Whamcloud - gitweb
LU-7345 obdclass: annotate locks in __local_file_create
[fs/lustre-release.git] / lustre / osp / osp_dev.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
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
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2012, 2015, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/osp/osp_dev.c
37  *
38  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
39  * Author: Mikhail Pershin <mike.pershin@intel.com>
40  * Author: Di Wang <di.wang@intel.com>
41  */
42 /*
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:
51  *
52  * - OSP devices allocate FIDs for the stripe sub-objects of a striped
53  *   file or directory.
54  *
55  * - OSP objects represent the remote MDT and OST objects that are
56  *   the stripes of a striped object.
57  *
58  * - OSP devices log, send, and track synchronous operations (setattr
59  *   and unlink) to remote targets.
60  *
61  * - OSP objects are the bottom slice of the compound LU object
62  *   representing a remote MDT object: MDT/MDD/LOD/OSP.
63  *
64  * - OSP objects are used by LFSCK to represent remote OST objects
65  *   during the verification of MDT-OST consistency.
66  *
67  * - OSP devices batch idempotent requests (declare_attr_get() and
68  *   declare_xattr_get()) to the remote target and cache their results.
69  *
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.
73  */
74
75 #define DEBUG_SUBSYSTEM S_MDS
76
77 #include <linux/kthread.h>
78 #include <obd_class.h>
79 #include <lustre_ioctl.h>
80 #include <lustre_param.h>
81 #include <lustre_log.h>
82
83 #include "osp_internal.h"
84
85 /* Slab for OSP object allocation */
86 struct kmem_cache *osp_object_kmem;
87
88 static struct lu_kmem_descr osp_caches[] = {
89         {
90                 .ckd_cache = &osp_object_kmem,
91                 .ckd_name  = "osp_obj",
92                 .ckd_size  = sizeof(struct osp_object)
93         },
94         {
95                 .ckd_cache = NULL
96         }
97 };
98
99 /**
100  * Implementation of lu_device_operations::ldo_object_alloc
101  *
102  * Allocates an OSP object in memory, whose FID is on the remote target.
103  *
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
108  *                      an OST object.
109  * \param[in] d         OSP device
110  *
111  * \retval object       object being created if the creation succeed.
112  * \retval NULL         NULL if the creation failed.
113  */
114 static struct lu_object *osp_object_alloc(const struct lu_env *env,
115                                           const struct lu_object_header *hdr,
116                                           struct lu_device *d)
117 {
118         struct lu_object_header *h = NULL;
119         struct osp_object       *o;
120         struct lu_object        *l;
121
122         OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
123         if (o != NULL) {
124                 l = &o->opo_obj.do_lu;
125
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 */
130                 if (hdr == NULL) {
131                         h = &o->opo_header;
132                         lu_object_header_init(h);
133                         dt_object_init(&o->opo_obj, h, d);
134                         lu_object_add_top(h, l);
135                 } else {
136                         dt_object_init(&o->opo_obj, h, d);
137                 }
138
139                 l->lo_ops = &osp_lu_obj_ops;
140
141                 return l;
142         } else {
143                 return NULL;
144         }
145 }
146
147 /**
148  * Find or create the local object
149  *
150  * Finds or creates the local file referenced by \a reg_id and return the
151  * attributes of the local file.
152  *
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.
159  *
160  * \retval object               object(dt_object) found or created
161  * \retval ERR_PTR(errno)       ERR_PTR(errno) if not get the object.
162  */
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)
166 {
167         struct osp_thread_info *osi = osp_env_info(env);
168         struct dt_object_format dof = { 0 };
169         struct dt_object       *dto;
170         int                  rc;
171         ENTRY;
172
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,
179                                 &dof, attr);
180         if (IS_ERR(dto))
181                 RETURN(dto);
182
183         /* Get attributes of the local object. */
184         rc = dt_attr_get(env, dto, attr);
185         if (rc) {
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);
189                 RETURN(ERR_PTR(rc));
190         }
191         RETURN(dto);
192 }
193
194 /**
195  * Write data buffer to a local file object.
196  *
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
202  *
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.
206  */
207 static int osp_write_local_file(const struct lu_env *env,
208                                 struct osp_device *osp,
209                                 struct dt_object *dt_obj,
210                                 struct lu_buf *buf,
211                                 loff_t offset)
212 {
213         struct thandle *th;
214         int rc;
215
216         th = dt_trans_create(env, osp->opd_storage);
217         if (IS_ERR(th))
218                 RETURN(PTR_ERR(th));
219
220         rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
221         if (rc)
222                 GOTO(out, rc);
223         rc = dt_trans_start_local(env, osp->opd_storage, th);
224         if (rc)
225                 GOTO(out, rc);
226
227         rc = dt_record_write(env, dt_obj, buf, &offset, th);
228 out:
229         dt_trans_stop(env, osp->opd_storage, th);
230         RETURN(rc);
231 }
232
233 /**
234  * Initialize last ID object.
235  *
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
239  * OSTs.
240  *
241  * \param[in] env       execution environment
242  * \param[in] osp       OSP device
243  *
244  * \retval 0            0 if initialization succeed
245  * \retval negative     negative errno if initialization failed
246  */
247 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
248 {
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;
252         int                     rc = -EFAULT;
253         ENTRY;
254
255         dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
256                                             MDD_LOV_OBJ_OID);
257         if (IS_ERR(dto))
258                 RETURN(PTR_ERR(dto));
259
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,
264                                    osp->opd_index);
265                 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
266                 if (rc != 0 && rc != -EFAULT)
267                         GOTO(out, rc);
268         }
269
270         if (rc == -EFAULT) { /* fresh LAST_ID */
271                 fid->f_oid = 0;
272                 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
273                                    osp->opd_index);
274                 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
275                                           osi->osi_off);
276                 if (rc != 0)
277                         GOTO(out, rc);
278         }
279         osp->opd_last_used_oid_file = dto;
280         RETURN(0);
281 out:
282         /* object will be released in device cleanup path */
283         CERROR("%s: can't initialize lov_objid: rc = %d\n",
284                osp->opd_obd->obd_name, rc);
285         lu_object_put(env, &dto->do_lu);
286         osp->opd_last_used_oid_file = NULL;
287         RETURN(rc);
288 }
289
290 /**
291  * Initialize last sequence object.
292  *
293  * This function initializes the LAST_SEQ file in the local OSD, which stores
294  * the current last used sequence of data objects. The MDT will use the last
295  * sequence and last id (\see osp_init_last_objid()) to synchronize the
296  * precreate object cache with OSTs.
297  *
298  * \param[in] env       execution environment
299  * \param[in] osp       OSP device
300  *
301  * \retval 0            0 if initialization succeed
302  * \retval negative     negative errno if initialization failed
303  */
304 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
305 {
306         struct osp_thread_info  *osi = osp_env_info(env);
307         struct lu_fid           *fid = &osp->opd_last_used_fid;
308         struct dt_object        *dto;
309         int                     rc = -EFAULT;
310         ENTRY;
311
312         dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
313                                             MDD_LOV_OBJ_OSEQ);
314         if (IS_ERR(dto))
315                 RETURN(PTR_ERR(dto));
316
317         /* object will be released in device cleanup path */
318         if (osi->osi_attr.la_size >=
319             sizeof(osi->osi_id) * (osp->opd_index + 1)) {
320                 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
321                                    osp->opd_index);
322                 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
323                 if (rc != 0 && rc != -EFAULT)
324                         GOTO(out, rc);
325         }
326
327         if (rc == -EFAULT) { /* fresh OSP */
328                 fid->f_seq = 0;
329                 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
330                                     osp->opd_index);
331                 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
332                                           osi->osi_off);
333                 if (rc != 0)
334                         GOTO(out, rc);
335         }
336         osp->opd_last_used_seq_file = dto;
337         RETURN(0);
338 out:
339         /* object will be released in device cleanup path */
340         CERROR("%s: can't initialize lov_seq: rc = %d\n",
341                osp->opd_obd->obd_name, rc);
342         lu_object_put(env, &dto->do_lu);
343         osp->opd_last_used_seq_file = NULL;
344         RETURN(rc);
345 }
346
347 /**
348  * Initialize last OID and sequence object.
