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) 2014, Intel Corporation.
26 * lustre/osp/osp_trans.c
29 * 1. OSP (Object Storage Proxy) transaction methods
31 * Implement OSP layer transaction related interfaces for the dt_device API
32 * dt_device_operations.
35 * 2. Handle asynchronous idempotent operations
37 * The OSP uses OUT (Object Unified Target) RPC to talk with other server
38 * (MDT or OST) for kinds of operations, such as create, unlink, insert,
39 * delete, lookup, set_(x)attr, get_(x)attr, and etc. To reduce the number
40 * of RPCs, we allow multiple operations to be packaged together in single
43 * For the asynchronous idempotent operations, such as get_(x)attr, related
44 * RPCs will be inserted into an osp_device based shared asynchronous request
45 * queue - osp_device::opd_async_requests. When the queue is full, all the
46 * requests in the queue will be packaged into a single OUT RPC and given to
47 * the ptlrpcd daemon (for sending), then the queue is purged and other new
48 * requests can be inserted into it.
50 * When the asynchronous idempotent operation inserts the request into the
51 * shared queue, it will register an interpreter. When the packaged OUT RPC
52 * is replied (or failed to be sent out), all the registered interpreters
53 * will be called one by one to handle each own result.
56 * There are three kinds of transactions
58 * 1. Local transaction, all of updates of the transaction are in the local MDT.
59 * 2. Remote transaction, all of updates of the transaction are in one remote
60 * MDT, which only happens in LFSCK now.
61 * 3. Distribute transaction, updates for the transaction are in mulitple MDTs.
63 * Author: Di Wang <di.wang@intel.com>
64 * Author: Fan, Yong <fan.yong@intel.com>
67 #define DEBUG_SUBSYSTEM S_MDS
69 #include <lustre_net.h>
70 #include "osp_internal.h"
73 * The argument for the interpreter callback of osp request.
75 struct osp_update_args {
76 struct osp_update_request *oaua_update;
78 wait_queue_head_t *oaua_waitq;
79 bool oaua_flow_control;
83 * Call back for each update request.
85 struct osp_update_callback {
86 /* list in the osp_update_request::our_cb_items */
87 struct list_head ouc_list;
89 /* The target of the async update request. */
90 struct osp_object *ouc_obj;
92 /* The data used by or_interpreter. */
95 /* The interpreter function called after the async request handled. */
96 osp_update_interpreter_t ouc_interpreter;
99 static struct object_update_request *object_update_request_alloc(size_t size)
101 struct object_update_request *ourq;
103 OBD_ALLOC_LARGE(ourq, size);
105 return ERR_PTR(-ENOMEM);
107 ourq->ourq_magic = UPDATE_REQUEST_MAGIC;
108 ourq->ourq_count = 0;
114 * Allocate new update request
116 * Allocate new update request and insert it to the req_update_list.
118 * \param [in] our osp_udate_request where to create a new
121 * \retval 0 if creation succeeds.
122 * \retval negative errno if creation fails.
124 int osp_object_update_request_create(struct osp_update_request *our,
127 struct osp_update_request_sub *ours;
133 if (size < OUT_UPDATE_INIT_BUFFER_SIZE)
134 size = OUT_UPDATE_INIT_BUFFER_SIZE;
136 ours->ours_req = object_update_request_alloc(size);
138 if (IS_ERR(ours->ours_req)) {
143 ours->ours_req_size = size;
144 INIT_LIST_HEAD(&ours->ours_list);
145 list_add_tail(&ours->ours_list, &our->our_req_list);
151 * Get current update request
153 * Get current object update request from our_req_list in
154 * osp_update_request, because we always insert the new update
155 * request in the last position, so the last update request
156 * in the list will be the current update req.
158 * \param[in] our osp update request where to get the
159 * current object update.
161 * \retval the current object update.
163 struct osp_update_request_sub *
164 osp_current_object_update_request(struct osp_update_request *our)
166 if (list_empty(&our->our_req_list))
169 return list_entry(our->our_req_list.prev, struct osp_update_request_sub,
174 * Allocate and initialize osp_update_request
176 * osp_update_request is being used to track updates being executed on
177 * this dt_device(OSD or OSP). The update buffer will be 4k initially,
178 * and increased if needed.
