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 * lustre/osp/osp_object.c
31 * Lustre OST Proxy Device (OSP) is the agent on the local MDT for the OST
34 * OSP object attributes cache
35 * ---------------------------
36 * OSP object is the stub of the remote OST-object or MDT-object. Both the
37 * attribute and the extended attributes are stored on the peer side remotely.
38 * It is inefficient to send RPC to peer to fetch those attributes when every
39 * get_attr()/get_xattr() called. For a large system, the LFSCK synchronous
40 * mode scanning is prohibitively inefficient.
42 * So the OSP maintains the OSP object attributes cache to cache some
43 * attributes on the local MDT. The cache is organized against the OSP
46 * struct osp_xattr_entry {
47 * struct list_head oxe_list;
53 * unsigned int oxe_exist:1,
60 * struct lu_attr opo_attr;
61 * struct list_head opo_xattr_list;
62 * spinlock_t opo_lock;
66 * The basic attributes, such as owner/mode/flags, are stored in the
67 * osp_object::opo_attr. The extended attributes will be stored
68 * as osp_xattr_entry. Every extended attribute has an independent
69 * osp_xattr_entry, and all the osp_xattr_entry are linked into the
70 * osp_object::opo_xattr_list. The OSP object attributes cache
71 * is protected by the osp_object::opo_lock.
73 * Not all OSP objects have an attributes cache because maintaining
74 * the cache requires some resources. Currently, the OSP object
75 * attributes cache will be initialized when the attributes or the
76 * extended attributes are pre-fetched via osp_declare_attr_get()
77 * or osp_declare_xattr_get(). That is usually for LFSCK purpose,
78 * but it also can be shared by others.
81 * XXX: NOT prepare out RPC for remote transaction. ((please refer to the
82 * comment of osp_trans_create() for remote transaction)
84 * According to our current transaction/dt_object_lock framework (to make
85 * the cross-MDTs modification for DNE1 to be workable), the transaction
86 * sponsor will start the transaction firstly, then try to acquire related
87 * dt_object_lock if needed. Under such rules, if we want to prepare the
88 * OUT RPC in the transaction declare phase, then related attr/xattr
89 * should be known without dt_object_lock. But such condition maybe not
90 * true for some remote transaction case. For example:
92 * For linkEA repairing (by LFSCK) case, before the LFSCK thread obtained
93 * the dt_object_lock on the target MDT-object, it cannot know whether
94 * the MDT-object has linkEA or not, neither invalid or not.
96 * Since the LFSCK thread cannot hold dt_object_lock before the remote
97 * transaction start (otherwise there will be some potential deadlock),
98 * it cannot prepare related OUT RPC for repairing during the declare
99 * phase as other normal transactions do.
101 * To resolve the trouble, we will make OSP to prepare related OUT RPC
102 * after remote transaction started, and trigger the remote updating
103 * (send RPC) when trans_stop. Then the up layer users, such as LFSCK,
104 * can follow the general rule to handle trans_start/dt_object_lock
105 * for repairing linkEA inconsistency without distinguishing remote
108 * In fact, above solution for remote transaction should be the normal
109 * model without considering DNE1. The trouble brought by DNE1 will be
110 * resolved in DNE2. At that time, this patch can be removed.
113 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
114 * Author: Mikhail Pershin <mike.tappro@intel.com>
117 #define DEBUG_SUBSYSTEM S_MDS
119 #include <lustre_obdo.h>
120 #include <lustre_swab.h>
122 #include "osp_internal.h"
124 static inline __u32 osp_dev2node(struct osp_device *osp)
126 return osp->opd_storage->dd_lu_dev.ld_site->ld_seq_site->ss_node_id;
129 static inline const char *osp_dto2name(struct osp_object *obj)
131 return obj->opo_obj.do_lu.lo_dev->ld_obd->obd_name;
134 static inline bool is_ost_obj(struct lu_object *lo)
136 return !lu2osp_dev(lo->lo_dev)->opd_connect_mdt;
139 static inline void __osp_oac_xattr_assignment(struct osp_object *obj,
140 struct osp_xattr_entry *oxe,
141 const struct lu_buf *buf)
144 memcpy(oxe->oxe_value, buf->lb_buf, buf->lb_len);
146 oxe->oxe_vallen = buf->lb_len;
152 * Assign FID to the OST object.
154 * This function will assign the FID to the OST object of a striped file.
156 * \param[in] env pointer to the thread context
157 * \param[in] d pointer to the OSP device
158 * \param[in] o pointer to the OSP object that the FID will be
161 static void osp_object_assign_fid(const struct lu_env *env,
162 struct osp_device *d, struct osp_object *o)
164 struct osp_thread_info *osi = osp_env_info(env);
166 LASSERT(fid_is_zero(lu_object_fid(&o->opo_obj.do_lu)));
167 LASSERT(o->opo_reserved);
170 osp_precreate_get_fid(env, d, &osi->osi_fid);
172 lu_object_assign_fid(env, &o->opo_obj.do_lu, &osi->osi_fid);
175 #define OXE_DEFAULT_LEN 16
178 * Release reference from the OSP object extended attribute entry.
180 * If it is the last reference, then free the entry.
182 * \param[in] oxe pointer to the OSP object extended attribute entry.
184 static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe)
186 if (atomic_dec_and_test(&oxe->oxe_ref)) {
187 LASSERT(list_empty(&oxe->oxe_list));
189 OBD_FREE_LARGE(oxe, oxe->oxe_buflen);
194 * Find the named extended attribute in the OSP object attributes cache.
196 * The caller should take the osp_object::opo_lock before calling
199 * \param[in] obj pointer to the OSP object
200 * \param[in] name the name of the extended attribute
201 * \param[in] namelen the name length of the extended attribute
203 * \retval pointer to the found extended attribute entry
204 * \retval NULL if the specified extended attribute is not
207 static struct osp_xattr_entry *
208 osp_oac_xattr_find_locked(struct osp_object *obj, const char *name,
211 struct osp_xattr_entry *oxe;
213 list_for_each_entry(oxe, &obj->opo_xattr_list, oxe_list) {
214 if (namelen == oxe->oxe_namelen &&
215 strncmp(name, oxe->oxe_buf, namelen) == 0)
223 * Find the named extended attribute in the OSP object attributes cache.
225 * Call osp_oac_xattr_find_locked() with the osp_object::opo_lock held.
227 * \param[in] obj pointer to the OSP object
228 * \param[in] name the name of the extended attribute
229 * \param[in] unlink true if the extended attribute entry is to be removed
232 * \retval pointer to the found extended attribute entry
233 * \retval NULL if the specified extended attribute is not
236 static struct osp_xattr_entry *osp_oac_xattr_find(struct osp_object *obj,
237 const char *name, bool unlink)
239 struct osp_xattr_entry *oxe = NULL;
241 spin_lock(&obj->opo_lock);
242 oxe = osp_oac_xattr_find_locked(obj, name, strlen(name));
245 list_del_init(&oxe->oxe_list);
247 atomic_inc(&oxe->oxe_ref);
249 spin_unlock(&obj->opo_lock);
255 * Find the named extended attribute in the OSP object attributes cache.
257 * If it is not in the cache, then add an empty entry (that will be
258 * filled later) to cache with the given name.
260 * \param[in] obj pointer to the OSP object
261 * \param[in] name the name of the extended attribute
262 * \param[in] len the length of the extended attribute value
264 * \retval pointer to the found or new-created extended
266 * \retval NULL if the specified extended attribute is not in the
267 * cache or fail to add new empty entry to the cache.
269 static struct osp_xattr_entry *
270 osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, size_t len)
272 struct osp_xattr_entry *oxe;
273 struct osp_xattr_entry *tmp = NULL;
274 size_t namelen = strlen(name);
275 size_t size = sizeof(*oxe) + namelen + 1 +
276 (len ? len : OXE_DEFAULT_LEN);
278 oxe = osp_oac_xattr_find(obj, name, false);
282 OBD_ALLOC_LARGE(oxe, size);
286 INIT_LIST_HEAD(&oxe->oxe_list);
287 oxe->oxe_buflen = size;
288 oxe->oxe_namelen = namelen;
289 memcpy(oxe->oxe_buf, name, namelen);
290 oxe->oxe_value = oxe->oxe_buf + namelen + 1;
291 /* One ref is for the caller, the other is for the entry on the list. */
292 atomic_set(&oxe->oxe_ref, 2);
294 spin_lock(&obj->opo_lock);
295 tmp = osp_oac_xattr_find_locked(obj, name, namelen);
297 list_add_tail(&oxe->oxe_list, &obj->opo_xattr_list);
299 atomic_inc(&tmp->oxe_ref);
300 spin_unlock(&obj->opo_lock);
303 OBD_FREE_LARGE(oxe, size);
311 * Assign the cached OST-object's EA with the given value.
313 * If the current EA entry in cache has not enough space to hold the new
314 * value, remove it, create a new one, then assign with the given value.