349  *
350  * If the MDT is just upgraded to 2.4 from the lower version, where the
351  * LAST_SEQ file does not exist, the file will be created and IDIF sequence
352  * will be written into the file.
353  *
354  * \param[in] env       execution environment
355  * \param[in] osp       OSP device
356  *
357  * \retval 0            0 if initialization succeed
358  * \retval negative     negative error if initialization failed
359  */
360 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
361 {
362         struct osp_thread_info *osi = osp_env_info(env);
363         int                  rc;
364         ENTRY;
365
366         fid_zero(&osp->opd_last_used_fid);
367         rc = osp_init_last_objid(env, osp);
368         if (rc < 0) {
369                 CERROR("%s: Can not get ids %d from old objid!\n",
370                        osp->opd_obd->obd_name, rc);
371                 RETURN(rc);
372         }
373
374         rc = osp_init_last_seq(env, osp);
375         if (rc < 0) {
376                 CERROR("%s: Can not get ids %d from old objid!\n",
377                        osp->opd_obd->obd_name, rc);
378                 GOTO(out, rc);
379         }
380
381         if (fid_oid(&osp->opd_last_used_fid) != 0 &&
382             fid_seq(&osp->opd_last_used_fid) == 0) {
383                 /* Just upgrade from the old version,
384                  * set the seq to be IDIF */
385                 osp->opd_last_used_fid.f_seq =
386                    fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
387                                 osp->opd_index);
388                 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
389                                     &osp->opd_last_used_fid.f_seq,
390                                     osp->opd_index);
391                 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
392                                           &osi->osi_lb, osi->osi_off);
393                 if (rc) {
394                         CERROR("%s : Can not write seq file: rc = %d\n",
395                                osp->opd_obd->obd_name, rc);
396                         GOTO(out, rc);
397                 }
398         }
399
400         if (!fid_is_zero(&osp->opd_last_used_fid) &&
401                  !fid_is_sane(&osp->opd_last_used_fid)) {
402                 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
403                         PFID(&osp->opd_last_used_fid));
404                 GOTO(out, rc = -EINVAL);
405         }
406
407         CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
408                osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
409 out:
410         if (rc != 0) {
411                 if (osp->opd_last_used_oid_file != NULL) {
412                         lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
413                         osp->opd_last_used_oid_file = NULL;
414                 }
415                 if (osp->opd_last_used_seq_file != NULL) {
416                         lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
417                         osp->opd_last_used_seq_file = NULL;
418                 }
419         }
420
421         RETURN(rc);
422 }
423
424 /**
425  * Release the last sequence and OID file objects in OSP device.
426  *
427  * \param[in] env       execution environment
428  * \param[in] osp       OSP device
429  */
430 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp)
431 {
432         /* release last_used file */
433         if (osp->opd_last_used_oid_file != NULL) {
434                 lu_object_put(env, &osp->opd_last_used_oid_file->do_lu);
435                 osp->opd_last_used_oid_file = NULL;
436         }
437
438         if (osp->opd_last_used_seq_file != NULL) {
439                 lu_object_put(env, &osp->opd_last_used_seq_file->do_lu);
440                 osp->opd_last_used_seq_file = NULL;
441         }
442 }
443
444 /**
445  * Disconnects the connection between OSP and its correspondent MDT or OST, and
446  * the import will be marked as inactive. It will only be called during OSP
447  * cleanup process.
448  *
449  * \param[in] d         OSP device being disconnected
450  *
451  * \retval 0            0 if disconnection succeed
452  * \retval negative     negative errno if disconnection failed
453  */
454 static int osp_disconnect(struct osp_device *d)
455 {
456         struct obd_import *imp;
457         int rc = 0;
458
459         imp = d->opd_obd->u.cli.cl_import;
460
461         /* Mark import deactivated now, so we don't try to reconnect if any
462          * of the cleanup RPCs fails (e.g. ldlm cancel, etc).  We don't
463          * fully deactivate the import, or that would drop all requests. */
464         LASSERT(imp != NULL);
465         spin_lock(&imp->imp_lock);
466         imp->imp_deactive = 1;
467         spin_unlock(&imp->imp_lock);
468
469         ptlrpc_deactivate_import(imp);
470
471         /* Some non-replayable imports (MDS's OSCs) are pinged, so just
472          * delete it regardless.  (It's safe to delete an import that was
473          * never added.) */
474         (void)ptlrpc_pinger_del_import(imp);
475
476         rc = ptlrpc_disconnect_import(imp, 0);
477         if (rc != 0)
478                 CERROR("%s: can't disconnect: rc = %d\n",
479                        d->opd_obd->obd_name, rc);
480
481         ptlrpc_invalidate_import(imp);
482
483         RETURN(rc);
484 }
485
486 /**
487  * Initialize the osp_update structure in OSP device
488  *
489  * Allocate osp update structure and start update thread.
490  *
491  * \param[in] osp       OSP device
492  *
493  * \retval              0 if initialization succeeds.
494  * \retval              negative errno if initialization fails.
495  */
496 static int osp_update_init(struct osp_device *osp)
497 {
498         struct l_wait_info      lwi = { 0 };
499         struct task_struct      *task;
500
501         ENTRY;
502
503         LASSERT(osp->opd_connect_mdt);
504
505         OBD_ALLOC_PTR(osp->opd_update);
506         if (osp->opd_update == NULL)
507                 RETURN(-ENOMEM);
508
509         init_waitqueue_head(&osp->opd_update_thread.t_ctl_waitq);
510         init_waitqueue_head(&osp->opd_update->ou_waitq);
511         spin_lock_init(&osp->opd_update->ou_lock);
512         INIT_LIST_HEAD(&osp->opd_update->ou_list);
513         osp->opd_update->ou_rpc_version = 1;
514         osp->opd_update->ou_version = 1;
515
516         /* start thread handling sending updates to the remote MDT */
517         task = kthread_run(osp_send_update_thread, osp,
518                            "osp_up%u-%u", osp->opd_index, osp->opd_group);
519         if (IS_ERR(task)) {
520                 int rc = PTR_ERR(task);
521
522                 OBD_FREE_PTR(osp->opd_update);
523                 osp->opd_update = NULL;
524                 CERROR("%s: can't start precreate thread: rc = %d\n",
525                        osp->opd_obd->obd_name, rc);
526                 RETURN(rc);
527         }
528
529         l_wait_event(osp->opd_update_thread.t_ctl_waitq,
530                      osp_send_update_thread_running(osp) ||
531                      osp_send_update_thread_stopped(osp), &lwi);
532
533         RETURN(0);
534 }
535
536 /**
537  * Finialize osp_update structure in OSP device
538  *
539  * Stop the OSP update sending thread, then delete the left
540  * osp thandle in the sending list.
541  *
542  * \param [in] osp      OSP device.
543  */
544 static void osp_update_fini(const struct lu_env *env, struct osp_device *osp)
545 {
546         struct osp_update_request *our;
547         struct osp_update_request *tmp;
548         struct osp_updates *ou = osp->opd_update;
549
550         if (ou == NULL)
551                 return;
552
553         osp->opd_update_thread.t_flags = SVC_STOPPING;
554         wake_up(&ou->ou_waitq);
555
556         wait_event(osp->opd_update_thread.t_ctl_waitq,
557                    osp->opd_update_thread.t_flags & SVC_STOPPED);
558
559         /* Remove the left osp thandle from the list */
560         spin_lock(&ou->ou_lock);
561         list_for_each_entry_safe(our, tmp, &ou->ou_list,
562                                  our_list) {
563                 list_del_init(&our->our_list);
564                 LASSERT(our->our_th != NULL);
565                 osp_trans_callback(env, our->our_th, -EIO);
566                 /* our will be destroyed in osp_thandle_put() */
567                 osp_thandle_put(our->our_th);
568         }
569         spin_unlock(&ou->ou_lock);
570
571         OBD_FREE_PTR(ou);
572         osp->opd_update = NULL;
573 }
574
575 /**
576  * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
577  * precreate threads etc.
578  *
579  * \param[in] env       execution environment.
580  * \param[in] d         OSP device being disconnected.