180 * \param [in] dt dt device
182 * \retval osp_update_request being allocated if succeed
183 * \retval ERR_PTR(errno) if failed
185 struct osp_update_request *osp_update_request_create(struct dt_device *dt)
187 struct osp_update_request *our;
191 return ERR_PTR(-ENOMEM);
193 INIT_LIST_HEAD(&our->our_req_list);
194 INIT_LIST_HEAD(&our->our_cb_items);
195 INIT_LIST_HEAD(&our->our_list);
197 osp_object_update_request_create(our, OUT_UPDATE_INIT_BUFFER_SIZE);
201 void osp_update_request_destroy(struct osp_update_request *our)
203 struct osp_update_request_sub *ours;
204 struct osp_update_request_sub *tmp;
209 list_for_each_entry_safe(ours, tmp, &our->our_req_list, ours_list) {
210 list_del(&ours->ours_list);
211 if (ours->ours_req != NULL)
212 OBD_FREE(ours->ours_req, ours->ours_req_size);
219 object_update_request_dump(const struct object_update_request *ourq,
223 size_t total_size = 0;
225 for (i = 0; i < ourq->ourq_count; i++) {
226 struct object_update *update;
229 update = object_update_request_get(ourq, i, &size);
230 LASSERT(update != NULL);
231 CDEBUG(mask, "i = %u fid = "DFID" op = %s master = %u"
232 "params = %d batchid = "LPU64" size = %zu\n",
233 i, PFID(&update->ou_fid),
234 update_op_str(update->ou_type),
235 update->ou_master_index, update->ou_params_count,
236 update->ou_batchid, size);
241 CDEBUG(mask, "updates = %p magic = %x count = %d size = %zu\n", ourq,
242 ourq->ourq_magic, ourq->ourq_count, total_size);
246 * Prepare inline update request
248 * Prepare OUT update ptlrpc inline request, and the request usually includes
249 * one update buffer, which does not need bulk transfer.
251 * \param[in] env execution environment
252 * \param[in] req ptlrpc request
253 * \param[in] ours sub osp_update_request to be packed
255 * \retval 0 if packing succeeds
256 * \retval negative errno if packing fails
258 int osp_prep_inline_update_req(const struct lu_env *env,
259 struct ptlrpc_request *req,
260 struct osp_update_request_sub *ours)
262 struct out_update_header *ouh;
263 __u32 update_req_size = object_update_request_size(ours->ours_req);
266 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_HEADER, RCL_CLIENT,
267 update_req_size + sizeof(*ouh));
269 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, OUT_UPDATE);
273 ouh = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_HEADER);
274 ouh->ouh_magic = OUT_UPDATE_HEADER_MAGIC;
276 ouh->ouh_inline_length = update_req_size;
278 memcpy(ouh->ouh_inline_data, ours->ours_req, update_req_size);
280 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_REPLY,
281 RCL_SERVER, OUT_UPDATE_REPLY_SIZE);
283 ptlrpc_request_set_replen(req);
284 req->rq_request_portal = OUT_PORTAL;
285 req->rq_reply_portal = OSC_REPLY_PORTAL;
291 * Prepare update request.
293 * Prepare OUT update ptlrpc request, and the request usually includes
294 * all of updates (stored in \param ureq) from one operation.
296 * \param[in] env execution environment
297 * \param[in] imp import on which ptlrpc request will be sent
298 * \param[in] ureq hold all of updates which will be packed into the req
299 * \param[in] reqp request to be created
301 * \retval 0 if preparation succeeds.
302 * \retval negative errno if preparation fails.
304 int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
305 struct osp_update_request *our,
306 struct ptlrpc_request **reqp)
308 struct ptlrpc_request *req;
309 struct ptlrpc_bulk_desc *desc;
310 struct osp_update_request_sub *ours;
311 struct out_update_header *ouh;
312 struct out_update_buffer *oub;
317 list_for_each_entry(ours, &our->our_req_list, ours_list) {
318 object_update_request_dump(ours->ours_req, D_INFO);
321 LASSERT(buf_count > 0);
323 req = ptlrpc_request_alloc(imp, &RQF_OUT_UPDATE);
327 if (buf_count == 1) {
328 ours = list_entry(our->our_req_list.next,
329 struct osp_update_request_sub, ours_list);
331 /* Let's check if it can be packed inline */
332 if (object_update_request_size(ours->ours_req) +
333 sizeof(struct out_update_header) <
334 OUT_UPDATE_MAX_INLINE_SIZE) {
335 rc = osp_prep_inline_update_req(env, req, ours);
342 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_HEADER, RCL_CLIENT,
343 sizeof(struct osp_update_request));
345 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_BUF, RCL_CLIENT,
346 buf_count * sizeof(*oub));
348 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, OUT_UPDATE);
352 ouh = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_HEADER);
353 ouh->ouh_magic = OUT_UPDATE_HEADER_MAGIC;
354 ouh->ouh_count = buf_count;
355 ouh->ouh_inline_length = 0;
356 oub = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_BUF);
357 list_for_each_entry(ours, &our->our_req_list, ours_list) {
358 oub->oub_size = ours->ours_req_size;
362 req->rq_bulk_write = 1;
363 desc = ptlrpc_prep_bulk_imp(req, buf_count,
364 MD_MAX_BRW_SIZE >> LNET_MTU_BITS,
365 PTLRPC_BULK_GET_SOURCE | PTLRPC_BULK_BUF_KVEC,
366 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
368 GOTO(out_req, rc = -ENOMEM);
370 /* NB req now owns desc and will free it when it gets freed */
371 list_for_each_entry(ours, &our->our_req_list, ours_list)
372 desc->bd_frag_ops->add_iov_frag(desc, ours->ours_req,
373 ours->ours_req_size);
375 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_REPLY,
376 RCL_SERVER, OUT_UPDATE_REPLY_SIZE);
378 ptlrpc_request_set_replen(req);
379 req->rq_request_portal = OUT_PORTAL;
380 req->rq_reply_portal = OSC_REPLY_PORTAL;
385 ptlrpc_req_finished(req);
393 * Send update request to the remote MDT synchronously.
