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, 2014, 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>
82 #include "osp_internal.h"
84 /* Slab for OSP object allocation */
85 struct kmem_cache *osp_object_kmem;
87 static struct lu_kmem_descr osp_caches[] = {
89 .ckd_cache = &osp_object_kmem,
90 .ckd_name = "osp_obj",
91 .ckd_size = sizeof(struct osp_object)
99 * Implementation of lu_device_operations::ldo_object_alloc
101 * Allocates an OSP object in memory, whose FID is on the remote target.
103 * \param[in] env execution environment
104 * \param[in] hdr The header of the object stack. If it is NULL, it
105 * means the object is not built from top device, i.e.
106 * it is a sub-stripe object of striped directory or
108 * \param[in] d OSP device
110 * \retval object object being created if the creation succeed.
111 * \retval NULL NULL if the creation failed.
113 static struct lu_object *osp_object_alloc(const struct lu_env *env,
114 const struct lu_object_header *hdr,
117 struct lu_object_header *h = NULL;
118 struct osp_object *o;
121 OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
123 l = &o->opo_obj.do_lu;
125 /* If hdr is NULL, it means the object is not built
126 * from the top dev(MDT/OST), usually it happens when
127 * building striped object, like data object on MDT or
128 * striped object for directory */
131 lu_object_header_init(h);
132 dt_object_init(&o->opo_obj, h, d);
133 lu_object_add_top(h, l);
135 dt_object_init(&o->opo_obj, h, d);
138 l->lo_ops = &osp_lu_obj_ops;
147 * Find or create the local object
149 * Finds or creates the local file referenced by \a reg_id and return the
150 * attributes of the local file.
152 * \param[in] env execution environment
153 * \param[in] osp OSP device
154 * \param[out] attr attributes of the object
155 * \param[in] reg_id the local object ID of the file. It will be used
156 * to compose a local FID{FID_SEQ_LOCAL_FILE, reg_id, 0}
157 * to identify the object.
159 * \retval object object(dt_object) found or created
160 * \retval ERR_PTR(errno) ERR_PTR(errno) if not get the object.
162 static struct dt_object
163 *osp_find_or_create_local_file(const struct lu_env *env, struct osp_device *osp,
164 struct lu_attr *attr, __u32 reg_id)
166 struct osp_thread_info *osi = osp_env_info(env);
167 struct dt_object_format dof = { 0 };
168 struct dt_object *dto;
172 lu_local_obj_fid(&osi->osi_fid, reg_id);
173 attr->la_valid = LA_MODE;
174 attr->la_mode = S_IFREG | 0644;
175 dof.dof_type = DFT_REGULAR;
176 /* Find or create the local object by osi_fid. */
177 dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
182 /* Get attributes of the local object. */
183 rc = dt_attr_get(env, dto, attr);
185 CERROR("%s: can't be initialized: rc = %d\n",
186 osp->opd_obd->obd_name, rc);
187 lu_object_put(env, &dto->do_lu);
194 * Write data buffer to a local file object.
196 * \param[in] env execution environment
197 * \param[in] osp OSP device
198 * \param[in] dt_obj object written to
199 * \param[in] buf buffer containing byte array and length
200 * \param[in] offset write offset in the object in bytes
202 * \retval 0 0 if write succeed
203 * \retval -EFAULT -EFAULT if only part of buffer is written.
204 * \retval negative other negative errno if write failed.
206 static int osp_write_local_file(const struct lu_env *env,
207 struct osp_device *osp,
208 struct dt_object *dt_obj,
215 th = dt_trans_create(env, osp->opd_storage);
219 rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
222 rc = dt_trans_start_local(env, osp->opd_storage, th);
226 rc = dt_record_write(env, dt_obj, buf, &offset, th);
228 dt_trans_stop(env, osp->opd_storage, th);
233 * Initialize last ID object.
235 * This function initializes the LAST_ID file, which stores the current last
236 * used id of data objects. The MDT will use the last used id and the last_seq
237 * (\see osp_init_last_seq()) to synchronize the precreate object cache with
240 * \param[in] env execution environment
241 * \param[in] osp OSP device
243 * \retval 0 0 if initialization succeed
244 * \retval negative negative errno if initialization failed
246 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
248 struct osp_thread_info *osi = osp_env_info(env);
249 struct lu_fid *fid = &osp->opd_last_used_fid;
250 struct dt_object *dto;
254 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
257 RETURN(PTR_ERR(dto));
259 /* object will be released in device cleanup path */
260 if (osi->osi_attr.la_size >=
261 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
262 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
264 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
265 if (rc != 0 && rc != -EFAULT)
269 if (rc == -EFAULT) { /* fresh LAST_ID */
271 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
273 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
278 osp->opd_last_used_oid_file = dto;
281 /* object will be released in device cleanup path */
282 CERROR("%s: can't initialize lov_objid: rc = %d\n",
283 osp->opd_obd->obd_name, rc);
284 lu_object_put(env, &dto->do_lu);
285 osp->opd_last_used_oid_file = NULL;
290 * Initialize last sequence object.
292 * This function initializes the LAST_SEQ file in the local OSD, which stores
293 * the current last used sequence of data objects. The MDT will use the last
294 * sequence and last id (\see osp_init_last_objid()) to synchronize the
295 * precreate object cache with OSTs.
