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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/osp/osp_dev.c
33 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
34 * Author: Mikhail Pershin <mike.pershin@intel.com>
35 * Author: Di Wang <di.wang@intel.com>
38 * The Object Storage Proxy (OSP) module provides an implementation of
39 * the DT API for remote MDTs and OSTs. Every local OSP device (or
40 * object) is a proxy for a remote OSD device (or object). Thus OSP
41 * converts DT operations into RPCs, which are sent to the OUT service
42 * on a remote target, converted back to DT operations, and
43 * executed. Of course there are many ways in which this description
44 * is inaccurate but it's a good enough mental model. OSP is used by
45 * the MDT stack in several ways:
47 * - OSP devices allocate FIDs for the stripe sub-objects of a striped
50 * - OSP objects represent the remote MDT and OST objects that are
51 * the stripes of a striped object.
53 * - OSP devices log, send, and track synchronous operations (setattr
54 * and unlink) to remote targets.
56 * - OSP objects are the bottom slice of the compound LU object
57 * representing a remote MDT object: MDT/MDD/LOD/OSP.
59 * - OSP objects are used by LFSCK to represent remote OST objects
60 * during the verification of MDT-OST consistency.
62 * - OSP devices batch idempotent requests (declare_attr_get() and
63 * declare_xattr_get()) to the remote target and cache their results.
65 * In addition the OSP layer implements a subset of the OBD device API
66 * to support being a client of a remote target, connecting to other
67 * layers, and FID allocation.
70 #define DEBUG_SUBSYSTEM S_MDS
72 #include <linux/kthread.h>
74 #include <uapi/linux/lustre/lustre_ioctl.h>
75 #include <lustre_log.h>
76 #include <lustre_obdo.h>
77 #include <uapi/linux/lustre/lustre_param.h>
78 #include <obd_class.h>
80 #include "osp_internal.h"
82 /* Slab for OSP object allocation */
83 struct kmem_cache *osp_object_kmem;
85 static struct lu_kmem_descr osp_caches[] = {
87 .ckd_cache = &osp_object_kmem,
88 .ckd_name = "osp_obj",
89 .ckd_size = sizeof(struct osp_object)
97 * Implementation of lu_device_operations::ldo_object_alloc
99 * Allocates an OSP object in memory, whose FID is on the remote target.
101 * \param[in] env execution environment
102 * \param[in] hdr The header of the object stack. If it is NULL, it
103 * means the object is not built from top device, i.e.
104 * it is a sub-stripe object of striped directory or
106 * \param[in] d OSP device
108 * \retval object object being created if the creation succeed.
109 * \retval NULL NULL if the creation failed.
111 static struct lu_object *osp_object_alloc(const struct lu_env *env,
112 const struct lu_object_header *hdr,
115 struct osp_object *o;
117 OBD_SLAB_ALLOC_PTR_GFP(o, osp_object_kmem, GFP_NOFS);
119 struct lu_object *l = &o->opo_obj.do_lu;
121 /* If hdr is NULL, it means the object is not built
122 * from the top dev(MDT/OST), usually it happens when
123 * building striped object, like data object on MDT or
124 * striped object for directory */
126 struct lu_object_header *h = &o->opo_header;
128 lu_object_header_init(h);
129 dt_object_init(&o->opo_obj, h, d);
130 lu_object_add_top(h, l);
132 dt_object_init(&o->opo_obj, NULL, d);
135 l->lo_ops = &osp_lu_obj_ops;
137 init_rwsem(&o->opo_sem);
138 INIT_LIST_HEAD(&o->opo_xattr_list);
139 INIT_LIST_HEAD(&o->opo_invalidate_cb_list);
140 spin_lock_init(&o->opo_lock);
141 init_rwsem(&o->opo_invalidate_sem);
150 * Find or create the local object
152 * Finds or creates the local file referenced by \a reg_id and return the
153 * attributes of the local file.
155 * \param[in] env execution environment
156 * \param[in] osp OSP device
157 * \param[out] attr attributes of the object
158 * \param[in] reg_id the local object ID of the file. It will be used
159 * to compose a local FID{FID_SEQ_LOCAL_FILE, reg_id, 0}
160 * to identify the object.
162 * \retval object object(dt_object) found or created
163 * \retval ERR_PTR(errno) ERR_PTR(errno) if not get the object.
165 static struct dt_object
166 *osp_find_or_create_local_file(const struct lu_env *env, struct osp_device *osp,
167 struct lu_attr *attr, __u32 reg_id)
169 struct osp_thread_info *osi = osp_env_info(env);
170 struct dt_object_format dof = { 0 };
171 struct dt_object *dto;
175 lu_local_obj_fid(&osi->osi_fid, reg_id);
176 attr->la_valid = LA_MODE;
177 attr->la_mode = S_IFREG | 0644;
178 dof.dof_type = DFT_REGULAR;
179 /* Find or create the local object by osi_fid. */
180 dto = dt_find_or_create(env, osp->opd_storage, &osi->osi_fid,
185 /* Get attributes of the local object. */
186 rc = dt_attr_get(env, dto, attr);
188 CERROR("%s: can't be initialized: rc = %d\n",
189 osp->opd_obd->obd_name, rc);
190 dt_object_put(env, dto);
197 * Write data buffer to a local file object.
199 * \param[in] env execution environment
200 * \param[in] osp OSP device
201 * \param[in] dt_obj object written to
202 * \param[in] buf buffer containing byte array and length
203 * \param[in] offset write offset in the object in bytes
205 * \retval 0 0 if write succeed
206 * \retval -EFAULT -EFAULT if only part of buffer is written.
207 * \retval negative other negative errno if write failed.
209 static int osp_write_local_file(const struct lu_env *env,
210 struct osp_device *osp,
211 struct dt_object *dt_obj,
218 if (osp->opd_storage->dd_rdonly)
221 th = dt_trans_create(env, osp->opd_storage);
225 rc = dt_declare_record_write(env, dt_obj, buf, offset, th);
228 rc = dt_trans_start_local(env, osp->opd_storage, th);
232 rc = dt_record_write(env, dt_obj, buf, &offset, th);
234 dt_trans_stop(env, osp->opd_storage, th);
239 * Initialize last ID object.