395 * \param[in] env execution environment
396 * \param[in] imp import on which ptlrpc request will be sent
397 * \param[in] our hold all of updates which will be packed into the req
398 * \param[in] reqp request to be created
400 * \retval 0 if RPC succeeds.
401 * \retval negative errno if RPC fails.
403 int osp_remote_sync(const struct lu_env *env, struct osp_device *osp,
404 struct osp_update_request *our,
405 struct ptlrpc_request **reqp)
407 struct obd_import *imp = osp->opd_obd->u.cli.cl_import;
408 struct ptlrpc_request *req = NULL;
412 rc = osp_prep_update_req(env, imp, our, &req);
416 /* This will only be called with read-only update, and these updates
417 * might be used to retrieve update log during recovery process, so
418 * it will be allowed to send during recovery process */
419 req->rq_allow_replay = 1;
421 /* Note: some dt index api might return non-zero result here, like
422 * osd_index_ea_lookup, so we should only check rc < 0 here */
423 rc = ptlrpc_queue_wait(req);
425 if (rc < 0 || reqp == NULL)
426 ptlrpc_req_finished(req);
433 static void osp_trans_stop_cb(struct osp_thandle *oth, int result)
435 struct dt_txn_commit_cb *dcb;
436 struct dt_txn_commit_cb *tmp;
438 /* call per-transaction stop callbacks if any */
439 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
441 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
442 "commit callback entry: magic=%x name='%s'\n",
443 dcb->dcb_magic, dcb->dcb_name);
444 list_del_init(&dcb->dcb_linkage);
445 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
450 * Allocate an osp request and initialize it with the given parameters.
452 * \param[in] obj pointer to the operation target
453 * \param[in] data pointer to the data used by the interpreter
454 * \param[in] interpreter pointer to the interpreter function
456 * \retval pointer to the asychronous request
457 * \retval NULL if the allocation failed
459 static struct osp_update_callback *
460 osp_update_callback_init(struct osp_object *obj, void *data,
461 osp_update_interpreter_t interpreter)
463 struct osp_update_callback *ouc;
469 lu_object_get(osp2lu_obj(obj));
470 INIT_LIST_HEAD(&ouc->ouc_list);
472 ouc->ouc_data = data;
473 ouc->ouc_interpreter = interpreter;
479 * Destroy the osp_update_callback.
481 * \param[in] env pointer to the thread context
482 * \param[in] ouc pointer to osp_update_callback
484 static void osp_update_callback_fini(const struct lu_env *env,
485 struct osp_update_callback *ouc)
487 LASSERT(list_empty(&ouc->ouc_list));
489 lu_object_put(env, osp2lu_obj(ouc->ouc_obj));
494 * Interpret the packaged OUT RPC results.
496 * For every packaged sub-request, call its registered interpreter function.
497 * Then destroy the sub-request.
499 * \param[in] env pointer to the thread context
500 * \param[in] req pointer to the RPC
501 * \param[in] arg pointer to data used by the interpreter
502 * \param[in] rc the RPC return value
504 * \retval 0 for success
505 * \retval negative error number on failure
507 static int osp_update_interpret(const struct lu_env *env,
508 struct ptlrpc_request *req, void *arg, int rc)
510 struct object_update_reply *reply = NULL;
511 struct osp_update_args *oaua = arg;
512 struct osp_update_request *our = oaua->oaua_update;
513 struct osp_thandle *oth;
514 struct osp_update_callback *ouc;
515 struct osp_update_callback *next;
525 oaua->oaua_update = NULL;
527 if (oaua->oaua_flow_control) {
528 struct osp_device *osp;
530 LASSERT(oth != NULL);
531 osp = dt2osp_dev(oth->ot_super.th_dev);
532 obd_put_request_slot(&osp->opd_obd->u.cli);
535 /* Unpack the results from the reply message. */
536 if (req->rq_repmsg != NULL) {
537 reply = req_capsule_server_sized_get(&req->rq_pill,
538 &RMF_OUT_UPDATE_REPLY,
539 OUT_UPDATE_REPLY_SIZE);
540 if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC)
543 count = reply->ourp_count;
548 list_for_each_entry_safe(ouc, next, &our->our_cb_items, ouc_list) {
549 list_del_init(&ouc->ouc_list);
551 /* The peer may only have handled some requests (indicated
552 * by the 'count') in the packaged OUT RPC, we can only get
553 * results for the handled part. */
554 if (index < count && reply->ourp_lens[index] > 0) {
555 struct object_update_result *result;
557 result = object_update_result_get(reply, index, NULL);
561 rc1 = result->our_rc;
564 if (unlikely(rc1 == 0))
568 if (ouc->ouc_interpreter != NULL)
569 ouc->ouc_interpreter(env, reply, req, ouc->ouc_obj,
570 ouc->ouc_data, index, rc1);
572 osp_update_callback_fini(env, ouc);
576 if (oaua->oaua_count != NULL && atomic_dec_and_test(oaua->oaua_count))
577 wake_up_all(oaua->oaua_waitq);
580 /* oth and osp_update_requests will be destoryed in
582 osp_trans_stop_cb(oth, rc);
583 osp_thandle_put(oth);
585 osp_update_request_destroy(our);
592 * Pack all the requests in the shared asynchronous idempotent request queue
593 * into a single OUT RPC that will be given to the background ptlrpcd daemon.