581  *
582  * \retval 0            0 if cleanup succeed
583  * \retval negative     negative errno if cleanup failed
584  */
585 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
586 {
587         int                      rc = 0;
588         ENTRY;
589
590         LASSERT(env);
591
592         rc = osp_disconnect(d);
593
594         if (!d->opd_connect_mdt) {
595                 /* stop sync thread */
596                 osp_sync_fini(d);
597
598                 /* stop precreate thread */
599                 osp_precreate_fini(d);
600
601                 /* release last_used file */
602                 osp_last_used_fini(env, d);
603         }
604
605         obd_fid_fini(d->opd_obd);
606
607         RETURN(rc);
608 }
609
610 /**
611  * Implementation of osp_lu_ops::ldo_process_config
612  *
613  * This function processes config log records in OSP layer. It is usually
614  * called from the top layer of MDT stack, and goes through the stack by calling
615  * ldo_process_config of next layer.
616  *
617  * \param[in] env       execution environment
618  * \param[in] dev       lu_device of OSP
619  * \param[in] lcfg      config log
620  *
621  * \retval 0            0 if the config log record is executed correctly.
622  * \retval negative     negative errno if the record execution fails.
623  */
624 static int osp_process_config(const struct lu_env *env,
625                               struct lu_device *dev, struct lustre_cfg *lcfg)
626 {
627         struct osp_device               *d = lu2osp_dev(dev);
628         struct obd_device               *obd = d->opd_obd;
629         int                              rc;
630
631         ENTRY;
632
633         switch (lcfg->lcfg_command) {
634         case LCFG_PRE_CLEANUP:
635                 rc = osp_disconnect(d);
636                 osp_update_fini(env, d);
637                 break;
638         case LCFG_CLEANUP:
639                 lu_dev_del_linkage(dev->ld_site, dev);
640                 rc = osp_shutdown(env, d);
641                 break;
642         case LCFG_PARAM:
643                 LASSERT(obd);
644                 rc = class_process_proc_param(d->opd_connect_mdt ?
645                                               PARAM_OSP : PARAM_OSC,
646                                               obd->obd_vars, lcfg, obd);
647                 if (rc > 0)
648                         rc = 0;
649                 if (rc == -ENOSYS) {
650                         /* class_process_proc_param() haven't found matching
651                          * parameter and returned ENOSYS so that layer(s)
652                          * below could use that. But OSP is the bottom, so
653                          * just ignore it */
654                         CERROR("%s: unknown param %s\n",
655                                (char *)lustre_cfg_string(lcfg, 0),
656                                (char *)lustre_cfg_string(lcfg, 1));
657                         rc = 0;
658                 }
659                 break;
660         default:
661                 CERROR("%s: unknown command %u\n",
662                        (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
663                 rc = 0;
664                 break;
665         }
666
667         RETURN(rc);
668 }
669
670 /**
671  * Implementation of osp_lu_ops::ldo_recovery_complete
672  *
673  * This function is called after recovery is finished, and OSP layer
674  * will wake up precreate thread here.
675  *
676  * \param[in] env       execution environment
677  * \param[in] dev       lu_device of OSP
678  *
679  * \retval 0            0 unconditionally
680  */
681 static int osp_recovery_complete(const struct lu_env *env,
682                                  struct lu_device *dev)
683 {
684         struct osp_device       *osp = lu2osp_dev(dev);
685
686         ENTRY;
687         osp->opd_recovery_completed = 1;
688
689         if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
690                 wake_up(&osp->opd_pre_waitq);
691
692         RETURN(0);
693 }
694
695 const struct lu_device_operations osp_lu_ops = {
696         .ldo_object_alloc       = osp_object_alloc,
697         .ldo_process_config     = osp_process_config,
698         .ldo_recovery_complete  = osp_recovery_complete,
699 };
700
701 /**
702  * Implementation of dt_device_operations::dt_statfs
703  *
704  * This function provides statfs status (for precreation) from
705  * corresponding OST. Note: this function only retrieves the status
706  * from the OSP device, and the real statfs RPC happens inside
707  * precreate thread (\see osp_statfs_update). Note: OSP for MDT does
708  * not need to retrieve statfs data for now.
709  *
710  * \param[in] env       execution environment.
711  * \param[in] dev       dt_device of OSP.
712  * \param[out] sfs      holds the retrieved statfs data.
713  *
714  * \retval 0            0 statfs data was retrieved successfully or
715  *                      retrieval was not needed
716  * \retval negative     negative errno if get statfs failed.
717  */
718 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
719                       struct obd_statfs *sfs)
720 {
721         struct osp_device *d = dt2osp_dev(dev);
722
723         ENTRY;
724
725         if (unlikely(d->opd_imp_active == 0))
726                 RETURN(-ENOTCONN);
727
728         if (d->opd_pre == NULL)
729                 RETURN(0);
730
731         /* return recently updated data */
732         *sfs = d->opd_statfs;
733
734         /*
735          * layer above osp (usually lod) can use ffree to estimate
736          * how many objects are available for immediate creation
737          */
738         spin_lock(&d->opd_pre_lock);
739         LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
740                  fid_seq(&d->opd_pre_used_fid),
741                  "last_created "DFID", next_fid "DFID"\n",
742                  PFID(&d->opd_pre_last_created_fid),
743                  PFID(&d->opd_pre_used_fid));
744         sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
745                               fid_oid(&d->opd_pre_used_fid);
746         sfs->os_fprecreated -= d->opd_pre_reserved;
747         spin_unlock(&d->opd_pre_lock);
748
749         LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
750
751         CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
752                LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
753                sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
754                sfs->os_files, sfs->os_ffree);
755         RETURN(0);
756 }
757
758 static int osp_sync_timeout(void *data)
759 {
760         return 1;
761 }
762
763 /**
764  * Implementation of dt_device_operations::dt_sync
765  *
766  * This function synchronizes the OSP cache to the remote target. It wakes
767  * up unlink log threads and sends out unlink records to the remote OST.
768  *
769  * \param[in] env       execution environment
770  * \param[in] dev       dt_device of OSP
771  *
772  * \retval 0            0 if synchronization succeeds
773  * \retval negative     negative errno if synchronization fails
774  */
775 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
776 {
777         struct osp_device *d = dt2osp_dev(dev);
778         cfs_time_t         expire;
779         struct l_wait_info lwi = { 0 };
780         unsigned long      id, old;
781         int                rc = 0;
782         unsigned long      start = cfs_time_current();
783         ENTRY;
784
785         /* No Sync between MDTs yet. */
786         if (d->opd_connect_mdt)
787                 RETURN(0);
788
789         if (unlikely(d->opd_imp_active == 0))
790                 RETURN(-ENOTCONN);
791
792         id = d->opd_syn_last_used_id;
793         down_write(&d->opd_async_updates_rwsem);
794
795         CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
796                atomic_read(&d->opd_async_updates_count));
797
798         /* make sure the connection is fine */
799         expire = cfs_time_shift(obd_timeout);
800         lwi = LWI_TIMEOUT(expire - cfs_time_current(), osp_sync_timeout, d);
801         rc = l_wait_event(d->opd_syn_barrier_waitq,
802                           atomic_read(&d->opd_async_updates_count) == 0,
803                           &lwi);
804         up_write(&d->opd_async_updates_rwsem);
805         if (rc != 0)
806                 GOTO(out, rc);
807
808         CDEBUG(D_CACHE, "%s: id: used %lu, processed %lu\n",
809                d->opd_obd->obd_name, id, d->opd_syn_last_processed_id);
810
811         /* wait till all-in-line are processed */
812         while (d->opd_syn_last_processed_id < id) {
813
814                 old = d->opd_syn_last_processed_id;
815
816                 /* make sure the connection is fine */
817                 expire = cfs_time_shift(obd_timeout);
818                 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
819                                   osp_sync_timeout, d);
820                 l_wait_event(d->opd_syn_barrier_waitq,
821                              d->opd_syn_last_processed_id >= id,
822                              &lwi);
823
824                 if (d->opd_syn_last_processed_id >= id)
825                         break;
826
827                 if (d->opd_syn_last_processed_id != old) {
828                         /* some progress have been made,
829                          * keep trying... */
830                         continue;
831                 }
832
833                 /* no changes and expired, something is wrong */
834                 GOTO(out, rc = -ETIMEDOUT);
835         }
836
837         /* block new processing (barrier>0 - few callers are possible */
838         atomic_inc(&d->opd_syn_barrier);
839
840         CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
841                d->opd_syn_rpc_in_flight);
842
843         /* wait till all-in-flight are replied, so executed by the target */
844         /* XXX: this is used by LFSCK at the moment, which doesn't require
845          *      all the changes to be committed, but in general it'd be
846          *      better to wait till commit */
847         while (d->opd_syn_rpc_in_flight > 0) {
848
849                 old = d->opd_syn_rpc_in_flight;
850
851                 expire = cfs_time_shift(obd_timeout);
852                 lwi = LWI_TIMEOUT(expire - cfs_time_current(),
853                                   osp_sync_timeout, d);
854                 l_wait_event(d->opd_syn_barrier_waitq,
855                                 d->opd_syn_rpc_in_flight == 0, &lwi);
856
857                 if (d->opd_syn_rpc_in_flight == 0)
858                         break;
859
860                 if (d->opd_syn_rpc_in_flight != old) {
861                         /* some progress have been made */
862                         continue;
863                 }
864
865                 /* no changes and expired, something is wrong */
866                 GOTO(out, rc = -ETIMEDOUT);
867         }
868
869 out:
870         /* resume normal processing (barrier=0) */
871         atomic_dec(&d->opd_syn_barrier);
872         __osp_sync_check_for_work(d);
873
874         CDEBUG(D_CACHE, "%s: done in %lu: rc = %d\n", d->opd_obd->obd_name,
875                cfs_time_current() - start, rc);
876
877         RETURN(rc);
878 }
879
880 const struct dt_device_operations osp_dt_ops = {
881         .dt_statfs       = osp_statfs,
882         .dt_sync         = osp_sync,
883         .dt_trans_create = osp_trans_create,
884         .dt_trans_start  = osp_trans_start,
885         .dt_trans_stop   = osp_trans_stop,
886         .dt_trans_cb_add   = osp_trans_cb_add,
887 };
888
889 /**
890  * Connect OSP to local OSD.