241 * This function initializes the LAST_ID file, which stores the current last
242 * used id of data objects. The MDT will use the last used id and the last_seq
243 * (\see osp_init_last_seq()) to synchronize the precreate object cache with
246 * \param[in] env execution environment
247 * \param[in] osp OSP device
249 * \retval 0 0 if initialization succeed
250 * \retval negative negative errno if initialization failed
252 static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp)
254 struct osp_thread_info *osi = osp_env_info(env);
255 struct lu_fid *fid = &osp->opd_last_used_fid;
256 struct dt_object *dto;
260 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
263 RETURN(PTR_ERR(dto));
265 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &osp->opd_last_id,
268 /* object will be released in device cleanup path */
269 if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) {
270 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
271 if (rc != 0 && rc != -EFAULT)
273 /* In case of idif bits 32-48 go to f_seq
274 * (see osp_init_last_seq). So don't care
275 * about u64->u32 convertion. */
276 fid->f_oid = osp->opd_last_id;
279 if (rc == -EFAULT) { /* fresh LAST_ID */
280 osp->opd_last_id = 0;
282 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
287 osp->opd_last_used_oid_file = dto;
290 /* object will be released in device cleanup path */
291 CERROR("%s: can't initialize lov_objid: rc = %d\n",
292 osp->opd_obd->obd_name, rc);
293 dt_object_put(env, dto);
294 osp->opd_last_used_oid_file = NULL;
299 * Initialize last sequence object.
301 * This function initializes the LAST_SEQ file in the local OSD, which stores
302 * the current last used sequence of data objects. The MDT will use the last
303 * sequence and last id (\see osp_init_last_objid()) to synchronize the
304 * precreate object cache with OSTs.
306 * \param[in] env execution environment
307 * \param[in] osp OSP device
309 * \retval 0 0 if initialization succeed
310 * \retval negative negative errno if initialization failed
312 static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp)
314 struct osp_thread_info *osi = osp_env_info(env);
315 struct lu_fid *fid = &osp->opd_last_used_fid;
316 struct dt_object *dto;
320 dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
323 RETURN(PTR_ERR(dto));
325 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
328 /* object will be released in device cleanup path */
329 if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) {
330 rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
331 if (rc != 0 && rc != -EFAULT)
333 if (fid_is_idif(fid))
334 fid->f_seq = fid_idif_seq(osp->opd_last_id,
338 if (rc == -EFAULT) { /* fresh OSP */
340 rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
345 osp->opd_last_used_seq_file = dto;
348 /* object will be released in device cleanup path */
349 CERROR("%s: can't initialize lov_seq: rc = %d\n",
350 osp->opd_obd->obd_name, rc);
351 dt_object_put(env, dto);
352 osp->opd_last_used_seq_file = NULL;
357 * Initialize last OID and sequence object.
359 * If the MDT is just upgraded to 2.4 from the lower version, where the
360 * LAST_SEQ file does not exist, the file will be created and IDIF sequence
361 * will be written into the file.
363 * \param[in] env execution environment
364 * \param[in] osp OSP device
366 * \retval 0 0 if initialization succeed
367 * \retval negative negative error if initialization failed
369 static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp)
371 struct osp_thread_info *osi = osp_env_info(env);
375 fid_zero(&osp->opd_last_used_fid);
376 rc = osp_init_last_objid(env, osp);
378 CERROR("%s: Can not get ids %d from old objid!\n",
379 osp->opd_obd->obd_name, rc);
383 rc = osp_init_last_seq(env, osp);
385 CERROR("%s: Can not get sequence %d from old objseq!\n",
386 osp->opd_obd->obd_name, rc);
390 if (fid_oid(&osp->opd_last_used_fid) != 0 &&
391 fid_seq(&osp->opd_last_used_fid) == 0) {
392 /* Just upgrade from the old version,
393 * set the seq to be IDIF */
394 osp->opd_last_used_fid.f_seq =
395 fid_idif_seq(fid_oid(&osp->opd_last_used_fid),
397 osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off,
398 &osp->opd_last_used_fid.f_seq,
400 rc = osp_write_local_file(env, osp, osp->opd_last_used_seq_file,
401 &osi->osi_lb, osi->osi_off);
403 CERROR("%s : Can not write seq file: rc = %d\n",
404 osp->opd_obd->obd_name, rc);
409 if (!fid_is_zero(&osp->opd_last_used_fid) &&
410 !fid_is_sane(&osp->opd_last_used_fid)) {
411 CERROR("%s: Got invalid FID "DFID"\n", osp->opd_obd->obd_name,
412 PFID(&osp->opd_last_used_fid));
413 GOTO(out, rc = -EINVAL);
416 osp_fid_to_obdid(&osp->opd_last_used_fid, &osp->opd_last_id);
417 CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
418 osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
421 if (osp->opd_last_used_oid_file != NULL) {
422 dt_object_put(env, osp->opd_last_used_oid_file);
423 osp->opd_last_used_oid_file = NULL;
425 if (osp->opd_last_used_seq_file != NULL) {
426 dt_object_put(env, osp->opd_last_used_seq_file);
427 osp->opd_last_used_seq_file = NULL;
435 * Release the last sequence and OID file objects in OSP device.
437 * \param[in] env execution environment
438 * \param[in] osp OSP device
440 static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp)
442 /* release last_used file */
443 if (osp->opd_last_used_oid_file != NULL) {
444 dt_object_put(env, osp->opd_last_used_oid_file);
445 osp->opd_last_used_oid_file = NULL;
448 if (osp->opd_last_used_seq_file != NULL) {
449 dt_object_put(env, osp->opd_last_used_seq_file);
450 osp->opd_last_used_seq_file = NULL;
455 * Disconnects the connection between OSP and its correspondent MDT or OST, and
456 * the import will be marked as inactive. It will only be called during OSP
459 * \param[in] d OSP device being disconnected
461 * \retval 0 0 if disconnection succeed
462 * \retval negative negative errno if disconnection failed
464 static int osp_disconnect(struct osp_device *d)
466 struct obd_device *obd = d->opd_obd;
467 struct obd_import *imp;
470 imp = obd->u.cli.cl_import;
472 /* Mark import deactivated now, so we don't try to reconnect if any
473 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
474 * fully deactivate the import, or that would drop all requests. */
475 LASSERT(imp != NULL);
476 spin_lock(&imp->imp_lock);
477 imp->imp_deactive = 1;
478 spin_unlock(&imp->imp_lock);
480 ptlrpc_deactivate_import(imp);
482 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
483 * delete it regardless. (It's safe to delete an import that was
485 (void)ptlrpc_pinger_del_import(imp);
487 /* Send disconnect on healthy import, do force disconnect otherwise */
488 spin_lock(&imp->imp_lock);
489 imp->imp_obd->obd_force = imp->imp_state != LUSTRE_IMP_FULL;
490 spin_unlock(&imp->imp_lock);
492 rc = ptlrpc_disconnect_import(imp, 0);
494 CERROR("%s: can't disconnect: rc = %d\n", obd->obd_name, rc);
496 ptlrpc_invalidate_import(imp);
502 * Initialize the osp_update structure in OSP device
504 * Allocate osp update structure and start update thread.
