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) 2013, 2017, Intel Corporation.
26 * lustre/osp/osp_md_object.c
28 * OST/MDT proxy device (OSP) Metadata methods
30 * This file implements methods for remote MD object, which include
31 * dt_object_operations, dt_index_operations and dt_body_operations.
33 * If there are multiple MDTs in one filesystem, one operation might
34 * include modifications in several MDTs. In such cases, clients
35 * send the RPC to the master MDT, then the operation is decomposed into
36 * object updates which will be dispatched to OSD or OSP. The local updates
37 * go to local OSD and the remote updates go to OSP. In OSP, these remote
38 * object updates will be packed into an update RPC, sent to the remote MDT
39 * and handled by Object Update Target (OUT).
41 * In DNE phase I, because of missing complete recovery solution, updates
42 * will be executed in order and synchronously.
43 * 1. The transaction is created.
44 * 2. In transaction declare, it collects and packs remote
45 * updates (in osp_md_declare_xxx()).
46 * 3. In transaction start, it sends these remote updates
47 * to remote MDTs, which will execute these updates synchronously.
48 * 4. In transaction execute phase, the local updates will be executed
51 * Author: Di Wang <di.wang@intel.com>
54 #define DEBUG_SUBSYSTEM S_MDS
56 #include <llog_swab.h>
57 #include <lustre_log.h>
58 #include "osp_internal.h"
60 #define OUT_UPDATE_BUFFER_SIZE_ADD 4096
61 #define OUT_UPDATE_BUFFER_SIZE_MAX (256 * 4096) /* 1M update size now */
64 * Interpreter call for object creation
66 * Object creation interpreter, which will be called after creating
67 * the remote object to set flags and status.
69 * \param[in] env execution environment
70 * \param[in] reply update reply
71 * \param[in] req ptlrpc update request for creating object
72 * \param[in] obj object to be created
73 * \param[in] data data used in this function.
74 * \param[in] index index(position) of create update in the whole
76 * \param[in] rc update result on the remote MDT.
78 * \retval only return 0 for now
80 static int osp_create_interpreter(const struct lu_env *env,
81 struct object_update_reply *reply,
82 struct ptlrpc_request *req,
83 struct osp_object *obj,
84 void *data, int index, int rc)
86 if (rc != 0 && rc != -EEXIST) {
87 obj->opo_obj.do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
88 obj->opo_non_exist = 1;
92 * invalidate opo cache for the object after the object is created, so
93 * attr_get will try to get attr from remote object.
95 osp_obj_invalidate_cache(obj);
101 * Implementation of dt_object_operations::do_declare_create
103 * Create the osp_update_request to track the update for this OSP
104 * in the transaction.
106 * \param[in] env execution environment
107 * \param[in] dt remote object to be created
108 * \param[in] attr attribute of the created object
109 * \param[in] hint creation hint
110 * \param[in] dof creation format information
111 * \param[in] th the transaction handle
113 * \retval 0 if preparation succeeds.
114 * \retval negative errno if preparation fails.
116 int osp_md_declare_create(const struct lu_env *env, struct dt_object *dt,
117 struct lu_attr *attr, struct dt_allocation_hint *hint,
118 struct dt_object_format *dof, struct thandle *th)
120 return osp_trans_update_request_create(th);
123 struct object_update *
124 update_buffer_get_update(struct object_update_request *request,
130 if (index > request->ourq_count)
133 ptr = &request->ourq_updates[0];
134 for (i = 0; i < index; i++)
135 ptr += object_update_size(ptr);
141 * Implementation of dt_object_operations::do_create
143 * It adds an OUT_CREATE sub-request into the OUT RPC that will be flushed
144 * when the transaction stop, and sets necessary flags for created object.
146 * \param[in] env execution environment
147 * \param[in] dt object to be created
148 * \param[in] attr attribute of the created object
149 * \param[in] hint creation hint
150 * \param[in] dof creation format information
151 * \param[in] th the transaction handle
153 * \retval 0 if packing creation succeeds.
154 * \retval negative errno if packing creation fails.
156 int osp_md_create(const struct lu_env *env, struct dt_object *dt,
157 struct lu_attr *attr, struct dt_allocation_hint *hint,
158 struct dt_object_format *dof, struct thandle *th)
160 struct osp_update_request *update;
161 struct osp_object *obj = dt2osp_obj(dt);
164 update = thandle_to_osp_update_request(th);
165 LASSERT(update != NULL);
167 LASSERT(attr->la_valid & LA_TYPE);
168 rc = OSP_UPDATE_RPC_PACK(env, out_create_pack, update,
169 lu_object_fid(&dt->do_lu), attr, hint, dof);
173 rc = osp_insert_update_callback(env, update, dt2osp_obj(dt), NULL,
174 osp_create_interpreter);
179 dt->do_lu.lo_header->loh_attr |= LOHA_EXISTS | (attr->la_mode & S_IFMT);
180 dt2osp_obj(dt)->opo_non_exist = 0;
182 obj->opo_attr = *attr;
188 * Implementation of dt_object_operations::do_declare_ref_del
190 * Create the osp_update_request to track the update for this OSP
191 * in the transaction.