316 * \param[in] obj pointer to the OSP object
317 * \param[in] oxe pointer to the cached EA entry to be assigned
318 * \param[in] buf pointer to the buffer with new EA value
320 * \retval pointer to the new created EA entry in cache if
321 * current entry is not big enough; otherwise, the
322 * input 'oxe' will be returned.
324 static struct osp_xattr_entry *
325 osp_oac_xattr_assignment(struct osp_object *obj, struct osp_xattr_entry *oxe,
326 const struct lu_buf *buf)
328 struct osp_xattr_entry *new = NULL;
329 struct osp_xattr_entry *old = NULL;
330 int namelen = oxe->oxe_namelen;
331 size_t size = sizeof(*oxe) + namelen + 1 + buf->lb_len;
332 bool unlink_only = false;
334 if (oxe->oxe_buflen < size) {
335 OBD_ALLOC_LARGE(new, size);
337 INIT_LIST_HEAD(&new->oxe_list);
338 new->oxe_buflen = size;
339 new->oxe_namelen = namelen;
340 memcpy(new->oxe_buf, oxe->oxe_buf, namelen);
341 new->oxe_value = new->oxe_buf + namelen + 1;
342 /* One ref is for the caller,
343 * the other is for the entry on the list. */
344 atomic_set(&new->oxe_ref, 2);
345 __osp_oac_xattr_assignment(obj, new, buf);
348 CWARN("%s: cannot update cached xattr %.*s of "DFID"\n",
349 osp_dto2name(obj), namelen, oxe->oxe_buf,
350 PFID(lu_object_fid(&obj->opo_obj.do_lu)));
354 spin_lock(&obj->opo_lock);
355 old = osp_oac_xattr_find_locked(obj, oxe->oxe_buf, namelen);
358 /* Unlink the 'old'. */
359 list_del_init(&old->oxe_list);
361 /* Drop the ref for 'old' on list. */
362 osp_oac_xattr_put(old);
364 /* Drop the ref for current using. */
365 osp_oac_xattr_put(oxe);
368 /* Insert 'new' into list. */
369 list_add_tail(&new->oxe_list, &obj->opo_xattr_list);
370 } else if (unlink_only) {
371 /* Unlink the 'old'. */
372 list_del_init(&old->oxe_list);
374 /* Drop the ref for 'old' on list. */
375 osp_oac_xattr_put(old);
377 __osp_oac_xattr_assignment(obj, oxe, buf);
380 /* Drop the ref for current using. */
381 osp_oac_xattr_put(oxe);
384 /* Someone unlinked the 'old' by race,
385 * insert the 'new' one into list. */
386 list_add_tail(&new->oxe_list, &obj->opo_xattr_list);
388 spin_unlock(&obj->opo_lock);
394 * Parse the OSP object attribute from the RPC reply.
396 * If the attribute is valid, then it will be added to the OSP object
399 * \param[in] env pointer to the thread context
400 * \param[in] reply pointer to the RPC reply
401 * \param[in] req pointer to the RPC request
402 * \param[out] attr pointer to buffer to hold the output attribute
403 * \param[in] obj pointer to the OSP object
404 * \param[in] index the index of the attribute buffer in the reply
406 * \retval 0 for success
407 * \retval negative error number on failure
409 static int osp_get_attr_from_reply(const struct lu_env *env,
410 struct object_update_reply *reply,
411 struct ptlrpc_request *req,
412 struct lu_attr *attr,
413 struct osp_object *obj, int index)
415 struct osp_thread_info *osi = osp_env_info(env);
416 struct lu_buf *rbuf = &osi->osi_lb2;
417 struct obdo *lobdo = &osi->osi_obdo;
421 rc = object_update_result_data_get(reply, rbuf, index);
425 wobdo = rbuf->lb_buf;
426 if (rbuf->lb_len != sizeof(*wobdo))
429 LASSERT(req != NULL);
430 if (ptlrpc_req_need_swab(req))
431 lustre_swab_obdo(wobdo);
433 lustre_get_wire_obdo(NULL, lobdo, wobdo);
435 spin_lock(&obj->opo_lock);
436 la_from_obdo(&obj->opo_attr, lobdo, lobdo->o_valid);
437 spin_unlock(&obj->opo_lock);
440 la_from_obdo(attr, lobdo, lobdo->o_valid);
446 * Interpreter function for getting OSP object attribute asynchronously.
448 * Called to interpret the result of an async mode RPC for getting the
449 * OSP object attribute.
451 * \param[in] env pointer to the thread context
452 * \param[in] reply pointer to the RPC reply
453 * \param[in] req pointer to the RPC request
454 * \param[in] obj pointer to the OSP object
455 * \param[out] data pointer to buffer to hold the output attribute
456 * \param[in] index the index of the attribute buffer in the reply
457 * \param[in] rc the result for handling the RPC
459 * \retval 0 for success
460 * \retval negative error number on failure
462 static int osp_attr_get_interpterer(const struct lu_env *env,
463 struct object_update_reply *reply,
464 struct ptlrpc_request *req,
465 struct osp_object *obj,
466 void *data, int index, int rc)
468 struct lu_attr *attr = data;
471 osp2lu_obj(obj)->lo_header->loh_attr |= LOHA_EXISTS;
472 obj->opo_non_exist = 0;
474 return osp_get_attr_from_reply(env, reply, req, NULL, obj,
478 osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS;
479 obj->opo_non_exist = 1;
482 spin_lock(&obj->opo_lock);
484 spin_unlock(&obj->opo_lock);
491 * Implement OSP layer dt_object_operations::do_declare_attr_get() interface.
493 * Declare that the caller will get attribute from the specified OST object.
495 * This function adds an Object Unified Target (OUT) sub-request to the per-OSP
496 * based shared asynchronous request queue. The osp_attr_get_interpterer()
497 * is registered as the interpreter function to handle the result of this
500 * \param[in] env pointer to the thread context
501 * \param[in] dt pointer to the OSP layer dt_object
503 * \retval 0 for success
504 * \retval negative error number on failure
506 static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt)
508 struct osp_object *obj = dt2osp_obj(dt);
509 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
512 mutex_lock(&osp->opd_async_requests_mutex);
513 rc = osp_insert_async_request(env, OUT_ATTR_GET, obj, 0, NULL, NULL,
514 &obj->opo_attr, sizeof(struct obdo),
515 osp_attr_get_interpterer);
516 mutex_unlock(&osp->opd_async_requests_mutex);
522 * Implement OSP layer dt_object_operations::do_attr_get() interface.
524 * Get attribute from the specified MDT/OST object.
526 * If the attribute is in the OSP object attributes cache, then return
527 * the cached attribute directly. Otherwise it will trigger an OUT RPC
528 * to the peer to get the attribute synchronously, if successful, add it
529 * to the OSP attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.)
531 * \param[in] env pointer to the thread context
532 * \param[in] dt pointer to the OSP layer dt_object
533 * \param[out] attr pointer to the buffer to hold the output attribute
535 * \retval 0 for success
536 * \retval negative error number on failure
538 int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
539 struct lu_attr *attr)
541 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
542 struct osp_object *obj = dt2osp_obj(dt);
543 struct dt_device *dev = &osp->opd_dt_dev;
544 struct osp_update_request *update;
545 struct object_update_reply *reply;
546 struct ptlrpc_request *req = NULL;
547 int invalidated, cache = 0, rc = 0;
550 if (is_ost_obj(&dt->do_lu) && obj->opo_non_exist)
553 spin_lock(&obj->opo_lock);
554 if (obj->opo_attr.la_valid != 0 && !obj->opo_stale) {
555 *attr = obj->opo_attr;
556 spin_unlock(&obj->opo_lock);
560 spin_unlock(&obj->opo_lock);
562 update = osp_update_request_create(dev);
564 RETURN(PTR_ERR(update));
566 rc = OSP_UPDATE_RPC_PACK(env, out_attr_get_pack, update,
567 lu_object_fid(&dt->do_lu));
569 CERROR("%s: Insert update error "DFID": rc = %d\n",
570 dev->dd_lu_dev.ld_obd->obd_name,
571 PFID(lu_object_fid(&dt->do_lu)), rc);
576 invalidated = atomic_read(&obj->opo_invalidate_seq);
578 rc = osp_remote_sync(env, osp, update, &req);
580 down_read(&obj->opo_invalidate_sem);
581 if (invalidated == atomic_read(&obj->opo_invalidate_seq)) {
582 /* no invalited has came so far, we can cache the attrs */
588 osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS;
590 obj->opo_non_exist = 1;
592 CERROR("%s: osp_attr_get update error "DFID": rc = %d\n",
593 dev->dd_lu_dev.ld_obd->obd_name,
594 PFID(lu_object_fid(&dt->do_lu)), rc);
600 osp2lu_obj(obj)->lo_header->loh_attr |= LOHA_EXISTS;
601 obj->opo_non_exist = 0;
602 reply = req_capsule_server_sized_get(&req->rq_pill,
603 &RMF_OUT_UPDATE_REPLY,
604 OUT_UPDATE_REPLY_SIZE);
605 if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC)
606 GOTO(out, rc = -EPROTO);
608 rc = osp_get_attr_from_reply(env, reply, req, attr,
609 cache ? obj : NULL, 0);
613 spin_lock(&obj->opo_lock);
616 spin_unlock(&obj->opo_lock);
621 up_read(&obj->opo_invalidate_sem);
625 ptlrpc_req_finished(req);
627 osp_update_request_destroy(env, update);
633 * Implement OSP layer dt_object_operations::do_declare_attr_set() interface.