506 * \param[in] osp OSP device
508 * \retval 0 if initialization succeeds.
509 * \retval negative errno if initialization fails.
511 static int osp_update_init(struct osp_device *osp)
513 struct task_struct *task;
518 LASSERT(osp->opd_connect_mdt);
520 if (osp->opd_storage->dd_rdonly)
523 OBD_ALLOC_PTR(osp->opd_update);
524 if (osp->opd_update == NULL)
527 init_waitqueue_head(&osp->opd_update->ou_waitq);
528 spin_lock_init(&osp->opd_update->ou_lock);
529 INIT_LIST_HEAD(&osp->opd_update->ou_list);
530 osp->opd_update->ou_rpc_version = 1;
531 osp->opd_update->ou_version = 1;
532 osp->opd_update->ou_generation = 0;
534 rc = lu_env_init(&osp->opd_update->ou_env,
535 osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
537 CERROR("%s: init env error: rc = %d\n", osp->opd_obd->obd_name,
539 OBD_FREE_PTR(osp->opd_update);
540 osp->opd_update = NULL;
543 /* start thread handling sending updates to the remote MDT */
544 task = kthread_create(osp_send_update_thread, osp,
545 "osp_up%u-%u", osp->opd_index, osp->opd_group);
547 int rc = PTR_ERR(task);
549 lu_env_fini(&osp->opd_update->ou_env);
550 OBD_FREE_PTR(osp->opd_update);
551 osp->opd_update = NULL;
552 CERROR("%s: can't start precreate thread: rc = %d\n",
553 osp->opd_obd->obd_name, rc);
557 osp->opd_update->ou_update_task = task;
558 wake_up_process(task);
564 * Finialize osp_update structure in OSP device
566 * Stop the OSP update sending thread, then delete the left
567 * osp thandle in the sending list.
569 * \param [in] osp OSP device.
571 static void osp_update_fini(const struct lu_env *env, struct osp_device *osp)
573 struct osp_update_request *our;
574 struct osp_update_request *tmp;
575 struct osp_updates *ou = osp->opd_update;
580 kthread_stop(ou->ou_update_task);
581 lu_env_fini(&ou->ou_env);
583 /* Remove the left osp thandle from the list */
584 spin_lock(&ou->ou_lock);
585 list_for_each_entry_safe(our, tmp, &ou->ou_list,
587 list_del_init(&our->our_list);
588 LASSERT(our->our_th != NULL);
589 osp_trans_callback(env, our->our_th, -EIO);
590 /* our will be destroyed in osp_thandle_put() */
591 osp_thandle_put(env, our->our_th);
593 spin_unlock(&ou->ou_lock);
596 osp->opd_update = NULL;
600 * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
601 * precreate threads etc.
603 * \param[in] env execution environment.
604 * \param[in] d OSP device being disconnected.
606 * \retval 0 0 if cleanup succeed
607 * \retval negative negative errno if cleanup failed
609 static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
616 rc = osp_disconnect(d);
620 if (!d->opd_connect_mdt) {
621 /* stop sync thread */
624 /* stop precreate thread */
625 osp_precreate_fini(d);
627 /* release last_used file */
628 osp_last_used_fini(env, d);
631 obd_fid_fini(d->opd_obd);
637 * Implementation of osp_lu_ops::ldo_process_config
639 * This function processes config log records in OSP layer. It is usually
640 * called from the top layer of MDT stack, and goes through the stack by calling
641 * ldo_process_config of next layer.
643 * \param[in] env execution environment
644 * \param[in] dev lu_device of OSP
645 * \param[in] lcfg config log
647 * \retval 0 0 if the config log record is executed correctly.
648 * \retval negative negative errno if the record execution fails.
650 static int osp_process_config(const struct lu_env *env,
651 struct lu_device *dev, struct lustre_cfg *lcfg)
653 struct osp_device *d = lu2osp_dev(dev);
654 struct dt_device *dt = lu2dt_dev(dev);
655 struct obd_device *obd = d->opd_obd;
661 switch (lcfg->lcfg_command) {
662 case LCFG_PRE_CLEANUP:
663 rc = osp_disconnect(d);
664 osp_update_fini(env, d);
665 if (obd->obd_namespace != NULL)
666 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
669 lu_dev_del_linkage(dev->ld_site, dev);
670 rc = osp_shutdown(env, d);
673 count = class_modify_config(lcfg, d->opd_connect_mdt ?
674 PARAM_OSP : PARAM_OSC,
677 /* class_modify_config() haven't found matching
678 * parameter and returned an error so that layer(s)
679 * below could use that. But OSP is the bottom, so
682 CERROR("%s: unknown param %s\n",
683 (char *)lustre_cfg_string(lcfg, 0),
684 (char *)lustre_cfg_string(lcfg, 1));
689 CERROR("%s: unknown command %u\n",
690 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
699 * Implementation of osp_lu_ops::ldo_recovery_complete
701 * This function is called after recovery is finished, and OSP layer
702 * will wake up precreate thread here.
704 * \param[in] env execution environment
705 * \param[in] dev lu_device of OSP
707 * \retval 0 0 unconditionally
709 static int osp_recovery_complete(const struct lu_env *env,
710 struct lu_device *dev)
712 struct osp_device *osp = lu2osp_dev(dev);
715 osp->opd_recovery_completed = 1;
717 if (!osp->opd_connect_mdt && osp->opd_pre != NULL)
718 wake_up(&osp->opd_pre_waitq);
724 * Implementation of lu_device_operations::ldo_fid_alloc() for OSP
726 * Allocate FID from remote MDT.
728 * see include/lu_object.h for the details.
730 static int osp_fid_alloc(const struct lu_env *env, struct lu_device *d,
731 struct lu_fid *fid, struct lu_object *parent,
732 const struct lu_name *name)
734 struct osp_device *osp = lu2osp_dev(d);
735 struct client_obd *cli = &osp->opd_obd->u.cli;
736 struct lu_client_seq *seq = cli->cl_seq;
741 /* Sigh, fid client is not ready yet */
742 if (!osp->opd_obd->u.cli.cl_seq)
745 if (!osp->opd_obd->u.cli.cl_seq->lcs_exp)
748 rc = seq_client_alloc_fid(env, seq, fid);
753 const struct lu_device_operations osp_lu_ops = {
754 .ldo_object_alloc = osp_object_alloc,
755 .ldo_process_config = osp_process_config,
756 .ldo_recovery_complete = osp_recovery_complete,
757 .ldo_fid_alloc = osp_fid_alloc,
761 * Implementation of dt_device_operations::dt_statfs
763 * This function provides statfs status (for precreation) from
764 * corresponding OST. Note: this function only retrieves the status
765 * from the OSP device, and the real statfs RPC happens inside
766 * precreate thread (\see osp_statfs_update). Note: OSP for MDT does
767 * not need to retrieve statfs data for now.
