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, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osp/osp_sync.c
34 * Lustre OST Proxy Device
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 #define DEBUG_SUBSYSTEM S_MDS
42 #include <linux/kthread.h>
43 #include <lustre_log.h>
44 #include <lustre_update.h>
45 #include "osp_internal.h"
47 static int osp_sync_id_traction_init(struct osp_device *d);
48 static void osp_sync_id_traction_fini(struct osp_device *d);
49 static __u64 osp_sync_id_get(struct osp_device *d, __u64 id);
50 static void osp_sync_remove_from_tracker(struct osp_device *d);
53 * this is a components of OSP implementing synchronization between MDS and OST
54 * it llogs all interesting changes (currently it's uig/gid change and object
55 * destroy) atomically, then makes sure changes hit OST storage
57 * we have 4 queues of work:
59 * the first queue is llog itself, once read a change is stored in 2nd queue
60 * in form of RPC (but RPC isn't fired yet).
62 * the second queue (opd_syn_waiting_for_commit) holds changes awaiting local
63 * commit. once change is committed locally it migrates onto 3rd queue.
65 * the third queue (opd_syn_committed_here) holds changes committed locally,
66 * but not sent to OST (as the pipe can be full). once pipe becomes non-full
67 * we take a change from the queue and fire corresponded RPC.
69 * once RPC is reported committed by OST (using regular last_committed mech.)
70 * the change jumps into 4th queue (opd_syn_committed_there), now we can
71 * cancel corresponded llog record and release RPC
73 * opd_syn_changes is a number of unread llog records (to be processed).
74 * notice this number doesn't include llog records from previous boots.
75 * with OSP_SYN_THRESHOLD we try to batch processing a bit (TO BE IMPLEMENTED)
77 * opd_syn_rpc_in_progress is a number of requests in 2-4 queues.
78 * we control this with OSP_MAX_IN_PROGRESS so that OSP don't consume
79 * too much memory -- how to deal with 1000th OSTs ? batching could help?
81 * opd_syn_rpc_in_flight is a number of RPC in flight.
82 * we control this with OSP_MAX_IN_FLIGHT
85 /* XXX: do math to learn reasonable threshold
86 * should it be ~ number of changes fitting bulk? */
88 #define OSP_SYN_THRESHOLD 10
89 #define OSP_MAX_IN_FLIGHT 8
90 #define OSP_MAX_IN_PROGRESS 4096
92 #define OSP_JOB_MAGIC 0x26112005
94 struct osp_job_req_args {
95 /** bytes reserved for ptlrpc_replay_req() */
96 struct ptlrpc_replay_async_args jra_raa;
97 struct list_head jra_committed_link;
98 struct list_head jra_inflight_link;
102 static inline int osp_sync_running(struct osp_device *d)
104 return !!(d->opd_syn_thread.t_flags & SVC_RUNNING);
108 * Check status: whether OSP thread has stopped
110 * \param[in] d OSP device
112 * \retval 0 still running
115 static inline int osp_sync_stopped(struct osp_device *d)
117 return !!(d->opd_syn_thread.t_flags & SVC_STOPPED);
121 ** Check for new changes to sync
123 * \param[in] d OSP device
125 * \retval 1 there are changes
126 * \retval 0 there are no changes
128 static inline int osp_sync_has_new_job(struct osp_device *d)
130 return ((d->opd_syn_last_processed_id < d->opd_syn_last_used_id) &&
131 (d->opd_syn_last_processed_id < d->opd_syn_last_committed_id))
132 || (d->opd_syn_prev_done == 0);
135 static inline int osp_sync_inflight_conflict(struct osp_device *d,
136 struct llog_rec_hdr *h)
138 struct osp_job_req_args *jra;
142 if (h == NULL || h->lrh_type == LLOG_GEN_REC ||
143 list_empty(&d->opd_syn_inflight_list))
146 memset(&ostid, 0, sizeof(ostid));
147 switch (h->lrh_type) {
149 ostid_set_seq(&ostid, ((struct llog_unlink_rec *)h)->lur_oseq);
150 ostid_set_id(&ostid, ((struct llog_unlink_rec *)h)->lur_oid);
152 case MDS_UNLINK64_REC:
153 fid_to_ostid(&((struct llog_unlink64_rec *)h)->lur_fid, &ostid);
155 case MDS_SETATTR64_REC:
156 ostid = ((struct llog_setattr64_rec *)h)->lsr_oi;
162 spin_lock(&d->opd_syn_lock);
163 list_for_each_entry(jra, &d->opd_syn_inflight_list, jra_inflight_link) {
164 struct ptlrpc_request *req;
165 struct ost_body *body;
167 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
169 req = container_of((void *)jra, struct ptlrpc_request,
171 body = req_capsule_client_get(&req->rq_pill,
175 if (memcmp(&ostid, &body->oa.o_oi, sizeof(ostid)) == 0) {
180 spin_unlock(&d->opd_syn_lock);
185 static inline int osp_sync_low_in_progress(struct osp_device *d)
