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, 2015, 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);
152 * Get current update request
154 * Get current object update request from our_req_list in
155 * osp_update_request, because we always insert the new update
156 * request in the last position, so the last update request
157 * in the list will be the current update req.
159 * \param[in] our osp update request where to get the
160 * current object update.
162 * \retval the current object update.
164 struct osp_update_request_sub *
165 osp_current_object_update_request(struct osp_update_request *our)
167 if (list_empty(&our->our_req_list))
170 return list_entry(our->our_req_list.prev, struct osp_update_request_sub,
175 * Allocate and initialize osp_update_request
177 * osp_update_request is being used to track updates being executed on
178 * this dt_device(OSD or OSP). The update buffer will be 4k initially,
179 * and increased if needed.
181 * \param [in] dt dt device
183 * \retval osp_update_request being allocated if succeed
184 * \retval ERR_PTR(errno) if failed
186 struct osp_update_request *osp_update_request_create(struct dt_device *dt)
188 struct osp_update_request *our;
192 return ERR_PTR(-ENOMEM);
194 INIT_LIST_HEAD(&our->our_req_list);
195 INIT_LIST_HEAD(&our->our_cb_items);
196 INIT_LIST_HEAD(&our->our_list);
197 spin_lock_init(&our->our_list_lock);
199 osp_object_update_request_create(our, OUT_UPDATE_INIT_BUFFER_SIZE);
203 void osp_update_request_destroy(struct osp_update_request *our)
205 struct osp_update_request_sub *ours;
206 struct osp_update_request_sub *tmp;
211 list_for_each_entry_safe(ours, tmp, &our->our_req_list, ours_list) {
212 list_del(&ours->ours_list);
213 if (ours->ours_req != NULL)
214 OBD_FREE_LARGE(ours->ours_req, ours->ours_req_size);
221 object_update_request_dump(const struct object_update_request *ourq,
225 size_t total_size = 0;
227 for (i = 0; i < ourq->ourq_count; i++) {
228 struct object_update *update;
231 update = object_update_request_get(ourq, i, &size);
232 LASSERT(update != NULL);
233 CDEBUG(mask, "i = %u fid = "DFID" op = %s "
234 "params = %d batchid = "LPU64" size = %zu repsize %u\n",
235 i, PFID(&update->ou_fid),
236 update_op_str(update->ou_type),
237 update->ou_params_count,
238 update->ou_batchid, size,
239 (unsigned)update->ou_result_size);
244 CDEBUG(mask, "updates = %p magic = %x count = %d size = %zu\n", ourq,
245 ourq->ourq_magic, ourq->ourq_count, total_size);
249 * Prepare inline update request
251 * Prepare OUT update ptlrpc inline request, and the request usually includes
252 * one update buffer, which does not need bulk transfer.
254 * \param[in] env execution environment
255 * \param[in] req ptlrpc request
256 * \param[in] ours sub osp_update_request to be packed
258 * \retval 0 if packing succeeds
259 * \retval negative errno if packing fails
261 int osp_prep_inline_update_req(const struct lu_env *env,
262 struct ptlrpc_request *req,
263 struct osp_update_request *our,
266 struct osp_update_request_sub *ours;
267 struct out_update_header *ouh;
268 __u32 update_req_size;
271 ours = list_entry(our->our_req_list.next,
272 struct osp_update_request_sub, ours_list);
273 update_req_size = object_update_request_size(ours->ours_req);
274 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_HEADER, RCL_CLIENT,
275 update_req_size + sizeof(*ouh));
277 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, OUT_UPDATE);
281 ouh = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_HEADER);
282 ouh->ouh_magic = OUT_UPDATE_HEADER_MAGIC;
284 ouh->ouh_inline_length = update_req_size;
285 ouh->ouh_reply_size = repsize;
287 memcpy(ouh->ouh_inline_data, ours->ours_req, update_req_size);
289 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_REPLY,
290 RCL_SERVER, repsize);
292 ptlrpc_request_set_replen(req);
293 req->rq_request_portal = OUT_PORTAL;
294 req->rq_reply_portal = OSC_REPLY_PORTAL;
300 * Prepare update request.
302 * Prepare OUT update ptlrpc request, and the request usually includes
303 * all of updates (stored in \param ureq) from one operation.
305 * \param[in] env execution environment
306 * \param[in] imp import on which ptlrpc request will be sent
307 * \param[in] ureq hold all of updates which will be packed into the req
308 * \param[in] reqp request to be created
310 * \retval 0 if preparation succeeds.
311 * \retval negative errno if preparation fails.