769 * \param[in] env execution environment.
770 * \param[in] dev dt_device of OSP.
771 * \param[out] sfs holds the retrieved statfs data.
773 * \retval 0 0 statfs data was retrieved successfully or
774 * retrieval was not needed
775 * \retval negative negative errno if get statfs failed.
777 static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
778 struct obd_statfs *sfs, struct obd_statfs_info *info)
780 struct osp_device *d = dt2osp_dev(dev);
781 struct obd_import *imp = d->opd_obd->u.cli.cl_import;
785 if (imp->imp_state == LUSTRE_IMP_CLOSED)
788 if (unlikely(d->opd_imp_active == 0))
791 /* return recently updated data */
792 *sfs = d->opd_statfs;
794 info->os_reserved_mb_low = d->opd_reserved_mb_low;
795 info->os_reserved_mb_high = d->opd_reserved_mb_high;
798 if (d->opd_pre == NULL)
801 CDEBUG(D_OTHER, "%s: %llu blocks, %llu free, %llu avail, "
802 "%u bsize, %u reserved mb low, %u reserved mb high, "
803 "%llu files, %llu free files\n", d->opd_obd->obd_name,
804 sfs->os_blocks, sfs->os_bfree, sfs->os_bavail, sfs->os_bsize,
805 d->opd_reserved_mb_low, d->opd_reserved_mb_high,
806 sfs->os_files, sfs->os_ffree);
809 if (info && !info->os_enable_pre)
813 * The layer above osp (usually lod) can use f_precreated to
814 * estimate how many objects are available for immediate usage.
816 spin_lock(&d->opd_pre_lock);
817 sfs->os_fprecreated = osp_fid_diff(&d->opd_pre_last_created_fid,
818 &d->opd_pre_used_fid);
819 sfs->os_fprecreated -= d->opd_pre_reserved;
820 LASSERTF(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2,
821 "last_created "DFID", next_fid "DFID", reserved %llu\n",
822 PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_pre_used_fid),
823 d->opd_pre_reserved);
824 spin_unlock(&d->opd_pre_lock);
829 * Implementation of dt_device_operations::dt_sync
831 * This function synchronizes the OSP cache to the remote target. It wakes
832 * up unlink log threads and sends out unlink records to the remote OST.
834 * \param[in] env execution environment
835 * \param[in] dev dt_device of OSP
837 * \retval 0 0 if synchronization succeeds
838 * \retval negative negative errno if synchronization fails
840 static int osp_sync(const struct lu_env *env, struct dt_device *dev)
842 struct osp_device *d = dt2osp_dev(dev);
843 time64_t start = ktime_get_seconds();
849 /* No Sync between MDTs yet. */
850 if (d->opd_connect_mdt)
853 recs = atomic_read(&d->opd_sync_changes);
854 old = atomic64_read(&d->opd_sync_processed_recs);
856 osp_sync_force(env, dt2osp_dev(dev));
858 if (unlikely(d->opd_imp_active == 0))
861 down_write(&d->opd_async_updates_rwsem);
863 CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
864 atomic_read(&d->opd_async_updates_count));
866 /* make sure the connection is fine */
867 rc = wait_event_idle_timeout(
868 d->opd_sync_barrier_waitq,
869 atomic_read(&d->opd_async_updates_count) == 0,
870 cfs_time_seconds(obd_timeout));
876 up_write(&d->opd_async_updates_rwsem);
880 CDEBUG(D_CACHE, "%s: processed %llu\n", d->opd_obd->obd_name,
881 (unsigned long long)atomic64_read(&d->opd_sync_processed_recs));
883 while (atomic64_read(&d->opd_sync_processed_recs) < old + recs) {
884 __u64 last = atomic64_read(&d->opd_sync_processed_recs);
885 /* make sure the connection is fine */
886 wait_event_idle_timeout(
887 d->opd_sync_barrier_waitq,
888 atomic64_read(&d->opd_sync_processed_recs)
890 cfs_time_seconds(obd_timeout));
892 if (atomic64_read(&d->opd_sync_processed_recs) >= old + recs)
895 if (atomic64_read(&d->opd_sync_processed_recs) != last) {
896 /* some progress have been made,
901 /* no changes and expired, something is wrong */
902 GOTO(out, rc = -ETIMEDOUT);
905 /* block new processing (barrier>0 - few callers are possible */
906 atomic_inc(&d->opd_sync_barrier);
908 CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
909 atomic_read(&d->opd_sync_rpcs_in_flight));
911 /* wait till all-in-flight are replied, so executed by the target */
912 /* XXX: this is used by LFSCK at the moment, which doesn't require
913 * all the changes to be committed, but in general it'd be
914 * better to wait till commit */
915 while (atomic_read(&d->opd_sync_rpcs_in_flight) > 0) {
916 old = atomic_read(&d->opd_sync_rpcs_in_flight);
918 wait_event_idle_timeout(
919 d->opd_sync_barrier_waitq,
920 atomic_read(&d->opd_sync_rpcs_in_flight) == 0,
921 cfs_time_seconds(obd_timeout));
923 if (atomic_read(&d->opd_sync_rpcs_in_flight) == 0)
926 if (atomic_read(&d->opd_sync_rpcs_in_flight) != old) {
927 /* some progress have been made */
931 /* no changes and expired, something is wrong */
932 GOTO(out, rc = -ETIMEDOUT);
936 /* resume normal processing (barrier=0) */
937 atomic_dec(&d->opd_sync_barrier);
938 osp_sync_check_for_work(d);
940 CDEBUG(D_CACHE, "%s: done in %lld: rc = %d\n", d->opd_obd->obd_name,
941 ktime_get_seconds() - start, rc);
946 static const struct dt_device_operations osp_dt_ops = {
947 .dt_statfs = osp_statfs,
949 .dt_trans_create = osp_trans_create,
950 .dt_trans_start = osp_trans_start,
951 .dt_trans_stop = osp_trans_stop,
952 .dt_trans_cb_add = osp_trans_cb_add,
956 * Connect OSP to local OSD.
958 * Locate the local OSD referenced by \a nextdev and connect to it. Sometimes,
959 * OSP needs to access the local OSD to store some information. For example,
960 * during precreate, it needs to update last used OID and sequence file
961 * (LAST_SEQ) in local OSD.