297 * \param[in] env execution environment
298 * \param[in] osp OSP device
300 * \retval 0 0 if initialization succeed
301 * \retval negative negative errno if initialization failed
303 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
305 struct osp_thread_info *osi = osp_env_info(env);
306 struct lu_fid *fid = &osp->opd_last_used_fid;
307 struct dt_object *dto;
311 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
314 RETURN(PTR_ERR(dto));
316 /* object will be released in device cleanup path */
317 if (osi->osi_attr.la_size >=
318 sizeof(osi->osi_id) * (osp->opd_index + 1)) {
319 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
321 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
322 if (rc != 0 && rc != -EFAULT)
326 if (rc == -EFAULT) { /* fresh OSP */
328 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
330 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
335 osp->opd_last_used_seq_file = dto;
338 /* object will be released in device cleanup path */
339 CERROR("%s: can't initialize lov_seq: rc = %d\n",
340 osp->opd_obd->obd_name, rc);
341 lu_object_put(env, &dto->do_lu);
342 osp->opd_last_used_seq_file = NULL;
347 * Initialize last OID and sequence object.
349 * If the MDT is just upgraded to 2.4 from the lower version, where the
350 * LAST_SEQ file does not exist, the file will be created and IDIF sequence
351 * will be written into the file.
353 * \param[in] env execution environment
354 * \param[in] osp OSP device
356 * \retval 0 0 if initialization succeed
357 * \retval negative negative error if initialization failed
359 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
361 struct osp_thread_info *osi = osp_env_info(env);
365 fid_zero(&osp->opd_last_used_fid);
366 rc = osp_init_last_objid(env, osp);
368 CERROR("%s: Can not get ids %d from old objid!\n",
369 osp->opd_obd->obd_name, rc);
373 rc = osp_init_last_seq(env, osp);
375 CERROR("%s: Can not get ids %d from old objid!\n",
376 osp->opd_obd->obd_name, rc);
380 if (fid_oid(&osp->opd_last_used_fid) != 0 &&
381 fid_seq(&osp->opd_last_used_fid) == 0) {
382 /* Just upgrade from the old version,
383 * set the seq to be IDIF */
384 osp->opd_last_used_fid.f_seq =
385 fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
387 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
388 &osp->opd_last_used_fid.f_seq,
390 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
391 &osi->osi_lb, osi->osi_off);
393 CERROR("%s : Can not write seq file: rc = %d\n",
394 osp->opd_obd->obd_name, rc);
399 if (!fid_is_zero(&osp->opd_last_used_fid) &&
400 !fid_is_sane(&osp->opd_last_used_fid)) {
401 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
402 PFID(&osp->opd_last_used_fid));
403 GOTO(out, rc = -EINVAL);
406 CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
407 osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
410 if (osp->opd_last_used_oid_file != NULL) {
411 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
412 osp->opd_last_used_oid_file = NULL;
414 if (osp->opd_last_used_seq_file != NULL) {
415 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
416 osp->opd_last_used_seq_file = NULL;
424 * Release the last sequence and OID file objects in OSP device.
426 * \param[in] env execution environment
427 * \param[in] osp OSP device
429 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp)
431 /* release last_used file */
432 if (osp->opd_last_used_oid_file != NULL) {
433 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
434 osp->opd_last_used_oid_file = NULL;
437 if (osp->opd_last_used_seq_file != NULL) {
438 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
439 osp->opd_last_used_seq_file = NULL;
444 * Disconnects the connection between OSP and its correspondent MDT or OST, and
445 * the import will be marked as inactive. It will only be called during OSP
448 * \param[in] d OSP device being disconnected
450 * \retval 0 0 if disconnection succeed
451 * \retval negative negative errno if disconnection failed
453 static int osp_disconnect(struct osp_device *d)
455 struct obd_import *imp;
458 imp = d->opd_obd->u.cli.cl_import;
460 /* Mark import deactivated now, so we don't try to reconnect if any
461 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
462 * fully deactivate the import, or that would drop all requests. */
463 LASSERT(imp != NULL);
464 spin_lock(&imp->imp_lock);
465 imp->imp_deactive = 1;
466 spin_unlock(&imp->imp_lock);
468 ptlrpc_deactivate_import(imp);
470 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
471 * delete it regardless. (It's safe to delete an import that was
473 (void)ptlrpc_pinger_del_import(imp);
475 rc = ptlrpc_disconnect_import(imp, 0);
477 CERROR("%s: can't disconnect: rc = %d\n",
478 d->opd_obd->obd_name, rc);
480 ptlrpc_invalidate_import(imp);
486 * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
487 * precreate threads etc.
489 * \param[in] env execution environment.
490 * \param[in] d OSP device being disconnected.
492 * \retval 0 0 if cleanup succeed
493 * \retval negative negative errno if cleanup failed
495 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
502 rc = osp_disconnect(d);
504 if (!d->opd_connect_mdt) {
505 /* stop sync thread */
508 /* stop precreate thread */
509 osp_precreate_fini(d);
511 /* release last_used file */
512 osp_last_used_fini(env, d);
515 obd_fid_fini(d->opd_obd);
521 * Implementation of osp_lu_ops::ldo_process_config
523 * This function processes config log records in OSP layer. It is usually
524 * called from the top layer of MDT stack, and goes through the stack by calling
525 * ldo_process_config of next layer.
527 * \param[in] env execution environment
528 * \param[in] dev lu_device of OSP
529 * \param[in] lcfg config log
531 * \retval 0 0 if the config log record is executed correctly.
532 * \retval negative negative errno if the record execution fails.
534 static int osp_process_config(const struct lu_env *env,
535 struct lu_device *dev, struct lustre_cfg *lcfg)
537 struct osp_device *d = lu2osp_dev(dev);
538 struct obd_device *obd = d->opd_obd;
543 switch (lcfg->lcfg_command) {
544 case LCFG_PRE_CLEANUP:
545 rc = osp_disconnect(d);
548 lu_dev_del_linkage(dev->ld_site, dev);
549 rc = osp_shutdown(env, d);
553 rc = class_process_proc_param(PARAM_OSC, obd->obd_vars,
558 /* class_process_proc_param() haven't found matching
559 * parameter and returned ENOSYS so that layer(s)
560 * below could use that. But OSP is the bottom, so
562 CERROR("%s: unknown param %s\n",
563 (char *)lustre_cfg_string(lcfg, 0),
564 (char *)lustre_cfg_string(lcfg, 1));
569 CERROR("%s: unknown command %u\n",
570 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
579 * Implementation of osp_lu_ops::ldo_recovery_complete
581 * This function is called after recovery is finished, and OSP layer
582 * will wake up precreate thread here.