193 * \param[in] env execution environment
194 * \param[in] dt object to decrease the reference count.
195 * \param[in] th the transaction handle of refcount decrease.
197 * \retval 0 if preparation succeeds.
198 * \retval negative errno if preparation fails.
200 static int osp_md_declare_ref_del(const struct lu_env *env,
201 struct dt_object *dt, struct thandle *th)
203 return osp_trans_update_request_create(th);
207 * Implementation of dt_object_operations::do_ref_del
209 * Add an OUT_REF_DEL sub-request into the OUT RPC that will be
210 * flushed when the transaction stop.
212 * \param[in] env execution environment
213 * \param[in] dt object to decrease the reference count
214 * \param[in] th the transaction handle
216 * \retval 0 if packing ref_del succeeds.
217 * \retval negative errno if packing fails.
219 static int osp_md_ref_del(const struct lu_env *env, struct dt_object *dt,
222 struct osp_update_request *update;
225 update = thandle_to_osp_update_request(th);
226 LASSERT(update != NULL);
228 rc = OSP_UPDATE_RPC_PACK(env, out_ref_del_pack, update,
229 lu_object_fid(&dt->do_lu));
234 * Implementation of dt_object_operations::do_declare_ref_del
236 * Create the osp_update_request to track the update for this OSP
237 * in the transaction.
239 * \param[in] env execution environment
240 * \param[in] dt object on which to increase the reference count.
241 * \param[in] th the transaction handle.
243 * \retval 0 if preparation succeeds.
244 * \retval negative errno if preparation fails.
246 static int osp_md_declare_ref_add(const struct lu_env *env,
247 struct dt_object *dt, struct thandle *th)
249 return osp_trans_update_request_create(th);
253 * Implementation of dt_object_operations::do_ref_add
255 * Add an OUT_REF_ADD sub-request into the OUT RPC that will be flushed
256 * when the transaction stop.
258 * \param[in] env execution environment
259 * \param[in] dt object on which to increase the reference count
260 * \param[in] th the transaction handle
262 * \retval 0 if packing ref_add succeeds.
263 * \retval negative errno if packing fails.
265 static int osp_md_ref_add(const struct lu_env *env, struct dt_object *dt,
268 struct osp_update_request *update;
271 update = thandle_to_osp_update_request(th);
272 LASSERT(update != NULL);
274 rc = OSP_UPDATE_RPC_PACK(env, out_ref_add_pack, update,
275 lu_object_fid(&dt->do_lu));
280 * Implementation of dt_object_operations::do_ah_init
282 * Initialize the allocation hint for object creation, which is usually called
283 * before the creation, and these hints (parent and child mode) will be sent to
284 * the remote Object Update Target (OUT) and used in the object create process,
285 * same as OSD object creation.
287 * \param[in] env execution environment
288 * \param[in] ah the hint to be initialized
289 * \param[in] parent the parent of the object
290 * \param[in] child the object to be created
291 * \param[in] child_mode the mode of the created object
293 static void osp_md_ah_init(const struct lu_env *env,
294 struct dt_allocation_hint *ah,
295 struct dt_object *parent,
296 struct dt_object *child,
301 ah->dah_parent = parent;
302 ah->dah_mode = child_mode;
306 * Implementation of dt_object_operations::do_declare_attr_get
308 * Create the osp_update_request to track the update for this OSP
309 * in the transaction.
311 * \param[in] env execution environment
312 * \param[in] dt object on which to set attributes
313 * \param[in] attr attributes to be set
314 * \param[in] th the transaction handle
316 * \retval 0 if preparation succeeds.
317 * \retval negative errno if preparation fails.
319 int osp_md_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
320 const struct lu_attr *attr, struct thandle *th)
322 return osp_trans_update_request_create(th);
326 * Implementation of dt_object_operations::do_attr_set
328 * Set attributes to the specified remote object.
330 * Add the OUT_ATTR_SET sub-request into the OUT RPC that will be flushed
331 * when the transaction stop.
333 * \param[in] env execution environment
334 * \param[in] dt object to set attributes
335 * \param[in] attr attributes to be set
336 * \param[in] th the transaction handle
338 * \retval 0 if packing attr_set succeeds.
339 * \retval negative errno if packing fails.
341 int osp_md_attr_set(const struct lu_env *env, struct dt_object *dt,
342 const struct lu_attr *attr, struct thandle *th)
344 struct osp_update_request *update;
347 update = thandle_to_osp_update_request(th);
348 LASSERT(update != NULL);
350 rc = OSP_UPDATE_RPC_PACK(env, out_attr_set_pack, update,
351 lu_object_fid(&dt->do_lu), attr);
356 * Implementation of dt_object_operations::do_read_lock
358 * osp_md_{read,write}_lock() will only lock the remote object in the
359 * local cache, which uses the semaphore (opo_sem) inside the osp_object to
360 * lock the object. Note: it will not lock the object in the whole cluster,
361 * which relies on the LDLM lock.