963 * \param[in] env execution environment
964 * \param[in] osp OSP device
965 * \param[in] nextdev the name of local OSD
967 * \retval 0 0 connection succeeded
968 * \retval negative negative errno connection failed
970 static int osp_connect_to_osd(const struct lu_env *env, struct osp_device *osp,
973 struct obd_connect_data *data = NULL;
974 struct obd_device *obd;
979 LASSERT(osp->opd_storage_exp == NULL);
985 obd = class_name2obd(nextdev);
987 CERROR("%s: can't locate next device: %s\n",
988 osp->opd_obd->obd_name, nextdev);
989 GOTO(out, rc = -ENOTCONN);
992 rc = obd_connect(env, &osp->opd_storage_exp, obd, &obd->obd_uuid, data,
995 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
996 osp->opd_obd->obd_name, nextdev, rc);
1000 osp->opd_dt_dev.dd_lu_dev.ld_site =
1001 osp->opd_storage_exp->exp_obd->obd_lu_dev->ld_site;
1002 LASSERT(osp->opd_dt_dev.dd_lu_dev.ld_site);
1003 osp->opd_storage = lu2dt_dev(osp->opd_storage_exp->exp_obd->obd_lu_dev);
1011 * Determine if the lock needs to be cancelled
1013 * Determine if the unused lock should be cancelled before replay, see
1014 * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists
1017 * \param[in] lock lock to be checked.
1019 * \retval 1 if the lock needs to be cancelled before replay.
1020 * \retval 0 if the lock does not need to be cancelled before
1023 static int osp_cancel_weight(struct ldlm_lock *lock)
1025 if (lock->l_resource->lr_type != LDLM_IBITS)
1032 * Initialize OSP device according to the parameters in the configuration
1035 * Reconstruct the local device name from the configuration profile, and
1036 * initialize necessary threads and structures according to the OSP type
1039 * Since there is no record in the MDT configuration for the local disk
1040 * device, we have to extract this from elsewhere in the profile.
1041 * The only information we get at setup is from the OSC records:
1042 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
1044 * Note: configs generated by Lustre 1.8 are missing the -MDTxxxx part,
1045 * so, we need to reconstruct the name of the underlying OSD from this:
1046 * {fsname}-{svname}-osd, for example "lustre-MDT0000-osd".
1048 * \param[in] env execution environment
1049 * \param[in] osp OSP device
1050 * \param[in] ldt lu device type of OSP
1051 * \param[in] cfg configuration log
1053 * \retval 0 0 if OSP initialization succeeded.
1054 * \retval negative negative errno if OSP initialization failed.
1056 static int osp_init0(const struct lu_env *env, struct osp_device *osp,
1057 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1059 struct obd_device *obd;
1060 struct obd_import *imp;
1061 char *src, *tgt, *osdname = NULL;
1068 mutex_init(&osp->opd_async_requests_mutex);
1069 INIT_LIST_HEAD(&osp->opd_async_updates);
1070 init_rwsem(&osp->opd_async_updates_rwsem);
1071 atomic_set(&osp->opd_async_updates_count, 0);
1073 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1075 CERROR("Cannot find obd with name %s\n",
1076 lustre_cfg_string(cfg, 0));
1081 src = lustre_cfg_string(cfg, 0);
1085 tgt = strrchr(src, '-');
1087 CERROR("%s: invalid target name %s: rc = %d\n",
1088 osp->opd_obd->obd_name, lustre_cfg_string(cfg, 0),
1093 if (strncmp(tgt, "-osc", 4) == 0) {
1094 /* Old OSC name fsname-OSTXXXX-osc */
1095 for (tgt--; tgt > src && *tgt != '-'; tgt--)
1098 CERROR("%s: invalid target name %s: rc = %d\n",
1099 osp->opd_obd->obd_name,
1100 lustre_cfg_string(cfg, 0), -EINVAL);
1104 if (strncmp(tgt, "-OST", 4) != 0) {
1105 CERROR("%s: invalid target name %s: rc = %d\n",
1106 osp->opd_obd->obd_name,
1107 lustre_cfg_string(cfg, 0), -EINVAL);
1111 rc = target_name2index(tgt + 1, &idx, &mdt);
1112 if (rc < 0 || rc & LDD_F_SV_ALL || mdt[0] != '-') {
1113 CERROR("%s: invalid OST index in '%s': rc = %d\n",
1114 osp->opd_obd->obd_name, src, -EINVAL);
1117 osp->opd_index = idx;
1121 /* New OSC name fsname-OSTXXXX-osc-MDTXXXX */
1122 if (strncmp(tgt, "-MDT", 4) != 0 &&
1123 strncmp(tgt, "-OST", 4) != 0) {
1124 CERROR("%s: invalid target name %s: rc = %d\n",
1125 osp->opd_obd->obd_name,
1126 lustre_cfg_string(cfg, 0), -EINVAL);
1130 rc = target_name2index(tgt + 1, &idx, &mdt);
1131 if (rc < 0 || rc & LDD_F_SV_ALL || *mdt != '\0') {
1132 CERROR("%s: invalid OST index in '%s': rc = %d\n",
1133 osp->opd_obd->obd_name, src, -EINVAL);
1137 /* Get MDT index from the name and set it to opd_group,
1138 * which will be used by OSP to connect with OST */
1139 osp->opd_group = idx;
1140 if (tgt - src <= 12) {
1141 CERROR("%s: invalid mdt index from %s: rc =%d\n",
1142 osp->opd_obd->obd_name,
1143 lustre_cfg_string(cfg, 0), -EINVAL);
1147 if (strncmp(tgt - 12, "-MDT", 4) == 0)
1148 osp->opd_connect_mdt = 1;
1150 rc = target_name2index(tgt - 11, &idx, &mdt);
1151 if (rc < 0 || rc & LDD_F_SV_ALL || mdt[0] != '-') {
1152 CERROR("%s: invalid OST index in '%s': rc =%d\n",
1153 osp->opd_obd->obd_name, src, -EINVAL);
1157 osp->opd_index = idx;
1158 idx = tgt - src - 12;
1160 /* check the fsname length, and after this everything else will fit */
1161 if (idx > MTI_NAME_MAXLEN) {
1162 CERROR("%s: fsname too long in '%s': rc = %d\n",
1163 osp->opd_obd->obd_name, src, -EINVAL);
1167 OBD_ALLOC(osdname, MAX_OBD_NAME);
1168 if (osdname == NULL)
1171 memcpy(osdname, src, idx); /* copy just the fsname part */
1172 osdname[idx] = '\0';
1174 mdt = strstr(mdt, "-MDT");
1175 if (mdt == NULL) /* 1.8 configs don't have "-MDT0000" at the end */
1176 strcat(osdname, "-MDT0000");
1178 strcat(osdname, mdt);
1179 strcat(osdname, "-osd");
1180 CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
1182 osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
1183 osp->opd_dt_dev.dd_ops = &osp_dt_ops;
1185 obd->obd_lu_dev = &osp->opd_dt_dev.dd_lu_dev;
1187 rc = osp_connect_to_osd(env, osp, osdname);
1191 rc = ptlrpcd_addref();
1193 GOTO(out_disconnect, rc);
1195 rc = client_obd_setup(obd, cfg);
1197 CERROR("%s: can't setup obd: rc = %d\n", osp->opd_obd->obd_name,
1202 osp_tunables_init(osp);
1204 rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
1205 LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
1207 CERROR("%s: fid init error: rc = %d\n",
1208 osp->opd_obd->obd_name, rc);
1212 if (!osp->opd_connect_mdt) {
1213 /* Initialize last id from the storage - will be
1214 * used in orphan cleanup. */
1215 if (!osp->opd_storage->dd_rdonly) {
1216 rc = osp_last_used_init(env, osp);
1221 /* Initialize precreation thread, it handles new
1222 * connections as well. */
1223 rc = osp_init_precreate(osp);
1225 GOTO(out_last_used, rc);
1228 * Initialize synhronization mechanism taking
1229 * care of propogating changes to OST in near
1230 * transactional manner.