584 * \param[in] env execution environment
585 * \param[in] dev lu_device of OSP
587 * \retval 0 0 unconditionally
589 static int osp_recovery_complete(const struct lu_env *env,
590 struct lu_device *dev)
592 struct osp_device *osp = lu2osp_dev(dev);
595 osp->opd_recovery_completed = 1;
597 if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
598 wake_up(&osp->opd_pre_waitq);
603 const struct lu_device_operations osp_lu_ops = {
604 .ldo_object_alloc = osp_object_alloc,
605 .ldo_process_config = osp_process_config,
606 .ldo_recovery_complete = osp_recovery_complete,
610 * Implementation of dt_device_operations::dt_statfs
612 * This function provides statfs status (for precreation) from
613 * corresponding OST. Note: this function only retrieves the status
614 * from the OSP device, and the real statfs RPC happens inside
615 * precreate thread (\see osp_statfs_update). Note: OSP for MDT does
616 * not need to retrieve statfs data for now.
618 * \param[in] env execution environment.
619 * \param[in] dev dt_device of OSP.
620 * \param[out] sfs holds the retrieved statfs data.
622 * \retval 0 0 statfs data was retrieved successfully or
623 * retrieval was not needed
624 * \retval negative negative errno if get statfs failed.
626 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
627 struct obd_statfs *sfs)
629 struct osp_device *d = dt2osp_dev(dev);
633 if (unlikely(d->opd_imp_active == 0))
636 if (d->opd_pre == NULL)
639 /* return recently updated data */
640 *sfs = d->opd_statfs;
643 * layer above osp (usually lod) can use ffree to estimate
644 * how many objects are available for immediate creation
646 spin_lock(&d->opd_pre_lock);
647 LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
648 fid_seq(&d->opd_pre_used_fid),
649 "last_created "DFID", next_fid "DFID"\n",
650 PFID(&d->opd_pre_last_created_fid),
651 PFID(&d->opd_pre_used_fid));
652 sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
653 fid_oid(&d->opd_pre_used_fid);
654 sfs->os_fprecreated -= d->opd_pre_reserved;
655 spin_unlock(&d->opd_pre_lock);
657 LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
659 CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
660 LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
661 sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
662 sfs->os_files, sfs->os_ffree);
666 static int osp_sync_timeout(void *data)
672 * Implementation of dt_device_operations::dt_sync
674 * This function synchronizes the OSP cache to the remote target. It wakes
675 * up unlink log threads and sends out unlink records to the remote OST.
677 * \param[in] env execution environment
678 * \param[in] dev dt_device of OSP
680 * \retval 0 0 if synchronization succeeds
681 * \retval negative negative errno if synchronization fails
683 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
685 struct osp_device *d = dt2osp_dev(dev);
687 struct l_wait_info lwi = { 0 };
688 unsigned long id, old;
690 unsigned long start = cfs_time_current();
693 /* No Sync between MDTs yet. */
694 if (d->opd_connect_mdt)
697 if (unlikely(d->opd_imp_active == 0))
700 id = d->opd_syn_last_used_id;
701 down_write(&d->opd_async_updates_rwsem);
703 CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
704 atomic_read(&d->opd_async_updates_count));
706 /* make sure the connection is fine */
707 expire = cfs_time_shift(obd_timeout);
708 lwi = LWI_TIMEOUT(expire - cfs_time_current(), osp_sync_timeout, d);
709 rc = l_wait_event(d->opd_syn_barrier_waitq,
710 atomic_read(&d->opd_async_updates_count) == 0,
712 up_write(&d->opd_async_updates_rwsem);
716 CDEBUG(D_CACHE, "%s: id: used %lu, processed %lu\n",
717 d->opd_obd->obd_name, id, d->opd_syn_last_processed_id);
719 /* wait till all-in-line are processed */
720 while (d->opd_syn_last_processed_id < id) {
722 old = d->opd_syn_last_processed_id;
724 /* make sure the connection is fine */
725 expire = cfs_time_shift(obd_timeout);
726 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
727 osp_sync_timeout, d);
728 l_wait_event(d->opd_syn_barrier_waitq,
729 d->opd_syn_last_processed_id >= id,
732 if (d->opd_syn_last_processed_id >= id)
735 if (d->opd_syn_last_processed_id != old) {
736 /* some progress have been made,
741 /* no changes and expired, something is wrong */
742 GOTO(out, rc = -ETIMEDOUT);
745 /* block new processing (barrier>0 - few callers are possible */
746 atomic_inc(&d->opd_syn_barrier);
748 CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
749 d->opd_syn_rpc_in_flight);
751 /* wait till all-in-flight are replied, so executed by the target */
752 /* XXX: this is used by LFSCK at the moment, which doesn't require
753 * all the changes to be committed, but in general it'd be
754 * better to wait till commit */
755 while (d->opd_syn_rpc_in_flight > 0) {
757 old = d->opd_syn_rpc_in_flight;
759 expire = cfs_time_shift(obd_timeout);
760 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
761 osp_sync_timeout, d);
762 l_wait_event(d->opd_syn_barrier_waitq,
763 d->opd_syn_rpc_in_flight == 0, &lwi);
765 if (d->opd_syn_rpc_in_flight == 0)
768 if (d->opd_syn_rpc_in_flight != old) {
769 /* some progress have been made */
773 /* no changes and expired, something is wrong */
774 GOTO(out, rc = -ETIMEDOUT);
778 /* resume normal processing (barrier=0) */
779 atomic_dec(&d->opd_syn_barrier);
780 __osp_sync_check_for_work(d);
782 CDEBUG(D_CACHE, "%s: done in %lu: rc = %d\n", d->opd_obd->obd_name,
783 cfs_time_current() - start, rc);
788 const struct dt_device_operations osp_dt_ops = {
789 .dt_statfs = osp_statfs,
791 .dt_trans_create = osp_trans_create,
792 .dt_trans_start = osp_trans_start,
793 .dt_trans_stop = osp_trans_stop,
794 .dt_trans_cb_add = osp_trans_cb_add,
798 * Connect OSP to local OSD.