891  *
892  * Locate the local OSD referenced by \a nextdev and connect to it. Sometimes,
893  * OSP needs to access the local OSD to store some information. For example,
894  * during precreate, it needs to update last used OID and sequence file
895  * (LAST_SEQ) in local OSD.
896  *
897  * \param[in] env       execution environment
898  * \param[in] osp       OSP device
899  * \param[in] nextdev   the name of local OSD
900  *
901  * \retval 0            0 connection succeeded
902  * \retval negative     negative errno connection failed
903  */
904 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *osp,
905                               const char *nextdev)
906 {
907         struct obd_connect_data *data = NULL;
908         struct obd_device       *obd;
909         int                      rc;
910
911         ENTRY;
912
913         LASSERT(osp->opd_storage_exp == NULL);
914
915         OBD_ALLOC_PTR(data);
916         if (data == NULL)
917                 RETURN(-ENOMEM);
918
919         obd = class_name2obd(nextdev);
920         if (obd == NULL) {
921                 CERROR("%s: can't locate next device: %s\n",
922                        osp->opd_obd->obd_name, nextdev);
923                 GOTO(out, rc = -ENOTCONN);
924         }
925
926         rc = obd_connect(env, &osp->opd_storage_exp, obd, &obd->obd_uuid, data,
927                          NULL);
928         if (rc) {
929                 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
930                        osp->opd_obd->obd_name, nextdev, rc);
931                 GOTO(out, rc);
932         }
933
934         osp->opd_dt_dev.dd_lu_dev.ld_site =
935                 osp->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
936         LASSERT(osp->opd_dt_dev.dd_lu_dev.ld_site);
937         osp->opd_storage = lu2dt_dev(osp->opd_storage_exp->exp_obd->obd_lu_dev);
938
939 out:
940         OBD_FREE_PTR(data);
941         RETURN(rc);
942 }
943
944 /**
945  * Determine if the lock needs to be cancelled
946  *
947  * Determine if the unused lock should be cancelled before replay, see
948  * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists
949  * between MDTs.
950  *
951  * \param[in] lock      lock to be checked.
952  *
953  * \retval              1 if the lock needs to be cancelled before replay.
954  * \retval              0 if the lock does not need to be cancelled before
955  *                      replay.
956  */
957 static int osp_cancel_weight(struct ldlm_lock *lock)
958 {
959         if (lock->l_resource->lr_type != LDLM_IBITS)
960                 RETURN(0);
961
962         RETURN(1);
963 }
964
965 /**
966  * Initialize OSP device according to the parameters in the configuration
967  * log \a cfg.
968  *
969  * Reconstruct the local device name from the configuration profile, and
970  * initialize necessary threads and structures according to the OSP type
971  * (MDT or OST).
972  *
973  * Since there is no record in the MDT configuration for the local disk
974  * device, we have to extract this from elsewhere in the profile.
975  * The only information we get at setup is from the OSC records:
976  * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
977  *
978  * Note: configs generated by Lustre 1.8 are missing the -MDTxxxx part,
979  * so, we need to reconstruct the name of the underlying OSD from this:
980  * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd".
981  *
982  * \param[in] env       execution environment
983  * \param[in] osp       OSP device
984  * \param[in] ldt       lu device type of OSP
985  * \param[in] cfg       configuration log
986  *
987  * \retval 0            0 if OSP initialization succeeded.
988  * \retval negative     negative errno if OSP initialization failed.
989  */
990 static int osp_init0(const struct lu_env *env, struct osp_device *osp,
991                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
992 {
993         struct obd_device       *obd;
994         struct obd_import       *imp;
995         class_uuid_t            uuid;
996         char                    *src, *tgt, *mdt, *osdname = NULL;
997         int                     rc;
998         long                    idx;
999
1000         ENTRY;
1001
1002         mutex_init(&osp->opd_async_requests_mutex);
1003         INIT_LIST_HEAD(&osp->opd_async_updates);
1004         init_rwsem(&osp->opd_async_updates_rwsem);
1005         atomic_set(&osp->opd_async_updates_count, 0);
1006
1007         obd = class_name2obd(lustre_cfg_string(cfg, 0));
1008         if (obd == NULL) {
1009                 CERROR("Cannot find obd with name %s\n",
1010                        lustre_cfg_string(cfg, 0));
1011                 RETURN(-ENODEV);
1012         }
1013         osp->opd_obd = obd;
1014
1015         src = lustre_cfg_string(cfg, 0);
1016         if (src == NULL)
1017                 RETURN(-EINVAL);
1018
1019         tgt = strrchr(src, '-');
1020         if (tgt == NULL) {
1021                 CERROR("%s: invalid target name %s: rc = %d\n",
1022                        osp->opd_obd->obd_name, lustre_cfg_string(cfg, 0),
1023                        -EINVAL);
1024                 RETURN(-EINVAL);
1025         }
1026
1027         if (strncmp(tgt, "-osc", 4) == 0) {
1028                 /* Old OSC name fsname-OSTXXXX-osc */
1029                 for (tgt--; tgt > src && *tgt != '-'; tgt--)
1030                         ;
1031                 if (tgt == src) {
1032                         CERROR("%s: invalid target name %s: rc = %d\n",
1033                                osp->opd_obd->obd_name,
1034                                lustre_cfg_string(cfg, 0), -EINVAL);
1035                         RETURN(-EINVAL);
1036                 }
1037
1038                 if (strncmp(tgt, "-OST", 4) != 0) {
1039                         CERROR("%s: invalid target name %s: rc = %d\n",
1040                                osp->opd_obd->obd_name,
1041                                lustre_cfg_string(cfg, 0), -EINVAL);
1042                         RETURN(-EINVAL);
1043                 }
1044
1045                 idx = simple_strtol(tgt + 4, &mdt, 16);
1046                 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
1047                         CERROR("%s: invalid OST index in '%s': rc = %d\n",
1048                                osp->opd_obd->obd_name, src, -EINVAL);
1049                         RETURN(-EINVAL);
1050                 }
1051                 osp->opd_index = idx;
1052                 osp->opd_group = 0;
1053                 idx = tgt - src;
1054         } else {
1055                 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
1056                 if (strncmp(tgt, "-MDT", 4) != 0 &&
1057                     strncmp(tgt, "-OST", 4) != 0) {
1058                         CERROR("%s: invalid target name %s: rc = %d\n",
1059                                osp->opd_obd->obd_name,
1060                                lustre_cfg_string(cfg, 0), -EINVAL);
1061                         RETURN(-EINVAL);
1062                 }
1063
1064                 idx = simple_strtol(tgt + 4, &mdt, 16);
1065                 if (*mdt != '\0' || idx > INT_MAX || idx < 0) {
1066                         CERROR("%s: invalid OST index in '%s': rc = %d\n",
1067                                osp->opd_obd->obd_name, src, -EINVAL);
1068                         RETURN(-EINVAL);
1069                 }
1070
1071                 /* Get MDT index from the name and set it to opd_group,
1072                  * which will be used by OSP to connect with OST */