1232 rc = osp_sync_init(env, osp);
1234 GOTO(out_precreat, rc);
1236 osp->opd_got_disconnected = 1;
1237 rc = osp_update_init(osp);
1242 rc = osp_init_statfs(osp);
1244 GOTO(out_precreat, rc);
1246 ns_register_cancel(obd->obd_namespace, osp_cancel_weight);
1249 * Initiate connect to OST
1251 imp = obd->u.cli.cl_import;
1253 rc = ptlrpc_init_import(imp);
1257 OBD_FREE(osdname, MAX_OBD_NAME);
1261 if (!osp->opd_connect_mdt)
1262 /* stop sync thread */
1265 /* stop precreate thread */
1266 if (!osp->opd_connect_mdt)
1267 osp_precreate_fini(osp);
1269 osp_update_fini(env, osp);
1271 if (!osp->opd_connect_mdt)
1272 osp_last_used_fini(env, osp);
1274 obd_fid_fini(osp->opd_obd);
1276 osp_tunables_fini(osp);
1277 client_obd_cleanup(obd);
1281 obd_disconnect(osp->opd_storage_exp);
1284 OBD_FREE(osdname, MAX_OBD_NAME);
1289 * Implementation of lu_device_type_operations::ldto_device_free
1291 * Free the OSP device in memory. No return value is needed for now,
1292 * so always return NULL to comply with the interface.
1294 * \param[in] env execution environment
1295 * \param[in] lu lu_device of OSP
1297 * \retval NULL NULL unconditionally
1299 static struct lu_device *osp_device_free(const struct lu_env *env,
1300 struct lu_device *lu)
1302 struct osp_device *osp = lu2osp_dev(lu);
1304 lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR,
1306 dt_device_fini(&osp->opd_dt_dev);
1313 * Implementation of lu_device_type_operations::ldto_device_alloc
1315 * This function allocates and initializes OSP device in memory according to
1318 * \param[in] env execution environment
1319 * \param[in] type device type of OSP
1320 * \param[in] lcfg config log
1322 * \retval pointer the pointer of allocated OSP if succeed.
1323 * \retval ERR_PTR(errno) ERR_PTR(errno) if failed.
1325 static struct lu_device *osp_device_alloc(const struct lu_env *env,
1326 struct lu_device_type *type,
1327 struct lustre_cfg *lcfg)
1329 struct osp_device *osp;
1330 struct lu_device *ld;
1334 ld = ERR_PTR(-ENOMEM);
1338 ld = osp2lu_dev(osp);
1339 dt_device_init(&osp->opd_dt_dev, type);
1340 rc = osp_init0(env, osp, type, lcfg);
1342 osp_device_free(env, ld);
1350 * Implementation of lu_device_type_operations::ldto_device_fini
1352 * This function cleans up the OSP device, i.e. release and free those
1353 * attached items in osp_device.
1355 * \param[in] env execution environment
1356 * \param[in] ld lu_device of OSP
1358 * \retval NULL NULL if cleanup succeeded.
1359 * \retval ERR_PTR(errno) ERR_PTR(errno) if cleanup failed.
1361 static struct lu_device *osp_device_fini(const struct lu_env *env,
1362 struct lu_device *ld)
1364 struct osp_device *osp = lu2osp_dev(ld);
1369 if (osp->opd_async_requests != NULL) {
1370 osp_update_request_destroy(env, osp->opd_async_requests);
1371 osp->opd_async_requests = NULL;
1374 if (osp->opd_storage_exp) {
1375 /* wait for the commit callbacks to complete */
1376 wait_event(osp->opd_sync_waitq,
1377 atomic_read(&osp->opd_commits_registered) == 0);
1378 obd_disconnect(osp->opd_storage_exp);
1381 LASSERT(osp->opd_obd);
1383 rc = client_obd_cleanup(osp->opd_obd);
1386 RETURN(ERR_PTR(rc));
1389 osp_tunables_fini(osp);
1397 * Implementation of obd_ops::o_reconnect
1399 * This function is empty and does not need to do anything for now.
1401 static int osp_reconnect(const struct lu_env *env,
1402 struct obd_export *exp, struct obd_device *obd,
1403 struct obd_uuid *cluuid,
1404 struct obd_connect_data *data,
1411 * Implementation of obd_ops::o_connect
1413 * Connect OSP to the remote target (MDT or OST). Allocate the
1414 * export and return it to the LOD, which calls this function
1415 * for each OSP to connect it to the remote target. This function
1416 * is currently only called once per OSP.
1418 * \param[in] env execution environment
1419 * \param[out] exp export connected to OSP
1420 * \param[in] obd OSP device
1421 * \param[in] cluuid OSP device client uuid
1422 * \param[in] data connect_data to be used to connect to the remote
1424 * \param[in] localdata necessary for the API interface, but not used in
1427 * \retval 0 0 if the connection succeeded.
1428 * \retval negative negative errno if the connection failed.
1430 static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp,
1431 struct obd_device *obd, struct obd_uuid *cluuid,
1432 struct obd_connect_data *data, void *localdata)
1434 struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
1439 LASSERT(data != NULL);
1440 LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
1442 rc = client_connect_import(env, &osp->opd_exp, obd, cluuid, data,
1447 osp->opd_obd->u.cli.cl_seq->lcs_exp = class_export_get(osp->opd_exp);
1448 *exp = osp->opd_exp;
1454 * Implementation of obd_ops::o_disconnect
1456 * Disconnect the export for the OSP. This is called by LOD to release the
1457 * OSP during cleanup (\see lod_del_device()). The OSP will be released after
1458 * the export is released.