800 * Locate the local OSD referenced by \a nextdev and connect to it. Sometimes,
801 * OSP needs to access the local OSD to store some information. For example,
802 * during precreate, it needs to update last used OID and sequence file
803 * (LAST_SEQ) in local OSD.
805 * \param[in] env execution environment
806 * \param[in] osp OSP device
807 * \param[in] nextdev the name of local OSD
809 * \retval 0 0 connection succeeded
810 * \retval negative negative errno connection failed
812 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *osp,
815 struct obd_connect_data *data = NULL;
816 struct obd_device *obd;
821 LASSERT(osp->opd_storage_exp == NULL);
827 obd = class_name2obd(nextdev);
829 CERROR("%s: can't locate next device: %s\n",
830 osp->opd_obd->obd_name, nextdev);
831 GOTO(out, rc = -ENOTCONN);
834 rc = obd_connect(env, &osp->opd_storage_exp, obd, &obd->obd_uuid, data,
837 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
838 osp->opd_obd->obd_name, nextdev, rc);
842 osp->opd_dt_dev.dd_lu_dev.ld_site =
843 osp->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
844 LASSERT(osp->opd_dt_dev.dd_lu_dev.ld_site);
845 osp->opd_storage = lu2dt_dev(osp->opd_storage_exp->exp_obd->obd_lu_dev);
853 * Determine if the lock needs to be cancelled
855 * Determine if the unused lock should be cancelled before replay, see
856 * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists
859 * \param[in] lock lock to be checked.
861 * \retval 1 if the lock needs to be cancelled before replay.
862 * \retval 0 if the lock does not need to be cancelled before
865 static int osp_cancel_weight(struct ldlm_lock *lock)
867 if (lock->l_resource->lr_type != LDLM_IBITS)
874 * Initialize OSP device according to the parameters in the configuration
877 * Reconstruct the local device name from the configuration profile, and
878 * initialize necessary threads and structures according to the OSP type
881 * Since there is no record in the MDT configuration for the local disk
882 * device, we have to extract this from elsewhere in the profile.
883 * The only information we get at setup is from the OSC records:
884 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
886 * Note: configs generated by Lustre 1.8 are missing the -MDTxxxx part,
887 * so, we need to reconstruct the name of the underlying OSD from this:
888 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd".
890 * \param[in] env execution environment
891 * \param[in] osp OSP device
892 * \param[in] ldt lu device type of OSP
893 * \param[in] cfg configuration log
895 * \retval 0 0 if OSP initialization succeeded.
896 * \retval negative negative errno if OSP initialization failed.
898 static int osp_init0(const struct lu_env *env, struct osp_device *osp,
899 struct lu_device_type *ldt, struct lustre_cfg *cfg)
901 struct obd_device *obd;
902 struct obd_import *imp;
904 char *src, *tgt, *mdt, *osdname = NULL;
910 mutex_init(&osp->opd_async_requests_mutex);
911 INIT_LIST_HEAD(&osp->opd_async_updates);
912 init_rwsem(&osp->opd_async_updates_rwsem);
913 atomic_set(&osp->opd_async_updates_count, 0);
915 obd = class_name2obd(lustre_cfg_string(cfg, 0));
917 CERROR("Cannot find obd with name %s\n",
918 lustre_cfg_string(cfg, 0));
923 src = lustre_cfg_string(cfg, 0);
927 tgt = strrchr(src, '-');
929 CERROR("%s: invalid target name %s: rc = %d\n",
930 osp->opd_obd->obd_name, lustre_cfg_string(cfg, 0),
935 if (strncmp(tgt, "-osc", 4) == 0) {
936 /* Old OSC name fsname-OSTXXXX-osc */
937 for (tgt--; tgt > src && *tgt != '-'; tgt--)
940 CERROR("%s: invalid target name %s: rc = %d\n",
941 osp->opd_obd->obd_name,
942 lustre_cfg_string(cfg, 0), -EINVAL);
946 if (strncmp(tgt, "-OST", 4) != 0) {
947 CERROR("%s: invalid target name %s: rc = %d\n",
948 osp->opd_obd->obd_name,
949 lustre_cfg_string(cfg, 0), -EINVAL);
953 idx = simple_strtol(tgt + 4, &mdt, 16);
954 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
955 CERROR("%s: invalid OST index in '%s': rc = %d\n",
956 osp->opd_obd->obd_name, src, -EINVAL);
959 osp->opd_index = idx;
963 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
964 if (strncmp(tgt, "-MDT", 4) != 0 &&
965 strncmp(tgt, "-OST", 4) != 0) {
966 CERROR("%s: invalid target name %s: rc = %d\n",
967 osp->opd_obd->obd_name,
968 lustre_cfg_string(cfg, 0), -EINVAL);
972 idx = simple_strtol(tgt + 4, &mdt, 16);
973 if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
974 CERROR("%s: invalid OST index in '%s': rc = %d\n",
975 osp->opd_obd->obd_name, src, -EINVAL);
979 /* Get MDT index from the name and set it to opd_group,
980 * which will be used by OSP to connect with OST */
981 osp->opd_group = idx;
982 if (tgt - src <= 12) {
983 CERROR("%s: invalid mdt index from %s: rc =%d\n",
984 osp->opd_obd->obd_name,
985 lustre_cfg_string(cfg, 0), -EINVAL);
989 if (strncmp(tgt - 12, "-MDT", 4) == 0)