1073                 osp->opd_group = idx;
1074                 if (tgt - src <= 12) {
1075                         CERROR("%s: invalid mdt index from %s: rc =%d\n",
1076                                osp->opd_obd->obd_name,
1077                                lustre_cfg_string(cfg, 0), -EINVAL);
1078                         RETURN(-EINVAL);
1079                 }
1080
1081                 if (strncmp(tgt - 12, "-MDT", 4) == 0)
1082                         osp->opd_connect_mdt = 1;
1083
1084                 idx = simple_strtol(tgt - 8, &mdt, 16);
1085                 if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
1086                         CERROR("%s: invalid OST index in '%s': rc =%d\n",
1087                                osp->opd_obd->obd_name, src, -EINVAL);
1088                         RETURN(-EINVAL);
1089                 }
1090
1091                 osp->opd_index = idx;
1092                 idx = tgt - src - 12;
1093         }
1094         /* check the fsname length, and after this everything else will fit */
1095         if (idx > MTI_NAME_MAXLEN) {
1096                 CERROR("%s: fsname too long in '%s': rc = %d\n",
1097                        osp->opd_obd->obd_name, src, -EINVAL);
1098                 RETURN(-EINVAL);
1099         }
1100
1101         OBD_ALLOC(osdname, MAX_OBD_NAME);
1102         if (osdname == NULL)
1103                 RETURN(-ENOMEM);
1104
1105         memcpy(osdname, src, idx); /* copy just the fsname part */
1106         osdname[idx] = '\0';
1107
1108         mdt = strstr(mdt, "-MDT");
1109         if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
1110                 strcat(osdname, "-MDT0000");
1111         else
1112                 strcat(osdname, mdt);
1113         strcat(osdname, "-osd");
1114         CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
1115
1116         if (osp->opd_connect_mdt) {
1117                 struct client_obd *cli = &osp->opd_obd->u.cli;
1118
1119                 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
1120                 if (!cli->cl_rpc_lock)
1121                         GOTO(out_fini, rc = -ENOMEM);
1122                 osp_init_rpc_lock(cli->cl_rpc_lock);
1123         }
1124
1125         osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
1126         osp->opd_dt_dev.dd_ops = &osp_dt_ops;
1127
1128         obd->obd_lu_dev = &osp->opd_dt_dev.dd_lu_dev;
1129
1130         rc = osp_connect_to_osd(env, osp, osdname);
1131         if (rc)
1132                 GOTO(out_fini, rc);
1133
1134         rc = ptlrpcd_addref();
1135         if (rc)
1136                 GOTO(out_disconnect, rc);
1137
1138         rc = client_obd_setup(obd, cfg);
1139         if (rc) {
1140                 CERROR("%s: can't setup obd: rc = %d\n", osp->opd_obd->obd_name,
1141                        rc);
1142                 GOTO(out_ref, rc);
1143         }
1144
1145         osp_lprocfs_init(osp);
1146
1147         rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
1148                           LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
1149         if (rc) {
1150                 CERROR("%s: fid init error: rc = %d\n",
1151                        osp->opd_obd->obd_name, rc);
1152                 GOTO(out_proc, rc);
1153         }
1154
1155         if (!osp->opd_connect_mdt) {
1156                 /* Initialize last id from the storage - will be
1157                  * used in orphan cleanup. */
1158                 rc = osp_last_used_init(env, osp);
1159                 if (rc)
1160                         GOTO(out_fid, rc);
1161
1162
1163                 /* Initialize precreation thread, it handles new
1164                  * connections as well. */
1165                 rc = osp_init_precreate(osp);
1166                 if (rc)
1167                         GOTO(out_last_used, rc);
1168
1169                 /*
1170                  * Initialize synhronization mechanism taking
1171                  * care of propogating changes to OST in near
1172                  * transactional manner.
1173                  */
1174                 rc = osp_sync_init(env, osp);
1175                 if (rc < 0)
1176                         GOTO(out_precreat, rc);
1177         } else {
1178                 rc = osp_update_init(osp);
1179                 if (rc != 0)
1180                         GOTO(out_fid, rc);
1181         }
1182
1183         ns_register_cancel(obd->obd_namespace, osp_cancel_weight);
1184
1185         /*
1186          * Initiate connect to OST
1187          */
1188         ll_generate_random_uuid(uuid);
1189         class_uuid_unparse(uuid, &osp->opd_cluuid);
1190
1191         imp = obd->u.cli.cl_import;
1192
1193         rc = ptlrpc_init_import(imp);
1194         if (rc)
1195                 GOTO(out, rc);
1196         if (osdname)
1197                 OBD_FREE(osdname, MAX_OBD_NAME);
1198         RETURN(0);
1199
1200 out:
1201         if (!osp->opd_connect_mdt)
1202                 /* stop sync thread */
1203                 osp_sync_fini(osp);
1204 out_precreat:
1205         /* stop precreate thread */
1206         if (!osp->opd_connect_mdt)
1207                 osp_precreate_fini(osp);
1208         else
1209                 osp_update_fini(env, osp);
1210 out_last_used:
1211         if (!osp->opd_connect_mdt)
1212                 osp_last_used_fini(env, osp);
1213 out_fid:
1214         obd_fid_fini(osp->opd_obd);
1215 out_proc:
1216         ptlrpc_lprocfs_unregister_obd(obd);
1217         lprocfs_obd_cleanup(obd);
1218         if (osp->opd_symlink)
1219                 lprocfs_remove(&osp->opd_symlink);
1220         client_obd_cleanup(obd);
1221 out_ref:
1222         ptlrpcd_decref();
1223 out_disconnect:
1224         if (osp->opd_connect_mdt) {
1225                 struct client_obd *cli = &osp->opd_obd->u.cli;
1226                 if (cli->cl_rpc_lock != NULL) {
1227                         OBD_FREE_PTR(cli->cl_rpc_lock);
1228                         cli->cl_rpc_lock = NULL;
1229                 }
1230         }
1231         obd_disconnect(osp->opd_storage_exp);
1232 out_fini:
1233         if (osdname)
1234                 OBD_FREE(osdname, MAX_OBD_NAME);
1235         RETURN(rc);
1236 }
1237
1238 /**
1239  * Implementation of lu_device_type_operations::ldto_device_free
1240  *
1241  * Free the OSP device in memory.  No return value is needed for now,
1242  * so always return NULL to comply with the interface.
1243  *
1244  * \param[in] env       execution environment
1245  * \param[in] lu        lu_device of OSP
1246  *
1247  * \retval NULL         NULL unconditionally
1248  */
1249 static struct lu_device *osp_device_free(const struct lu_env *env,
1250                                          struct lu_device *lu)
1251 {
1252         struct osp_device *osp = lu2osp_dev(lu);
1253
1254         if (atomic_read(&lu->ld_ref) && lu->ld_site) {
1255                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1256                 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1257         }
1258         dt_device_fini(&osp->opd_dt_dev);
1259         OBD_FREE_PTR(osp);
1260
1261         return NULL;
1262 }
1263
1264 /**
1265  * Implementation of lu_device_type_operations::ldto_device_alloc
1266  *
1267  * This function allocates and initializes OSP device in memory according to
1268  * the config log.
1269  *
1270  * \param[in] env       execution environment
1271  * \param[in] type      device type of OSP
1272  * \param[in] lcfg      config log
1273  *
1274  * \retval pointer              the pointer of allocated OSP if succeed.