595 * \param[in] env pointer to the thread context
596 * \param[in] osp pointer to the OSP device
597 * \param[in] our pointer to the shared queue
599 * \retval 0 for success
600 * \retval negative error number on failure
602 int osp_unplug_async_request(const struct lu_env *env,
603 struct osp_device *osp,
604 struct osp_update_request *our)
606 struct osp_update_args *args;
607 struct ptlrpc_request *req = NULL;
610 rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import,
613 struct osp_update_callback *ouc;
614 struct osp_update_callback *next;
616 list_for_each_entry_safe(ouc, next,
617 &our->our_cb_items, ouc_list) {
618 list_del_init(&ouc->ouc_list);
619 if (ouc->ouc_interpreter != NULL)
620 ouc->ouc_interpreter(env, NULL, NULL,
622 ouc->ouc_data, 0, rc);
623 osp_update_callback_fini(env, ouc);
625 osp_update_request_destroy(our);
627 args = ptlrpc_req_async_args(req);
628 args->oaua_update = our;
629 args->oaua_count = NULL;
630 args->oaua_waitq = NULL;
631 args->oaua_flow_control = false;
632 req->rq_interpret_reply = osp_update_interpret;
633 ptlrpcd_add_req(req);
640 * Find or create (if NOT exist or purged) the shared asynchronous idempotent
641 * request queue - osp_device::opd_async_requests.
643 * If the osp_device::opd_async_requests is not NULL, then return it directly;
644 * otherwise create new osp_update_request and attach it to opd_async_requests.
646 * \param[in] osp pointer to the OSP device
648 * \retval pointer to the shared queue
649 * \retval negative error number on failure
651 static struct osp_update_request *
652 osp_find_or_create_async_update_request(struct osp_device *osp)
654 struct osp_update_request *our = osp->opd_async_requests;
659 our = osp_update_request_create(&osp->opd_dt_dev);
663 osp->opd_async_requests = our;
669 * Insert an osp_update_callback into the osp_update_request.
671 * Insert an osp_update_callback to the osp_update_request. Usually each update
672 * in the osp_update_request will have one correspondent callback, and these
673 * callbacks will be called in rq_interpret_reply.
675 * \param[in] env pointer to the thread context
676 * \param[in] obj pointer to the operation target object
677 * \param[in] data pointer to the data used by the interpreter
678 * \param[in] interpreter pointer to the interpreter function
680 * \retval 0 for success
681 * \retval negative error number on failure
683 int osp_insert_update_callback(const struct lu_env *env,
684 struct osp_update_request *our,
685 struct osp_object *obj, void *data,
686 osp_update_interpreter_t interpreter)
688 struct osp_update_callback *ouc;
690 ouc = osp_update_callback_init(obj, data, interpreter);
694 list_add_tail(&ouc->ouc_list, &our->our_cb_items);
700 * Insert an asynchronous idempotent request to the shared request queue that
701 * is attached to the osp_device.
703 * This function generates a new osp_async_request with the given parameters,
704 * then tries to insert the request into the osp_device-based shared request
705 * queue. If the queue is full, then triggers the packaged OUT RPC to purge
706 * the shared queue firstly, and then re-tries.
708 * NOTE: must hold the osp::opd_async_requests_mutex to serialize concurrent
709 * osp_insert_async_request call from others.
711 * \param[in] env pointer to the thread context
712 * \param[in] op operation type, see 'enum update_type'
713 * \param[in] obj pointer to the operation target
714 * \param[in] count array size of the subsequent \a lens and \a bufs
715 * \param[in] lens buffer length array for the subsequent \a bufs
716 * \param[in] bufs the buffers to compose the request
717 * \param[in] data pointer to the data used by the interpreter
718 * \param[in] interpreter pointer to the interpreter function
720 * \retval 0 for success
721 * \retval negative error number on failure
723 int osp_insert_async_request(const struct lu_env *env, enum update_type op,
724 struct osp_object *obj, int count,
725 __u16 *lens, const void **bufs, void *data,
726 osp_update_interpreter_t interpreter)
728 struct osp_device *osp;
729 struct osp_update_request *our;
730 struct object_update *object_update;
731 size_t max_update_size;
732 struct object_update_request *ureq;
733 struct osp_update_request_sub *ours;
737 osp = lu2osp_dev(osp2lu_obj(obj)->lo_dev);
738 our = osp_find_or_create_async_update_request(osp);
740 RETURN(PTR_ERR(our));
743 ours = osp_current_object_update_request(our);
745 ureq = ours->ours_req;
746 max_update_size = ours->ours_req_size -
747 object_update_request_size(ureq);
749 object_update = update_buffer_get_update(ureq, ureq->ourq_count);
750 rc = out_update_pack(env, object_update, &max_update_size, op,
751 lu_object_fid(osp2lu_obj(obj)), count, lens, bufs);
752 /* The queue is full. */
754 osp->opd_async_requests = NULL;
755 mutex_unlock(&osp->opd_async_requests_mutex);
757 rc = osp_unplug_async_request(env, osp, our);
758 mutex_lock(&osp->opd_async_requests_mutex);
762 our = osp_find_or_create_async_update_request(osp);
764 RETURN(PTR_ERR(our));
774 rc = osp_insert_update_callback(env, our, obj, data, interpreter);
779 int osp_trans_update_request_create(struct thandle *th)
781 struct osp_thandle *oth = thandle_to_osp_thandle(th);
782 struct osp_update_request *our;
784 if (oth->ot_our != NULL)
787 our = osp_update_request_create(th->th_dev);
789 th->th_result = PTR_ERR(our);
799 void osp_thandle_destroy(struct osp_thandle *oth)
801 LASSERT(oth->ot_magic == OSP_THANDLE_MAGIC);
802 LASSERT(list_empty(&oth->ot_commit_dcb_list));
803 LASSERT(list_empty(&oth->ot_stop_dcb_list));
804 if (oth->ot_our != NULL)
805 osp_update_request_destroy(oth->ot_our);
810 * The OSP layer dt_device_operations::dt_trans_create() interface
811 * to create a transaction.