990 osp->opd_connect_mdt = 1;
992 idx = simple_strtol(tgt - 8, &mdt, 16);
993 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
994 CERROR("%s: invalid OST index in '%s': rc =%d\n",
995 osp->opd_obd->obd_name, src, -EINVAL);
999 osp->opd_index = idx;
1000 idx = tgt - src - 12;
1002 /* check the fsname length, and after this everything else will fit */
1003 if (idx > MTI_NAME_MAXLEN) {
1004 CERROR("%s: fsname too long in '%s': rc = %d\n",
1005 osp->opd_obd->obd_name, src, -EINVAL);
1009 OBD_ALLOC(osdname, MAX_OBD_NAME);
1010 if (osdname == NULL)
1013 memcpy(osdname, src, idx); /* copy just the fsname part */
1014 osdname[idx] = '\0';
1016 mdt = strstr(mdt, "-MDT");
1017 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
1018 strcat(osdname, "-MDT0000");
1020 strcat(osdname, mdt);
1021 strcat(osdname, "-osd");
1022 CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
1024 if (osp->opd_connect_mdt) {
1025 struct client_obd *cli = &osp->opd_obd->u.cli;
1027 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
1028 if (!cli->cl_rpc_lock)
1029 GOTO(out_fini, rc = -ENOMEM);
1030 osp_init_rpc_lock(cli->cl_rpc_lock);
1033 osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
1034 osp->opd_dt_dev.dd_ops = &osp_dt_ops;
1036 obd->obd_lu_dev = &osp->opd_dt_dev.dd_lu_dev;
1038 rc = osp_connect_to_osd(env, osp, osdname);
1042 rc = ptlrpcd_addref();
1044 GOTO(out_disconnect, rc);
1046 rc = client_obd_setup(obd, cfg);
1048 CERROR("%s: can't setup obd: rc = %d\n", osp->opd_obd->obd_name,
1053 osp_lprocfs_init(osp);
1055 rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
1056 LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
1058 CERROR("%s: fid init error: rc = %d\n",
1059 osp->opd_obd->obd_name, rc);
1063 if (!osp->opd_connect_mdt) {
1064 /* Initialize last id from the storage - will be
1065 * used in orphan cleanup. */
1066 rc = osp_last_used_init(env, osp);
1071 /* Initialize precreation thread, it handles new
1072 * connections as well. */
1073 rc = osp_init_precreate(osp);
1075 GOTO(out_last_used, rc);
1078 * Initialize synhronization mechanism taking
1079 * care of propogating changes to OST in near
1080 * transactional manner.
1082 rc = osp_sync_init(env, osp);
1084 GOTO(out_precreat, rc);
1087 ns_register_cancel(obd->obd_namespace, osp_cancel_weight);
1090 * Initiate connect to OST
1092 ll_generate_random_uuid(uuid);
1093 class_uuid_unparse(uuid, &osp->opd_cluuid);
1095 imp = obd->u.cli.cl_import;
1097 rc = ptlrpc_init_import(imp);
1101 OBD_FREE(osdname, MAX_OBD_NAME);
1105 if (!osp->opd_connect_mdt)
1106 /* stop sync thread */
1109 /* stop precreate thread */
1110 if (!osp->opd_connect_mdt)
1111 osp_precreate_fini(osp);
1113 if (!osp->opd_connect_mdt)
1114 osp_last_used_fini(env, osp);
1116 obd_fid_fini(osp->opd_obd);
1118 ptlrpc_lprocfs_unregister_obd(obd);
1119 lprocfs_obd_cleanup(obd);
1120 if (osp->opd_symlink)
1121 lprocfs_remove(&osp->opd_symlink);
1122 client_obd_cleanup(obd);
1126 if (osp->opd_connect_mdt) {
1127 struct client_obd *cli = &osp->opd_obd->u.cli;
1128 if (cli->cl_rpc_lock != NULL) {
1129 OBD_FREE_PTR(cli->cl_rpc_lock);
1130 cli->cl_rpc_lock = NULL;
1133 obd_disconnect(osp->opd_storage_exp);
1136 OBD_FREE(osdname, MAX_OBD_NAME);
1141 * Implementation of lu_device_type_operations::ldto_device_free
1143 * Free the OSP device in memory. No return value is needed for now,
1144 * so always return NULL to comply with the interface.
1146 * \param[in] env execution environment
1147 * \param[in] lu lu_device of OSP
1149 * \retval NULL NULL unconditionally
1151 static struct lu_device *osp_device_free(const struct lu_env *env,
1152 struct lu_device *lu)
1154 struct osp_device *osp = lu2osp_dev(lu);
1156 if (atomic_read(&lu->ld_ref) && lu->ld_site) {
1157 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1158 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1160 dt_device_fini(&osp->opd_dt_dev);
1167 * Implementation of lu_device_type_operations::ldto_device_alloc
1169 * This function allocates and initializes OSP device in memory according to
1172 * \param[in] env execution environment
1173 * \param[in] type device type of OSP
1174 * \param[in] lcfg config log
1176 * \retval pointer the pointer of allocated OSP if succeed.
1177 * \retval ERR_PTR(errno) ERR_PTR(errno) if failed.
1179 static struct lu_device *osp_device_alloc(const struct lu_env *env,
1180 struct lu_device_type *type,
1181 struct lustre_cfg *lcfg)
1183 struct osp_device *osp;
1184 struct lu_device *ld;
1188 ld = ERR_PTR(-ENOMEM);
1192 ld = osp2lu_dev(osp);
1193 dt_device_init(&osp->opd_dt_dev, type);
1194 rc = osp_init0(env, osp, type, lcfg);
1196 osp_device_free(env, ld);
1204 * Implementation of lu_device_type_operations::ldto_device_fini
1206 * This function cleans up the OSP device, i.e. release and free those
1207 * attached items in osp_device.