635 * If the transaction is not remote one, then declare the credits that will
636 * be used for the subsequent llog record for the object's attributes.
638 * \param[in] env pointer to the thread context
639 * \param[in] dt pointer to the OSP layer dt_object
640 * \param[in] attr pointer to the attribute to be set
641 * \param[in] th pointer to the transaction handler
643 * \retval 0 for success
644 * \retval negative error number on failure
646 static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
647 const struct lu_attr *attr, struct thandle *th)
649 struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
650 struct osp_object *o = dt2osp_obj(dt);
653 if (is_only_remote_trans(th))
654 return osp_md_declare_attr_set(env, dt, attr, th);
656 * Usually we don't allow server stack to manipulate size
657 * but there is a special case when striping is created
658 * late, after stripeless file got truncated to non-zero.
660 * In this case we do the following:
662 * 1) grab id in declare - this can lead to leaked OST objects
663 * but we don't currently have proper mechanism and the only
664 * options we have are to do truncate RPC holding transaction
665 * open (very bad) or to grab id in declare at cost of leaked
666 * OST object in same very rare unfortunate case (just bad)
667 * notice 1.6-2.0 do assignment outside of running transaction
668 * all the time, meaning many more chances for leaked objects.
670 * 2) send synchronous truncate RPC with just assigned id
673 /* there are few places in MDD code still passing NULL
674 * XXX: to be fixed soon */
678 if (attr->la_valid & LA_SIZE && attr->la_size > 0 &&
679 fid_is_zero(lu_object_fid(&o->opo_obj.do_lu))) {
680 LASSERT(!dt_object_exists(dt));
681 osp_object_assign_fid(env, d, o);
682 rc = osp_object_truncate(env, dt, attr->la_size);
687 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
690 /* track all UID/GID, projid, and layout version changes via llog */
691 rc = osp_sync_declare_add(env, o, MDS_SETATTR64_REC, th);
697 * Implement OSP layer dt_object_operations::do_attr_set() interface.
699 * Set attribute to the specified OST object.
701 * If the transaction is a remote one, then add OUT_ATTR_SET sub-request
702 * in the OUT RPC that will be flushed when the remote transaction stop.
703 * Otherwise, it will generate a MDS_SETATTR64_REC record in the llog that
704 * will be handled by a dedicated thread asynchronously.
706 * If the attribute entry exists in the OSP object attributes cache,
707 * then update the cached attribute according to given attribute.
709 * \param[in] env pointer to the thread context
710 * \param[in] dt pointer to the OSP layer dt_object
711 * \param[in] attr pointer to the attribute to be set
712 * \param[in] th pointer to the transaction handler
714 * \retval 0 for success
715 * \retval negative error number on failure
717 static int osp_attr_set(const struct lu_env *env, struct dt_object *dt,
718 const struct lu_attr *attr, struct thandle *th)
720 struct osp_object *o = dt2osp_obj(dt);
724 /* we're interested in uid/gid/projid/layout version changes only */
725 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
728 if (!is_only_remote_trans(th)) {
729 if (attr->la_flags & LUSTRE_SET_SYNC_FL) {
730 struct ptlrpc_request *req = NULL;
731 struct osp_update_request *update = NULL;
732 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
734 update = osp_update_request_create(&osp->opd_dt_dev);
736 RETURN(PTR_ERR(update));
738 rc = OSP_UPDATE_RPC_PACK(env, out_attr_set_pack, update,
739 lu_object_fid(&dt->do_lu),
742 CERROR("%s: update error "DFID": rc = %d\n",
743 osp->opd_obd->obd_name,
744 PFID(lu_object_fid(&dt->do_lu)), rc);
746 osp_update_request_destroy(env, update);
750 rc = osp_remote_sync(env, osp, update, &req);
752 ptlrpc_req_finished(req);
754 osp_update_request_destroy(env, update);
756 rc = osp_sync_add(env, o, MDS_SETATTR64_REC, th, attr);
757 /* XXX: send new uid/gid to OST ASAP? */
762 /* It is for OST-object attr_set directly without updating
763 * local MDT-object attribute. It is usually used by LFSCK. */
764 rc = osp_md_attr_set(env, dt, attr, th);
765 CDEBUG(D_INFO, "(1) set attr "DFID": rc = %d\n",
766 PFID(&dt->do_lu.lo_header->loh_fid), rc);
771 /* Update the OSP object attributes cache. */
773 spin_lock(&o->opo_lock);
774 if (attr->la_valid & LA_UID) {
775 la->la_uid = attr->la_uid;
776 la->la_valid |= LA_UID;
779 if (attr->la_valid & LA_GID) {
780 la->la_gid = attr->la_gid;
781 la->la_valid |= LA_GID;
783 if (attr->la_valid & LA_PROJID) {
784 la->la_projid = attr->la_projid;
785 la->la_valid |= LA_PROJID;
787 spin_unlock(&o->opo_lock);
794 * Interpreter function for getting OSP object extended attribute asynchronously
796 * Called to interpret the result of an async mode RPC for getting the
797 * OSP object extended attribute.
799 * \param[in] env pointer to the thread context
800 * \param[in] reply pointer to the RPC reply
801 * \param[in] req pointer to the RPC request
802 * \param[in] obj pointer to the OSP object
803 * \param[out] data pointer to OSP object attributes cache
804 * \param[in] index the index of the attribute buffer in the reply
805 * \param[in] rc the result for handling the RPC
807 * \retval 0 for success
808 * \retval negative error number on failure
810 static int osp_xattr_get_interpterer(const struct lu_env *env,
811 struct object_update_reply *reply,
812 struct ptlrpc_request *req,
813 struct osp_object *obj,
814 void *data, int index, int rc)
816 struct osp_xattr_entry *oxe = data;
818 spin_lock(&obj->opo_lock);
820 struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2;
821 size_t len = sizeof(*oxe) + oxe->oxe_namelen + 1;
823 rc = object_update_result_data_get(reply, rbuf, index);
824 if (rc == -ENOENT || rc == -ENODATA || rc == 0) {
830 if (unlikely(rc < 0) ||
831 rbuf->lb_len > (oxe->oxe_buflen - len)) {
836 __osp_oac_xattr_assignment(obj, oxe, rbuf);
837 } else if (rc == -ENOENT || rc == -ENODATA) {
845 spin_unlock(&obj->opo_lock);
847 /* Put the reference obtained in the osp_declare_xattr_get(). */
848 osp_oac_xattr_put(oxe);
854 * Implement OSP dt_object_operations::do_declare_xattr_get() interface.
856 * Declare that the caller will get extended attribute from the specified
859 * This function will add an OUT_XATTR_GET sub-request to the per OSP
860 * based shared asynchronous request queue with the interpreter function:
861 * osp_xattr_get_interpterer().
863 * \param[in] env pointer to the thread context
864 * \param[in] dt pointer to the OSP layer dt_object
865 * \param[out] buf pointer to the lu_buf to hold the extended attribute
866 * \param[in] name the name for the expected extended attribute
868 * \retval 0 for success
869 * \retval negative error number on failure
871 static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt,
872 struct lu_buf *buf, const char *name)
874 struct osp_object *obj = dt2osp_obj(dt);
875 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
876 struct osp_xattr_entry *oxe;
880 LASSERT(buf != NULL);
881 LASSERT(name != NULL);
883 if (unlikely(buf->lb_len == 0))
886 oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
890 len = strlen(name) + 1;
891 mutex_lock(&osp->opd_async_requests_mutex);
892 rc = osp_insert_async_request(env, OUT_XATTR_GET, obj, 1,
893 &len, (const void **)&name,
895 osp_xattr_get_interpterer);
897 mutex_unlock(&osp->opd_async_requests_mutex);
898 osp_oac_xattr_put(oxe);
900 struct osp_update_request *our;
901 struct osp_update_request_sub *ours;
903 /* XXX: Currently, we trigger the batched async OUT
904 * RPC via dt_declare_xattr_get(). It is not
905 * perfect solution, but works well now.
907 * We will improve it in the future. */
908 our = osp->opd_async_requests;
909 ours = osp_current_object_update_request(our);
910 if (ours != NULL && ours->ours_req != NULL &&
911 ours->ours_req->ourq_count > 0) {
912 osp->opd_async_requests = NULL;
913 mutex_unlock(&osp->opd_async_requests_mutex);
914 rc = osp_unplug_async_request(env, osp, our);
916 mutex_unlock(&osp->opd_async_requests_mutex);
924 * Implement OSP layer dt_object_operations::do_xattr_get() interface.