187 return atomic_read(&d->opd_syn_rpc_in_progress) <
188 d->opd_syn_max_rpc_in_progress;
192 * Check for room in the network pipe to OST
194 * \param[in] d OSP device
196 * \retval 1 there is room
197 * \retval 0 no room, the pipe is full
199 static inline int osp_sync_low_in_flight(struct osp_device *d)
201 return atomic_read(&d->opd_syn_rpc_in_flight) <
202 d->opd_syn_max_rpc_in_flight;
206 * Wake up check for the main sync thread
208 * \param[in] d OSP device
210 * \retval 1 time to wake up
211 * \retval 0 no need to wake up
213 static inline int osp_sync_has_work(struct osp_device *d)
215 /* has new/old changes and low in-progress? */
216 if (osp_sync_has_new_job(d) && osp_sync_low_in_progress(d) &&
217 osp_sync_low_in_flight(d) && d->opd_imp_connected)
220 /* has remotely committed? */
221 if (!list_empty(&d->opd_syn_committed_there))
227 #define osp_sync_check_for_work(d) \
229 if (osp_sync_has_work(d)) { \
230 wake_up(&d->opd_syn_waitq); \
234 void __osp_sync_check_for_work(struct osp_device *d)
236 osp_sync_check_for_work(d);
239 static inline __u64 osp_sync_correct_id(struct osp_device *d,
240 struct llog_rec_hdr *rec)
243 * llog use cyclic store with 32 bit lrh_id
244 * so overflow lrh_id is possible. Range between
245 * last_processed and last_committed is less than
246 * 64745 ^ 2 and less than 2^32 - 1
248 __u64 correct_id = d->opd_syn_last_committed_id;
250 if ((correct_id & 0xffffffffULL) < rec->lrh_id)
251 correct_id -= 0x100000000ULL;
253 correct_id &= ~0xffffffffULL;
254 correct_id |= rec->lrh_id;
259 * Check and return ready-for-new status.
261 * The thread processing llog record uses this function to check whether
262 * it's time to take another record and process it. The number of conditions
263 * must be met: the connection should be ready, RPCs in flight not exceeding
264 * the limit, the record is committed locally, etc (see the lines below).
266 * \param[in] d OSP device
267 * \param[in] rec next llog record to process
269 * \retval 0 not ready
272 static inline int osp_sync_can_process_new(struct osp_device *d,
273 struct llog_rec_hdr *rec)
277 if (unlikely(atomic_read(&d->opd_syn_barrier) > 0))
279 if (unlikely(osp_sync_inflight_conflict(d, rec)))
281 if (!osp_sync_low_in_progress(d))
283 if (!osp_sync_low_in_flight(d))
285 if (!d->opd_imp_connected)
287 if (d->opd_syn_prev_done == 0)
289 if (atomic_read(&d->opd_syn_changes) == 0)
292 osp_sync_correct_id(d, rec) <= d->opd_syn_last_committed_id)
298 * Declare intention to add a new change.
300 * With regard to OSD API, we have to declare any changes ahead. In this
301 * case we declare an intention to add a llog record representing the
302 * change on the local storage.
304 * \param[in] env LU environment provided by the caller
305 * \param[in] o OSP object
306 * \param[in] type type of change: MDS_UNLINK64_REC or MDS_SETATTR64_REC
307 * \param[in] th transaction handle (local)
309 * \retval 0 on success
310 * \retval negative negated errno on error
312 int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
313 llog_op_type type, struct thandle *th)
315 struct osp_thread_info *osi = osp_env_info(env);
316 struct osp_device *d = lu2osp_dev(o->opo_obj.do_lu.lo_dev);
317 struct llog_ctxt *ctxt;
318 struct thandle *storage_th;
323 /* it's a layering violation, to access internals of th,
324 * but we can do this as a sanity check, for a while */
325 LASSERT(th->th_top != NULL);
326 storage_th = thandle_get_sub_by_dt(env, th->th_top, d->opd_storage);
327 if (IS_ERR(storage_th))
328 RETURN(PTR_ERR(storage_th));
331 case MDS_UNLINK64_REC:
332 osi->osi_hdr.lrh_len = sizeof(struct llog_unlink64_rec);
334 case MDS_SETATTR64_REC:
335 osi->osi_hdr.lrh_len = sizeof(struct llog_setattr64_rec);
341 /* we want ->dt_trans_start() to allocate per-thandle structure */
342 storage_th->th_tags |= LCT_OSP_THREAD;
344 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
347 rc = llog_declare_add(env, ctxt->loc_handle, &osi->osi_hdr,
355 * Generate a llog record for a given change.
357 * Generates a llog record for the change passed. The change can be of two
358 * types: unlink and setattr. The record gets an ID which later will be
359 * used to track commit status of the change. For unlink changes, the caller
360 * can supply a starting FID and the count of the objects to destroy. For
361 * setattr the caller should apply attributes to apply.