1209 * \param[in] env execution environment
1210 * \param[in] ld lu_device of OSP
1212 * \retval NULL NULL if cleanup succeeded.
1213 * \retval ERR_PTR(errno) ERR_PTR(errno) if cleanup failed.
1215 static struct lu_device *osp_device_fini(const struct lu_env *env,
1216 struct lu_device *ld)
1218 struct osp_device *osp = lu2osp_dev(ld);
1219 struct obd_import *imp;
1224 if (osp->opd_async_requests != NULL) {
1225 dt_update_request_destroy(osp->opd_async_requests);
1226 osp->opd_async_requests = NULL;
1229 if (osp->opd_storage_exp)
1230 obd_disconnect(osp->opd_storage_exp);
1232 imp = osp->opd_obd->u.cli.cl_import;
1234 if (imp->imp_rq_pool) {
1235 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1236 imp->imp_rq_pool = NULL;
1239 if (osp->opd_symlink)
1240 lprocfs_remove(&osp->opd_symlink);
1242 LASSERT(osp->opd_obd);
1243 ptlrpc_lprocfs_unregister_obd(osp->opd_obd);
1244 lprocfs_obd_cleanup(osp->opd_obd);
1246 if (osp->opd_connect_mdt) {
1247 struct client_obd *cli = &osp->opd_obd->u.cli;
1248 if (cli->cl_rpc_lock != NULL) {
1249 OBD_FREE_PTR(cli->cl_rpc_lock);
1250 cli->cl_rpc_lock = NULL;
1254 rc = client_obd_cleanup(osp->opd_obd);
1257 RETURN(ERR_PTR(rc));
1266 * Implementation of obd_ops::o_reconnect
1268 * This function is empty and does not need to do anything for now.
1270 static int osp_reconnect(const struct lu_env *env,
1271 struct obd_export *exp, struct obd_device *obd,
1272 struct obd_uuid *cluuid,
1273 struct obd_connect_data *data,
1280 * Implementation of obd_ops::o_connect
1282 * Connect OSP to the remote target (MDT or OST). Allocate the
1283 * export and return it to the LOD, which calls this function
1284 * for each OSP to connect it to the remote target. This function
1285 * is currently only called once per OSP.
1287 * \param[in] env execution environment
1288 * \param[out] exp export connected to OSP
1289 * \param[in] obd OSP device
1290 * \param[in] cluuid OSP device client uuid
1291 * \param[in] data connect_data to be used to connect to the remote
1293 * \param[in] localdata necessary for the API interface, but not used in
1296 * \retval 0 0 if the connection succeeded.
1297 * \retval negative negative errno if the connection failed.
1299 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
1300 struct obd_device *obd, struct obd_uuid *cluuid,
1301 struct obd_connect_data *data, void *localdata)
1303 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1304 struct obd_connect_data *ocd;
1305 struct obd_import *imp;
1306 struct lustre_handle conn;
1311 CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
1313 rc = class_connect(&conn, obd, cluuid);
1317 *exp = class_conn2export(&conn);
1318 /* Why should there ever be more than 1 connect? */
1319 osp->opd_connects++;
1320 LASSERT(osp->opd_connects == 1);
1322 osp->opd_exp = *exp;
1324 imp = osp->opd_obd->u.cli.cl_import;
1325 imp->imp_dlm_handle = conn;
1327 LASSERT(data != NULL);
1328 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
1329 ocd = &imp->imp_connect_data;
1332 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1334 ocd->ocd_version = LUSTRE_VERSION_CODE;
1335 ocd->ocd_index = data->ocd_index;
1336 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1338 rc = ptlrpc_connect_import(imp);
1340 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
1343 osp->opd_obd->u.cli.cl_seq->lcs_exp =
1344 class_export_get(osp->opd_exp);
1347 ptlrpc_pinger_add_import(imp);
1353 * Implementation of obd_ops::o_disconnect
1355 * Disconnect the export for the OSP. This is called by LOD to release the
1356 * OSP during cleanup (\see lod_del_device()). The OSP will be released after
1357 * the export is released.
1359 * \param[in] exp export to be disconnected.
1361 * \retval 0 0 if disconnection succeed
1362 * \retval negative negative errno if disconnection failed
1364 static int osp_obd_disconnect(struct obd_export *exp)
1366 struct obd_device *obd = exp->exp_obd;
1367 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1371 /* Only disconnect the underlying layers on the final disconnect. */
1372 LASSERT(osp->opd_connects == 1);
1373 osp->opd_connects--;
1375 rc = class_disconnect(exp);
1377 CERROR("%s: class disconnect error: rc = %d\n",
1382 /* destroy the device */
1383 class_manual_cleanup(obd);
1389 * Implementation of obd_ops::o_statfs
1391 * Send a RPC to the remote target to get statfs status. This is only used
1392 * in lprocfs helpers by obd_statfs.
1394 * \param[in] env execution environment
1395 * \param[in] exp connection state from this OSP to the parent (LOD)
1397 * \param[out] osfs hold the statfs result
1398 * \param[in] unused Not used in this function for now
1399 * \param[in] flags flags to indicate how OSP will issue the RPC
1401 * \retval 0 0 if statfs succeeded.
1402 * \retval negative negative errno if statfs failed.