313 int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
314 struct osp_update_request *our,
315 struct ptlrpc_request **reqp)
317 struct ptlrpc_request *req;
318 struct ptlrpc_bulk_desc *desc;
319 struct osp_update_request_sub *ours;
320 const struct object_update_request *ourq;
321 struct out_update_header *ouh;
322 struct out_update_buffer *oub;
325 struct object_update_reply *reply;
330 list_for_each_entry(ours, &our->our_req_list, ours_list) {
331 object_update_request_dump(ours->ours_req, D_INFO);
333 ourq = ours->ours_req;
334 for (i = 0; i < ourq->ourq_count; i++) {
335 struct object_update *update;
339 /* XXX: it's very inefficient to lookup update
340 * this way, iterating from the beginning
342 update = object_update_request_get(ourq, i, &size);
343 LASSERT(update != NULL);
345 repsize += sizeof(reply->ourp_lens[0]);
346 repsize += sizeof(struct object_update_result);
347 repsize += update->ou_result_size;
352 repsize += sizeof(*reply);
353 repsize = (repsize + OUT_UPDATE_REPLY_SIZE - 1) &
354 ~(OUT_UPDATE_REPLY_SIZE - 1);
355 LASSERT(buf_count > 0);
357 req = ptlrpc_request_alloc(imp, &RQF_OUT_UPDATE);
361 if (buf_count == 1) {
362 ours = list_entry(our->our_req_list.next,
363 struct osp_update_request_sub, ours_list);
365 /* Let's check if it can be packed inline */
366 if (object_update_request_size(ours->ours_req) +
367 sizeof(struct out_update_header) <
368 OUT_UPDATE_MAX_INLINE_SIZE) {
369 rc = osp_prep_inline_update_req(env, req, our, repsize);
376 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_HEADER, RCL_CLIENT,
377 sizeof(struct osp_update_request));
379 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_BUF, RCL_CLIENT,
380 buf_count * sizeof(*oub));
382 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, OUT_UPDATE);
386 ouh = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_HEADER);
387 ouh->ouh_magic = OUT_UPDATE_HEADER_MAGIC;
388 ouh->ouh_count = buf_count;
389 ouh->ouh_inline_length = 0;
390 ouh->ouh_reply_size = repsize;
391 oub = req_capsule_client_get(&req->rq_pill, &RMF_OUT_UPDATE_BUF);
392 list_for_each_entry(ours, &our->our_req_list, ours_list) {
393 oub->oub_size = ours->ours_req_size;
397 req->rq_bulk_write = 1;
398 desc = ptlrpc_prep_bulk_imp(req, buf_count,
399 MD_MAX_BRW_SIZE >> LNET_MTU_BITS,
400 PTLRPC_BULK_GET_SOURCE | PTLRPC_BULK_BUF_KVEC,
401 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
403 GOTO(out_req, rc = -ENOMEM);
405 /* NB req now owns desc and will free it when it gets freed */
406 list_for_each_entry(ours, &our->our_req_list, ours_list) {
407 desc->bd_frag_ops->add_iov_frag(desc, ours->ours_req,
408 ours->ours_req_size);
409 total += ours->ours_req_size;
411 CDEBUG(D_OTHER, "total %d in %u\n", total, our->our_update_nr);
413 req_capsule_set_size(&req->rq_pill, &RMF_OUT_UPDATE_REPLY,
414 RCL_SERVER, repsize);
416 ptlrpc_request_set_replen(req);
417 req->rq_request_portal = OUT_PORTAL;
418 req->rq_reply_portal = OSC_REPLY_PORTAL;
423 ptlrpc_req_finished(req);
431 * Send update request to the remote MDT synchronously.
433 * \param[in] env execution environment
434 * \param[in] imp import on which ptlrpc request will be sent
435 * \param[in] our hold all of updates which will be packed into the req
436 * \param[in] reqp request to be created
438 * \retval 0 if RPC succeeds.
439 * \retval negative errno if RPC fails.
441 int osp_remote_sync(const struct lu_env *env, struct osp_device *osp,
442 struct osp_update_request *our,
443 struct ptlrpc_request **reqp)
445 struct obd_import *imp = osp->opd_obd->u.cli.cl_import;
446 struct ptlrpc_request *req = NULL;
450 rc = osp_prep_update_req(env, imp, our, &req);
454 /* This will only be called with read-only update, and these updates
455 * might be used to retrieve update log during recovery process, so
456 * it will be allowed to send during recovery process */
457 req->rq_allow_replay = 1;
458 req->rq_allow_intr = 1;
460 /* Note: some dt index api might return non-zero result here, like
461 * osd_index_ea_lookup, so we should only check rc < 0 here */
462 rc = ptlrpc_queue_wait(req);
464 if (rc < 0 || reqp == NULL)
465 ptlrpc_req_finished(req);
473 * Invalidate all objects in the osp thandle
475 * invalidate all of objects in the update request, which will be called
476 * when the transaction is aborted.
478 * \param[in] oth osp thandle.
480 static void osp_thandle_invalidate_object(const struct lu_env *env,
481 struct osp_thandle *oth)
483 struct osp_update_request *our = oth->ot_our;
484 struct osp_update_request_sub *ours;
489 list_for_each_entry(ours, &our->our_req_list, ours_list) {
490 struct object_update_request *our_req = ours->ours_req;
492 struct lu_object *obj;
493 struct osp_object *osp_obj;
495 for (i = 0; i < our_req->ourq_count; i++) {
496 struct object_update *update;
498 update = object_update_request_get(our_req, i, NULL);
502 if (update->ou_type != OUT_WRITE)
505 if (!fid_is_sane(&update->ou_fid))
508 obj = lu_object_find_slice(env,
509 &oth->ot_super.th_dev->dd_lu_dev,
510 &update->ou_fid, NULL);
514 osp_obj = lu2osp_obj(obj);
515 if (osp_obj->opo_ooa != NULL) {
516 spin_lock(&osp_obj->opo_lock);
517 osp_obj->opo_ooa->ooa_attr.la_valid = 0;
518 osp_obj->opo_stale = 1;
519 spin_unlock(&osp_obj->opo_lock);
521 lu_object_put(env, obj);
526 static void osp_trans_stop_cb(const struct lu_env *env,
527 struct osp_thandle *oth, int result)
529 struct dt_txn_commit_cb *dcb;
530 struct dt_txn_commit_cb *tmp;
532 /* call per-transaction stop callbacks if any */
533 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
535 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
536 "commit callback entry: magic=%x name='%s'\n",
537 dcb->dcb_magic, dcb->dcb_name);
538 list_del_init(&dcb->dcb_linkage);
539 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
543 osp_thandle_invalidate_object(env, oth);
547 * Allocate an osp request and initialize it with the given parameters.