926 * Get extended attribute from the specified MDT/OST object.
928 * If the extended attribute is in the OSP object attributes cache, then
929 * return the cached extended attribute directly. Otherwise it will get
930 * the extended attribute synchronously, if successful, add it to the OSP
931 * attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.)
933 * There is a race condition: some other thread has added the named extended
934 * attributed entry to the OSP object attributes cache during the current
935 * OUT_XATTR_GET handling. If such case happens, the OSP will replace the
936 * (just) existing extended attribute entry with the new replied one.
938 * \param[in] env pointer to the thread context
939 * \param[in] dt pointer to the OSP layer dt_object
940 * \param[out] buf pointer to the lu_buf to hold the extended attribute
941 * \param[in] name the name for the expected extended attribute
943 * \retval 0 for success
944 * \retval negative error number on failure
946 int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
947 struct lu_buf *buf, const char *name)
949 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
950 struct osp_object *obj = dt2osp_obj(dt);
951 struct dt_device *dev = &osp->opd_dt_dev;
952 struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2;
953 struct osp_update_request *update = NULL;
954 struct ptlrpc_request *req = NULL;
955 struct object_update_reply *reply;
956 struct osp_xattr_entry *oxe = NULL;
957 const char *dname = osp_dto2name(obj);
958 int invalidated, rc = 0;
961 LASSERT(buf != NULL);
962 LASSERT(name != NULL);
964 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NETWORK) &&
965 osp->opd_index == cfs_fail_val) {
966 if (is_ost_obj(&dt->do_lu)) {
967 if (osp_dev2node(osp) == cfs_fail_val)
970 if (strcmp(name, XATTR_NAME_LINK) == 0)
975 if (unlikely(obj->opo_non_exist))
978 invalidated = atomic_read(&obj->opo_invalidate_seq);
980 oxe = osp_oac_xattr_find(obj, name, false);
982 spin_lock(&obj->opo_lock);
983 if (oxe->oxe_ready) {
985 GOTO(unlock, rc = -ENODATA);
987 if (buf->lb_buf == NULL)
988 GOTO(unlock, rc = oxe->oxe_vallen);
990 if (buf->lb_len < oxe->oxe_vallen)
991 GOTO(unlock, rc = -ERANGE);
993 memcpy(buf->lb_buf, oxe->oxe_value,
996 GOTO(unlock, rc = oxe->oxe_vallen);
999 spin_unlock(&obj->opo_lock);
1000 osp_oac_xattr_put(oxe);
1004 spin_unlock(&obj->opo_lock);
1006 update = osp_update_request_create(dev);
1008 GOTO(out_req, rc = PTR_ERR(update));
1010 rc = OSP_UPDATE_RPC_PACK(env, out_xattr_get_pack, update,
1011 lu_object_fid(&dt->do_lu), name, buf->lb_len);
1013 CERROR("%s: Insert update error "DFID": rc = %d\n",
1014 dname, PFID(lu_object_fid(&dt->do_lu)), rc);
1018 rc = osp_remote_sync(env, osp, update, &req);
1020 down_read(&obj->opo_invalidate_sem);
1021 if (invalidated != atomic_read(&obj->opo_invalidate_seq)) {
1022 /* invalidated has been requested, we can't cache the result */
1025 dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
1028 reply = req_capsule_server_sized_get(&req->rq_pill,
1029 &RMF_OUT_UPDATE_REPLY,
1030 OUT_UPDATE_REPLY_SIZE);
1031 if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
1032 CERROR("%s: Wrong version %x expected %x "DFID
1033 ": rc = %d\n", dname, reply->ourp_magic,
1035 PFID(lu_object_fid(&dt->do_lu)), -EPROTO);
1036 GOTO(out, rc = -EPROTO);
1038 rc = object_update_result_data_get(reply, rbuf, 0);
1043 if (rc == -ENOENT) {
1044 dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
1045 obj->opo_non_exist = 1;
1049 oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
1052 CWARN("%s: Fail to add xattr (%s) to cache for "
1053 DFID" (1): rc = %d\n", dname, name,
1054 PFID(lu_object_fid(&dt->do_lu)), rc);
1059 spin_lock(&obj->opo_lock);
1060 if (rc == -ENOENT || rc == -ENODATA) {
1066 spin_unlock(&obj->opo_lock);
1071 reply = req_capsule_server_sized_get(&req->rq_pill,
1072 &RMF_OUT_UPDATE_REPLY,
1073 OUT_UPDATE_REPLY_SIZE);
1074 if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
1075 CERROR("%s: Wrong version %x expected %x "DFID": rc = %d\n",
1076 dname, reply->ourp_magic, UPDATE_REPLY_MAGIC,
1077 PFID(lu_object_fid(&dt->do_lu)), -EPROTO);
1079 GOTO(out, rc = -EPROTO);
1082 rc = object_update_result_data_get(reply, rbuf, 0);
1083 if (rc < 0 || rbuf->lb_len == 0) {
1084 if (oxe == NULL && rc == -ENODATA) {
1085 oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
1088 CWARN("%s: Fail to add xattr (%s) to cache for "
1089 DFID" (1): rc = %d\n", dname, name,
1090 PFID(lu_object_fid(&dt->do_lu)), rc);
1096 spin_lock(&obj->opo_lock);
1097 if (unlikely(rc == -ENODATA)) {
1103 spin_unlock(&obj->opo_lock);
1109 /* For detecting EA size. */
1114 oxe = osp_oac_xattr_find_or_add(obj, name, rbuf->lb_len);
1116 CWARN("%s: Fail to add xattr (%s) to "
1117 "cache for "DFID" (2): rc = %d\n",
1118 dname, name, PFID(lu_object_fid(&dt->do_lu)), rc);
1124 oxe = osp_oac_xattr_assignment(obj, oxe, rbuf);
1129 up_read(&obj->opo_invalidate_sem);
1132 if (rc > 0 && buf->lb_buf) {
1133 if (unlikely(buf->lb_len < rbuf->lb_len))
1136 memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len);
1140 ptlrpc_req_finished(req);
1142 if (update && !IS_ERR(update))
1143 osp_update_request_destroy(env, update);
1146 osp_oac_xattr_put(oxe);
1152 * Implement OSP layer dt_object_operations::do_declare_xattr_set() interface.
1154 * Declare that the caller will set extended attribute to the specified
1157 * If it is non-remote transaction, it will add an OUT_XATTR_SET sub-request
1158 * to the OUT RPC that will be flushed when the transaction start. And if the
1159 * OSP attributes cache is initialized, then check whether the name extended
1160 * attribute entry exists in the cache or not. If yes, replace it; otherwise,
1161 * add the extended attribute to the cache.
1163 * \param[in] env pointer to the thread context
1164 * \param[in] dt pointer to the OSP layer dt_object
1165 * \param[in] buf pointer to the lu_buf to hold the extended attribute
1166 * \param[in] name the name of the extended attribute to be set
1167 * \param[in] flag to indicate the detailed set operation: LU_XATTR_CREATE
1168 * or LU_XATTR_REPLACE or others
1169 * \param[in] th pointer to the transaction handler
1171 * \retval 0 for success
1172 * \retval negative error number on failure
1174 int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
1175 const struct lu_buf *buf, const char *name,
1176 int flag, struct thandle *th)
1178 return osp_trans_update_request_create(th);
1182 * Implement OSP layer dt_object_operations::do_xattr_set() interface.
1184 * Set extended attribute to the specified MDT/OST object.
1186 * Add an OUT_XATTR_SET sub-request into the OUT RPC that will be flushed in
1187 * the transaction stop. And if the OSP attributes cache is initialized, then
1188 * check whether the name extended attribute entry exists in the cache or not.
1189 * If yes, replace it; otherwise, add the extended attribute to the cache.
1191 * \param[in] env pointer to the thread context
1192 * \param[in] dt pointer to the OSP layer dt_object
1193 * \param[in] buf pointer to the lu_buf to hold the extended attribute
1194 * \param[in] name the name of the extended attribute to be set
1195 * \param[in] fl to indicate the detailed set operation: LU_XATTR_CREATE
1196 * or LU_XATTR_REPLACE or others
1197 * \param[in] th pointer to the transaction handler
1199 * \retval 0 for success
1200 * \retval negative error number on failure
1202 int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
1203 const struct lu_buf *buf, const char *name, int fl,
1206 struct osp_object *o = dt2osp_obj(dt);
1207 struct osp_update_request *update;
1208 struct osp_xattr_entry *oxe;
1212 update = thandle_to_osp_update_request(th);
1213 LASSERT(update != NULL);
1215 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
1216 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
1218 rc = OSP_UPDATE_RPC_PACK(env, out_xattr_set_pack, update,
1219 lu_object_fid(&dt->do_lu), buf, name, fl);
1223 /* Do not cache linkEA that may be self-adjusted by peers
1224 * under EA overflow case. */
1225 if (strcmp(name, XATTR_NAME_LINK) == 0) {
1226 oxe = osp_oac_xattr_find(o, name, true);
1228 osp_oac_xattr_put(oxe);
1233 oxe = osp_oac_xattr_find_or_add(o, name, buf->lb_len);
1235 CWARN("%s: cannot cache xattr '%s' of "DFID"\n",
1236 osp_dto2name(o), name, PFID(lu_object_fid(&dt->do_lu)));
1241 oxe = osp_oac_xattr_assignment(o, oxe, buf);
1243 osp_oac_xattr_put(oxe);
1249 * Implement OSP layer dt_object_operations::do_declare_xattr_del() interface.