363 * \param[in] env execution environment
364 * \param[in] dt object to be locked
365 * \param[in] role lock role from MDD layer, see mdd_object_role().
367 static void osp_md_read_lock(const struct lu_env *env, struct dt_object *dt,
370 struct osp_object *obj = dt2osp_obj(dt);
372 LASSERT(obj->opo_owner != env);
373 down_read_nested(&obj->opo_sem, role);
375 LASSERT(obj->opo_owner == NULL);
379 * Implementation of dt_object_operations::do_write_lock
381 * Lock the remote object in write mode.
383 * \param[in] env execution environment
384 * \param[in] dt object to be locked
385 * \param[in] role lock role from MDD layer, see mdd_object_role().
387 static void osp_md_write_lock(const struct lu_env *env, struct dt_object *dt,
390 struct osp_object *obj = dt2osp_obj(dt);
392 down_write_nested(&obj->opo_sem, role);
394 LASSERT(obj->opo_owner == NULL);
395 obj->opo_owner = env;
399 * Implementation of dt_object_operations::do_read_unlock
401 * Unlock the read lock of remote object.
403 * \param[in] env execution environment
404 * \param[in] dt object to be unlocked
406 static void osp_md_read_unlock(const struct lu_env *env, struct dt_object *dt)
408 struct osp_object *obj = dt2osp_obj(dt);
410 up_read(&obj->opo_sem);
414 * Implementation of dt_object_operations::do_write_unlock
416 * Unlock the write lock of remote object.
418 * \param[in] env execution environment
419 * \param[in] dt object to be unlocked
421 static void osp_md_write_unlock(const struct lu_env *env, struct dt_object *dt)
423 struct osp_object *obj = dt2osp_obj(dt);
425 LASSERT(obj->opo_owner == env);
426 obj->opo_owner = NULL;
427 up_write(&obj->opo_sem);
431 * Implementation of dt_object_operations::do_write_locked
433 * Test if the object is locked in write mode.
435 * \param[in] env execution environment
436 * \param[in] dt object to be tested
438 static int osp_md_write_locked(const struct lu_env *env, struct dt_object *dt)
440 struct osp_object *obj = dt2osp_obj(dt);
442 return obj->opo_owner == env;
446 * Implementation of dt_index_operations::dio_lookup
448 * Look up record by key under a remote index object. It packs lookup update
449 * into RPC, sends to the remote OUT and waits for the lookup result.
451 * \param[in] env execution environment
452 * \param[in] dt index object to lookup
453 * \param[out] rec record in which to return lookup result
454 * \param[in] key key of index which will be looked up
456 * \retval 1 if the lookup succeeds.
457 * \retval negative errno if the lookup fails.
459 static int osp_md_index_lookup(const struct lu_env *env, struct dt_object *dt,
460 struct dt_rec *rec, const struct dt_key *key)
462 struct lu_buf *lbuf = &osp_env_info(env)->osi_lb2;
463 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
464 struct dt_device *dt_dev = &osp->opd_dt_dev;
465 struct osp_update_request *update;
466 struct object_update_reply *reply;
467 struct ptlrpc_request *req = NULL;
472 /* Because it needs send the update buffer right away,
473 * just create an update buffer, instead of attaching the
474 * update_remote list of the thandle.
476 update = osp_update_request_create(dt_dev);
478 RETURN(PTR_ERR(update));
480 rc = OSP_UPDATE_RPC_PACK(env, out_index_lookup_pack, update,
481 lu_object_fid(&dt->do_lu), rec, key);
483 CERROR("%s: Insert update error: rc = %d\n",
484 dt_dev->dd_lu_dev.ld_obd->obd_name, rc);
488 rc = osp_remote_sync(env, osp, update, &req);
492 reply = req_capsule_server_sized_get(&req->rq_pill,
493 &RMF_OUT_UPDATE_REPLY,
494 OUT_UPDATE_REPLY_SIZE);
495 if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
496 CERROR("%s: Wrong version %x expected %x: rc = %d\n",
497 dt_dev->dd_lu_dev.ld_obd->obd_name,
498 reply->ourp_magic, UPDATE_REPLY_MAGIC, -EPROTO);
499 GOTO(out, rc = -EPROTO);
502 rc = object_update_result_data_get(reply, lbuf, 0);
506 if (lbuf->lb_len != sizeof(*fid)) {
507 CERROR("%s: lookup "DFID" %s wrong size %d\n",
508 dt_dev->dd_lu_dev.ld_obd->obd_name,
509 PFID(lu_object_fid(&dt->do_lu)), (char *)key,
511 GOTO(out, rc = -EINVAL);
515 if (ptlrpc_rep_need_swab(req))
516 lustre_swab_lu_fid(fid);
517 if (!fid_is_sane(fid)) {
518 CERROR("%s: lookup "DFID" %s invalid fid "DFID"\n",
519 dt_dev->dd_lu_dev.ld_obd->obd_name,
520 PFID(lu_object_fid(&dt->do_lu)), (char *)key, PFID(fid));
521 GOTO(out, rc = -EINVAL);
524 memcpy(rec, fid, sizeof(*fid));
530 ptlrpc_req_finished(req);
532 osp_update_request_destroy(env, update);
538 * Implementation of dt_index_operations::dio_declare_insert
540 * Create the osp_update_request to track the update for this OSP
541 * in the transaction.