1460 * \param[in] exp export to be disconnected.
1462 * \retval 0 0 if disconnection succeed
1463 * \retval negative negative errno if disconnection failed
1465 static int osp_obd_disconnect(struct obd_export *exp)
1467 struct obd_device *obd = exp->exp_obd;
1471 rc = class_disconnect(exp);
1473 CERROR("%s: class disconnect error: rc = %d\n",
1478 /* destroy the device */
1479 class_manual_cleanup(obd);
1485 * Implementation of obd_ops::o_statfs
1487 * Send a RPC to the remote target to get statfs status. This is only used
1488 * in lprocfs helpers by obd_statfs.
1490 * \param[in] env execution environment
1491 * \param[in] exp connection state from this OSP to the parent (LOD)
1493 * \param[out] osfs hold the statfs result
1494 * \param[in] unused Not used in this function for now
1495 * \param[in] flags flags to indicate how OSP will issue the RPC
1497 * \retval 0 0 if statfs succeeded.
1498 * \retval negative negative errno if statfs failed.
1500 static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
1501 struct obd_statfs *osfs, time64_t unused, __u32 flags)
1503 struct obd_statfs *msfs;
1504 struct ptlrpc_request *req;
1505 struct obd_import *imp = NULL, *imp0;
1510 /* Since the request might also come from lprocfs, so we need
1511 * sync this with client_disconnect_export Bug15684
1513 with_imp_locked(exp->exp_obd, imp0, rc)
1514 imp = class_import_get(imp0);
1518 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
1520 class_import_put(imp);
1525 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
1527 ptlrpc_request_free(req);
1530 ptlrpc_request_set_replen(req);
1531 req->rq_request_portal = OST_CREATE_PORTAL;
1532 ptlrpc_at_set_req_timeout(req);
1534 if (flags & OBD_STATFS_NODELAY) {
1535 /* procfs requests not want stat in wait for avoid deadlock */
1536 req->rq_no_resend = 1;
1537 req->rq_no_delay = 1;
1540 rc = ptlrpc_queue_wait(req);
1544 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1546 GOTO(out, rc = -EPROTO);
1552 ptlrpc_req_finished(req);
1557 * Implementation of obd_ops::o_import_event
1559 * This function is called when some related import event happens. It will
1560 * mark the necessary flags according to the event and notify the necessary
1561 * threads (mainly precreate thread).
1563 * \param[in] obd OSP OBD device
1564 * \param[in] imp import attached from OSP to remote (OST/MDT) service
1565 * \param[in] event event related to remote service (IMP_EVENT_*)
1567 * \retval 0 0 if the event handling succeeded.
1568 * \retval negative negative errno if the event handling failed.
1570 static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
1571 enum obd_import_event event)
1573 struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
1577 case IMP_EVENT_DISCON:
1578 d->opd_got_disconnected = 1;
1579 d->opd_imp_connected = 0;
1580 if (d->opd_connect_mdt)
1583 if (d->opd_pre != NULL) {
1584 osp_pre_update_status(d, -ENODEV);
1585 wake_up(&d->opd_pre_waitq);
1588 CDEBUG(D_HA, "got disconnected\n");
1590 case IMP_EVENT_INACTIVE:
1591 d->opd_imp_active = 0;
1592 d->opd_imp_connected = 0;
1593 d->opd_obd->obd_inactive = 1;
1594 if (d->opd_connect_mdt)
1596 if (d->opd_pre != NULL) {
1597 /* Import is invalid, we can`t get stripes so
1599 rc = imp->imp_deactive ? -ESHUTDOWN : -ENODEV;
1600 osp_pre_update_status(d, rc);
1601 wake_up(&d->opd_pre_waitq);
1604 CDEBUG(D_HA, "got inactive\n");
1606 case IMP_EVENT_ACTIVE:
1607 d->opd_imp_active = 1;
1609 d->opd_new_connection = 1;
1610 d->opd_imp_connected = 1;
1611 d->opd_imp_seen_connected = 1;
1612 d->opd_obd->obd_inactive = 0;
1613 wake_up(&d->opd_pre_waitq);
1614 if (d->opd_connect_mdt)
1617 osp_sync_check_for_work(d);
1618 CDEBUG(D_HA, "got connected\n");
1620 case IMP_EVENT_INVALIDATE:
1621 if (d->opd_connect_mdt)
1622 osp_invalidate_request(d);
1624 if (obd->obd_namespace == NULL)
1626 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
1629 case IMP_EVENT_DEACTIVATE:
1630 case IMP_EVENT_ACTIVATE:
1633 CERROR("%s: unsupported import event: %#x\n",
1634 obd->obd_name, event);
1640 * Implementation of obd_ops: o_iocontrol
1642 * This function is the ioctl handler for OSP. Note: lctl will access the OSP
1643 * directly by ioctl, instead of through the MDS stack.
1645 * param[in] cmd ioctl command.
1646 * param[in] exp export of this OSP.
1647 * param[in] len data length of \a karg.
1648 * param[in] karg input argument which is packed as
1650 * param[out] uarg pointer to userspace buffer (must access by
1653 * \retval 0 0 if the ioctl handling succeeded.
1654 * \retval negative negative errno if the ioctl handling failed.
1656 static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1657 void *karg, void __user *uarg)
1659 struct obd_device *obd = exp->exp_obd;
1660 struct osp_device *d;
1661 struct obd_ioctl_data *data = karg;
1666 LASSERT(obd->obd_lu_dev);
1667 d = lu2osp_dev(obd->obd_lu_dev);
1668 LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
1670 if (!try_module_get(THIS_MODULE)) {
1671 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1672 module_name(THIS_MODULE));
1677 case OBD_IOC_CLIENT_RECOVER:
1678 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
1679 data->ioc_inlbuf1, 0);
1683 case IOC_OSC_SET_ACTIVE:
1684 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
1688 CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
1689 cmd, current->comm);
1692 module_put(THIS_MODULE);
1698 * Implementation of obd_ops::o_get_info
1700 * Retrieve information by key. Retrieval starts from the top layer
1701 * (MDT) of the MDS stack and traverses the stack by calling the
1702 * obd_get_info() method of the next sub-layer.
1704 * \param[in] env execution environment
1705 * \param[in] exp export of this OSP
1706 * \param[in] keylen length of \a key
1707 * \param[in] key the key
1708 * \param[out] vallen length of \a val
1709 * \param[out] val holds the value returned by the key
1711 * \retval 0 0 if getting information succeeded.