1251 * Declare that the caller will delete extended attribute on the specified
1254 * If it is non-remote transaction, it will add an OUT_XATTR_DEL sub-request
1255 * to the OUT RPC that will be flushed when the transaction start. And if the
1256 * name extended attribute entry exists in the OSP attributes cache, then remove
1257 * it from the cache.
1259 * \param[in] env pointer to the thread context
1260 * \param[in] dt pointer to the OSP layer dt_object
1261 * \param[in] name the name of the extended attribute to be set
1262 * \param[in] th pointer to the transaction handler
1264 * \retval 0 for success
1265 * \retval negative error number on failure
1267 int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
1268 const char *name, struct thandle *th)
1270 return osp_trans_update_request_create(th);
1274 * Implement OSP layer dt_object_operations::do_xattr_del() interface.
1276 * Delete extended attribute on the specified MDT/OST object.
1278 * If it is remote transaction, it will add an OUT_XATTR_DEL sub-request into
1279 * the OUT RPC that will be flushed when the transaction stop. And if the name
1280 * extended attribute entry exists in the OSP attributes cache, then remove it
1283 * \param[in] env pointer to the thread context
1284 * \param[in] dt pointer to the OSP layer dt_object
1285 * \param[in] name the name of the extended attribute to be set
1286 * \param[in] th pointer to the transaction handler
1288 * \retval 0 for success
1289 * \retval negative error number on failure
1291 int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
1292 const char *name, struct thandle *th)
1294 struct osp_update_request *update;
1295 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1296 struct osp_object *o = dt2osp_obj(dt);
1297 struct osp_xattr_entry *oxe;
1300 update = thandle_to_osp_update_request(th);
1301 LASSERT(update != NULL);
1303 rc = OSP_UPDATE_RPC_PACK(env, out_xattr_del_pack, update, fid, name);
1307 oxe = osp_oac_xattr_find(o, name, true);
1309 /* Drop the ref for entry on list. */
1310 osp_oac_xattr_put(oxe);
1315 void osp_obj_invalidate_cache(struct osp_object *obj)
1317 struct osp_xattr_entry *oxe;
1318 struct osp_xattr_entry *tmp;
1320 spin_lock(&obj->opo_lock);
1321 list_for_each_entry_safe(oxe, tmp, &obj->opo_xattr_list, oxe_list) {
1323 list_del_init(&oxe->oxe_list);
1324 osp_oac_xattr_put(oxe);
1326 obj->opo_attr.la_valid = 0;
1327 spin_unlock(&obj->opo_lock);
1331 * Implement OSP layer dt_object_operations::do_invalidate() interface.
1333 * Invalidate attributes cached on the specified MDT/OST object.
1335 * \param[in] env pointer to the thread context
1336 * \param[in] dt pointer to the OSP layer dt_object
1338 * \retval 0 for success
1339 * \retval negative error number on failure
1341 int osp_invalidate(const struct lu_env *env, struct dt_object *dt)
1343 struct osp_object *obj = dt2osp_obj(dt);
1346 CDEBUG(D_HA, "Invalidate osp_object "DFID"\n",
1347 PFID(lu_object_fid(&dt->do_lu)));
1349 /* serialize attr/EA set vs. invalidation */
1350 down_write(&obj->opo_invalidate_sem);
1352 /* this should invalidate all in-flights */
1353 atomic_inc(&obj->opo_invalidate_seq);
1355 spin_lock(&obj->opo_lock);
1356 /* do not mark new objects stale */
1357 if (obj->opo_attr.la_valid)
1359 obj->opo_non_exist = 0;
1360 spin_unlock(&obj->opo_lock);
1362 osp_obj_invalidate_cache(obj);
1364 up_write(&obj->opo_invalidate_sem);
1369 bool osp_check_stale(struct dt_object *dt)
1371 struct osp_object *obj = dt2osp_obj(dt);
1373 if (is_ost_obj(&dt->do_lu) && obj->opo_non_exist)
1376 return obj->opo_stale;
1381 * Implement OSP layer dt_object_operations::do_declare_create() interface.
1383 * Declare that the caller will create the OST object.
1385 * If the transaction is a remote transaction and the FID for the OST-object
1386 * has been assigned already, then handle it as creating (remote) MDT object
1387 * via osp_md_declare_create(). This function is usually used for LFSCK
1388 * to re-create the lost OST object. Otherwise, if it is not replay case, the
1389 * OSP will reserve pre-created object for the subsequent create operation;
1390 * if the MDT side cached pre-created objects are less than some threshold,
1391 * then it will wakeup the pre-create thread.
1393 * \param[in] env pointer to the thread context
1394 * \param[in] dt pointer to the OSP layer dt_object
1395 * \param[in] attr the attribute for the object to be created
1396 * \param[in] hint pointer to the hint for creating the object, such as
1398 * \param[in] dof pointer to the dt_object_format for help the creation
1399 * \param[in] th pointer to the transaction handler
1401 * \retval 0 for success
1402 * \retval negative error number on failure
1404 static int osp_declare_create(const struct lu_env *env, struct dt_object *dt,
1405 struct lu_attr *attr,
1406 struct dt_allocation_hint *hint,
1407 struct dt_object_format *dof, struct thandle *th)
1409 struct osp_thread_info *osi = osp_env_info(env);
1410 struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
1411 struct osp_object *o = dt2osp_obj(dt);
1412 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1413 struct thandle *local_th;
1418 if (is_only_remote_trans(th) && !fid_is_zero(fid)) {
1419 LASSERT(fid_is_sane(fid));
1421 rc = osp_md_declare_create(env, dt, attr, hint, dof, th);
1426 /* should happen to non-0 OSP only so that at least one object
1427 * has been already declared in the scenario and LOD should
1429 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OSC_CREATE_FAIL) && d->opd_index == 1)
1432 LASSERT(d->opd_last_used_oid_file);
1435 * There can be gaps in precreated ids and record to unlink llog
1436 * XXX: we do not handle gaps yet, implemented before solution
1437 * was found to be racy, so we disabled that. there is no
1438 * point in making useless but expensive llog declaration.
1440 /* rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th); */
1442 local_th = osp_get_storage_thandle(env, th, d);
1443 if (IS_ERR(local_th))
1444 RETURN(PTR_ERR(local_th));
1446 if (unlikely(!fid_is_zero(fid))) {
1447 /* replay case: caller knows fid */
1448 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, NULL,
1450 rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
1451 &osi->osi_lb, osi->osi_off,
1457 * in declaration we need to reserve object so that we don't block
1458 * awaiting precreation RPC to complete
1460 rc = osp_precreate_reserve(env, d, !hint || hint->dah_can_block);
1462 * we also need to declare update to local "last used id" file for
1463 * recovery if object isn't used for a reason, we need to release
1464 * reservation, this can be made in osd_object_release()
1467 /* mark id is reserved: in create we don't want to talk
1469 LASSERT(o->opo_reserved == 0);
1470 o->opo_reserved = 1;
1472 /* common for all OSPs file hystorically */
1473 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, NULL,
1475 rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
1476 &osi->osi_lb, osi->osi_off,
1479 /* not needed in the cache anymore */
1480 set_bit(LU_OBJECT_HEARD_BANSHEE,
1481 &dt->do_lu.lo_header->loh_flags);
1487 * Implement OSP layer dt_object_operations::do_create() interface.
1489 * Create the OST object.
1491 * If the transaction is a remote transaction and the FID for the OST-object
1492 * has been assigned already, then handle it as handling MDT object via the
1493 * osp_md_create(). For other cases, the OSP will assign FID to the
1494 * object to be created, and update last_used Object ID (OID) file.