364 * \param[in] env LU environment provided by the caller
365 * \param[in] d OSP device
366 * \param[in] fid fid of the object the change should be applied to
367 * \param[in] type type of change: MDS_UNLINK64_REC or MDS_SETATTR64_REC
368 * \param[in] count count of objects to destroy
369 * \param[in] th transaction handle (local)
370 * \param[in] attr attributes for setattr
372 * \retval 0 on success
373 * \retval negative negated errno on error
375 static int osp_sync_add_rec(const struct lu_env *env, struct osp_device *d,
376 const struct lu_fid *fid, llog_op_type type,
377 int count, struct thandle *th,
378 const struct lu_attr *attr)
380 struct osp_thread_info *osi = osp_env_info(env);
381 struct llog_ctxt *ctxt;
382 struct osp_txn_info *txn;
383 struct thandle *storage_th;
388 /* it's a layering violation, to access internals of th,
389 * but we can do this as a sanity check, for a while */
390 LASSERT(th->th_top != NULL);
391 storage_th = thandle_get_sub_by_dt(env, th->th_top, d->opd_storage);
392 if (IS_ERR(storage_th))
393 RETURN(PTR_ERR(storage_th));
396 case MDS_UNLINK64_REC:
397 osi->osi_hdr.lrh_len = sizeof(osi->osi_unlink);
398 osi->osi_hdr.lrh_type = MDS_UNLINK64_REC;
399 osi->osi_unlink.lur_fid = *fid;
400 osi->osi_unlink.lur_count = count;
402 case MDS_SETATTR64_REC:
403 rc = fid_to_ostid(fid, &osi->osi_oi);
405 osi->osi_hdr.lrh_len = sizeof(osi->osi_setattr);
406 osi->osi_hdr.lrh_type = MDS_SETATTR64_REC;
407 osi->osi_setattr.lsr_oi = osi->osi_oi;
409 osi->osi_setattr.lsr_uid = attr->la_uid;
410 osi->osi_setattr.lsr_gid = attr->la_gid;
411 osi->osi_setattr.lsr_valid =
412 ((attr->la_valid & LA_UID) ? OBD_MD_FLUID : 0) |
413 ((attr->la_valid & LA_GID) ? OBD_MD_FLGID : 0);
419 txn = osp_txn_info(&storage_th->th_ctx);
422 txn->oti_current_id = osp_sync_id_get(d, txn->oti_current_id);
423 osi->osi_hdr.lrh_id = (txn->oti_current_id & 0xffffffffULL);
424 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
428 rc = llog_add(env, ctxt->loc_handle, &osi->osi_hdr, &osi->osi_cookie,
432 if (likely(rc >= 0)) {
433 CDEBUG(D_OTHER, "%s: new record "DOSTID":%lu/%lu: %d\n",
434 d->opd_obd->obd_name,
435 POSTID(&osi->osi_cookie.lgc_lgl.lgl_oi),
436 (unsigned long)osi->osi_cookie.lgc_lgl.lgl_ogen,
437 (unsigned long)osi->osi_cookie.lgc_index, rc);
438 atomic_inc(&d->opd_syn_changes);
440 /* return 0 always here, error case just cause no llog record */
444 int osp_sync_add(const struct lu_env *env, struct osp_object *o,
445 llog_op_type type, struct thandle *th,
446 const struct lu_attr *attr)
448 return osp_sync_add_rec(env, lu2osp_dev(o->opo_obj.do_lu.lo_dev),
449 lu_object_fid(&o->opo_obj.do_lu), type, 1,
453 int osp_sync_gap(const struct lu_env *env, struct osp_device *d,
454 struct lu_fid *fid, int lost, struct thandle *th)
456 return osp_sync_add_rec(env, d, fid, MDS_UNLINK64_REC, lost, th, NULL);
460 * it's quite obvious we can't maintain all the structures in the memory:
461 * while OST is down, MDS can be processing thousands and thousands of unlinks
462 * filling persistent llogs and in-core respresentation
464 * this doesn't scale at all. so we need basically the following:
465 * a) destroy/setattr append llog records
466 * b) once llog has grown to X records, we process first Y committed records
468 * once record R is found via llog_process(), it becomes committed after any
469 * subsequent commit callback (at the most)
473 * ptlrpc commit callback.
475 * The callback is called by PTLRPC when a RPC is reported committed by the
476 * target (OST). We register the callback for the every RPC applying a change
477 * from the llog. This way we know then the llog records can be cancelled.
478 * Notice the callback can be called when OSP is finishing. We can detect this
479 * checking that actual transno in the request is less or equal of known
480 * committed transno (see osp_sync_process_committed() for the details).
481 * XXX: this is pretty expensive and can be improved later using batching.
483 * \param[in] req request
485 static void osp_sync_request_commit_cb(struct ptlrpc_request *req)
487 struct osp_device *d = req->rq_cb_data;
488 struct osp_job_req_args *jra;
490 CDEBUG(D_HA, "commit req %p, transno %llu\n", req, req->rq_transno);
492 if (unlikely(req->rq_transno == 0))
495 /* do not do any opd_dyn_rpc_* accounting here
496 * it's done in osp_sync_interpret sooner or later */
499 jra = ptlrpc_req_async_args(req);
500 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
501 LASSERT(list_empty(&jra->jra_committed_link));
503 ptlrpc_request_addref(req);
505 spin_lock(&d->opd_syn_lock);
506 list_add(&jra->jra_committed_link, &d->opd_syn_committed_there);
507 spin_unlock(&d->opd_syn_lock);
509 /* XXX: some batching wouldn't hurt */
510 wake_up(&d->opd_syn_waitq);
514 * RPC interpretation callback.