1275  * \retval ERR_PTR(errno)       ERR_PTR(errno) if failed.
1276  */
1277 static struct lu_device *osp_device_alloc(const struct lu_env *env,
1278                                           struct lu_device_type *type,
1279                                           struct lustre_cfg *lcfg)
1280 {
1281         struct osp_device *osp;
1282         struct lu_device  *ld;
1283
1284         OBD_ALLOC_PTR(osp);
1285         if (osp == NULL) {
1286                 ld = ERR_PTR(-ENOMEM);
1287         } else {
1288                 int rc;
1289
1290                 ld = osp2lu_dev(osp);
1291                 dt_device_init(&osp->opd_dt_dev, type);
1292                 rc = osp_init0(env, osp, type, lcfg);
1293                 if (rc != 0) {
1294                         osp_device_free(env, ld);
1295                         ld = ERR_PTR(rc);
1296                 }
1297         }
1298         return ld;
1299 }
1300
1301 /**
1302  * Implementation of lu_device_type_operations::ldto_device_fini
1303  *
1304  * This function cleans up the OSP device, i.e. release and free those
1305  * attached items in osp_device.
1306  *
1307  * \param[in] env       execution environment
1308  * \param[in] ld        lu_device of OSP
1309  *
1310  * \retval NULL                 NULL if cleanup succeeded.
1311  * \retval ERR_PTR(errno)       ERR_PTR(errno) if cleanup failed.
1312  */
1313 static struct lu_device *osp_device_fini(const struct lu_env *env,
1314                                          struct lu_device *ld)
1315 {
1316         struct osp_device *osp = lu2osp_dev(ld);
1317         struct obd_import *imp;
1318         int                rc;
1319
1320         ENTRY;
1321
1322         if (osp->opd_async_requests != NULL) {
1323                 osp_update_request_destroy(osp->opd_async_requests);
1324                 osp->opd_async_requests = NULL;
1325         }
1326
1327         if (osp->opd_storage_exp)
1328                 obd_disconnect(osp->opd_storage_exp);
1329
1330         imp = osp->opd_obd->u.cli.cl_import;
1331
1332         if (osp->opd_symlink)
1333                 lprocfs_remove(&osp->opd_symlink);
1334
1335         LASSERT(osp->opd_obd);
1336         ptlrpc_lprocfs_unregister_obd(osp->opd_obd);
1337         lprocfs_obd_cleanup(osp->opd_obd);
1338
1339         if (osp->opd_connect_mdt) {
1340                 struct client_obd *cli = &osp->opd_obd->u.cli;
1341                 if (cli->cl_rpc_lock != NULL) {
1342                         OBD_FREE_PTR(cli->cl_rpc_lock);
1343                         cli->cl_rpc_lock = NULL;
1344                 }
1345         }
1346
1347         rc = client_obd_cleanup(osp->opd_obd);
1348         if (rc != 0) {
1349                 ptlrpcd_decref();
1350                 RETURN(ERR_PTR(rc));
1351         }
1352
1353         ptlrpcd_decref();
1354
1355         RETURN(NULL);
1356 }
1357
1358 /**
1359  * Implementation of obd_ops::o_reconnect
1360  *
1361  * This function is empty and does not need to do anything for now.
1362  */
1363 static int osp_reconnect(const struct lu_env *env,
1364                          struct obd_export *exp, struct obd_device *obd,
1365                          struct obd_uuid *cluuid,
1366                          struct obd_connect_data *data,
1367                          void *localdata)
1368 {
1369         return 0;
1370 }
1371
1372 /*
1373  * Implementation of obd_ops::o_connect
1374  *
1375  * Connect OSP to the remote target (MDT or OST). Allocate the
1376  * export and return it to the LOD, which calls this function
1377  * for each OSP to connect it to the remote target. This function
1378  * is currently only called once per OSP.
1379  *
1380  * \param[in] env       execution environment
1381  * \param[out] exp      export connected to OSP
1382  * \param[in] obd       OSP device
1383  * \param[in] cluuid    OSP device client uuid
1384  * \param[in] data      connect_data to be used to connect to the remote
1385  *                      target
1386  * \param[in] localdata necessary for the API interface, but not used in
1387  *                      this function
1388  *
1389  * \retval 0            0 if the connection succeeded.
1390  * \retval negative     negative errno if the connection failed.
1391  */
1392 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
1393                            struct obd_device *obd, struct obd_uuid *cluuid,
1394                            struct obd_connect_data *data, void *localdata)
1395 {
1396         struct osp_device       *osp = lu2osp_dev(obd->obd_lu_dev);
1397         struct obd_connect_data *ocd;
1398         struct obd_import       *imp;
1399         struct lustre_handle     conn;
1400         int                      rc;
1401
1402         ENTRY;
1403
1404         CDEBUG(D_CONFIG, "connect #%d\n", osp->opd_connects);
1405
1406         rc = class_connect(&conn, obd, cluuid);
1407         if (rc)
1408                 RETURN(rc);
1409
1410         *exp = class_conn2export(&conn);
1411         /* Why should there ever be more than 1 connect? */
1412         osp->opd_connects++;
1413         LASSERT(osp->opd_connects == 1);
1414
1415         osp->opd_exp = *exp;
1416
1417         imp = osp->opd_obd->u.cli.cl_import;
1418         imp->imp_dlm_handle = conn;
1419
1420         LASSERT(data != NULL);
1421         LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
1422         ocd = &imp->imp_connect_data;
1423         *ocd = *data;
1424
1425         imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1426
1427         ocd->ocd_version = LUSTRE_VERSION_CODE;
1428         ocd->ocd_index = data->ocd_index;
1429         imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
1430
1431         rc = ptlrpc_connect_import(imp);
1432         if (rc) {
1433                 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
1434                 GOTO(out, rc);
1435         } else {
1436                 osp->opd_obd->u.cli.cl_seq->lcs_exp =
1437                                 class_export_get(osp->opd_exp);
1438         }
1439
1440         ptlrpc_pinger_add_import(imp);
1441 out:
1442         RETURN(rc);
1443 }
1444
1445 /**
1446  * Implementation of obd_ops::o_disconnect
1447  *
1448  * Disconnect the export for the OSP.  This is called by LOD to release the
1449  * OSP during cleanup (\see lod_del_device()). The OSP will be released after
1450  * the export is released.
1451  *
1452  * \param[in] exp       export to be disconnected.
1453  *
1454  * \retval 0            0 if disconnection succeed
1455  * \retval negative     negative errno if disconnection failed
1456  */
1457 static int osp_obd_disconnect(struct obd_export *exp)
1458 {
1459         struct obd_device *obd = exp->exp_obd;
1460         struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1461         int                rc;
1462         ENTRY;
1463
1464         /* Only disconnect the underlying layers on the final disconnect. */
1465         LASSERT(osp->opd_connects == 1);
1466         osp->opd_connects--;
1467
1468         rc = class_disconnect(exp);
1469         if (rc) {
1470                 CERROR("%s: class disconnect error: rc = %d\n",
1471                        obd->obd_name, rc);
1472                 RETURN(rc);
1473         }
1474
1475         /* destroy the device */
1476         class_manual_cleanup(obd);
1477
1478         RETURN(rc);
1479 }
1480
1481 /**
1482  * Implementation of obd_ops::o_statfs
1483  *
1484  * Send a RPC to the remote target to get statfs status. This is only used
1485  * in lprocfs helpers by obd_statfs.
1486  *
1487  * \param[in] env       execution environment
1488  * \param[in] exp       connection state from this OSP to the parent (LOD)
1489  *                      device
1490  * \param[out] osfs     hold the statfs result
1491  * \param[in] unused    Not used in this function for now
1492  * \param[in] flags     flags to indicate how OSP will issue the RPC
1493  *
1494  * \retval 0            0 if statfs succeeded.
1495  * \retval negative     negative errno if statfs failed.