543 * \param[in] env execution environment
544 * \param[in] dt object for which to insert index
545 * \param[in] rec record of the index which will be inserted
546 * \param[in] key key of the index which will be inserted
547 * \param[in] th the transaction handle
549 * \retval 0 if preparation succeeds.
550 * \retval negative errno if preparation fails.
552 static int osp_md_declare_index_insert(const struct lu_env *env,
553 struct dt_object *dt,
554 const struct dt_rec *rec,
555 const struct dt_key *key,
558 return osp_trans_update_request_create(th);
562 * Implementation of dt_index_operations::dio_insert
564 * Add an OUT_INDEX_INSERT sub-request into the OUT RPC that will
565 * be flushed when the transaction stop.
567 * \param[in] env execution environment
568 * \param[in] dt object for which to insert index
569 * \param[in] rec record of the index to be inserted
570 * \param[in] key key of the index to be inserted
571 * \param[in] th the transaction handle
572 * \param[in] ignore_quota quota enforcement for insert
574 * \retval 0 if packing index insert succeeds.
575 * \retval negative errno if packing fails.
577 static int osp_md_index_insert(const struct lu_env *env, struct dt_object *dt,
578 const struct dt_rec *rec,
579 const struct dt_key *key, struct thandle *th,
582 struct osp_update_request *update;
585 update = thandle_to_osp_update_request(th);
586 LASSERT(update != NULL);
588 rc = OSP_UPDATE_RPC_PACK(env, out_index_insert_pack, update,
589 lu_object_fid(&dt->do_lu), rec, key);
594 * Implementation of dt_index_operations::dio_declare_delete
596 * Create the osp_update_request to track the update for this OSP
597 * in the transaction.
599 * \param[in] env execution environment
600 * \param[in] dt object for which to delete index
601 * \param[in] key key of the index
602 * \param[in] th the transaction handle
604 * \retval 0 if preparation succeeds.
605 * \retval negative errno if preparation fails.
607 static int osp_md_declare_index_delete(const struct lu_env *env,
608 struct dt_object *dt,
609 const struct dt_key *key,
612 return osp_trans_update_request_create(th);
616 * Implementation of dt_index_operations::dio_delete
618 * Add an OUT_INDEX_DELETE sub-request into the OUT RPC that will
619 * be flushed when the transaction stop.
621 * \param[in] env execution environment
622 * \param[in] dt object for which to delete index
623 * \param[in] key key of the index which will be deleted
624 * \param[in] th the transaction handle
626 * \retval 0 if packing index delete succeeds.
627 * \retval negative errno if packing fails.
629 static int osp_md_index_delete(const struct lu_env *env,
630 struct dt_object *dt,
631 const struct dt_key *key,
634 struct osp_update_request *update;
637 update = thandle_to_osp_update_request(th);
638 LASSERT(update != NULL);
640 rc = OSP_UPDATE_RPC_PACK(env, out_index_delete_pack, update,
641 lu_object_fid(&dt->do_lu), key);
647 * Implementation of dt_index_operations::dio_it.next
649 * Advance the pointer of the iterator to the next entry. It shares a similar
650 * internal implementation with osp_orphan_it_next(), which is being used for
651 * remote orphan index object. This method will be used for remote directory.
653 * \param[in] env execution environment
654 * \param[in] di iterator of this iteration
656 * \retval 0 if the pointer is advanced successfully.
657 * \retval 1 if it reaches to the end of the index object.
658 * \retval negative errno if the pointer cannot be advanced.
660 static int osp_md_index_it_next(const struct lu_env *env, struct dt_it *di)
662 struct osp_it *it = (struct osp_it *)di;
663 struct lu_idxpage *idxpage;
664 struct lu_dirent *ent = (struct lu_dirent *)it->ooi_ent;
669 idxpage = it->ooi_cur_idxpage;
670 if (idxpage != NULL) {
671 if (idxpage->lip_nr == 0)
677 (struct lu_dirent *)idxpage->lip_entries;
679 } else if (le16_to_cpu(ent->lde_reclen) != 0 &&
680 it->ooi_pos_ent < idxpage->lip_nr) {
681 ent = (struct lu_dirent *)(((char *)ent) +
682 le16_to_cpu(ent->lde_reclen));
690 rc = osp_it_next_page(env, di);
698 * Implementation of dt_index_operations::dio_it.key
700 * Get the key at current iterator poisiton. These iteration methods
701 * (dio_it) will only be used for iterating the remote directory, so
702 * the key is the name of the directory entry.