516 * The callback is called by ptlrpc when RPC is replied. Now we have to decide
518 * - put request on a special list to wait until it's committed by the target,
519 * if the request is successful
520 * - schedule llog record cancel if no target object is found
521 * - try later (essentially after reboot) in case of unexpected error
523 * \param[in] env LU environment provided by the caller
524 * \param[in] req request replied
525 * \param[in] aa callback data
526 * \param[in] rc result of RPC
530 static int osp_sync_interpret(const struct lu_env *env,
531 struct ptlrpc_request *req, void *aa, int rc)
533 struct osp_device *d = req->rq_cb_data;
534 struct osp_job_req_args *jra = aa;
536 if (jra->jra_magic != OSP_JOB_MAGIC) {
537 DEBUG_REQ(D_ERROR, req, "bad magic %u\n", jra->jra_magic);
542 CDEBUG(D_HA, "reply req %p/%d, rc %d, transno %u\n", req,
543 atomic_read(&req->rq_refcount),
544 rc, (unsigned) req->rq_transno);
545 LASSERT(rc || req->rq_transno);
549 * we tried to destroy object or update attributes,
550 * but object doesn't exist anymore - cancell llog record
552 LASSERT(req->rq_transno == 0);
553 LASSERT(list_empty(&jra->jra_committed_link));
555 ptlrpc_request_addref(req);
557 spin_lock(&d->opd_syn_lock);
558 list_add(&jra->jra_committed_link, &d->opd_syn_committed_there);
559 spin_unlock(&d->opd_syn_lock);
561 wake_up(&d->opd_syn_waitq);
563 struct obd_import *imp = req->rq_import;
565 * error happened, we'll try to repeat on next boot ?
567 LASSERTF(req->rq_transno == 0 ||
568 req->rq_import_generation < imp->imp_generation,
569 "transno %llu, rc %d, gen: req %d, imp %d\n",
570 req->rq_transno, rc, req->rq_import_generation,
571 imp->imp_generation);
572 if (req->rq_transno == 0) {
573 /* this is the last time we see the request
574 * if transno is not zero, then commit cb
575 * will be called at some point */
576 LASSERT(atomic_read(&d->opd_syn_rpc_in_progress) > 0);
577 atomic_dec(&d->opd_syn_rpc_in_progress);
580 wake_up(&d->opd_syn_waitq);
581 } else if (d->opd_pre != NULL &&
582 unlikely(d->opd_pre_status == -ENOSPC)) {
584 * if current status is -ENOSPC (lack of free space on OST)
585 * then we should poll OST immediately once object destroy
588 osp_statfs_need_now(d);
591 spin_lock(&d->opd_syn_lock);
592 list_del_init(&jra->jra_inflight_link);
593 spin_unlock(&d->opd_syn_lock);
594 LASSERT(atomic_read(&d->opd_syn_rpc_in_flight) > 0);
595 atomic_dec(&d->opd_syn_rpc_in_flight);
596 if (unlikely(atomic_read(&d->opd_syn_barrier) > 0))
597 wake_up(&d->opd_syn_barrier_waitq);
598 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
599 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
600 atomic_read(&d->opd_syn_rpc_in_progress));
602 osp_sync_check_for_work(d);
608 ** Add request to ptlrpc queue.
610 * This is just a tiny helper function to put the request on the sending list
612 * \param[in] d OSP device
613 * \param[in] req request
615 static void osp_sync_send_new_rpc(struct osp_device *d,
616 struct ptlrpc_request *req)
618 struct osp_job_req_args *jra;
620 LASSERT(atomic_read(&d->opd_syn_rpc_in_flight) <=
621 d->opd_syn_max_rpc_in_flight);
623 jra = ptlrpc_req_async_args(req);
624 jra->jra_magic = OSP_JOB_MAGIC;
625 INIT_LIST_HEAD(&jra->jra_committed_link);
626 spin_lock(&d->opd_syn_lock);
627 list_add_tail(&jra->jra_inflight_link, &d->opd_syn_inflight_list);
628 spin_unlock(&d->opd_syn_lock);
630 ptlrpcd_add_req(req);
635 * Allocate and prepare RPC for a new change.
637 * The function allocates and initializes an RPC which will be sent soon to
638 * apply the change to the target OST. The request is initialized from the
639 * llog record passed. Notice only the fields common to all type of changes
642 * \param[in] d OSP device
643 * \param[in] llh llog handle where the record is stored
644 * \param[in] h llog record
645 * \param[in] op type of the change
646 * \param[in] format request format to be used
648 * \retval pointer new request on success
649 * \retval ERR_PTR(errno) on error
651 static struct ptlrpc_request *osp_sync_new_job(struct osp_device *d,
652 struct llog_handle *llh,
653 struct llog_rec_hdr *h,
655 const struct req_format *format)
657 struct ptlrpc_request *req;
658 struct ost_body *body;
659 struct obd_import *imp;
662 /* Prepare the request */
663 imp = d->opd_obd->u.cli.cl_import;
666 if (OBD_FAIL_CHECK(OBD_FAIL_OSP_CHECK_ENOMEM))
667 RETURN(ERR_PTR(-ENOMEM));
669 req = ptlrpc_request_alloc(imp, format);
671 RETURN(ERR_PTR(-ENOMEM));
673 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, op);
675 ptlrpc_req_finished(req);
680 * this is a trick: to save on memory allocations we put cookie
681 * into the request, but don't set corresponded flag in o_valid
682 * so that OST doesn't interpret this cookie. once the request
683 * is committed on OST we take cookie from the request and cancel
685 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
687 body->oa.o_lcookie.lgc_lgl = llh->lgh_id;
688 body->oa.o_lcookie.lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
689 body->oa.o_lcookie.lgc_index = h->lrh_index;
691 req->rq_interpret_reply = osp_sync_interpret;
692 req->rq_commit_cb = osp_sync_request_commit_cb;
695 ptlrpc_request_set_replen(req);
701 * Generate a request for setattr change.