1404 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
1405 struct obd_statfs *osfs, __u64 unused, __u32 flags)
1407 struct obd_statfs *msfs;
1408 struct ptlrpc_request *req;
1409 struct obd_import *imp = NULL;
1414 /* Since the request might also come from lprocfs, so we need
1415 * sync this with client_disconnect_export Bug15684 */
1416 down_read(&exp->exp_obd->u.cli.cl_sem);
1417 if (exp->exp_obd->u.cli.cl_import)
1418 imp = class_import_get(exp->exp_obd->u.cli.cl_import);
1419 up_read(&exp->exp_obd->u.cli.cl_sem);
1423 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
1425 class_import_put(imp);
1430 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
1432 ptlrpc_request_free(req);
1435 ptlrpc_request_set_replen(req);
1436 req->rq_request_portal = OST_CREATE_PORTAL;
1437 ptlrpc_at_set_req_timeout(req);
1439 if (flags & OBD_STATFS_NODELAY) {
1440 /* procfs requests not want stat in wait for avoid deadlock */
1441 req->rq_no_resend = 1;
1442 req->rq_no_delay = 1;
1445 rc = ptlrpc_queue_wait(req);
1449 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1451 GOTO(out, rc = -EPROTO);
1457 ptlrpc_req_finished(req);
1462 * Implementation of obd_ops::o_import_event
1464 * This function is called when some related import event happens. It will
1465 * mark the necessary flags according to the event and notify the necessary
1466 * threads (mainly precreate thread).
1468 * \param[in] obd OSP OBD device
1469 * \param[in] imp import attached from OSP to remote (OST/MDT) service
1470 * \param[in] event event related to remote service (IMP_EVENT_*)
1472 * \retval 0 0 if the event handling succeeded.
1473 * \retval negative negative errno if the event handling failed.
1475 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1476 enum obd_import_event event)
1478 struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1481 case IMP_EVENT_DISCON:
1482 d->opd_got_disconnected = 1;
1483 d->opd_imp_connected = 0;
1484 if (d->opd_connect_mdt)
1487 if (d->opd_pre != NULL) {
1488 osp_pre_update_status(d, -ENODEV);
1489 wake_up(&d->opd_pre_waitq);
1492 CDEBUG(D_HA, "got disconnected\n");
1494 case IMP_EVENT_INACTIVE:
1495 d->opd_imp_active = 0;
1496 if (d->opd_connect_mdt)
1499 if (d->opd_pre != NULL) {
1500 osp_pre_update_status(d, -ENODEV);
1501 wake_up(&d->opd_pre_waitq);
1504 CDEBUG(D_HA, "got inactive\n");
1506 case IMP_EVENT_ACTIVE:
1507 d->opd_imp_active = 1;
1509 if (d->opd_got_disconnected)
1510 d->opd_new_connection = 1;
1511 d->opd_imp_connected = 1;
1512 d->opd_imp_seen_connected = 1;
1513 if (d->opd_connect_mdt)
1516 if (d->opd_pre != NULL)
1517 wake_up(&d->opd_pre_waitq);
1519 __osp_sync_check_for_work(d);
1520 CDEBUG(D_HA, "got connected\n");
1522 case IMP_EVENT_INVALIDATE:
1523 if (obd->obd_namespace == NULL)
1525 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1528 case IMP_EVENT_DEACTIVATE:
1529 case IMP_EVENT_ACTIVATE:
1532 CERROR("%s: unsupported import event: %#x\n",
1533 obd->obd_name, event);
1539 * Implementation of obd_ops: o_iocontrol
1541 * This function is the ioctl handler for OSP. Note: lctl will access the OSP
1542 * directly by ioctl, instead of through the MDS stack.
1544 * param[in] cmd ioctl command.
1545 * param[in] exp export of this OSP.
1546 * param[in] len data length of \a karg.
1547 * param[in] karg input argument which is packed as
1549 * param[out] uarg pointer to userspace buffer (must access by
1552 * \retval 0 0 if the ioctl handling succeeded.
1553 * \retval negative negative errno if the ioctl handling failed.
1555 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1556 void *karg, void *uarg)
1558 struct obd_device *obd = exp->exp_obd;
1559 struct osp_device *d;
1560 struct obd_ioctl_data *data = karg;
1565 LASSERT(obd->obd_lu_dev);
1566 d = lu2osp_dev(obd->obd_lu_dev);
1567 LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1569 if (!try_module_get(THIS_MODULE)) {
1570 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1571 module_name(THIS_MODULE));
1576 case OBD_IOC_CLIENT_RECOVER:
1577 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1578 data->ioc_inlbuf1, 0);
1582 case IOC_OSC_SET_ACTIVE:
1583 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1586 case OBD_IOC_PING_TARGET:
1587 rc = ptlrpc_obd_ping(obd);
1590 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1591 cmd, current_comm());
1594 module_put(THIS_MODULE);
1599 * Implementation of obd_ops::o_get_info
1601 * Retrieve information by key. Retrieval starts from the top layer
1602 * (MDT) of the MDS stack and traverses the stack by calling the
1603 * obd_get_info() method of the next sub-layer.
1605 * \param[in] env execution environment
1606 * \param[in] exp export of this OSP
1607 * \param[in] keylen length of \a key
1608 * \param[in] key the key
1609 * \param[out] vallen length of \a val
1610 * \param[out] val holds the value returned by the key
1612 * \retval 0 0 if getting information succeeded.
1613 * \retval negative negative errno if getting information failed.
1615 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1616 __u32 keylen, void *key, __u32 *vallen, void *val)
1620 if (KEY_IS(KEY_OSP_CONNECTED)) {
1621 struct obd_device *obd = exp->exp_obd;
1622 struct osp_device *osp;
1624 if (!obd->obd_set_up || obd->obd_stopping)
1627 osp = lu2osp_dev(obd->obd_lu_dev);
1630 * 1.8/2.0 behaviour is that OST being connected once at least
1631 * is considered "healthy". and one "healthy" OST is enough to
1632 * allow lustre clients to connect to MDS
1634 RETURN(!osp->opd_imp_seen_connected);
1641 * Implementation of obd_ops: o_fid_alloc
1643 * Allocate a FID. There are two cases in which OSP performs
1646 * 1. FID precreation for data objects, which is done in
1647 * osp_precreate_fids() instead of this function.