1496 * \param[in] env pointer to the thread context
1497 * \param[in] dt pointer to the OSP layer dt_object
1498 * \param[in] attr the attribute for the object to be created
1499 * \param[in] hint pointer to the hint for creating the object, such as
1501 * \param[in] dof pointer to the dt_object_format for help the creation
1502 * \param[in] th pointer to the transaction handler
1504 * \retval 0 for success
1505 * \retval negative error number on failure
1507 static int osp_create(const struct lu_env *env, struct dt_object *dt,
1508 struct lu_attr *attr, struct dt_allocation_hint *hint,
1509 struct dt_object_format *dof, struct thandle *th)
1511 struct osp_thread_info *osi = osp_env_info(env);
1512 struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
1513 struct osp_object *o = dt2osp_obj(dt);
1515 struct lu_fid *fid = &osi->osi_fid;
1516 struct thandle *local_th;
1519 if (is_only_remote_trans(th) &&
1520 !fid_is_zero(lu_object_fid(&dt->do_lu))) {
1521 LASSERT(fid_is_sane(lu_object_fid(&dt->do_lu)));
1523 rc = osp_md_create(env, dt, attr, hint, dof, th);
1525 o->opo_non_exist = 0;
1530 o->opo_non_exist = 0;
1531 if (o->opo_reserved) {
1532 /* regular case, fid is assigned holding transaction open */
1533 osp_object_assign_fid(env, d, o);
1536 memcpy(fid, lu_object_fid(&dt->do_lu), sizeof(*fid));
1538 LASSERTF(fid_is_sane(fid), "fid for osp_object %p is insane"DFID"!\n",
1541 if (!o->opo_reserved) {
1542 /* special case, id was assigned outside of transaction
1543 * see comments in osp_declare_attr_set */
1544 LASSERT(d->opd_pre != NULL);
1545 spin_lock(&d->opd_pre_lock);
1546 osp_update_last_fid(d, fid);
1547 spin_unlock(&d->opd_pre_lock);
1550 CDEBUG(D_INODE, "fid for osp_object %p is "DFID"\n", o, PFID(fid));
1552 /* If the precreate ends, it means it will be ready to rollover to
1553 * the new sequence soon, all the creation should be synchronized,
1554 * otherwise during replay, the replay fid will be inconsistent with
1555 * last_used/create fid */
1556 if (osp_precreate_end_seq(env, d) && osp_is_fid_client(d))
1559 local_th = osp_get_storage_thandle(env, th, d);
1560 if (IS_ERR(local_th))
1561 RETURN(PTR_ERR(local_th));
1563 * it's OK if the import is inactive by this moment - id was created
1564 * by OST earlier, we just need to maintain it consistently on the disk
1565 * once import is reconnected, OSP will claim this and other objects
1566 * used and OST either keep them, if they exist or recreate
1569 /* we might have lost precreated objects */
1570 if (unlikely(d->opd_gap_count) > 0) {
1571 LASSERT(d->opd_pre != NULL);
1572 spin_lock(&d->opd_pre_lock);
1573 if (d->opd_gap_count > 0) {
1574 int count = d->opd_gap_count;
1576 rc = ostid_set_id(&osi->osi_oi,
1577 fid_oid(&d->opd_gap_start_fid));
1579 spin_unlock(&d->opd_pre_lock);
1582 d->opd_gap_count = 0;
1583 spin_unlock(&d->opd_pre_lock);
1585 CDEBUG(D_HA, "Writing gap "DFID"+%d in llog\n",
1586 PFID(&d->opd_gap_start_fid), count);
1587 /* real gap handling is disabled intil ORI-692 will be
1588 * fixed, now we only report gaps */
1590 spin_unlock(&d->opd_pre_lock);
1594 /* Only need update last_used oid file, seq file will only be update
1595 * during seq rollover */
1596 osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off,
1597 &d->opd_last_id, d->opd_index);
1599 rc = dt_record_write(env, d->opd_last_used_oid_file, &osi->osi_lb,
1600 &osi->osi_off, local_th);
1602 CDEBUG(D_HA, "%s: Wrote last used FID: "DFID", index %d: %d\n",
1603 d->opd_obd->obd_name, PFID(fid), d->opd_index, rc);
1609 * Implement OSP layer dt_object_operations::do_declare_destroy() interface.
1611 * Declare that the caller will destroy the specified OST object.
1613 * The OST object destroy will be handled via llog asynchronously. This
1614 * function will declare the credits for generating MDS_UNLINK64_REC llog.
1616 * \param[in] env pointer to the thread context
1617 * \param[in] dt pointer to the OSP layer dt_object to be destroyed
1618 * \param[in] th pointer to the transaction handler
1620 * \retval 0 for success
1621 * \retval negative error number on failure
1623 int osp_declare_destroy(const struct lu_env *env, struct dt_object *dt,
1626 struct osp_object *o = dt2osp_obj(dt);
1627 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
1632 LASSERT(!osp->opd_connect_mdt);
1634 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
1635 rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th);
1641 * Implement OSP layer dt_object_operations::do_destroy() interface.
1643 * Destroy the specified OST object.
1645 * The OSP generates a MDS_UNLINK64_REC record in the llog. There
1646 * will be some dedicated thread to handle the llog asynchronously.
1648 * It also marks the object as non-cached.
1650 * \param[in] env pointer to the thread context
1651 * \param[in] dt pointer to the OSP layer dt_object to be destroyed
1652 * \param[in] th pointer to the transaction handler
1654 * \retval 0 for success
1655 * \retval negative error number on failure
1657 static int osp_destroy(const struct lu_env *env, struct dt_object *dt,
1660 struct osp_object *o = dt2osp_obj(dt);
1661 struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
1666 o->opo_non_exist = 1;
1668 LASSERT(!osp->opd_connect_mdt);
1670 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ)) {
1671 /* once transaction is committed put proper command on
1672 * the queue going to our OST. */
1673 rc = osp_sync_add(env, o, MDS_UNLINK64_REC, th, NULL);
1678 /* not needed in cache any more */
1679 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
1684 static int osp_orphan_index_lookup(const struct lu_env *env,
1685 struct dt_object *dt,
1687 const struct dt_key *key)
1692 static int osp_orphan_index_declare_insert(const struct lu_env *env,
1693 struct dt_object *dt,
1694 const struct dt_rec *rec,
1695 const struct dt_key *key,
1696 struct thandle *handle)
1701 static int osp_orphan_index_insert(const struct lu_env *env,
1702 struct dt_object *dt,
1703 const struct dt_rec *rec,
1704 const struct dt_key *key,
1705 struct thandle *handle)
1710 static int osp_orphan_index_declare_delete(const struct lu_env *env,
1711 struct dt_object *dt,
1712 const struct dt_key *key,
1713 struct thandle *handle)
1718 static int osp_orphan_index_delete(const struct lu_env *env,
1719 struct dt_object *dt,
1720 const struct dt_key *key,
1721 struct thandle *handle)
1727 * Initialize the OSP layer index iteration.
1729 * \param[in] env pointer to the thread context
1730 * \param[in] dt pointer to the index object to be iterated
1731 * \param[in] attr unused
1733 * \retval pointer to the iteration structure
1734 * \retval negative error number on failure
1736 struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
1743 return ERR_PTR(-ENOMEM);
1745 it->ooi_pos_ent = -1;
1747 it->ooi_attr = attr;
1749 return (struct dt_it *)it;
1753 * Finalize the OSP layer index iteration.
1755 * \param[in] env pointer to the thread context
1756 * \param[in] di pointer to the iteration structure
1758 void osp_it_fini(const struct lu_env *env, struct dt_it *di)
1760 struct osp_it *it = (struct osp_it *)di;
1761 struct page **pages = it->ooi_pages;
1762 int npages = it->ooi_total_npages;
1765 if (pages != NULL) {
1766 for (i = 0; i < npages; i++) {
1767 if (pages[i] != NULL) {
1768 if (pages[i] == it->ooi_cur_page) {
1770 it->ooi_cur_page = NULL;
1772 __free_page(pages[i]);
1775 OBD_FREE_PTR_ARRAY(pages, npages);
1781 * Get more records for the iteration from peer.
1783 * The new records will be filled in an array of pages. The OSP side
1784 * allows 1MB bulk data to be transferred.