703 * The function prepares a new RPC, initializes it with setattr specific
704 * bits and send the RPC.
706 * \param[in] d OSP device
707 * \param[in] llh llog handle where the record is stored
708 * \param[in] h llog record
710 * \retval 0 on success
711 * \retval 1 on invalid record
712 * \retval negative negated errno on error
714 static int osp_sync_new_setattr_job(struct osp_device *d,
715 struct llog_handle *llh,
716 struct llog_rec_hdr *h)
718 struct llog_setattr64_rec *rec = (struct llog_setattr64_rec *)h;
719 struct ptlrpc_request *req;
720 struct ost_body *body;
723 LASSERT(h->lrh_type == MDS_SETATTR64_REC);
725 if (OBD_FAIL_CHECK(OBD_FAIL_OSP_CHECK_INVALID_REC))
727 /* lsr_valid can only be 0 or have OBD_MD_{FLUID,FLGID} set,
728 * so no bits other than these should be set. */
729 if ((rec->lsr_valid & ~(OBD_MD_FLUID | OBD_MD_FLGID)) != 0) {
730 CERROR("%s: invalid setattr record, lsr_valid:%llu\n",
731 d->opd_obd->obd_name, rec->lsr_valid);
732 /* return 1 on invalid record */
736 req = osp_sync_new_job(d, llh, h, OST_SETATTR, &RQF_OST_SETATTR);
738 RETURN(PTR_ERR(req));
740 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
742 body->oa.o_oi = rec->lsr_oi;
743 body->oa.o_uid = rec->lsr_uid;
744 body->oa.o_gid = rec->lsr_gid;
745 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID;
746 /* old setattr record (prior 2.6.0) doesn't have 'valid' stored,
747 * we assume that both UID and GID are valid in that case. */
748 if (rec->lsr_valid == 0)
749 body->oa.o_valid |= (OBD_MD_FLUID | OBD_MD_FLGID);
751 body->oa.o_valid |= rec->lsr_valid;
753 osp_sync_send_new_rpc(d, req);
758 * Generate a request for unlink change.
760 * The function prepares a new RPC, initializes it with unlink(destroy)
761 * specific bits and sends the RPC. The function is used to handle
762 * llog_unlink_rec which were used in the older versions of Lustre.
763 * Current version uses llog_unlink_rec64.
765 * \param[in] d OSP device
766 * \param[in] llh llog handle where the record is stored
767 * \param[in] h llog record
769 * \retval 0 on success
770 * \retval negative negated errno on error
772 static int osp_sync_new_unlink_job(struct osp_device *d,
773 struct llog_handle *llh,
774 struct llog_rec_hdr *h)
776 struct llog_unlink_rec *rec = (struct llog_unlink_rec *)h;
777 struct ptlrpc_request *req;
778 struct ost_body *body;
781 LASSERT(h->lrh_type == MDS_UNLINK_REC);
783 req = osp_sync_new_job(d, llh, h, OST_DESTROY, &RQF_OST_DESTROY);
785 RETURN(PTR_ERR(req));
787 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
789 ostid_set_seq(&body->oa.o_oi, rec->lur_oseq);
790 ostid_set_id(&body->oa.o_oi, rec->lur_oid);
791 body->oa.o_misc = rec->lur_count;
792 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID;
794 body->oa.o_valid |= OBD_MD_FLOBJCOUNT;
796 osp_sync_send_new_rpc(d, req);
801 * Generate a request for unlink change.
803 * The function prepares a new RPC, initializes it with unlink(destroy)
804 * specific bits and sends the RPC. Depending on the target (MDT or OST)
805 * two different protocols are used. For MDT we use OUT (basically OSD API
806 * updates transferred via a network). For OST we still use the old
807 * protocol (OBD?), originally for compatibility. Later we can start to
808 * use OUT for OST as well, this will allow batching and better code
811 * \param[in] d OSP device
812 * \param[in] llh llog handle where the record is stored
813 * \param[in] h llog record
815 * \retval 0 on success
816 * \retval negative negated errno on error
818 static int osp_sync_new_unlink64_job(struct osp_device *d,
819 struct llog_handle *llh,
820 struct llog_rec_hdr *h)
822 struct llog_unlink64_rec *rec = (struct llog_unlink64_rec *)h;
823 struct ptlrpc_request *req = NULL;
824 struct ost_body *body;
828 LASSERT(h->lrh_type == MDS_UNLINK64_REC);
829 req = osp_sync_new_job(d, llh, h, OST_DESTROY,
832 RETURN(PTR_ERR(req));
834 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
837 rc = fid_to_ostid(&rec->lur_fid, &body->oa.o_oi);
840 body->oa.o_misc = rec->lur_count;
841 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID |
843 osp_sync_send_new_rpc(d, req);
848 * Process llog records.
850 * This function is called to process the llog records committed locally.
851 * In the recovery model used by OSP we can apply a change to a remote
852 * target once corresponding transaction (like posix unlink) is committed
853 * locally so can't revert.
854 * Depending on the llog record type, a given handler is called that is
855 * responsible for preparing and sending the RPC to apply the change.