1712 * \retval negative negative errno if getting information failed.
1714 static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1715 __u32 keylen, void *key, __u32 *vallen, void *val)
1719 if (KEY_IS(KEY_OSP_CONNECTED)) {
1720 struct obd_device *obd = exp->exp_obd;
1721 struct osp_device *osp;
1723 if (!obd->obd_set_up || obd->obd_stopping)
1726 osp = lu2osp_dev(obd->obd_lu_dev);
1729 * 1.8/2.0 behaviour is that OST being connected once at least
1730 * is considered "healthy". and one "healthy" OST is enough to
1731 * allow lustre clients to connect to MDS
1733 RETURN(!osp->opd_imp_seen_connected);
1739 static int osp_obd_set_info_async(const struct lu_env *env,
1740 struct obd_export *exp,
1741 u32 keylen, void *key,
1742 u32 vallen, void *val,
1743 struct ptlrpc_request_set *set)
1745 struct obd_device *obd = exp->exp_obd;
1746 struct obd_import *imp = obd->u.cli.cl_import;
1747 struct osp_device *osp;
1748 struct ptlrpc_request *req;
1752 if (KEY_IS(KEY_SPTLRPC_CONF)) {
1753 sptlrpc_conf_client_adapt(exp->exp_obd);
1757 LASSERT(set != NULL);
1758 if (!obd->obd_set_up || obd->obd_stopping)
1760 osp = lu2osp_dev(obd->obd_lu_dev);
1762 req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
1766 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1767 RCL_CLIENT, keylen);
1768 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1769 RCL_CLIENT, vallen);
1770 if (osp->opd_connect_mdt)
1771 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1773 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
1775 ptlrpc_request_free(req);
1779 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1780 memcpy(tmp, key, keylen);
1781 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1782 memcpy(tmp, val, vallen);
1784 ptlrpc_request_set_replen(req);
1785 ptlrpc_set_add_req(set, req);
1786 ptlrpc_check_set(NULL, set);
1791 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
1792 LU_KEY_INIT_FINI(osp, struct osp_thread_info);
1793 static void osp_key_exit(const struct lu_context *ctx,
1794 struct lu_context_key *key, void *data)
1796 struct osp_thread_info *info = data;
1798 info->osi_attr.la_valid = 0;
1801 struct lu_context_key osp_thread_key = {
1802 .lct_tags = LCT_MD_THREAD,
1803 .lct_init = osp_key_init,
1804 .lct_fini = osp_key_fini,
1805 .lct_exit = osp_key_exit
1808 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
1809 LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
1811 struct lu_context_key osp_txn_key = {
1812 .lct_tags = LCT_OSP_THREAD,
1813 .lct_init = osp_txn_key_init,
1814 .lct_fini = osp_txn_key_fini
1816 LU_TYPE_INIT_FINI(osp, &osp_thread_key, &osp_txn_key);
1818 static const struct lu_device_type_operations osp_device_type_ops = {
1819 .ldto_init = osp_type_init,
1820 .ldto_fini = osp_type_fini,
1822 .ldto_start = osp_type_start,
1823 .ldto_stop = osp_type_stop,
1825 .ldto_device_alloc = osp_device_alloc,
1826 .ldto_device_free = osp_device_free,
1828 .ldto_device_fini = osp_device_fini
1831 static struct lu_device_type osp_device_type = {
1832 .ldt_tags = LU_DEVICE_DT,
1833 .ldt_name = LUSTRE_OSP_NAME,
1834 .ldt_ops = &osp_device_type_ops,
1835 .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD,
1838 static const struct obd_ops osp_obd_device_ops = {
1839 .o_owner = THIS_MODULE,
1840 .o_add_conn = client_import_add_conn,
1841 .o_del_conn = client_import_del_conn,
1842 .o_reconnect = osp_reconnect,
1843 .o_connect = osp_obd_connect,
1844 .o_disconnect = osp_obd_disconnect,
1845 .o_get_info = osp_obd_get_info,
1846 .o_set_info_async = osp_obd_set_info_async,
1847 .o_import_event = osp_import_event,
1848 .o_iocontrol = osp_iocontrol,
1849 .o_statfs = osp_obd_statfs,
1850 .o_fid_init = client_fid_init,
1851 .o_fid_fini = client_fid_fini,
1855 * Initialize OSP module.
1857 * Register device types OSP and Light Weight Proxy (LWP) (\see lwp_dev.c)
1858 * in obd_types (\see class_obd.c). Initialize procfs for the
1859 * the OSP device. Note: OSP was called OSC before Lustre 2.4,
1860 * so for compatibility it still uses the name "osc" in procfs.
1861 * This is called at module load time.
1863 * \retval 0 0 if initialization succeeds.
1864 * \retval negative negative errno if initialization failed.
1866 static int __init osp_init(void)
1868 struct obd_type *sym;
1871 rc = lu_kmem_init(osp_caches);
1875 rc = class_register_type(&osp_obd_device_ops, NULL, false,
1876 LUSTRE_OSP_NAME, &osp_device_type);
1878 lu_kmem_fini(osp_caches);
1882 rc = class_register_type(&lwp_obd_device_ops, NULL, false,
1883 LUSTRE_LWP_NAME, &lwp_device_type);
1885 class_unregister_type(LUSTRE_OSP_NAME);
1886 lu_kmem_fini(osp_caches);
1890 /* create "osc" entry for compatibility purposes */
1891 sym = class_add_symlinks(LUSTRE_OSC_NAME, true);
1894 /* does real "osc" already exist ? */
1903 * Finalize OSP module.
1905 * This callback is called when kernel unloads OSP module from memory, and
1906 * it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
1908 static void __exit osp_exit(void)
1910 struct obd_type *sym = class_search_type(LUSTRE_OSC_NAME);
1912 /* if this was never fully initialized by the osc layer
1913 * then we are responsible for freeing this obd_type
1916 /* final put if we manage this obd type */
1917 if (sym->typ_sym_filter)
1918 kobject_put(&sym->typ_kobj);
1919 /* put reference taken by class_search_type */
1920 kobject_put(&sym->typ_kobj);
1923 class_unregister_type(LUSTRE_LWP_NAME);
1924 class_unregister_type(LUSTRE_OSP_NAME);
1925 lu_kmem_fini(osp_caches);
1928 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1929 MODULE_DESCRIPTION("Lustre OSD Storage Proxy ("LUSTRE_OSP_NAME")");
1930 MODULE_VERSION(LUSTRE_VERSION_STRING);
1931 MODULE_LICENSE("GPL");
1933 module_init(osp_init);
1934 module_exit(osp_exit);