813 * There are two kinds of transactions that will involve OSP:
815 * 1) If the transaction only contains the updates on remote server
816 * (MDT or OST), such as re-generating the lost OST-object for
817 * LFSCK, then it is a remote transaction. For remote transaction,
818 * the upper layer caller (such as the LFSCK engine) will call the
819 * dt_trans_create() (with the OSP dt_device as the parameter),
820 * then the call will be directed to the osp_trans_create() that
821 * creates the transaction handler and returns it to the caller.
823 * 2) If the transcation contains both local and remote updates,
824 * such as cross MDTs create under DNE mode, then the upper layer
825 * caller will not trigger osp_trans_create(). Instead, it will
826 * call dt_trans_create() on other dt_device, such as LOD that
827 * will generate the transaction handler. Such handler will be
828 * used by the whole transaction in subsequent sub-operations.
830 * \param[in] env pointer to the thread context
831 * \param[in] d pointer to the OSP dt_device
833 * \retval pointer to the transaction handler
834 * \retval negative error number on failure
836 struct thandle *osp_trans_create(const struct lu_env *env, struct dt_device *d)
838 struct osp_thandle *oth;
839 struct thandle *th = NULL;
843 if (unlikely(oth == NULL))
844 RETURN(ERR_PTR(-ENOMEM));
846 oth->ot_magic = OSP_THANDLE_MAGIC;
849 th->th_tags = LCT_TX_HANDLE;
851 atomic_set(&oth->ot_refcount, 1);
852 INIT_LIST_HEAD(&oth->ot_commit_dcb_list);
853 INIT_LIST_HEAD(&oth->ot_stop_dcb_list);
859 * Add commit callback to transaction.
861 * Add commit callback to the osp thandle, which will be called
862 * when the thandle is committed remotely.
864 * \param[in] th the thandle
865 * \param[in] dcb commit callback structure
867 * \retval only return 0 for now.
869 int osp_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
871 struct osp_thandle *oth = thandle_to_osp_thandle(th);
873 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
874 LASSERT(&dcb->dcb_func != NULL);
875 if (dcb->dcb_flags & DCB_TRANS_STOP)
876 list_add(&dcb->dcb_linkage, &oth->ot_stop_dcb_list);
878 list_add(&dcb->dcb_linkage, &oth->ot_commit_dcb_list);
882 static void osp_trans_commit_cb(struct osp_thandle *oth, int result)
884 struct dt_txn_commit_cb *dcb;
885 struct dt_txn_commit_cb *tmp;
887 LASSERT(atomic_read(&oth->ot_refcount) > 0);
888 /* call per-transaction callbacks if any */
889 list_for_each_entry_safe(dcb, tmp, &oth->ot_commit_dcb_list,
891 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
892 "commit callback entry: magic=%x name='%s'\n",
893 dcb->dcb_magic, dcb->dcb_name);
894 list_del_init(&dcb->dcb_linkage);
895 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
899 static void osp_request_commit_cb(struct ptlrpc_request *req)
901 struct thandle *th = req->rq_cb_data;
902 struct osp_thandle *oth;
903 __u64 last_committed_transno = 0;
904 int result = req->rq_status;
910 oth = thandle_to_osp_thandle(th);
911 if (lustre_msg_get_last_committed(req->rq_repmsg))
912 last_committed_transno =
913 lustre_msg_get_last_committed(req->rq_repmsg);
915 if (last_committed_transno <
916 req->rq_import->imp_peer_committed_transno)
917 last_committed_transno =
918 req->rq_import->imp_peer_committed_transno;
920 CDEBUG(D_HA, "trans no "LPU64" committed transno "LPU64"\n",
921 req->rq_transno, last_committed_transno);
923 /* If the transaction is not really committed, mark result = 1 */
924 if (req->rq_transno != 0 &&
925 (req->rq_transno > last_committed_transno) && result == 0)
928 osp_trans_commit_cb(oth, result);
929 req->rq_committed = 1;
930 osp_thandle_put(oth);
935 * callback of osp transaction
937 * Call all of callbacks for this osp thandle. This will only be
938 * called in error handler path. In the normal processing path,
939 * these callback will be called in osp_request_commit_cb() and
940 * osp_update_interpret().