1786 * \param[in] env pointer to the thread context
1787 * \param[in] it pointer to the iteration structure
1789 * \retval 0 for success
1790 * \retval negative error number on failure
1792 static int osp_it_fetch(const struct lu_env *env, struct osp_it *it)
1794 struct lu_device *dev = it->ooi_obj->do_lu.lo_dev;
1795 struct osp_device *osp = lu2osp_dev(dev);
1796 struct page **pages;
1797 struct ptlrpc_request *req = NULL;
1798 struct ptlrpc_bulk_desc *desc;
1799 struct idx_info *ii;
1806 npages = min_t(unsigned int, OFD_MAX_BRW_SIZE, 1 << 20);
1807 npages /= PAGE_SIZE;
1809 OBD_ALLOC_PTR_ARRAY(pages, npages);
1813 it->ooi_pages = pages;
1814 it->ooi_total_npages = npages;
1815 for (i = 0; i < npages; i++) {
1816 pages[i] = alloc_page(GFP_NOFS);
1817 if (pages[i] == NULL)
1821 req = ptlrpc_request_alloc(osp->opd_obd->u.cli.cl_import,
1826 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, OBD_IDX_READ);
1828 ptlrpc_request_free(req);
1832 osp_set_req_replay(osp, req);
1833 req->rq_request_portal = OUT_PORTAL;
1834 ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO);
1835 memset(ii, 0, sizeof(*ii));
1836 if (fid_is_last_id(lu_object_fid(&it->ooi_obj->do_lu))) {
1837 /* LFSCK will iterate orphan object[FID_SEQ_LAYOUT_BTREE,
1838 * ost_index, 0] with LAST_ID FID, so it needs to replace
1839 * the FID with orphan FID here */
1840 ii->ii_fid.f_seq = FID_SEQ_LAYOUT_RBTREE;
1841 ii->ii_fid.f_oid = osp->opd_index;
1842 ii->ii_fid.f_ver = 0;
1843 ii->ii_flags = II_FL_NOHASH;
1844 ii->ii_attrs = osp_dev2node(osp);
1846 ii->ii_fid = *lu_object_fid(&it->ooi_obj->do_lu);
1847 ii->ii_flags = II_FL_NOHASH | II_FL_NOKEY | II_FL_VARKEY |
1849 ii->ii_attrs = it->ooi_attr;
1851 ii->ii_magic = IDX_INFO_MAGIC;
1852 ii->ii_count = npages * LU_PAGE_COUNT;
1853 ii->ii_hash_start = it->ooi_next;
1855 ptlrpc_at_set_req_timeout(req);
1857 desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1858 PTLRPC_BULK_PUT_SINK,
1860 &ptlrpc_bulk_kiov_pin_ops);
1862 GOTO(out, rc = -ENOMEM);
1864 for (i = 0; i < npages; i++)
1865 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1868 ptlrpc_request_set_replen(req);
1869 rc = ptlrpc_queue_wait(req);
1873 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1874 req->rq_bulk->bd_nob_transferred);
1879 ii = req_capsule_server_get(&req->rq_pill, &RMF_IDX_INFO);
1880 if (ii->ii_magic != IDX_INFO_MAGIC)
1881 GOTO(out, rc = -EPROTO);
1883 npages = (ii->ii_count + LU_PAGE_COUNT - 1) >>
1884 (PAGE_SHIFT - LU_PAGE_SHIFT);
1885 if (npages > it->ooi_total_npages) {
1886 CERROR("%s: returned more pages than expected, %u > %u\n",
1887 osp->opd_obd->obd_name, npages, it->ooi_total_npages);
1888 GOTO(out, rc = -EINVAL);
1891 it->ooi_rec_size = ii->ii_recsize;
1892 it->ooi_valid_npages = npages;
1893 if (ptlrpc_rep_need_swab(req))
1896 it->ooi_next = ii->ii_hash_end;
1899 ptlrpc_req_finished(req);
1905 * Move the iteration cursor to the next lu_page.
1907 * One system page (PAGE_SIZE) may contain multiple lu_page (4KB),
1908 * that depends on the LU_PAGE_COUNT. If it is not the last lu_page
1909 * in current system page, then move the iteration cursor to the next
1910 * lu_page in current system page. Otherwise, if there are more system
1911 * pages in the cache, then move the iteration cursor to the next system
1912 * page. If all the cached records (pages) have been iterated, then fetch
1913 * more records via osp_it_fetch().
1915 * \param[in] env pointer to the thread context
1916 * \param[in] di pointer to the iteration structure
1918 * \retval positive for end of the directory
1919 * \retval 0 for success
1920 * \retval negative error number on failure
1922 int osp_it_next_page(const struct lu_env *env, struct dt_it *di)
1924 struct osp_it *it = (struct osp_it *)di;
1925 struct lu_idxpage *idxpage;
1926 struct page **pages;
1932 idxpage = it->ooi_cur_idxpage;
1933 if (idxpage != NULL) {
1934 if (idxpage->lip_nr == 0)
1937 if (it->ooi_pos_ent < idxpage->lip_nr) {
1938 CDEBUG(D_INFO, "ooi_pos %d nr %d\n",
1939 (int)it->ooi_pos_ent, (int)idxpage->lip_nr);
1942 it->ooi_cur_idxpage = NULL;
1943 it->ooi_pos_lu_page++;
1946 if (it->ooi_pos_lu_page < LU_PAGE_COUNT) {
1947 it->ooi_cur_idxpage = (void *)it->ooi_cur_page +
1948 LU_PAGE_SIZE * it->ooi_pos_lu_page;
1950 lustre_swab_lip_header(it->ooi_cur_idxpage);
1951 if (it->ooi_cur_idxpage->lip_magic != LIP_MAGIC) {
1952 struct osp_device *osp =
1953 lu2osp_dev(it->ooi_obj->do_lu.lo_dev);
1955 CERROR("%s: invalid magic (%x != %x) for page "
1956 "%d/%d while read layout orphan index\n",
1957 osp->opd_obd->obd_name,
1958 it->ooi_cur_idxpage->lip_magic,
1959 LIP_MAGIC, it->ooi_pos_page,
1960 it->ooi_pos_lu_page);
1961 /* Skip this lu_page next time. */
1962 it->ooi_pos_ent = idxpage->lip_nr - 1;
1965 it->ooi_pos_ent = -1;
1969 kunmap(it->ooi_cur_page);
1970 it->ooi_cur_page = NULL;
1974 pages = it->ooi_pages;
1975 if (it->ooi_pos_page < it->ooi_valid_npages) {
1976 it->ooi_cur_page = kmap(pages[it->ooi_pos_page]);
1977 it->ooi_pos_lu_page = 0;
1981 for (i = 0; i < it->ooi_total_npages; i++) {
1982 if (pages[i] != NULL)
1983 __free_page(pages[i]);
1985 OBD_FREE_PTR_ARRAY(pages, it->ooi_total_npages);
1987 it->ooi_pos_page = 0;
1988 it->ooi_total_npages = 0;
1989 it->ooi_valid_npages = 0;
1992 it->ooi_cur_page = NULL;
1993 it->ooi_cur_idxpage = NULL;
1994 it->ooi_pages = NULL;
1997 if (it->ooi_next == II_END_OFF)
2000 rc = osp_it_fetch(env, it);
2008 * Move the iteration cursor to the next record.
2010 * If there are more records in the lu_page, then move the iteration
2011 * cursor to the next record directly. Otherwise, move the iteration
2012 * cursor to the record in the next lu_page via osp_it_next_page()
2014 * \param[in] env pointer to the thread context
2015 * \param[in] di pointer to the iteration structure
2017 * \retval positive for end of the directory
2018 * \retval 0 for success
2019 * \retval negative error number on failure
2021 static int osp_orphan_it_next(const struct lu_env *env, struct dt_it *di)
2023 struct osp_it *it = (struct osp_it *)di;
2024 struct lu_idxpage *idxpage;
2029 idxpage = it->ooi_cur_idxpage;
2030 if (idxpage != NULL) {
2031 if (idxpage->lip_nr == 0)
2035 if (it->ooi_pos_ent < idxpage->lip_nr) {
2036 if (it->ooi_rec_size ==
2037 sizeof(struct lu_orphan_rec_v3)) {
2039 (struct lu_orphan_ent_v3 *)idxpage->lip_entries+
2042 lustre_swab_orphan_ent_v3(it->ooi_ent);
2043 } else if (it->ooi_rec_size ==
2044 sizeof(struct lu_orphan_rec_v2)) {
2046 (struct lu_orphan_ent_v2 *)idxpage->lip_entries+
2049 lustre_swab_orphan_ent_v2(it->ooi_ent);
2052 (struct lu_orphan_ent *)idxpage->lip_entries +
2055 lustre_swab_orphan_ent(it->ooi_ent);
2061 rc = osp_it_next_page(env, di);
2068 int osp_it_get(const struct lu_env *env, struct dt_it *di,
2069 const struct dt_key *key)
2074 void osp_it_put(const struct lu_env *env, struct dt_it *di)
2078 static struct dt_key *osp_orphan_it_key(const struct lu_env *env,
2079 const struct dt_it *di)
2081 struct osp_it *it = (struct osp_it *)di;
2082 struct lu_orphan_ent *ent = (struct lu_orphan_ent *)it->ooi_ent;
2084 if (likely(ent != NULL))
2085 return (struct dt_key *)(&ent->loe_key);
2090 static int osp_orphan_it_key_size(const struct lu_env *env,
2091 const struct dt_it *di)
2093 return sizeof(struct lu_fid);
2096 static int osp_orphan_it_rec(const struct lu_env *env, const struct dt_it *di,
2097 struct dt_rec *rec, __u32 attr)
2099 struct osp_it *it = (struct osp_it *)di;
2101 if (likely(it->ooi_ent)) {
2102 if (it->ooi_rec_size == sizeof(struct lu_orphan_rec_v3)) {
2103 struct lu_orphan_ent_v3 *ent =
2104 (struct lu_orphan_ent_v3 *)it->ooi_ent;
2106 *(struct lu_orphan_rec_v3 *)rec = ent->loe_rec;
2107 } else if (it->ooi_rec_size ==
2108 sizeof(struct lu_orphan_rec_v2)) {
2109 struct lu_orphan_ent_v2 *ent =
2110 (struct lu_orphan_ent_v2 *)it->ooi_ent;
2112 *(struct lu_orphan_rec_v2 *)rec = ent->loe_rec;
2114 struct lu_orphan_ent *ent =
2115 (struct lu_orphan_ent *)it->ooi_ent;
2117 *(struct lu_orphan_rec *)rec = ent->loe_rec;
2125 __u64 osp_it_store(const struct lu_env *env, const struct dt_it *di)
2127 struct osp_it *it = (struct osp_it *)di;
2129 return it->ooi_next;
2133 * Locate the iteration cursor to the specified position (cookie).
