_del_conn = client_import_del_conn,
1695 .o_reconnect = osp_reconnect,
1696 .o_connect = osp_obd_connect,
1697 .o_disconnect = osp_obd_disconnect,
1698 .o_get_info = osp_obd_get_info,
1699 .o_import_event = osp_import_event,
1700 .o_iocontrol = osp_iocontrol,
1701 .o_statfs = osp_obd_statfs,
1702 .o_fid_init = client_fid_init,
1703 .o_fid_fini = client_fid_fini,
1704 .o_fid_alloc = osp_fid_alloc,
1707 struct llog_operations osp_mds_ost_orig_logops;
1710 * Initialize OSP module.
1712 * Register device types OSP and Light Weight Proxy (LWP) (\see lwp_dev.c)
1713 * in obd_types (\see class_obd.c). Initialize procfs for the
1714 * the OSP device. Note: OSP was called OSC before Lustre 2.4,
1715 * so for compatibility it still uses the name "osc" in procfs.
1716 * This is called at module load time.
1718 * \retval 0 0 if initialization succeeds.
1719 * \retval negative negative errno if initialization failed.
1721 static int __init osp_mod_init(void)
1723 struct obd_type *type;
1726 rc = lu_kmem_init(osp_caches);
1731 rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL,
1732 LUSTRE_OSP_NAME, &osp_device_type);
1734 lu_kmem_fini(osp_caches);
1738 rc = class_register_type(&lwp_obd_device_ops, NULL, true, NULL,
1739 LUSTRE_LWP_NAME, &lwp_device_type);
1741 class_unregister_type(LUSTRE_OSP_NAME);
1742 lu_kmem_fini(osp_caches);
1746 /* Note: add_rec/delcare_add_rec will be only used by catalogs */
1747 osp_mds_ost_orig_logops = llog_osd_ops;
1748 osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
1749 osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1751 /* create "osc" entry in procfs for compatibility purposes */
1752 type = class_search_type(LUSTRE_OSC_NAME);
1753 if (type != NULL && type->typ_procroot != NULL)
1756 type = class_search_type(LUSTRE_OSP_NAME);
1757 type->typ_procsym = lprocfs_seq_register("osc", proc_lustre_root,
1759 if (IS_ERR(type->typ_procsym)) {
1760 CERROR("osp: can't create compat entry \"osc\": %d\n",
1761 (int) PTR_ERR(type->typ_procsym));
1762 type->typ_procsym = NULL;
1768 * Finalize OSP module.
1770 * This callback is called when kernel unloads OSP module from memory, and
1771 * it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
1773 static void __exit osp_mod_exit(void)
1775 class_unregister_type(LUSTRE_LWP_NAME);
1776 class_unregister_type(LUSTRE_OSP_NAME);
1777 lu_kmem_fini(osp_caches);
1780 MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
1781 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
1782 MODULE_LICENSE("GPL");
1784 cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);