704 * \param[in] env execution environment
705 * \param[in] di iterator of this iteration
707 * \retval name of the current entry
709 static struct dt_key *osp_it_key(const struct lu_env *env,
710 const struct dt_it *di)
712 struct osp_it *it = (struct osp_it *)di;
713 struct lu_dirent *ent = (struct lu_dirent *)it->ooi_ent;
715 return (struct dt_key *)ent->lde_name;
719 * Implementation of dt_index_operations::dio_it.key_size
721 * Get the key size at current iterator poisiton. These iteration methods
722 * (dio_it) will only be used for iterating the remote directory, so the key
723 * size is the name size of the directory entry.
725 * \param[in] env execution environment
726 * \param[in] di iterator of this iteration
728 * \retval name size of the current entry
731 static int osp_it_key_size(const struct lu_env *env, const struct dt_it *di)
733 struct osp_it *it = (struct osp_it *)di;
734 struct lu_dirent *ent = (struct lu_dirent *)it->ooi_ent;
736 return (int)le16_to_cpu(ent->lde_namelen);
740 * Implementation of dt_index_operations::dio_it.rec
742 * Get the record at current iterator position. These iteration methods
743 * (dio_it) will only be used for iterating the remote directory, so it
744 * uses lu_dirent_calc_size() to calculate the record size.
746 * \param[in] env execution environment
747 * \param[in] di iterator of this iteration
748 * \param[out] rec the record to be returned
749 * \param[in] attr attributes of the index object, so it knows
750 * how to pack the entry.
752 * \retval only return 0 for now
754 static int osp_md_index_it_rec(const struct lu_env *env, const struct dt_it *di,
755 struct dt_rec *rec, __u32 attr)
757 struct osp_it *it = (struct osp_it *)di;
758 struct lu_dirent *ent = (struct lu_dirent *)it->ooi_ent;
761 reclen = lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen), attr);
762 memcpy(rec, ent, reclen);
767 * Implementation of dt_index_operations::dio_it.load
769 * Locate the iteration cursor to the specified position (cookie).
771 * \param[in] env pointer to the thread context
772 * \param[in] di pointer to the iteration structure
773 * \param[in] hash the specified position
775 * \retval positive number for locating to the exactly position
777 * \retval 0 for arriving at the end of the iteration
778 * \retval negative error number on failure
780 static int osp_it_load(const struct lu_env *env, const struct dt_it *di,
783 struct osp_it *it = (struct osp_it *)di;
787 rc = osp_md_index_it_next(env, (struct dt_it *)di);
797 const struct dt_index_operations osp_md_index_ops = {
798 .dio_lookup = osp_md_index_lookup,
799 .dio_declare_insert = osp_md_declare_index_insert,
800 .dio_insert = osp_md_index_insert,
801 .dio_declare_delete = osp_md_declare_index_delete,
802 .dio_delete = osp_md_index_delete,
808 .next = osp_md_index_it_next,
810 .key_size = osp_it_key_size,
811 .rec = osp_md_index_it_rec,
812 .store = osp_it_store,
814 .key_rec = osp_it_key_rec,
819 * Implement OSP layer dt_object_operations::do_xattr_list() interface.
821 * List extended attribute from the specified MDT/OST object, result is not
822 * cached because this is called by directory migration only.