549 * \param[in] obj pointer to the operation target
550 * \param[in] data pointer to the data used by the interpreter
551 * \param[in] interpreter pointer to the interpreter function
553 * \retval pointer to the asychronous request
554 * \retval NULL if the allocation failed
556 static struct osp_update_callback *
557 osp_update_callback_init(struct osp_object *obj, void *data,
558 osp_update_interpreter_t interpreter)
560 struct osp_update_callback *ouc;
566 lu_object_get(osp2lu_obj(obj));
567 INIT_LIST_HEAD(&ouc->ouc_list);
569 ouc->ouc_data = data;
570 ouc->ouc_interpreter = interpreter;
576 * Destroy the osp_update_callback.
578 * \param[in] env pointer to the thread context
579 * \param[in] ouc pointer to osp_update_callback
581 static void osp_update_callback_fini(const struct lu_env *env,
582 struct osp_update_callback *ouc)
584 LASSERT(list_empty(&ouc->ouc_list));
586 lu_object_put(env, osp2lu_obj(ouc->ouc_obj));
591 * Interpret the packaged OUT RPC results.
593 * For every packaged sub-request, call its registered interpreter function.
594 * Then destroy the sub-request.
596 * \param[in] env pointer to the thread context
597 * \param[in] req pointer to the RPC
598 * \param[in] arg pointer to data used by the interpreter
599 * \param[in] rc the RPC return value
601 * \retval 0 for success
602 * \retval negative error number on failure
604 static int osp_update_interpret(const struct lu_env *env,
605 struct ptlrpc_request *req, void *arg, int rc)
607 struct object_update_reply *reply = NULL;
608 struct osp_update_args *oaua = arg;
609 struct osp_update_request *our = oaua->oaua_update;
610 struct osp_thandle *oth;
611 struct osp_update_callback *ouc;
612 struct osp_update_callback *next;
622 oaua->oaua_update = NULL;
624 if (oaua->oaua_flow_control) {
625 struct osp_device *osp;
627 LASSERT(oth != NULL);
628 osp = dt2osp_dev(oth->ot_super.th_dev);
629 obd_put_request_slot(&osp->opd_obd->u.cli);
632 /* Unpack the results from the reply message. */
633 if (req->rq_repmsg != NULL) {
634 reply = req_capsule_server_sized_get(&req->rq_pill,
635 &RMF_OUT_UPDATE_REPLY,
636 OUT_UPDATE_REPLY_SIZE);
637 if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC) {
641 count = reply->ourp_count;
645 list_for_each_entry_safe(ouc, next, &our->our_cb_items, ouc_list) {
646 list_del_init(&ouc->ouc_list);
648 /* The peer may only have handled some requests (indicated
649 * by the 'count') in the packaged OUT RPC, we can only get
650 * results for the handled part. */
651 if (index < count && reply->ourp_lens[index] > 0 && rc >= 0) {
652 struct object_update_result *result;
654 result = object_update_result_get(reply, index, NULL);
658 rc1 = rc = result->our_rc;
659 } else if (rc1 >= 0) {
660 /* The peer did not handle these request, let's return
661 * -EINVAL to update interpret for now */
668 if (ouc->ouc_interpreter != NULL)
669 ouc->ouc_interpreter(env, reply, req, ouc->ouc_obj,
670 ouc->ouc_data, index, rc1);
672 osp_update_callback_fini(env, ouc);
676 if (oaua->oaua_count != NULL && atomic_dec_and_test(oaua->oaua_count))
677 wake_up_all(oaua->oaua_waitq);
680 /* oth and osp_update_requests will be destoryed in
682 osp_trans_stop_cb(env, oth, rc);
683 osp_thandle_put(oth);
685 osp_update_request_destroy(our);
692 * Pack all the requests in the shared asynchronous idempotent request queue
693 * into a single OUT RPC that will be given to the background ptlrpcd daemon.
695 * \param[in] env pointer to the thread context
696 * \param[in] osp pointer to the OSP device
697 * \param[in] our pointer to the shared queue
699 * \retval 0 for success
700 * \retval negative error number on failure
702 int osp_unplug_async_request(const struct lu_env *env,
703 struct osp_device *osp,
704 struct osp_update_request *our)
706 struct osp_update_args *args;
707 struct ptlrpc_request *req = NULL;
710 rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import,
713 struct osp_update_callback *ouc;
714 struct osp_update_callback *next;
716 list_for_each_entry_safe(ouc, next,
717 &our->our_cb_items, ouc_list) {
718 list_del_init(&ouc->ouc_list);
719 if (ouc->ouc_interpreter != NULL)
720 ouc->ouc_interpreter(env, NULL, NULL,
722 ouc->ouc_data, 0, rc);
723 osp_update_callback_fini(env, ouc);
725 osp_update_request_destroy(our);
727 args = ptlrpc_req_async_args(req);
728 args->oaua_update = our;
729 args->oaua_count = NULL;
730 args->oaua_waitq = NULL;
731 args->oaua_flow_control = false;
732 req->rq_interpret_reply = osp_update_interpret;
733 ptlrpcd_add_req(req);
740 * Find or create (if NOT exist or purged) the shared asynchronous idempotent
741 * request queue - osp_device::opd_async_requests.