856 * Special record type LLOG_GEN_REC marking a reboot is cancelled right away.
858 * \param[in] env LU environment provided by the caller
859 * \param[in] d OSP device
860 * \param[in] llh llog handle where the record is stored
861 * \param[in] rec llog record
863 static void osp_sync_process_record(const struct lu_env *env,
864 struct osp_device *d,
865 struct llog_handle *llh,
866 struct llog_rec_hdr *rec)
868 struct llog_handle *cathandle = llh->u.phd.phd_cat_handle;
869 struct llog_cookie cookie;
874 cookie.lgc_lgl = llh->lgh_id;
875 cookie.lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
876 cookie.lgc_index = rec->lrh_index;
878 if (unlikely(rec->lrh_type == LLOG_GEN_REC)) {
879 struct llog_gen_rec *gen = (struct llog_gen_rec *)rec;
881 /* we're waiting for the record generated by this instance */
882 LASSERT(d->opd_syn_prev_done == 0);
883 if (!memcmp(&d->opd_syn_generation, &gen->lgr_gen,
884 sizeof(gen->lgr_gen))) {
885 CDEBUG(D_HA, "processed all old entries\n");
886 d->opd_syn_prev_done = 1;
889 /* cancel any generation record */
890 rc = llog_cat_cancel_records(env, cathandle, 1, &cookie);
896 * now we prepare and fill requests to OST, put them on the queue
897 * and fire after next commit callback
900 /* notice we increment counters before sending RPC, to be consistent
901 * in RPC interpret callback which may happen very quickly */
902 atomic_inc(&d->opd_syn_rpc_in_flight);
903 atomic_inc(&d->opd_syn_rpc_in_progress);
905 switch (rec->lrh_type) {
906 /* case MDS_UNLINK_REC is kept for compatibility */
908 rc = osp_sync_new_unlink_job(d, llh, rec);
910 case MDS_UNLINK64_REC:
911 rc = osp_sync_new_unlink64_job(d, llh, rec);
913 case MDS_SETATTR64_REC:
914 rc = osp_sync_new_setattr_job(d, llh, rec);
917 CERROR("%s: unknown record type: %x\n", d->opd_obd->obd_name,
919 /* treat "unknown record type" as "invalid" */
924 /* For all kinds of records, not matter successful or not,
925 * we should decrease changes and bump last_processed_id.
927 if (d->opd_syn_prev_done) {
928 __u64 correct_id = osp_sync_correct_id(d, rec);
929 LASSERT(atomic_read(&d->opd_syn_changes) > 0);
930 LASSERT(correct_id <= d->opd_syn_last_committed_id);
931 /* NOTE: it's possible to meet same id if
932 * OST stores few stripes of same file
935 /* another thread may be trying to set new value */
937 if (correct_id > d->opd_syn_last_processed_id) {
938 d->opd_syn_last_processed_id = correct_id;
939 wake_up(&d->opd_syn_barrier_waitq);
943 atomic_dec(&d->opd_syn_changes);
946 atomic_dec(&d->opd_syn_rpc_in_flight);
947 atomic_dec(&d->opd_syn_rpc_in_progress);
950 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
951 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
952 atomic_read(&d->opd_syn_rpc_in_progress));
954 /* Delete the invalid record */
956 rc = llog_cat_cancel_records(env, cathandle, 1, &cookie);
958 CERROR("%s: can't delete invalid record: "
959 "fid = "DFID", rec_id = %u, rc = %d\n",
960 d->opd_obd->obd_name,
961 PFID(lu_object_fid(&cathandle->lgh_obj->do_lu)),
965 CDEBUG(D_HA, "found record %x, %d, idx %u, id %u\n",
966 rec->lrh_type, rec->lrh_len, rec->lrh_index, rec->lrh_id);
972 * Cancel llog records for the committed changes.
974 * The function walks through the list of the committed RPCs and cancels
975 * corresponding llog records. see osp_sync_request_commit_cb() for the
978 * \param[in] env LU environment provided by the caller
979 * \param[in] d OSP device
981 static void osp_sync_process_committed(const struct lu_env *env,
982 struct osp_device *d)
984 struct obd_device *obd = d->opd_obd;
985 struct obd_import *imp = obd->u.cli.cl_import;
986 struct ost_body *body;
987 struct ptlrpc_request *req;
988 struct llog_ctxt *ctxt;
989 struct llog_handle *llh;
990 struct list_head list;
995 if (list_empty(&d->opd_syn_committed_there))
999 * if current status is -ENOSPC (lack of free space on OST)
1000 * then we should poll OST immediately once object destroy
1002 * notice: we do this upon commit as well because some backends
1003 * (like DMU) do not release space right away.
1005 if (d->opd_pre != NULL && unlikely(d->opd_pre_status == -ENOSPC))
1006 osp_statfs_need_now(d);
1009 * now cancel them all
1010 * XXX: can we improve this using some batching?
1011 * with batch RPC that'll happen automatically?
1012 * XXX: can we store ctxt in lod_device and save few cycles ?