942 * \param [in] env execution environment
943 * \param [in] oth osp thandle
944 * \param [in] rc result of the osp thandle
946 void osp_trans_callback(const struct lu_env *env,
947 struct osp_thandle *oth, int rc)
949 struct osp_update_callback *ouc;
950 struct osp_update_callback *next;
952 if (oth->ot_our != NULL) {
953 list_for_each_entry_safe(ouc, next,
954 &oth->ot_our->our_cb_items, ouc_list) {
955 list_del_init(&ouc->ouc_list);
956 if (ouc->ouc_interpreter != NULL)
957 ouc->ouc_interpreter(env, NULL, NULL,
959 ouc->ouc_data, 0, rc);
960 osp_update_callback_fini(env, ouc);
963 osp_trans_stop_cb(oth, rc);
964 osp_trans_commit_cb(oth, rc);
968 * Send the request for remote updates.
970 * Send updates to the remote MDT. Prepare the request by osp_update_req
971 * and send them to remote MDT, for sync request, it will wait
972 * until the reply return, otherwise hand it to ptlrpcd.
974 * Please refer to osp_trans_create() for transaction type.
976 * \param[in] env pointer to the thread context
977 * \param[in] osp pointer to the OSP device
978 * \param[in] our pointer to the osp_update_request
980 * \retval 0 for success
981 * \retval negative error number on failure
983 static int osp_send_update_req(const struct lu_env *env,
984 struct osp_device *osp,
985 struct osp_update_request *our)
987 struct osp_update_args *args;
988 struct ptlrpc_request *req;
989 struct lu_device *top_device;
990 struct osp_thandle *oth = our->our_th;
994 LASSERT(oth != NULL);
995 LASSERT(our->our_req_sent == 0);
996 rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import,
999 osp_trans_callback(env, oth, rc);
1003 args = ptlrpc_req_async_args(req);
1004 args->oaua_update = our;
1005 osp_thandle_get(oth); /* hold for update interpret */
1006 req->rq_interpret_reply = osp_update_interpret;
1007 if (!oth->ot_super.th_wait_submit && !oth->ot_super.th_sync) {
1008 if (!osp->opd_imp_active || !osp->opd_imp_connected) {
1009 osp_trans_callback(env, oth, rc);
1010 osp_thandle_put(oth);
1011 GOTO(out, rc = -ENOTCONN);
1014 rc = obd_get_request_slot(&osp->opd_obd->u.cli);
1016 osp_trans_callback(env, oth, rc);
1017 osp_thandle_put(oth);
1018 GOTO(out, rc = -ENOTCONN);
1020 args->oaua_flow_control = true;
1022 if (!osp->opd_connect_mdt) {
1023 down_read(&osp->opd_async_updates_rwsem);
1024 args->oaua_count = &osp->opd_async_updates_count;
1025 args->oaua_waitq = &osp->opd_syn_barrier_waitq;
1026 up_read(&osp->opd_async_updates_rwsem);
1027 atomic_inc(args->oaua_count);
1030 ptlrpcd_add_req(req);
1033 osp_thandle_get(oth); /* hold for commit callback */
1034 req->rq_commit_cb = osp_request_commit_cb;
1035 req->rq_cb_data = &oth->ot_super;
1036 args->oaua_flow_control = false;
1038 /* If the transaction is created during MDT recoverying
1039 * process, it means this is an recovery update, we need
1040 * to let OSP send it anyway without checking recoverying
1041 * status, in case the other target is being recoveried
1042 * at the same time, and if we wait here for the import
1043 * to be recoveryed, it might cause deadlock */
1044 top_device = osp->opd_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
1045 if (top_device->ld_obd->obd_recovering)
1046 req->rq_allow_replay = 1;
1048 if (osp->opd_connect_mdt)
1049 osp_get_rpc_lock(osp);
1050 rc = ptlrpc_queue_wait(req);
1051 if (osp->opd_connect_mdt)
1052 osp_put_rpc_lock(osp);
1053 if ((rc == -ENOMEM && req->rq_set == NULL) ||
1054 (req->rq_transno == 0 && !req->rq_committed)) {
1055 if (args->oaua_update != NULL) {
1056 /* If osp_update_interpret is not being called,
1057 * release the osp_thandle */
1058 args->oaua_update = NULL;
1059 osp_thandle_put(oth);
1062 req->rq_cb_data = NULL;
1063 rc = rc == 0 ? req->rq_status : rc;
1064 osp_trans_callback(env, oth, rc);
1065 osp_thandle_put(oth);
1071 ptlrpc_req_finished(req);
1077 * Get local thandle for osp_thandle
1079 * Get the local OSD thandle from the OSP thandle. Currently, there
1080 * are a few OSP API (osp_object_create() and osp_sync_add()) needs
1081 * to update the object on local OSD device.