743 * If the osp_device::opd_async_requests is not NULL, then return it directly;
744 * otherwise create new osp_update_request and attach it to opd_async_requests.
746 * \param[in] osp pointer to the OSP device
748 * \retval pointer to the shared queue
749 * \retval negative error number on failure
751 static struct osp_update_request *
752 osp_find_or_create_async_update_request(struct osp_device *osp)
754 struct osp_update_request *our = osp->opd_async_requests;
759 our = osp_update_request_create(&osp->opd_dt_dev);
763 osp->opd_async_requests = our;
769 * Insert an osp_update_callback into the osp_update_request.
771 * Insert an osp_update_callback to the osp_update_request. Usually each update
772 * in the osp_update_request will have one correspondent callback, and these
773 * callbacks will be called in rq_interpret_reply.
775 * \param[in] env pointer to the thread context
776 * \param[in] obj pointer to the operation target object
777 * \param[in] data pointer to the data used by the interpreter
778 * \param[in] interpreter pointer to the interpreter function
780 * \retval 0 for success
781 * \retval negative error number on failure
783 int osp_insert_update_callback(const struct lu_env *env,
784 struct osp_update_request *our,
785 struct osp_object *obj, void *data,
786 osp_update_interpreter_t interpreter)
788 struct osp_update_callback *ouc;
790 ouc = osp_update_callback_init(obj, data, interpreter);
794 list_add_tail(&ouc->ouc_list, &our->our_cb_items);
800 * Insert an asynchronous idempotent request to the shared request queue that
801 * is attached to the osp_device.
803 * This function generates a new osp_async_request with the given parameters,
804 * then tries to insert the request into the osp_device-based shared request
805 * queue. If the queue is full, then triggers the packaged OUT RPC to purge
806 * the shared queue firstly, and then re-tries.
808 * NOTE: must hold the osp::opd_async_requests_mutex to serialize concurrent
809 * osp_insert_async_request call from others.
811 * \param[in] env pointer to the thread context
812 * \param[in] op operation type, see 'enum update_type'
813 * \param[in] obj pointer to the operation target
814 * \param[in] count array size of the subsequent \a lens and \a bufs
815 * \param[in] lens buffer length array for the subsequent \a bufs
816 * \param[in] bufs the buffers to compose the request
817 * \param[in] data pointer to the data used by the interpreter
818 * \param[in] repsize how many bytes the caller allocated for \a data
819 * \param[in] interpreter pointer to the interpreter function
821 * \retval 0 for success
822 * \retval negative error number on failure
824 int osp_insert_async_request(const struct lu_env *env, enum update_type op,
825 struct osp_object *obj, int count,
826 __u16 *lens, const void **bufs,
827 void *data, __u32 repsize,
828 osp_update_interpreter_t interpreter)
830 struct osp_device *osp;
831 struct osp_update_request *our;
832 struct object_update *object_update;
833 size_t max_update_size;
834 struct object_update_request *ureq;
835 struct osp_update_request_sub *ours;
839 osp = lu2osp_dev(osp2lu_obj(obj)->lo_dev);
840 our = osp_find_or_create_async_update_request(osp);
842 RETURN(PTR_ERR(our));
845 ours = osp_current_object_update_request(our);
847 ureq = ours->ours_req;
848 max_update_size = ours->ours_req_size -
849 object_update_request_size(ureq);
851 object_update = update_buffer_get_update(ureq, ureq->ourq_count);
852 rc = out_update_pack(env, object_update, &max_update_size, op,
853 lu_object_fid(osp2lu_obj(obj)), count, lens, bufs,
855 /* The queue is full. */
857 osp->opd_async_requests = NULL;
858 mutex_unlock(&osp->opd_async_requests_mutex);
860 rc = osp_unplug_async_request(env, osp, our);
861 mutex_lock(&osp->opd_async_requests_mutex);
865 our = osp_find_or_create_async_update_request(osp);
867 RETURN(PTR_ERR(our));
875 our->our_update_nr++;
878 rc = osp_insert_update_callback(env, our, obj, data, interpreter);
883 int osp_trans_update_request_create(struct thandle *th)
885 struct osp_thandle *oth = thandle_to_osp_thandle(th);
886 struct osp_update_request *our;
888 if (oth->ot_our != NULL)
891 our = osp_update_request_create(th->th_dev);
893 th->th_result = PTR_ERR(our);
903 void osp_thandle_destroy(struct osp_thandle *oth)
905 LASSERT(oth->ot_magic == OSP_THANDLE_MAGIC);
906 LASSERT(list_empty(&oth->ot_commit_dcb_list));
907 LASSERT(list_empty(&oth->ot_stop_dcb_list));
908 if (oth->ot_our != NULL)
909 osp_update_request_destroy(oth->ot_our);
914 * The OSP layer dt_device_operations::dt_trans_create() interface
915 * to create a transaction.