824 * \param[in] env pointer to the thread context
825 * \param[in] dt pointer to the OSP layer dt_object
826 * \param[out] buf pointer to the lu_buf to hold the extended attribute
828 * \retval positive bytes used/required in the buffer
829 * \retval negative error number on failure
831 static int osp_md_xattr_list(const struct lu_env *env, struct dt_object *dt,
832 const struct lu_buf *buf)
834 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
835 struct osp_object *obj = dt2osp_obj(dt);
836 struct dt_device *dev = &osp->opd_dt_dev;
837 struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2;
838 struct osp_update_request *update = NULL;
839 struct ptlrpc_request *req = NULL;
840 struct object_update_reply *reply;
841 const char *dname = dt->do_lu.lo_dev->ld_obd->obd_name;
848 if (unlikely(obj->opo_non_exist))
851 update = osp_update_request_create(dev);
853 RETURN(PTR_ERR(update));
855 rc = OSP_UPDATE_RPC_PACK(env, out_xattr_list_pack, update,
856 lu_object_fid(&dt->do_lu), buf->lb_len);
858 CERROR("%s: Insert update error "DFID": rc = %d\n",
859 dname, PFID(lu_object_fid(&dt->do_lu)), rc);
863 rc = osp_remote_sync(env, osp, update, &req);
866 dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
867 obj->opo_non_exist = 1;
872 reply = req_capsule_server_sized_get(&req->rq_pill,
873 &RMF_OUT_UPDATE_REPLY,
874 OUT_UPDATE_REPLY_SIZE);
875 if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
876 DEBUG_REQ(D_ERROR, req,
877 "%s: Wrong version %x expected %x "DFID": rc = %d\n",
878 dname, reply->ourp_magic, UPDATE_REPLY_MAGIC,
879 PFID(lu_object_fid(&dt->do_lu)), -EPROTO);
881 GOTO(out, rc = -EPROTO);
884 rc = object_update_result_data_get(reply, rbuf, 0);
891 if (unlikely(buf->lb_len < rbuf->lb_len))
892 GOTO(out, rc = -ERANGE);
894 memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len);
899 ptlrpc_req_finished(req);
901 if (update && !IS_ERR(update))
902 osp_update_request_destroy(env, update);
908 * Implementation of dt_object_operations::do_index_try
910 * Try to initialize the index API pointer for the given object. This
911 * is the entry point of the index API, i.e. we must call this method
912 * to initialize the index object before calling other index methods.
914 * \param[in] env execution environment
915 * \param[in] dt index object to be initialized
916 * \param[in] feat the index feature of the object
918 * \retval 0 if the initialization succeeds.
919 * \retval negative errno if the initialization fails.
921 static int osp_md_index_try(const struct lu_env *env,
922 struct dt_object *dt,
923 const struct dt_index_features *feat)
925 dt->do_index_ops = &osp_md_index_ops;
930 * Implementation of dt_object_operations::do_object_lock
932 * Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT,
933 * which will lock the object in the global namespace. And because the
934 * cross-MDT locks are relatively rare compared with normal local MDT operation,
935 * let's release it right away, instead of putting it into the LRU list.
937 * \param[in] env execution environment
938 * \param[in] dt object to be locked
939 * \param[out] lh lock handle
940 * \param[in] einfo enqueue information
941 * \param[in] policy lock policy
943 * \retval ELDLM_OK if locking the object succeeds.
944 * \retval negative errno if locking fails.
946 static int osp_md_object_lock(const struct lu_env *env,
947 struct dt_object *dt,
948 struct lustre_handle *lh,
949 struct ldlm_enqueue_info *einfo,
950 union ldlm_policy_data *policy)
952 struct ldlm_res_id *res_id;
953 struct osp_device *osp = dt2osp_dev(lu2dt_dev(dt->do_lu.lo_dev));
954 struct ptlrpc_request *req;
956 __u64 flags = LDLM_FL_NO_LRU;
959 res_id = einfo->ei_res_id;
960 LASSERT(res_id != NULL);
962 if (einfo->ei_mode & (LCK_EX | LCK_PW))
963 flags |= LDLM_FL_COS_INCOMPAT;
965 req = ldlm_enqueue_pack(osp->opd_exp, 0);
967 RETURN(PTR_ERR(req));
969 osp_set_req_replay(osp, req);
970 rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id,
971 (const union ldlm_policy_data *)policy,
972 &flags, NULL, 0, LVB_T_NONE, lh, 0);
974 ptlrpc_req_finished(req);
976 RETURN(rc == ELDLM_OK ? 0 : -EIO);
980 * Implementation of dt_object_operations::do_object_unlock
982 * Cancel a lock of a remote object.
984 * \param[in] env execution environment
985 * \param[in] dt object to be unlocked
986 * \param[in] einfo lock enqueue information
987 * \param[in] policy lock policy
989 * \retval Only return 0 for now.
991 static int osp_md_object_unlock(const struct lu_env *env,
992 struct dt_object *dt,
993 struct ldlm_enqueue_info *einfo,
994 union ldlm_policy_data *policy)
996 struct lustre_handle *lockh = einfo->ei_cbdata;
999 ldlm_lock_decref(lockh, einfo->ei_mode);
1005 * Implement OSP layer dt_object_operations::do_declare_destroy() interface.
1007 * Create the dt_update_request to track the update for this OSP
1008 * in the transaction.
1010 * \param[in] env pointer to the thread context
1011 * \param[in] dt pointer to the OSP layer dt_object to be destroyed
1012 * \param[in] th pointer to the transaction handler
1014 * \retval 0 for success
1015 * \retval negative error number on failure
1017 int osp_md_declare_destroy(const struct lu_env *env, struct dt_object *dt,
1020 return osp_trans_update_request_create(th);
1024 * Implement OSP layer dt_object_operations::do_destroy() interface.
1026 * Pack the destroy update into the RPC buffer, which will be sent
1027 * to the remote MDT during transaction stop.
1029 * It also marks the object as non-cached.