1083 * If the osp_thandle comes from normal stack (MDD->LOD->OSP), then
1084 * we will get local thandle by thandle_get_sub_by_dt.
1086 * If the osp_thandle is remote thandle (th_top == NULL, only used
1087 * by LFSCK), then it will create a local thandle, and stop it in
1088 * osp_trans_stop(). And this only happens on OSP for OST.
1090 * These are temporary solution, once OSP accessing OSD object is
1091 * being fixed properly, this function should be removed. XXX
1093 * \param[in] env pointer to the thread context
1094 * \param[in] th pointer to the transaction handler
1095 * \param[in] dt pointer to the OSP device
1097 * \retval pointer to the local thandle
1098 * \retval ERR_PTR(errno) if it fails.
1100 struct thandle *osp_get_storage_thandle(const struct lu_env *env,
1102 struct osp_device *osp)
1104 struct osp_thandle *oth;
1105 struct thandle *local_th;
1107 if (th->th_top != NULL)
1108 return thandle_get_sub_by_dt(env, th->th_top,
1111 LASSERT(!osp->opd_connect_mdt);
1112 oth = thandle_to_osp_thandle(th);
1113 if (oth->ot_storage_th != NULL)
1114 return oth->ot_storage_th;
1116 local_th = dt_trans_create(env, osp->opd_storage);
1117 if (IS_ERR(local_th))
1120 oth->ot_storage_th = local_th;
1126 * Set version for the transaction
1128 * Set the version for the transaction, then the osp RPC will be
1129 * sent in the order of version, i.e. the transaction with lower
1130 * version will be sent first.
1132 * \param [in] oth osp thandle to be set version.
1134 * \retval 0 if set version succeeds
1135 * negative errno if set version fails.
1137 int osp_check_and_set_rpc_version(struct osp_thandle *oth)
1139 struct osp_device *osp = dt2osp_dev(oth->ot_super.th_dev);
1140 struct osp_updates *ou = osp->opd_update;
1145 if (oth->ot_version != 0)
1148 spin_lock(&ou->ou_lock);
1149 oth->ot_version = ou->ou_version++;
1150 spin_unlock(&ou->ou_lock);
1152 CDEBUG(D_INFO, "%s: version "LPU64" oth:version %p:"LPU64"\n",
1153 osp->opd_obd->obd_name, ou->ou_version, oth, oth->ot_version);
1159 * Get next OSP update request in the sending list
1160 * Get next OSP update request in the sending list by version number, next
1162 * 1. transaction which does not have a version number.
1163 * 2. transaction whose version == opd_rpc_version.
1165 * \param [in] ou osp update structure.
1166 * \param [out] ourp the pointer holding the next update request.
1168 * \retval true if getting the next transaction.
1169 * \retval false if not getting the next transaction.
1172 osp_get_next_request(struct osp_updates *ou, struct osp_update_request **ourp)
1174 struct osp_update_request *our;
1175 struct osp_update_request *tmp;
1176 bool got_req = false;
1178 spin_lock(&ou->ou_lock);
1179 list_for_each_entry_safe(our, tmp, &ou->ou_list, our_list) {
1180 LASSERT(our->our_th != NULL);
1181 CDEBUG(D_INFO, "our %p version "LPU64" rpc_version "LPU64"\n",
1182 our, our->our_th->ot_version, ou->ou_rpc_version);
1183 if (our->our_th->ot_version == 0) {
1184 list_del_init(&our->our_list);
1190 /* Find next osp_update_request in the list */
1191 if (our->our_th->ot_version == ou->ou_rpc_version) {
1192 list_del_init(&our->our_list);
1198 spin_unlock(&ou->ou_lock);
1203 static void osp_update_rpc_version(struct osp_updates *ou,
1204 struct osp_thandle *oth)
1206 if (oth->ot_version == 0)
1209 LASSERT(oth->ot_version == ou->ou_rpc_version);
1210 spin_lock(&ou->ou_lock);
1211 ou->ou_rpc_version++;
1212 spin_unlock(&ou->ou_lock);
1216 * Sending update thread
1218 * Create thread to send update request to other MDTs, this thread will pull
1219 * out update request from the list in OSP by version number, i.e. it will
1220 * make sure the update request with lower version number will be sent first.
1222 * \param[in] arg hold the OSP device.
1224 * \retval 0 if the thread is created successfully.