917 * There are two kinds of transactions that will involve OSP:
919 * 1) If the transaction only contains the updates on remote server
920 * (MDT or OST), such as re-generating the lost OST-object for
921 * LFSCK, then it is a remote transaction. For remote transaction,
922 * the upper layer caller (such as the LFSCK engine) will call the
923 * dt_trans_create() (with the OSP dt_device as the parameter),
924 * then the call will be directed to the osp_trans_create() that
925 * creates the transaction handler and returns it to the caller.
927 * 2) If the transcation contains both local and remote updates,
928 * such as cross MDTs create under DNE mode, then the upper layer
929 * caller will not trigger osp_trans_create(). Instead, it will
930 * call dt_trans_create() on other dt_device, such as LOD that
931 * will generate the transaction handler. Such handler will be
932 * used by the whole transaction in subsequent sub-operations.
934 * \param[in] env pointer to the thread context
935 * \param[in] d pointer to the OSP dt_device
937 * \retval pointer to the transaction handler
938 * \retval negative error number on failure
940 struct thandle *osp_trans_create(const struct lu_env *env, struct dt_device *d)
942 struct osp_thandle *oth;
943 struct thandle *th = NULL;
947 if (unlikely(oth == NULL))
948 RETURN(ERR_PTR(-ENOMEM));
950 oth->ot_magic = OSP_THANDLE_MAGIC;
953 th->th_tags = LCT_TX_HANDLE;
955 atomic_set(&oth->ot_refcount, 1);
956 INIT_LIST_HEAD(&oth->ot_commit_dcb_list);
957 INIT_LIST_HEAD(&oth->ot_stop_dcb_list);
963 * Add commit callback to transaction.
965 * Add commit callback to the osp thandle, which will be called
966 * when the thandle is committed remotely.
968 * \param[in] th the thandle
969 * \param[in] dcb commit callback structure
971 * \retval only return 0 for now.
973 int osp_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
975 struct osp_thandle *oth = thandle_to_osp_thandle(th);
977 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
978 LASSERT(&dcb->dcb_func != NULL);
979 if (dcb->dcb_flags & DCB_TRANS_STOP)
980 list_add(&dcb->dcb_linkage, &oth->ot_stop_dcb_list);
982 list_add(&dcb->dcb_linkage, &oth->ot_commit_dcb_list);
986 static void osp_trans_commit_cb(struct osp_thandle *oth, int result)
988 struct dt_txn_commit_cb *dcb;
989 struct dt_txn_commit_cb *tmp;
991 LASSERT(atomic_read(&oth->ot_refcount) > 0);
992 /* call per-transaction callbacks if any */
993 list_for_each_entry_safe(dcb, tmp, &oth->ot_commit_dcb_list,
995 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
996 "commit callback entry: magic=%x name='%s'\n",
997 dcb->dcb_magic, dcb->dcb_name);
998 list_del_init(&dcb->dcb_linkage);
999 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1003 static void osp_request_commit_cb(struct ptlrpc_request *req)
1005 struct thandle *th = req->rq_cb_data;
1006 struct osp_thandle *oth;
1007 __u64 last_committed_transno = 0;
1008 int result = req->rq_status;
1014 oth = thandle_to_osp_thandle(th);
1015 if (req->rq_repmsg != NULL &&
1016 lustre_msg_get_last_committed(req->rq_repmsg))
1017 last_committed_transno =
1018 lustre_msg_get_last_committed(req->rq_repmsg);
1020 if (last_committed_transno <
1021 req->rq_import->imp_peer_committed_transno)
1022 last_committed_transno =
1023 req->rq_import->imp_peer_committed_transno;
1025 CDEBUG(D_HA, "trans no "LPU64" committed transno "LPU64"\n",
1026 req->rq_transno, last_committed_transno);
1028 /* If the transaction is not really committed, mark result = 1 */
1029 if (req->rq_transno != 0 &&
1030 (req->rq_transno > last_committed_transno) && result == 0)
1033 osp_trans_commit_cb(oth, result);
1034 req->rq_committed = 1;
1035 osp_thandle_put(oth);
1040 * callback of osp transaction
1042 * Call all of callbacks for this osp thandle. This will only be
1043 * called in error handler path. In the normal processing path,
1044 * these callback will be called in osp_request_commit_cb() and
1045 * osp_update_interpret().
1047 * \param [in] env execution environment
1048 * \param [in] oth osp thandle
1049 * \param [in] rc result of the osp thandle
1051 void osp_trans_callback(const struct lu_env *env,
1052 struct osp_thandle *oth, int rc)
1054 struct osp_update_callback *ouc;
1055 struct osp_update_callback *next;
1057 if (oth->ot_our != NULL) {
1058 list_for_each_entry_safe(ouc, next,
1059 &oth->ot_our->our_cb_items, ouc_list) {
1060 list_del_init(&ouc->ouc_list);
1061 if (ouc->ouc_interpreter != NULL)
1062 ouc->ouc_interpreter(env, NULL, NULL,
1064 ouc->ouc_data, 0, rc);
1065 osp_update_callback_fini(env, ouc);
1068 osp_trans_stop_cb(env, oth, rc);
1069 osp_trans_commit_cb(oth, rc);
1073 * Send the request for remote updates.
1075 * Send updates to the remote MDT. Prepare the request by osp_update_req
1076 * and send them to remote MDT, for sync request, it will wait
1077 * until the reply return, otherwise hand it to ptlrpcd.
1079 * Please refer to osp_trans_create() for transaction type.