1496  */
1497 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
1498                           struct obd_statfs *osfs, __u64 unused, __u32 flags)
1499 {
1500         struct obd_statfs       *msfs;
1501         struct ptlrpc_request   *req;
1502         struct obd_import       *imp = NULL;
1503         int                      rc;
1504
1505         ENTRY;
1506
1507         /* Since the request might also come from lprocfs, so we need
1508          * sync this with client_disconnect_export Bug15684 */
1509         down_read(&exp->exp_obd->u.cli.cl_sem);
1510         if (exp->exp_obd->u.cli.cl_import)
1511                 imp = class_import_get(exp->exp_obd->u.cli.cl_import);
1512         up_read(&exp->exp_obd->u.cli.cl_sem);
1513         if (!imp)
1514                 RETURN(-ENODEV);
1515
1516         req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
1517
1518         class_import_put(imp);
1519
1520         if (req == NULL)
1521                 RETURN(-ENOMEM);
1522
1523         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
1524         if (rc) {
1525                 ptlrpc_request_free(req);
1526                 RETURN(rc);
1527         }
1528         ptlrpc_request_set_replen(req);
1529         req->rq_request_portal = OST_CREATE_PORTAL;
1530         ptlrpc_at_set_req_timeout(req);
1531
1532         if (flags & OBD_STATFS_NODELAY) {
1533                 /* procfs requests not want stat in wait for avoid deadlock */
1534                 req->rq_no_resend = 1;
1535                 req->rq_no_delay = 1;
1536         }
1537
1538         rc = ptlrpc_queue_wait(req);
1539         if (rc)
1540                 GOTO(out, rc);
1541
1542         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1543         if (msfs == NULL)
1544                 GOTO(out, rc = -EPROTO);
1545
1546         *osfs = *msfs;
1547
1548         EXIT;
1549 out:
1550         ptlrpc_req_finished(req);
1551         return rc;
1552 }
1553
1554 /**
1555  * Implementation of obd_ops::o_import_event
1556  *
1557  * This function is called when some related import event happens. It will
1558  * mark the necessary flags according to the event and notify the necessary
1559  * threads (mainly precreate thread).
1560  *
1561  * \param[in] obd       OSP OBD device
1562  * \param[in] imp       import attached from OSP to remote (OST/MDT) service
1563  * \param[in] event     event related to remote service (IMP_EVENT_*)
1564  *
1565  * \retval 0            0 if the event handling succeeded.
1566  * \retval negative     negative errno if the event handling failed.
1567  */
1568 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1569                             enum obd_import_event event)
1570 {
1571         struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1572
1573         switch (event) {
1574         case IMP_EVENT_DISCON:
1575                 d->opd_got_disconnected = 1;
1576                 d->opd_imp_connected = 0;
1577                 if (d->opd_connect_mdt)
1578                         break;
1579
1580                 if (d->opd_pre != NULL) {
1581                         osp_pre_update_status(d, -ENODEV);
1582                         wake_up(&d->opd_pre_waitq);
1583                 }
1584
1585                 CDEBUG(D_HA, "got disconnected\n");
1586                 break;
1587         case IMP_EVENT_INACTIVE:
1588                 d->opd_imp_active = 0;
1589                 d->opd_imp_connected = 0;
1590                 if (d->opd_connect_mdt)
1591                         break;
1592
1593                 if (d->opd_pre != NULL) {
1594                         osp_pre_update_status(d, -ENODEV);
1595                         wake_up(&d->opd_pre_waitq);
1596                 }
1597
1598                 CDEBUG(D_HA, "got inactive\n");
1599                 break;
1600         case IMP_EVENT_ACTIVE:
1601                 d->opd_imp_active = 1;
1602
1603                 if (d->opd_got_disconnected)
1604                         d->opd_new_connection = 1;
1605                 d->opd_imp_connected = 1;
1606                 d->opd_imp_seen_connected = 1;
1607                 if (d->opd_connect_mdt)
1608                         break;
1609
1610                 if (d->opd_pre != NULL)
1611                         wake_up(&d->opd_pre_waitq);
1612
1613                 __osp_sync_check_for_work(d);
1614                 CDEBUG(D_HA, "got connected\n");
1615                 break;
1616         case IMP_EVENT_INVALIDATE:
1617                 if (obd->obd_namespace == NULL)
1618                         break;
1619                 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1620                 break;
1621         case IMP_EVENT_OCD:
1622         case IMP_EVENT_DEACTIVATE:
1623         case IMP_EVENT_ACTIVATE:
1624                 break;
1625         default:
1626                 CERROR("%s: unsupported import event: %#x\n",
1627                        obd->obd_name, event);
1628         }
1629         return 0;
1630 }
1631
1632 /**
1633  * Implementation of obd_ops: o_iocontrol
1634  *
1635  * This function is the ioctl handler for OSP. Note: lctl will access the OSP
1636  * directly by ioctl, instead of through the MDS stack.
1637  *
1638  * param[in] cmd        ioctl command.
1639  * param[in] exp        export of this OSP.
1640  * param[in] len        data length of \a karg.
1641  * param[in] karg       input argument which is packed as
1642  *                      obd_ioctl_data
1643  * param[out] uarg      pointer to userspace buffer (must access by
1644  *                      copy_to_user()).
1645  *
1646  * \retval 0            0 if the ioctl handling succeeded.
1647  * \retval negative     negative errno if the ioctl handling failed.
1648  */
1649 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1650                          void *karg, void *uarg)
1651 {
1652         struct obd_device       *obd = exp->exp_obd;
1653         struct osp_device       *d;
1654         struct obd_ioctl_data   *data = karg;
1655         int                      rc = 0;
1656
1657         ENTRY;
1658
1659         LASSERT(obd->obd_lu_dev);
1660         d = lu2osp_dev(obd->obd_lu_dev);
1661         LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1662
1663         if (!try_module_get(THIS_MODULE)) {
1664                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1665                        module_name(THIS_MODULE));
1666                 return -EINVAL;
1667         }
1668
1669         switch (cmd) {
1670         case OBD_IOC_CLIENT_RECOVER:
1671                 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1672                                            data->ioc_inlbuf1, 0);
1673                 if (rc > 0)
1674                         rc = 0;
1675                 break;
1676         case IOC_OSC_SET_ACTIVE:
1677                 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1678                                               data->ioc_offset);
1679                 break;
1680         case OBD_IOC_PING_TARGET:
1681                 rc = ptlrpc_obd_ping(obd);
1682                 break;
1683         default:
1684                 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1685                        cmd, current_comm());
1686                 rc = -ENOTTY;
1687         }
1688         module_put(THIS_MODULE);
1689         return rc;
1690 }
1691
1692
1693 /**
1694  * Implementation of obd_ops::o_get_info
1695  *
1696  * Retrieve information by key. Retrieval starts from the top layer
1697  * (MDT) of the MDS stack and traverses the stack by calling the
1698  * obd_get_info() method of the next sub-layer.
1699  *
1700  * \param[in] env       execution environment
1701  * \param[in] exp       export of this OSP
1702  * \param[in] keylen    length of \a key
1703  * \param[in] key       the key
1704  * \param[out] vallen   length of \a val
1705  * \param[out] val      holds the value returned by the key
1706  *
1707  * \retval 0            0 if getting information succeeded.
1708  * \retval negative     negative errno if getting information failed.