2135 * \param[in] env pointer to the thread context
2136 * \param[in] di pointer to the iteration structure
2137 * \param[in] hash the specified position
2139 * \retval positive number for locating to the exactly position
2141 * \retval 0 for arriving at the end of the iteration
2142 * \retval negative error number on failure
2144 int osp_orphan_it_load(const struct lu_env *env, const struct dt_it *di,
2147 struct osp_it *it = (struct osp_it *)di;
2150 it->ooi_next = hash;
2151 rc = osp_orphan_it_next(env, (struct dt_it *)di);
2161 int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
2167 static const struct dt_index_operations osp_orphan_index_ops = {
2168 .dio_lookup = osp_orphan_index_lookup,
2169 .dio_declare_insert = osp_orphan_index_declare_insert,
2170 .dio_insert = osp_orphan_index_insert,
2171 .dio_declare_delete = osp_orphan_index_declare_delete,
2172 .dio_delete = osp_orphan_index_delete,
2174 .init = osp_it_init,
2175 .fini = osp_it_fini,
2176 .next = osp_orphan_it_next,
2179 .key = osp_orphan_it_key,
2180 .key_size = osp_orphan_it_key_size,
2181 .rec = osp_orphan_it_rec,
2182 .store = osp_it_store,
2183 .load = osp_orphan_it_load,
2184 .key_rec = osp_it_key_rec,
2189 * Implement OSP layer dt_object_operations::do_index_try() interface.
2191 * Negotiate the index type.
2193 * If the target index is an IDIF object, then use osp_orphan_index_ops.
2194 * Otherwise, assign osp_md_index_ops to the dt_object::do_index_ops.
2195 * (\see lustre/include/lustre_fid.h for IDIF.)
2197 * \param[in] env pointer to the thread context
2198 * \param[in] dt pointer to the OSP layer dt_object
2199 * \param[in] feat unused
2201 * \retval 0 for success
2203 static int osp_index_try(const struct lu_env *env,
2204 struct dt_object *dt,
2205 const struct dt_index_features *feat)
2207 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2209 if (fid_is_last_id(fid) && fid_is_idif(fid))
2210 dt->do_index_ops = &osp_orphan_index_ops;
2212 dt->do_index_ops = &osp_md_index_ops;
2216 static const struct dt_object_operations osp_obj_ops = {
2217 .do_declare_attr_get = osp_declare_attr_get,
2218 .do_attr_get = osp_attr_get,
2219 .do_declare_attr_set = osp_declare_attr_set,
2220 .do_attr_set = osp_attr_set,
2221 .do_declare_xattr_get = osp_declare_xattr_get,
2222 .do_xattr_get = osp_xattr_get,
2223 .do_declare_xattr_set = osp_declare_xattr_set,
2224 .do_xattr_set = osp_xattr_set,
2225 .do_declare_create = osp_declare_create,
2226 .do_create = osp_create,
2227 .do_declare_destroy = osp_declare_destroy,
2228 .do_destroy = osp_destroy,
2229 .do_index_try = osp_index_try,
2233 * Implement OSP layer lu_object_operations::loo_object_init() interface.
2235 * Initialize the object.
2237 * If it is a remote MDT object, then call do_attr_get() to fetch
2238 * the attribute from the peer.
2240 * \param[in] env pointer to the thread context
2241 * \param[in] o pointer to the OSP layer lu_object
2242 * \param[in] conf unused
2244 * \retval 0 for success
2245 * \retval negative error number on failure
2247 static int osp_object_init(const struct lu_env *env, struct lu_object *o,
2248 const struct lu_object_conf *conf)
2250 struct osp_object *po = lu2osp_obj(o);
2254 spin_lock_init(&po->opo_lock);
2255 o->lo_header->loh_attr |= LOHA_REMOTE;
2256 INIT_LIST_HEAD(&po->opo_xattr_list);
2257 INIT_LIST_HEAD(&po->opo_invalidate_cb_list);
2258 init_rwsem(&po->opo_invalidate_sem);
2260 if (is_ost_obj(o)) {
2261 po->opo_obj.do_ops = &osp_obj_ops;
2263 struct lu_attr *la = &osp_env_info(env)->osi_attr;
2265 po->opo_obj.do_ops = &osp_md_obj_ops;
2266 po->opo_obj.do_body_ops = &osp_md_body_ops;
2268 if (conf != NULL && conf->loc_flags & LOC_F_NEW) {
2269 po->opo_non_exist = 1;
2271 rc = po->opo_obj.do_ops->do_attr_get(env, lu2dt_obj(o),
2274 o->lo_header->loh_attr |=
2275 LOHA_EXISTS | (la->la_mode & S_IFMT);
2276 if (rc == -ENOENT) {
2277 po->opo_non_exist = 1;
2281 init_rwsem(&po->opo_sem);
2286 static void osp_object_free_rcu(struct rcu_head *head)
2288 struct osp_object *obj = container_of(head, struct osp_object,
2289 opo_header.loh_rcu);
2291 kmem_cache_free(osp_object_kmem, obj);
2295 * Implement OSP layer lu_object_operations::loo_object_free() interface.
2297 * Finalize the object.
2299 * If the OSP object has attributes cache, then destroy the cache.
2300 * Free the object finally.
2302 * \param[in] env pointer to the thread context
2303 * \param[in] o pointer to the OSP layer lu_object
2305 static void osp_object_free(const struct lu_env *env, struct lu_object *o)
2307 struct osp_object *obj = lu2osp_obj(o);
2308 struct lu_object_header *h = o->lo_header;
2309 struct osp_xattr_entry *oxe;
2310 struct osp_xattr_entry *tmp;
2313 dt_object_fini(&obj->opo_obj);
2314 lu_object_header_fini(h);
2315 list_for_each_entry_safe(oxe, tmp, &obj->opo_xattr_list, oxe_list) {
2316 list_del(&oxe->oxe_list);
2317 count = atomic_read(&oxe->oxe_ref);
2318 LASSERTF(count == 1,
2319 "Still has %d users on the xattr entry %.*s\n",
2320 count-1, (int)oxe->oxe_namelen, oxe->oxe_buf);
2322 OBD_FREE_LARGE(oxe, oxe->oxe_buflen);
2324 OBD_FREE_PRE(obj, sizeof(*obj), "slab-freed");
2325 call_rcu(&obj->opo_header.loh_rcu, osp_object_free_rcu);
2329 * Implement OSP layer lu_object_operations::loo_object_release() interface.
2331 * Cleanup (not free) the object.
2333 * If it is a reserved object but failed to be created, or it is an OST
2334 * object, then mark the object as non-cached.
2336 * \param[in] env pointer to the thread context
2337 * \param[in] o pointer to the OSP layer lu_object
2339 static void osp_object_release(const struct lu_env *env, struct lu_object *o)
2341 struct osp_object *po = lu2osp_obj(o);
2342 struct osp_device *d = lu2osp_dev(o->lo_dev);
2347 * release reservation if object was declared but not created
2348 * this may require lu_object_put() in LOD
2350 if (unlikely(po->opo_reserved)) {
2351 LASSERT(d->opd_pre != NULL);
2352 LASSERT(d->opd_pre_reserved > 0);
2353 spin_lock(&d->opd_pre_lock);
2354 d->opd_pre_reserved--;
2355 spin_unlock(&d->opd_pre_lock);
2358 * Check that osp_precreate_cleanup_orphans is not blocked
2359 * due to opd_pre_reserved > 0.
2361 if (unlikely(d->opd_pre_reserved == 0 &&
2362 (d->opd_pre_recovering || d->opd_pre_status)))
2363 wake_up(&d->opd_pre_waitq);
2365 /* not needed in cache any more */
2366 set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
2370 /* XXX: Currently, NOT cache OST-object on MDT because:
2371 * 1. it is not often accessed on MDT.
2372 * 2. avoid up layer (such as LFSCK) to load too many
2373 * once-used OST-objects. */
2374 set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
2379 static int osp_object_print(const struct lu_env *env, void *cookie,
2380 lu_printer_t p, const struct lu_object *l)
2382 const struct osp_object *o = lu2osp_obj((struct lu_object *)l);
2384 return (*p)(env, cookie, LUSTRE_OSP_NAME"-object@%p", o);
2387 static int osp_object_invariant(const struct lu_object *o)
2392 const struct lu_object_operations osp_lu_obj_ops = {
2393 .loo_object_init = osp_object_init,
2394 .loo_object_free = osp_object_free,
2395 .loo_object_release = osp_object_release,
2396 .loo_object_print = osp_object_print,
2397 .loo_object_invariant = osp_object_invariant