1014 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1017 llh = ctxt->loc_handle;
1020 INIT_LIST_HEAD(&list);
1021 spin_lock(&d->opd_syn_lock);
1022 list_splice(&d->opd_syn_committed_there, &list);
1023 INIT_LIST_HEAD(&d->opd_syn_committed_there);
1024 spin_unlock(&d->opd_syn_lock);
1026 while (!list_empty(&list)) {
1027 struct osp_job_req_args *jra;
1029 jra = list_entry(list.next, struct osp_job_req_args,
1030 jra_committed_link);
1031 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
1032 list_del_init(&jra->jra_committed_link);
1034 req = container_of((void *)jra, struct ptlrpc_request,
1036 body = req_capsule_client_get(&req->rq_pill,
1039 /* import can be closing, thus all commit cb's are
1040 * called we can check committness directly */
1041 if (req->rq_import_generation == imp->imp_generation) {
1042 rc = llog_cat_cancel_records(env, llh, 1,
1043 &body->oa.o_lcookie);
1045 CERROR("%s: can't cancel record: %d\n",
1048 DEBUG_REQ(D_OTHER, req, "imp_committed = %llu",
1049 imp->imp_peer_committed_transno);
1051 ptlrpc_req_finished(req);
1055 llog_ctxt_put(ctxt);
1057 LASSERT(atomic_read(&d->opd_syn_rpc_in_progress) >= done);
1058 atomic_sub(done, &d->opd_syn_rpc_in_progress);
1059 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
1060 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
1061 atomic_read(&d->opd_syn_rpc_in_progress));
1063 osp_sync_check_for_work(d);
1065 /* wake up the thread if requested to stop:
1066 * it might be waiting for in-progress to complete */
1067 if (unlikely(osp_sync_running(d) == 0))
1068 wake_up(&d->opd_syn_waitq);
1074 * The core of the syncing mechanism.
1076 * This is a callback called by the llog processing function. Essentially it
1077 * suspends llog processing until there is a record to process (it's supposed
1078 * to be committed locally). The function handles RPCs committed by the target
1079 * and cancels corresponding llog records.
1081 * \param[in] env LU environment provided by the caller
1082 * \param[in] llh llog handle we're processing
1083 * \param[in] rec current llog record
1084 * \param[in] data callback data containing a pointer to the device
1086 * \retval 0 to ask the caller (llog_process()) to continue
1087 * \retval LLOG_PROC_BREAK to ask the caller to break
1089 static int osp_sync_process_queues(const struct lu_env *env,
1090 struct llog_handle *llh,
1091 struct llog_rec_hdr *rec,
1094 struct osp_device *d = data;
1097 struct l_wait_info lwi = { 0 };
1099 if (!osp_sync_running(d)) {
1100 CDEBUG(D_HA, "stop llog processing\n");
1101 return LLOG_PROC_BREAK;
1104 /* process requests committed by OST */
1105 osp_sync_process_committed(env, d);
1107 /* if we there are changes to be processed and we have
1108 * resources for this ... do now */
1109 if (osp_sync_can_process_new(d, rec)) {
1111 /* ask llog for another record */
1112 CDEBUG(D_HA, "%u changes, %u in progress,"
1114 atomic_read(&d->opd_syn_changes),
1115 atomic_read(&d->opd_syn_rpc_in_progress),
1116 atomic_read(&d->opd_syn_rpc_in_flight));
1119 osp_sync_process_record(env, d, llh, rec);
1124 if (d->opd_syn_last_processed_id == d->opd_syn_last_used_id)
1125 osp_sync_remove_from_tracker(d);
1127 l_wait_event(d->opd_syn_waitq,
1128 !osp_sync_running(d) ||
1129 osp_sync_can_process_new(d, rec) ||
1130 !list_empty(&d->opd_syn_committed_there),
1138 * This thread runs llog_cat_process() scanner calling our callback
1139 * to process llog records. in the callback we implement tricky
1140 * state machine as we don't want to start scanning of the llog again
1141 * and again, also we don't want to process too many records and send
1142 * too many RPCs a time. so, depending on current load (num of changes
1143 * being synced to OST) the callback can suspend awaiting for some
1144 * new conditions, like syncs completed.
1146 * In order to process llog records left by previous boots and to allow
1147 * llog_process_thread() to find something (otherwise it'd just exit
1148 * immediately) we add a special GENERATATION record on each boot.