1081 * \param[in] env pointer to the thread context
1082 * \param[in] osp pointer to the OSP device
1083 * \param[in] our pointer to the osp_update_request
1085 * \retval 0 for success
1086 * \retval negative error number on failure
1088 static int osp_send_update_req(const struct lu_env *env,
1089 struct osp_device *osp,
1090 struct osp_update_request *our)
1092 struct osp_update_args *args;
1093 struct ptlrpc_request *req;
1094 struct lu_device *top_device;
1095 struct osp_thandle *oth = our->our_th;
1099 LASSERT(oth != NULL);
1100 rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import,
1103 osp_trans_callback(env, oth, rc);
1107 args = ptlrpc_req_async_args(req);
1108 args->oaua_update = our;
1109 osp_thandle_get(oth); /* hold for update interpret */
1110 req->rq_interpret_reply = osp_update_interpret;
1111 if (!oth->ot_super.th_wait_submit && !oth->ot_super.th_sync) {
1112 if (!osp->opd_imp_active || !osp->opd_imp_connected) {
1113 osp_trans_callback(env, oth, rc);
1114 osp_thandle_put(oth);
1115 GOTO(out, rc = -ENOTCONN);
1118 rc = obd_get_request_slot(&osp->opd_obd->u.cli);
1120 osp_trans_callback(env, oth, rc);
1121 osp_thandle_put(oth);
1122 GOTO(out, rc = -ENOTCONN);
1124 args->oaua_flow_control = true;
1126 if (!osp->opd_connect_mdt) {
1127 down_read(&osp->opd_async_updates_rwsem);
1128 args->oaua_count = &osp->opd_async_updates_count;
1129 args->oaua_waitq = &osp->opd_syn_barrier_waitq;
1130 up_read(&osp->opd_async_updates_rwsem);
1131 atomic_inc(args->oaua_count);
1134 ptlrpcd_add_req(req);
1137 osp_thandle_get(oth); /* hold for commit callback */
1138 req->rq_commit_cb = osp_request_commit_cb;
1139 req->rq_cb_data = &oth->ot_super;
1140 args->oaua_flow_control = false;
1142 /* If the transaction is created during MDT recoverying
1143 * process, it means this is an recovery update, we need
1144 * to let OSP send it anyway without checking recoverying
1145 * status, in case the other target is being recoveried
1146 * at the same time, and if we wait here for the import
1147 * to be recoveryed, it might cause deadlock */
1148 top_device = osp->opd_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
1149 if (top_device->ld_obd->obd_recovering)
1150 req->rq_allow_replay = 1;
1152 if (osp->opd_connect_mdt)
1153 osp_get_rpc_lock(osp);
1154 rc = ptlrpc_queue_wait(req);
1155 if (osp->opd_connect_mdt)
1156 osp_put_rpc_lock(osp);
1157 if ((rc == -ENOMEM && req->rq_set == NULL) ||
1158 (req->rq_transno == 0 && !req->rq_committed)) {
1159 if (args->oaua_update != NULL) {
1160 /* If osp_update_interpret is not being called,
1161 * release the osp_thandle */
1162 args->oaua_update = NULL;
1163 osp_thandle_put(oth);
1166 req->rq_cb_data = NULL;
1167 rc = rc == 0 ? req->rq_status : rc;
1168 osp_trans_callback(env, oth, rc);
1169 osp_thandle_put(oth);
1175 ptlrpc_req_finished(req);
1181 * Get local thandle for osp_thandle
1183 * Get the local OSD thandle from the OSP thandle. Currently, there
1184 * are a few OSP API (osp_object_create() and osp_sync_add()) needs
1185 * to update the object on local OSD device.
1187 * If the osp_thandle comes from normal stack (MDD->LOD->OSP), then
1188 * we will get local thandle by thandle_get_sub_by_dt.
1190 * If the osp_thandle is remote thandle (th_top == NULL, only used
1191 * by LFSCK), then it will create a local thandle, and stop it in
1192 * osp_trans_stop(). And this only happens on OSP for OST.
1194 * These are temporary solution, once OSP accessing OSD object is
1195 * being fixed properly, this function should be removed. XXX
1197 * \param[in] env pointer to the thread context
1198 * \param[in] th pointer to the transaction handler
1199 * \param[in] dt pointer to the OSP device
1201 * \retval pointer to the local thandle
1202 * \retval ERR_PTR(errno) if it fails.
1204 struct thandle *osp_get_storage_thandle(const struct lu_env *env,
1206 struct osp_device *osp)
1208 struct osp_thandle *oth;
1209 struct thandle *local_th;
1211 if (th->th_top != NULL)
1212 return thandle_get_sub_by_dt(env, th->th_top,
1215 LASSERT(!osp->opd_connect_mdt);
1216 oth = thandle_to_osp_thandle(th);
1217 if (oth->ot_storage_th != NULL)
1218 return oth->ot_storage_th;
1220 local_th = dt_trans_create(env, osp->opd_storage);
1221 if (IS_ERR(local_th))
1224 oth->ot_storage_th = local_th;
1230 * Set version for the transaction
1232 * Set the version for the transaction and add the request to
1233 * the sending list, then after transaction stop, the request
1234 * will be picked in the order of version, by sending thread.
1236 * \param [in] oth osp thandle to be set version.
1238 * \retval 0 if set version succeeds
1239 * negative errno if set version fails.