1031 * \param[in] env pointer to the thread context
1032 * \param[in] dt pointer to the OSP layer dt_object to be destroyed
1033 * \param[in] th pointer to the transaction handler
1035 * \retval 0 for success
1036 * \retval negative error number on failure
1038 int osp_md_destroy(const struct lu_env *env, struct dt_object *dt,
1041 struct osp_object *o = dt2osp_obj(dt);
1042 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
1043 struct osp_update_request *update;
1047 o->opo_non_exist = 1;
1049 LASSERT(osp->opd_connect_mdt);
1050 update = thandle_to_osp_update_request(th);
1051 LASSERT(update != NULL);
1053 rc = OSP_UPDATE_RPC_PACK(env, out_destroy_pack, update,
1054 lu_object_fid(&dt->do_lu));
1058 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
1059 rc = osp_insert_update_callback(env, update, dt2osp_obj(dt), NULL,
1065 struct dt_object_operations osp_md_obj_ops = {
1066 .do_read_lock = osp_md_read_lock,
1067 .do_write_lock = osp_md_write_lock,
1068 .do_read_unlock = osp_md_read_unlock,
1069 .do_write_unlock = osp_md_write_unlock,
1070 .do_write_locked = osp_md_write_locked,
1071 .do_declare_create = osp_md_declare_create,
1072 .do_create = osp_md_create,
1073 .do_declare_ref_add = osp_md_declare_ref_add,
1074 .do_ref_add = osp_md_ref_add,
1075 .do_declare_ref_del = osp_md_declare_ref_del,
1076 .do_ref_del = osp_md_ref_del,
1077 .do_declare_destroy = osp_md_declare_destroy,
1078 .do_destroy = osp_md_destroy,
1079 .do_ah_init = osp_md_ah_init,
1080 .do_attr_get = osp_attr_get,
1081 .do_declare_attr_set = osp_md_declare_attr_set,
1082 .do_attr_set = osp_md_attr_set,
1083 .do_xattr_get = osp_xattr_get,
1084 .do_xattr_list = osp_md_xattr_list,
1085 .do_declare_xattr_set = osp_declare_xattr_set,
1086 .do_xattr_set = osp_xattr_set,
1087 .do_declare_xattr_del = osp_declare_xattr_del,
1088 .do_xattr_del = osp_xattr_del,
1089 .do_index_try = osp_md_index_try,
1090 .do_object_lock = osp_md_object_lock,
1091 .do_object_unlock = osp_md_object_unlock,
1092 .do_invalidate = osp_invalidate,
1096 * Implementation of dt_body_operations::dbo_declare_write
1098 * Create the osp_update_request to track the update for this OSP
1099 * in the transaction.
1101 * \param[in] env execution environment
1102 * \param[in] dt object to be written
1103 * \param[in] buf buffer to write which includes an embedded size field
1104 * \param[in] pos offet in the object to start writing at
1105 * \param[in] th transaction handle
1107 * \retval 0 if preparation succeeds.
1108 * \retval negative errno if preparation fails.
1110 static ssize_t osp_md_declare_write(const struct lu_env *env,
1111 struct dt_object *dt,
1112 const struct lu_buf *buf,
1113 loff_t pos, struct thandle *th)
1115 struct osp_device *osp = dt2osp_dev(th->th_dev);
1118 rc = osp_trans_update_request_create(th);
1122 if (osp->opd_update == NULL)
1125 if (dt2osp_obj(dt)->opo_stale)
1132 * Implementation of dt_body_operations::dbo_write
1134 * Pack the write object update into the RPC buffer, which will be sent
1135 * to the remote MDT during transaction stop.
1137 * \param[in] env execution environment
1138 * \param[in] dt object to be written
1139 * \param[in] buf buffer to write which includes an embedded size field
1140 * \param[in] pos offet in the object to start writing at
1141 * \param[in] th transaction handle
1142 * \param[in] ignore_quota quota enforcement for this write
1144 * \retval the buffer size in bytes if packing succeeds.
1145 * \retval negative errno if packing fails.