1225 * \retal negative error if the thread is not created
1228 int osp_send_update_thread(void *arg)
1231 struct osp_device *osp = arg;
1232 struct l_wait_info lwi = { 0 };
1233 struct osp_updates *ou = osp->opd_update;
1234 struct ptlrpc_thread *thread = &osp->opd_update_thread;
1235 struct osp_update_request *our = NULL;
1239 LASSERT(ou != NULL);
1240 rc = lu_env_init(&env, osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
1242 CERROR("%s: init env error: rc = %d\n", osp->opd_obd->obd_name,
1247 thread->t_flags = SVC_RUNNING;
1248 wake_up(&thread->t_ctl_waitq);
1251 l_wait_event(ou->ou_waitq,
1252 !osp_send_update_thread_running(osp) ||
1253 osp_get_next_request(ou, &our),
1256 if (!osp_send_update_thread_running(osp)) {
1257 if (our != NULL && our->our_th != NULL) {
1258 osp_trans_callback(&env, our->our_th, -EINTR);
1259 osp_thandle_put(our->our_th);
1264 if (our->our_req_sent == 0) {
1265 if (our->our_th != NULL &&
1266 our->our_th->ot_super.th_result != 0)
1267 osp_trans_callback(&env, our->our_th,
1268 our->our_th->ot_super.th_result);
1270 rc = osp_send_update_req(&env, osp, our);
1273 if (our->our_th != NULL) {
1274 /* Update the rpc version */
1275 osp_update_rpc_version(ou, our->our_th);
1276 /* Balanced for thandle_get in osp_trans_trigger() */
1277 osp_thandle_put(our->our_th);
1281 thread->t_flags = SVC_STOPPED;
1283 wake_up(&thread->t_ctl_waitq);
1289 * Trigger the request for remote updates.
1291 * Add the request to the sending list, and wake up osp update
1294 * \param[in] env pointer to the thread context
1295 * \param[in] osp pointer to the OSP device
1296 * \param[in] oth pointer to the transaction handler
1299 static void osp_trans_trigger(const struct lu_env *env,
1300 struct osp_device *osp,
1301 struct osp_thandle *oth)
1304 CDEBUG(D_INFO, "%s: add oth %p with version "LPU64"\n",
1305 osp->opd_obd->obd_name, oth, oth->ot_version);
1307 LASSERT(oth->ot_magic == OSP_THANDLE_MAGIC);
1308 osp_thandle_get(oth);
1309 LASSERT(oth->ot_our != NULL);
1310 spin_lock(&osp->opd_update->ou_lock);
1311 list_add_tail(&oth->ot_our->our_list,
1312 &osp->opd_update->ou_list);
1313 spin_unlock(&osp->opd_update->ou_lock);
1315 wake_up(&osp->opd_update->ou_waitq);
1319 * The OSP layer dt_device_operations::dt_trans_start() interface
1320 * to start the transaction.
1322 * If the transaction is a remote transaction, then related remote
1323 * updates will be triggered in the osp_trans_stop().
1324 * Please refer to osp_trans_create() for transaction type.
1326 * \param[in] env pointer to the thread context
1327 * \param[in] dt pointer to the OSP dt_device
1328 * \param[in] th pointer to the transaction handler
1330 * \retval 0 for success
1331 * \retval negative error number on failure
1333 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
1336 struct osp_thandle *oth = thandle_to_osp_thandle(th);
1338 if (oth->ot_super.th_sync)
1339 oth->ot_our->our_flags |= UPDATE_FL_SYNC;
1340 /* For remote thandle, if there are local thandle, start it here*/
1341 if (is_only_remote_trans(th) && oth->ot_storage_th != NULL)
1342 return dt_trans_start(env, oth->ot_storage_th->th_dev,
1343 oth->ot_storage_th);
1348 * The OSP layer dt_device_operations::dt_trans_stop() interface
1349 * to stop the transaction.
1351 * If the transaction is a remote transaction, related remote
1352 * updates will be triggered here via osp_trans_trigger().
1354 * For synchronous mode update or any failed update, the request
1355 * will be destroyed explicitly when the osp_trans_stop().
1357 * Please refer to osp_trans_create() for transaction type.
1359 * \param[in] env pointer to the thread context
1360 * \param[in] dt pointer to the OSP dt_device
1361 * \param[in] th pointer to the transaction handler
1363 * \retval 0 for success
1364 * \retval negative error number on failure
1366 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
1369 struct osp_thandle *oth = thandle_to_osp_thandle(th);
1370 struct osp_update_request *our = oth->ot_our;
1371 struct osp_device *osp = dt2osp_dev(dt);
1375 /* For remote transaction, if there is local storage thandle,
1377 if (oth->ot_storage_th != NULL && th->th_top == NULL) {
1378 dt_trans_stop(env, oth->ot_storage_th->th_dev,
1379 oth->ot_storage_th);
1380 oth->ot_storage_th = NULL;
1383 if (our == NULL || list_empty(&our->our_req_list)) {
1384 osp_trans_callback(env, oth, th->th_result);
1385 GOTO(out, rc = th->th_result);
1388 if (!osp->opd_connect_mdt) {
1389 osp_trans_callback(env, oth, th->th_result);
1390 rc = osp_send_update_req(env, osp, oth->ot_our);
1394 if (osp->opd_update == NULL ||
1395 !osp_send_update_thread_running(osp)) {
1396 osp_trans_callback(env, oth, -EIO);
1397 GOTO(out, rc = -EIO);
1401 /* if th_sync is set, then it needs to be sent
1402 * right away. Note: even thought the RPC has been
1403 * sent, it still needs to be added to the sending
1404 * list (see osp_trans_trigger()), so ou_rpc_version
1405 * can be updated correctly. */
1406 rc = osp_send_update_req(env, osp, our);
1407 our->our_req_sent = 1;
1410 osp_trans_trigger(env, osp, oth);
1412 osp_thandle_put(oth);