1241 int osp_check_and_set_rpc_version(struct osp_thandle *oth,
1242 struct osp_object *obj)
1244 struct osp_device *osp = dt2osp_dev(oth->ot_super.th_dev);
1245 struct osp_updates *ou = osp->opd_update;
1250 if (oth->ot_our->our_version != 0)
1253 spin_lock(&ou->ou_lock);
1254 spin_lock(&oth->ot_our->our_list_lock);
1255 if (obj->opo_stale) {
1256 spin_unlock(&oth->ot_our->our_list_lock);
1257 spin_unlock(&ou->ou_lock);
1261 /* Assign the version and add it to the sending list */
1262 osp_thandle_get(oth);
1263 oth->ot_our->our_version = ou->ou_version++;
1264 list_add_tail(&oth->ot_our->our_list,
1265 &osp->opd_update->ou_list);
1266 oth->ot_our->our_req_ready = 0;
1267 spin_unlock(&oth->ot_our->our_list_lock);
1268 spin_unlock(&ou->ou_lock);
1270 LASSERT(oth->ot_super.th_wait_submit == 1);
1271 CDEBUG(D_INFO, "%s: version "LPU64" oth:version %p:"LPU64"\n",
1272 osp->opd_obd->obd_name, ou->ou_version, oth,
1273 oth->ot_our->our_version);
1279 * Get next OSP update request in the sending list
1280 * Get next OSP update request in the sending list by version number, next
1282 * 1. transaction which does not have a version number.
1283 * 2. transaction whose version == opd_rpc_version.
1285 * \param [in] ou osp update structure.
1286 * \param [out] ourp the pointer holding the next update request.
1288 * \retval true if getting the next transaction.
1289 * \retval false if not getting the next transaction.
1292 osp_get_next_request(struct osp_updates *ou, struct osp_update_request **ourp)
1294 struct osp_update_request *our;
1295 struct osp_update_request *tmp;
1296 bool got_req = false;
1298 spin_lock(&ou->ou_lock);
1299 list_for_each_entry_safe(our, tmp, &ou->ou_list, our_list) {
1300 LASSERT(our->our_th != NULL);
1301 CDEBUG(D_HA, "ou %p version "LPU64" rpc_version "LPU64"\n",
1302 ou, our->our_version, ou->ou_rpc_version);
1303 spin_lock(&our->our_list_lock);
1304 /* Find next osp_update_request in the list */
1305 if (our->our_version == ou->ou_rpc_version &&
1306 our->our_req_ready) {
1307 list_del_init(&our->our_list);
1308 spin_unlock(&our->our_list_lock);
1313 spin_unlock(&our->our_list_lock);
1315 spin_unlock(&ou->ou_lock);
1321 * Invalidate update request
1323 * Invalidate update request in the OSP sending list, so all of
1324 * requests in the sending list will return error, which happens
1325 * when it finds one update (with writing llog) requests fails or
1326 * the OSP is evicted by remote target. see osp_send_update_thread().
1328 * \param[in] osp OSP device whose update requests will be
1331 void osp_invalidate_request(struct osp_device *osp)
1334 struct osp_updates *ou = osp->opd_update;
1335 struct osp_update_request *our;
1336 struct osp_update_request *tmp;
1344 rc = lu_env_init(&env, osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
1346 CERROR("%s: init env error: rc = %d\n", osp->opd_obd->obd_name,
1351 INIT_LIST_HEAD(&list);
1353 spin_lock(&ou->ou_lock);
1354 /* invalidate all of request in the sending list */
1355 list_for_each_entry_safe(our, tmp, &ou->ou_list, our_list) {
1356 spin_lock(&our->our_list_lock);
1357 if (our->our_req_ready)
1358 list_move(&our->our_list, &list);
1360 list_del_init(&our->our_list);
1362 if (our->our_th->ot_super.th_result == 0)
1363 our->our_th->ot_super.th_result = -EIO;
1365 if (our->our_version >= ou->ou_rpc_version)
1366 ou->ou_rpc_version = our->our_version + 1;
1367 spin_unlock(&our->our_list_lock);
1369 CDEBUG(D_HA, "%s invalidate our %p\n", osp->opd_obd->obd_name,
1373 spin_unlock(&ou->ou_lock);
1375 /* invalidate all of request in the sending list */
1376 list_for_each_entry_safe(our, tmp, &list, our_list) {
1377 spin_lock(&our->our_list_lock);
1378 list_del_init(&our->our_list);
1379 spin_unlock(&our->our_list_lock);
1380 osp_trans_callback(&env, our->our_th,
1381 our->our_th->ot_super.th_result);
1382 osp_thandle_put(our->our_th);
1388 * Sending update thread
1390 * Create thread to send update request to other MDTs, this thread will pull
1391 * out update request from the list in OSP by version number, i.e. it will
1392 * make sure the update request with lower version number will be sent first.
1394 * \param[in] arg hold the OSP device.
1396 * \retval 0 if the thread is created successfully.