1147 static ssize_t osp_md_write(const struct lu_env *env, struct dt_object *dt,
1148 const struct lu_buf *buf, loff_t *pos,
1149 struct thandle *th, int ignore_quota)
1151 struct osp_object *obj = dt2osp_obj(dt);
1152 struct osp_update_request *update;
1153 struct osp_thandle *oth = thandle_to_osp_thandle(th);
1157 update = thandle_to_osp_update_request(th);
1158 LASSERT(update != NULL);
1160 CDEBUG(D_INFO, "write "DFID" offset = %llu length = %zu\n",
1161 PFID(lu_object_fid(&dt->do_lu)), *pos, buf->lb_len);
1163 rc = OSP_UPDATE_RPC_PACK(env, out_write_pack, update,
1164 lu_object_fid(&dt->do_lu), buf, *pos);
1168 rc = osp_check_and_set_rpc_version(oth, obj);
1172 /* XXX: how about the write error happened later? */
1173 *pos += buf->lb_len;
1175 if (obj->opo_attr.la_valid & LA_SIZE && obj->opo_attr.la_size < *pos)
1176 obj->opo_attr.la_size = *pos;
1178 spin_lock(&obj->opo_lock);
1179 if (list_empty(&obj->opo_invalidate_cb_list)) {
1180 lu_object_get(&obj->opo_obj.do_lu);
1182 list_add_tail(&obj->opo_invalidate_cb_list,
1183 &update->our_invalidate_cb_list);
1185 spin_unlock(&obj->opo_lock);
1187 RETURN(buf->lb_len);
1190 static inline void orr_le_to_cpu(struct out_read_reply *orr_dst,
1191 const struct out_read_reply *orr_src)
1193 orr_dst->orr_size = le32_to_cpu(orr_src->orr_size);
1194 orr_dst->orr_padding = le32_to_cpu(orr_src->orr_padding);
1195 orr_dst->orr_offset = le64_to_cpu(orr_dst->orr_offset);
1200 static ssize_t osp_md_read(const struct lu_env *env, struct dt_object *dt,
1201 struct lu_buf *rbuf, loff_t *pos)
1203 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
1204 struct dt_device *dt_dev = &osp->opd_dt_dev;
1205 struct lu_buf *lbuf = &osp_env_info(env)->osi_lb2;
1206 char *ptr = rbuf->lb_buf;
1207 struct osp_update_request *update;
1208 struct ptlrpc_request *req = NULL;
1209 struct out_read_reply *orr;
1210 struct ptlrpc_bulk_desc *desc;
1211 struct object_update_reply *reply;
1218 /* Because it needs send the update buffer right away,
1219 * just create an update buffer, instead of attaching the
1220 * update_remote list of the thandle. */
1221 update = osp_update_request_create(dt_dev);
1223 RETURN(PTR_ERR(update));
1225 rc = OSP_UPDATE_RPC_PACK(env, out_read_pack, update,
1226 lu_object_fid(&dt->do_lu),
1227 rbuf->lb_len, *pos);
1229 CERROR("%s: cannot insert update: rc = %d\n",
1230 dt_dev->dd_lu_dev.ld_obd->obd_name, rc);
1231 GOTO(out_update, rc);
1234 CDEBUG(D_INFO, "%s "DFID" read offset %llu size %zu\n",
1235 dt_dev->dd_lu_dev.ld_obd->obd_name,
1236 PFID(lu_object_fid(&dt->do_lu)), *pos, rbuf->lb_len);
1237 rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import, update,
1240 GOTO(out_update, rc);
1242 nbufs = (rbuf->lb_len + OUT_BULK_BUFFER_SIZE - 1) /
1243 OUT_BULK_BUFFER_SIZE;
1244 /* allocate bulk descriptor */
1245 desc = ptlrpc_prep_bulk_imp(req, nbufs, 1,
1246 PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KVEC,
1247 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1249 GOTO(out, rc = -ENOMEM);
1251 /* split the buffer into small chunk size */
1252 left_size = rbuf->lb_len;
1253 for (i = 0; i < nbufs; i++) {
1256 read_size = left_size > OUT_BULK_BUFFER_SIZE ?
1257 OUT_BULK_BUFFER_SIZE : left_size;
1258 desc->bd_frag_ops->add_iov_frag(desc, ptr, read_size);
1263 osp_set_req_replay(osp, req);
1264 req->rq_bulk_read = 1;
1265 /* send request to master and wait for RPC to complete */
1266 rc = ptlrpc_queue_wait(req);
1270 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1271 req->rq_bulk->bd_nob_transferred);
1275 reply = req_capsule_server_sized_get(&req->rq_pill,
1276 &RMF_OUT_UPDATE_REPLY,
1277 OUT_UPDATE_REPLY_SIZE);
1279 if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
1280 CERROR("%s: invalid update reply magic %x expected %x:"
1281 " rc = %d\n", dt_dev->dd_lu_dev.ld_obd->obd_name,
1282 reply->ourp_magic, UPDATE_REPLY_MAGIC, -EPROTO);
1283 GOTO(out, rc = -EPROTO);
1286 rc = object_update_result_data_get(reply, lbuf, 0);
1290 if (lbuf->lb_len < sizeof(*orr))
1291 GOTO(out, rc = -EPROTO);
1294 orr_le_to_cpu(orr, orr);
1296 *pos = orr->orr_offset;
1298 ptlrpc_req_finished(req);
1301 osp_update_request_destroy(env, update);
1306 /* These body operation will be used to write symlinks during migration etc */
1307 struct dt_body_operations osp_md_body_ops = {
1308 .dbo_declare_write = osp_md_declare_write,
1309 .dbo_write = osp_md_write,
1310 .dbo_read = osp_md_read,