1709  */
1710 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1711                             __u32 keylen, void *key, __u32 *vallen, void *val)
1712 {
1713         int rc = -EINVAL;
1714
1715         if (KEY_IS(KEY_OSP_CONNECTED)) {
1716                 struct obd_device       *obd = exp->exp_obd;
1717                 struct osp_device       *osp;
1718
1719                 if (!obd->obd_set_up || obd->obd_stopping)
1720                         RETURN(-EAGAIN);
1721
1722                 osp = lu2osp_dev(obd->obd_lu_dev);
1723                 LASSERT(osp);
1724                 /*
1725                  * 1.8/2.0 behaviour is that OST being connected once at least
1726                  * is considered "healthy". and one "healthy" OST is enough to
1727                  * allow lustre clients to connect to MDS
1728                  */
1729                 RETURN(!osp->opd_imp_seen_connected);
1730         }
1731
1732         RETURN(rc);
1733 }
1734
1735 static int osp_obd_set_info_async(const struct lu_env *env,
1736                                   struct obd_export *exp,
1737                                   u32 keylen, void *key,
1738                                   u32 vallen, void *val,
1739                                   struct ptlrpc_request_set *set)
1740 {
1741         struct obd_device       *obd = exp->exp_obd;
1742         struct obd_import       *imp = obd->u.cli.cl_import;
1743         struct osp_device       *osp;
1744         struct ptlrpc_request   *req;
1745         char                    *tmp;
1746         int                      rc;
1747
1748         if (KEY_IS(KEY_SPTLRPC_CONF)) {
1749                 sptlrpc_conf_client_adapt(exp->exp_obd);
1750                 RETURN(0);
1751         }
1752
1753         LASSERT(set != NULL);
1754         if (!obd->obd_set_up || obd->obd_stopping)
1755                 RETURN(-EAGAIN);
1756         osp = lu2osp_dev(obd->obd_lu_dev);
1757
1758         req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
1759         if (req == NULL)
1760                 RETURN(-ENOMEM);
1761
1762         req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1763                              RCL_CLIENT, keylen);
1764         req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1765                              RCL_CLIENT, vallen);
1766         if (osp->opd_connect_mdt)
1767                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1768         else
1769                 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
1770         if (rc) {
1771                 ptlrpc_request_free(req);
1772                 RETURN(rc);
1773         }
1774
1775         tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1776         memcpy(tmp, key, keylen);
1777         tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1778         memcpy(tmp, val, vallen);
1779
1780         ptlrpc_request_set_replen(req);
1781         ptlrpc_set_add_req(set, req);
1782         ptlrpc_check_set(NULL, set);
1783
1784         RETURN(0);
1785 }
1786
1787 /**
1788  * Implementation of obd_ops: o_fid_alloc
1789  *
1790  * Allocate a FID. There are two cases in which OSP performs
1791  * FID allocation.
1792  *
1793  * 1. FID precreation for data objects, which is done in
1794  *    osp_precreate_fids() instead of this function.
1795  * 2. FID allocation for each sub-stripe of a striped directory.
1796  *    Similar to other FID clients, OSP requests the sequence
1797  *    from its corresponding remote MDT, which in turn requests
1798  *    sequences from the sequence controller (MDT0).
1799  *
1800  * \param[in] env       execution environment
1801  * \param[in] exp       export of the OSP
1802  * \param[out] fid      FID being allocated
1803  * \param[in] unused    necessary for the interface but unused.
1804  *
1805  * \retval 0            0 FID allocated successfully.
1806  * \retval 1            1 FID allocated successfully and new sequence
1807  *                      requested from seq meta server
1808  * \retval negative     negative errno if FID allocation failed.
1809  */
1810 static int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1811                          struct lu_fid *fid, struct md_op_data *unused)
1812 {
1813         struct client_obd       *cli = &exp->exp_obd->u.cli;
1814         struct osp_device       *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev);
1815         struct lu_client_seq    *seq = cli->cl_seq;
1816         ENTRY;
1817
1818         LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
1819         /* Sigh, fid client is not ready yet */
1820         LASSERT(osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL);
1821
1822         RETURN(seq_client_alloc_fid(env, seq, fid));
1823 }
1824
1825 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1826 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1827 static void osp_key_exit(const struct lu_context *ctx,
1828                          struct lu_context_key *key, void *data)
1829 {
1830         struct osp_thread_info *info = data;
1831
1832         info->osi_attr.la_valid = 0;
1833 }
1834
1835 struct lu_context_key osp_thread_key = {
1836         .lct_tags = LCT_MD_THREAD,
1837         .lct_init = osp_key_init,
1838         .lct_fini = osp_key_fini,
1839         .lct_exit = osp_key_exit
1840 };
1841
1842 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1843 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1844
1845 struct lu_context_key osp_txn_key = {
1846         .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE,
1847         .lct_init = osp_txn_key_init,
1848         .lct_fini = osp_txn_key_fini
1849 };
1850 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1851
1852 static struct lu_device_type_operations osp_device_type_ops = {
1853         .ldto_init           = osp_type_init,
1854         .ldto_fini           = osp_type_fini,
1855
1856         .ldto_start          = osp_type_start,
1857         .ldto_stop           = osp_type_stop,
1858
1859         .ldto_device_alloc   = osp_device_alloc,
1860         .ldto_device_free    = osp_device_free,
1861
1862         .ldto_device_fini    = osp_device_fini
1863 };
1864
1865 static struct lu_device_type osp_device_type = {
1866         .ldt_tags     = LU_DEVICE_DT,
1867         .ldt_name     = LUSTRE_OSP_NAME,
1868         .ldt_ops      = &osp_device_type_ops,
1869         .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
1870 };
1871
1872 static struct obd_ops osp_obd_device_ops = {
1873         .o_owner        = THIS_MODULE,
1874         .o_add_conn     = client_import_add_conn,
1875         .o_del_conn     = client_import_del_conn,
1876         .o_reconnect    = osp_reconnect,
1877         .o_connect      = osp_obd_connect,
1878         .o_disconnect   = osp_obd_disconnect,
1879         .o_get_info     = osp_obd_get_info,
1880         .o_set_info_async = osp_obd_set_info_async,
1881         .o_import_event = osp_import_event,
1882         .o_iocontrol    = osp_iocontrol,
1883         .o_statfs       = osp_obd_statfs,
1884         .o_fid_init     = client_fid_init,
1885         .o_fid_fini     = client_fid_fini,
1886         .o_fid_alloc    = osp_fid_alloc,
1887 };
1888
1889 struct llog_operations osp_mds_ost_orig_logops;
1890
1891 /**
1892  * Initialize OSP module.
1893  *
1894  * Register device types OSP and Light Weight Proxy (LWP) (\see lwp_dev.c)
1895  * in obd_types (\see class_obd.c).  Initialize procfs for the
1896  * the OSP device.  Note: OSP was called OSC before Lustre 2.4,
1897  * so for compatibility it still uses the name "osc" in procfs.
1898  * This is called at module load time.
1899  *
1900  * \retval 0            0 if initialization succeeds.
1901  * \retval negative     negative errno if initialization failed.
1902  */
1903 static int __init osp_mod_init(void)
1904 {
1905         struct obd_type *type;
1906         int rc;
1907
1908         rc = lu_kmem_init(osp_caches);
1909         if (rc)
1910                 return rc;
1911
1912
1913         rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL,
1914                                  LUSTRE_OSP_NAME, &osp_device_type);
1915         if (rc != 0) {
1916                 lu_kmem_fini(osp_caches);
1917                 return rc;
1918         }
1919
1920         rc = class_register_type(&lwp_obd_device_ops, NULL, true, NULL,
1921                                  LUSTRE_LWP_NAME, &lwp_device_type);
1922         if (rc != 0) {
1923                 class_unregister_type(LUSTRE_OSP_NAME);
1924                 lu_kmem_fini(osp_caches);
1925                 return rc;
1926         }
1927
1928         /* Note: add_rec/delcare_add_rec will be only used by catalogs */
1929         osp_mds_ost_orig_logops = llog_osd_ops;
1930         osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
1931         osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1932
1933         /* create "osc" entry in procfs for compatibility purposes */
1934         type = class_search_type(LUSTRE_OSC_NAME);
1935         if (type != NULL && type->typ_procroot != NULL)
1936                 return rc;
1937
1938         type = class_search_type(LUSTRE_OSP_NAME);
1939         type->typ_procsym = lprocfs_register("osc", proc_lustre_root,
1940                                              NULL, NULL);
1941         if (IS_ERR(type->typ_procsym)) {
1942                 CERROR("osp: can't create compat entry \"osc\": %d\n",
1943                        (int) PTR_ERR(type->typ_procsym));
1944                 type->typ_procsym = NULL;
1945         }
1946         return rc;
1947 }
1948
1949 /**
1950  * Finalize OSP module.
1951  *
1952  * This callback is called when kernel unloads OSP module from memory, and
1953  * it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
1954  */
1955 static void __exit osp_mod_exit(void)
1956 {
1957         class_unregister_type(LUSTRE_LWP_NAME);
1958         class_unregister_type(LUSTRE_OSP_NAME);
1959         lu_kmem_fini(osp_caches);
1960 }
1961
1962 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1963 MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
1964 MODULE_VERSION(LUSTRE_VERSION_STRING);
1965 MODULE_LICENSE("GPL");
1966
1967 module_init(osp_mod_init);
1968 module_exit(osp_mod_exit);