1397 * \retal negative error if the thread is not created
1400 int osp_send_update_thread(void *arg)
1403 struct osp_device *osp = arg;
1404 struct l_wait_info lwi = { 0 };
1405 struct osp_updates *ou = osp->opd_update;
1406 struct ptlrpc_thread *thread = &osp->opd_update_thread;
1407 struct osp_update_request *our = NULL;
1411 LASSERT(ou != NULL);
1412 rc = lu_env_init(&env, osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
1414 CERROR("%s: init env error: rc = %d\n", osp->opd_obd->obd_name,
1419 thread->t_flags = SVC_RUNNING;
1420 wake_up(&thread->t_ctl_waitq);
1423 l_wait_event(ou->ou_waitq,
1424 !osp_send_update_thread_running(osp) ||
1425 osp_get_next_request(ou, &our), &lwi);
1427 if (!osp_send_update_thread_running(osp)) {
1429 osp_trans_callback(&env, our->our_th, -EINTR);
1430 osp_thandle_put(our->our_th);
1435 LASSERT(our->our_th != NULL);
1436 if (our->our_th->ot_super.th_result != 0) {
1437 osp_trans_callback(&env, our->our_th,
1438 our->our_th->ot_super.th_result);
1439 rc = our->our_th->ot_super.th_result;
1440 } else if (OBD_FAIL_CHECK(OBD_FAIL_INVALIDATE_UPDATE)) {
1442 osp_trans_callback(&env, our->our_th, rc);
1444 rc = osp_send_update_req(&env, osp, our);
1447 /* Update the rpc version */
1448 spin_lock(&ou->ou_lock);
1449 if (our->our_version == ou->ou_rpc_version)
1450 ou->ou_rpc_version++;
1451 spin_unlock(&ou->ou_lock);
1453 /* If one update request fails, let's fail all of the requests
1454 * in the sending list, because the request in the sending
1455 * list are dependent on either other, continue sending these
1456 * request might cause llog or filesystem corruption */
1458 osp_invalidate_request(osp);
1460 /* Balanced for thandle_get in osp_check_and_set_rpc_version */
1461 osp_thandle_put(our->our_th);
1464 thread->t_flags = SVC_STOPPED;
1466 wake_up(&thread->t_ctl_waitq);
1472 * The OSP layer dt_device_operations::dt_trans_start() interface
1473 * to start the transaction.
1475 * If the transaction is a remote transaction, then related remote
1476 * updates will be triggered in the osp_trans_stop().
1477 * Please refer to osp_trans_create() for transaction type.
1479 * \param[in] env pointer to the thread context
1480 * \param[in] dt pointer to the OSP dt_device
1481 * \param[in] th pointer to the transaction handler
1483 * \retval 0 for success
1484 * \retval negative error number on failure
1486 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
1489 struct osp_thandle *oth = thandle_to_osp_thandle(th);
1491 if (oth->ot_super.th_sync)
1492 oth->ot_our->our_flags |= UPDATE_FL_SYNC;
1493 /* For remote thandle, if there are local thandle, start it here*/
1494 if (is_only_remote_trans(th) && oth->ot_storage_th != NULL)
1495 return dt_trans_start(env, oth->ot_storage_th->th_dev,
1496 oth->ot_storage_th);
1501 * The OSP layer dt_device_operations::dt_trans_stop() interface
1502 * to stop the transaction.
1504 * If the transaction is a remote transaction, related remote
1505 * updates will be triggered at the end of this function.
1507 * For synchronous mode update or any failed update, the request
1508 * will be destroyed explicitly when the osp_trans_stop().
1510 * Please refer to osp_trans_create() for transaction type.
1512 * \param[in] env pointer to the thread context
1513 * \param[in] dt pointer to the OSP dt_device
1514 * \param[in] th pointer to the transaction handler
1516 * \retval 0 for success
1517 * \retval negative error number on failure
1519 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
1522 struct osp_thandle *oth = thandle_to_osp_thandle(th);
1523 struct osp_update_request *our = oth->ot_our;
1524 struct osp_device *osp = dt2osp_dev(dt);
1528 /* For remote transaction, if there is local storage thandle,
1530 if (oth->ot_storage_th != NULL && th->th_top == NULL) {
1531 dt_trans_stop(env, oth->ot_storage_th->th_dev,
1532 oth->ot_storage_th);
1533 oth->ot_storage_th = NULL;
1536 if (our == NULL || list_empty(&our->our_req_list)) {
1537 osp_trans_callback(env, oth, th->th_result);
1538 GOTO(out, rc = th->th_result);
1541 if (!osp->opd_connect_mdt) {
1542 osp_trans_callback(env, oth, th->th_result);
1543 rc = osp_send_update_req(env, osp, oth->ot_our);
1547 if (osp->opd_update == NULL ||
1548 !osp_send_update_thread_running(osp)) {
1549 osp_trans_callback(env, oth, -EIO);
1550 GOTO(out, rc = -EIO);
1553 CDEBUG(D_HA, "%s: add oth %p with version "LPU64"\n",
1554 osp->opd_obd->obd_name, oth, our->our_version);
1556 LASSERT(our->our_req_ready == 0);
1557 spin_lock(&our->our_list_lock);
1558 if (likely(!list_empty(&our->our_list))) {
1559 /* notify sending thread */
1560 our->our_req_ready = 1;
1561 wake_up(&osp->opd_update->ou_waitq);
1562 spin_unlock(&our->our_list_lock);
1563 } else if (th->th_result == 0) {
1564 /* if the request does not needs to be serialized,
1565 * read-only request etc, let's send it right away */
1566 spin_unlock(&our->our_list_lock);
1567 rc = osp_send_update_req(env, osp, our);
1569 spin_unlock(&our->our_list_lock);
1570 osp_trans_callback(env, oth, th->th_result);
1573 osp_thandle_put(oth);