1648 * 2. FID allocation for each sub-stripe of a striped directory.
1649 * Similar to other FID clients, OSP requests the sequence
1650 * from its corresponding remote MDT, which in turn requests
1651 * sequences from the sequence controller (MDT0).
1653 * \param[in] env execution environment
1654 * \param[in] exp export of the OSP
1655 * \param[out] fid FID being allocated
1656 * \param[in] unused necessary for the interface but unused.
1658 * \retval 0 0 FID allocated successfully.
1659 * \retval 1 1 FID allocated successfully and new sequence
1660 * requested from seq meta server
1661 * \retval negative negative errno if FID allocation failed.
1663 static int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1664 struct lu_fid *fid, struct md_op_data *unused)
1666 struct client_obd *cli = &exp->exp_obd->u.cli;
1667 struct osp_device *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev);
1668 struct lu_client_seq *seq = cli->cl_seq;
1671 LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
1672 /* Sigh, fid client is not ready yet */
1673 LASSERT(osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL);
1675 RETURN(seq_client_alloc_fid(env, seq, fid));
1678 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1679 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1680 static void osp_key_exit(const struct lu_context *ctx,
1681 struct lu_context_key *key, void *data)
1683 struct osp_thread_info *info = data;
1685 info->osi_attr.la_valid = 0;
1688 struct lu_context_key osp_thread_key = {
1689 .lct_tags = LCT_MD_THREAD,
1690 .lct_init = osp_key_init,
1691 .lct_fini = osp_key_fini,
1692 .lct_exit = osp_key_exit
1695 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1696 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1698 struct lu_context_key osp_txn_key = {
1699 .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE,
1700 .lct_init = osp_txn_key_init,
1701 .lct_fini = osp_txn_key_fini
1703 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1705 static struct lu_device_type_operations osp_device_type_ops = {
1706 .ldto_init = osp_type_init,
1707 .ldto_fini = osp_type_fini,
1709 .ldto_start = osp_type_start,
1710 .ldto_stop = osp_type_stop,
1712 .ldto_device_alloc = osp_device_alloc,
1713 .ldto_device_free = osp_device_free,
1715 .ldto_device_fini = osp_device_fini
1718 static struct lu_device_type osp_device_type = {
1719 .ldt_tags = LU_DEVICE_DT,
1720 .ldt_name = LUSTRE_OSP_NAME,
1721 .ldt_ops = &osp_device_type_ops,
1722 .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
1725 static struct obd_ops osp_obd_device_ops = {
1726 .o_owner = THIS_MODULE,
1727 .o_add_conn = client_import_add_conn,
1728 .o_del_conn = client_import_del_conn,
1729 .o_reconnect = osp_reconnect,
1730 .o_connect = osp_obd_connect,
1731 .o_disconnect = osp_obd_disconnect,
1732 .o_get_info = osp_obd_get_info,
1733 .o_import_event = osp_import_event,
1734 .o_iocontrol = osp_iocontrol,
1735 .o_statfs = osp_obd_statfs,
1736 .o_fid_init = client_fid_init,
1737 .o_fid_fini = client_fid_fini,
1738 .o_fid_alloc = osp_fid_alloc,
1741 struct llog_operations osp_mds_ost_orig_logops;
1744 * Initialize OSP module.
1746 * Register device types OSP and Light Weight Proxy (LWP) (\see lwp_dev.c)
1747 * in obd_types (\see class_obd.c). Initialize procfs for the
1748 * the OSP device. Note: OSP was called OSC before Lustre 2.4,
1749 * so for compatibility it still uses the name "osc" in procfs.
1750 * This is called at module load time.
1752 * \retval 0 0 if initialization succeeds.
1753 * \retval negative negative errno if initialization failed.
1755 static int __init osp_mod_init(void)
1757 struct obd_type *type;
1760 rc = lu_kmem_init(osp_caches);
1765 rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL,
1766 LUSTRE_OSP_NAME, &osp_device_type);
1768 lu_kmem_fini(osp_caches);
1772 rc = class_register_type(&lwp_obd_device_ops, NULL, true, NULL,
1773 LUSTRE_LWP_NAME, &lwp_device_type);
1775 class_unregister_type(LUSTRE_OSP_NAME);
1776 lu_kmem_fini(osp_caches);
1780 /* Note: add_rec/delcare_add_rec will be only used by catalogs */
1781 osp_mds_ost_orig_logops = llog_osd_ops;
1782 osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
1783 osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1785 /* create "osc" entry in procfs for compatibility purposes */
1786 type = class_search_type(LUSTRE_OSC_NAME);
1787 if (type != NULL && type->typ_procroot != NULL)
1790 type = class_search_type(LUSTRE_OSP_NAME);
1791 type->typ_procsym = lprocfs_register("osc", proc_lustre_root,
1793 if (IS_ERR(type->typ_procsym)) {
1794 CERROR("osp: can't create compat entry \"osc\": %d\n",
1795 (int) PTR_ERR(type->typ_procsym));
1796 type->typ_procsym = NULL;
1802 * Finalize OSP module.
1804 * This callback is called when kernel unloads OSP module from memory, and
1805 * it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
1807 static void __exit osp_mod_exit(void)
1809 class_unregister_type(LUSTRE_LWP_NAME);
1810 class_unregister_type(LUSTRE_OSP_NAME);
1811 lu_kmem_fini(osp_caches);
1814 MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
1815 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
1816 MODULE_VERSION(LUSTRE_VERSION_STRING);
1817 MODULE_LICENSE("GPL");
1819 module_init(osp_mod_init);
1820 module_exit(osp_mod_exit);