1150 * \param[in] _arg a pointer to thread's arguments
1152 * \retval 0 on success
1153 * \retval negative negated errno on error
1155 static int osp_sync_thread(void *_arg)
1157 struct osp_device *d = _arg;
1158 struct ptlrpc_thread *thread = &d->opd_syn_thread;
1159 struct l_wait_info lwi = { 0 };
1160 struct llog_ctxt *ctxt;
1161 struct obd_device *obd = d->opd_obd;
1162 struct llog_handle *llh;
1168 rc = lu_env_init(&env, LCT_LOCAL);
1170 CERROR("%s: can't initialize env: rc = %d\n",
1175 spin_lock(&d->opd_syn_lock);
1176 thread->t_flags = SVC_RUNNING;
1177 spin_unlock(&d->opd_syn_lock);
1178 wake_up(&thread->t_ctl_waitq);
1180 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1182 CERROR("can't get appropriate context\n");
1183 GOTO(out, rc = -EINVAL);
1186 llh = ctxt->loc_handle;
1188 CERROR("can't get llh\n");
1189 llog_ctxt_put(ctxt);
1190 GOTO(out, rc = -EINVAL);
1193 rc = llog_cat_process(&env, llh, osp_sync_process_queues, d, 0, 0);
1195 CERROR("%s: llog process with osp_sync_process_queues "
1196 "failed: %d\n", d->opd_obd->obd_name, rc);
1199 LASSERTF(rc == 0 || rc == LLOG_PROC_BREAK,
1200 "%u changes, %u in progress, %u in flight: %d\n",
1201 atomic_read(&d->opd_syn_changes),
1202 atomic_read(&d->opd_syn_rpc_in_progress),
1203 atomic_read(&d->opd_syn_rpc_in_flight), rc);
1205 /* we don't expect llog_process_thread() to exit till umount */
1206 LASSERTF(thread->t_flags != SVC_RUNNING,
1207 "%u changes, %u in progress, %u in flight\n",
1208 atomic_read(&d->opd_syn_changes),
1209 atomic_read(&d->opd_syn_rpc_in_progress),
1210 atomic_read(&d->opd_syn_rpc_in_flight));
1212 /* wait till all the requests are completed */
1214 while (atomic_read(&d->opd_syn_rpc_in_progress) > 0) {
1215 osp_sync_process_committed(&env, d);
1217 lwi = LWI_TIMEOUT(cfs_time_seconds(5), NULL, NULL);
1218 rc = l_wait_event(d->opd_syn_waitq,
1219 atomic_read(&d->opd_syn_rpc_in_progress) == 0,
1221 if (rc == -ETIMEDOUT)
1223 LASSERTF(count < 10, "%s: %d %d %sempty\n",
1224 d->opd_obd->obd_name,
1225 atomic_read(&d->opd_syn_rpc_in_progress),
1226 atomic_read(&d->opd_syn_rpc_in_flight),
1227 list_empty(&d->opd_syn_committed_there) ? "" : "!");
1232 llog_cat_close(&env, llh);
1233 rc = llog_cleanup(&env, ctxt);
1235 CERROR("can't cleanup llog: %d\n", rc);
1237 LASSERTF(atomic_read(&d->opd_syn_rpc_in_progress) == 0,
1238 "%s: %d %d %sempty\n",
1239 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_progress),
1240 atomic_read(&d->opd_syn_rpc_in_flight),
1241 list_empty(&d->opd_syn_committed_there) ? "" : "!");
1243 thread->t_flags = SVC_STOPPED;
1245 wake_up(&thread->t_ctl_waitq);
1255 * Initializes the llog. Specific llog to be used depends on the type of the
1256 * target OSP represents (OST or MDT). The function adds appends a new llog
1257 * record to mark the place where the records associated with this boot
1260 * \param[in] env LU environment provided by the caller
1261 * \param[in] d OSP device
1263 * \retval 0 on success
1264 * \retval negative negated errno on error
1266 static int osp_sync_llog_init(const struct lu_env *env, struct osp_device *d)
1268 struct osp_thread_info *osi = osp_env_info(env);
1269 struct lu_fid *fid = &osi->osi_fid;
1270 struct llog_handle *lgh = NULL;
1271 struct obd_device *obd = d->opd_obd;
1272 struct llog_ctxt *ctxt;
1280 * open llog corresponding to our OST
1282 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1283 obd->obd_lvfs_ctxt.dt = d->opd_storage;
1285 lu_local_obj_fid(fid, LLOG_CATALOGS_OID);
1287 rc = llog_osd_get_cat_list(env, d->opd_storage, d->opd_index, 1,
1288 &osi->osi_cid, fid);
1290 if (rc != -EFAULT) {
1291 CERROR("%s: can't get id from catalogs: rc = %d\n",
1296 /* After sparse OST indices is supported, the CATALOG file
1297 * may become a sparse file that results in failure on
1298 * reading. Skip this error as the llog will be created
1300 memset(&osi->osi_cid, 0, sizeof(osi->osi_cid));
1304 CDEBUG(D_INFO, "%s: Init llog for %d - catid "DOSTID":%x\n",
1305 obd->obd_name, d->opd_index,
1306 POSTID(&osi->osi_cid.lci_logid.lgl_oi),
1307 osi->osi_cid.lci_logid.lgl_ogen);
1309 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_MDS_OST_ORIG_CTXT,
1310 d->opd_storage->dd_lu_dev.ld_obd,
1311 &osp_mds_ost_orig_logops);
1315 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1318 if (likely(logid_id(&osi->osi_cid.lci_logid) != 0)) {
1319 rc = llog_open(env, ctxt, &lgh, &osi->osi_cid.lci_logid, NULL,
1321 /* re-create llog if it is missing */
1323 logid_set_id(&osi->osi_cid.lci_logid, 0);
1325 GOTO(out_cleanup, rc);
1328 if (unlikely(logid_id(&osi->osi_cid.lci_logid) == 0)) {
1329 rc = llog_open_create(env, ctxt, &lgh, NULL, NULL);
1331 GOTO(out_cleanup, rc);
1332 osi->osi_cid.lci_logid = lgh->lgh_id;
1335 LASSERT(lgh != NULL);
1336 ctxt->loc_handle = lgh;
1338 rc = llog_init_handle(env, lgh, LLOG_F_IS_CAT, NULL);
1340 GOTO(out_close, rc);
1342 rc = llog_osd_put_cat_list(env, d->opd_storage, d->opd_index, 1,
1343 &osi->osi_cid, fid);
1345 GOTO(out_close, rc);
1348 * put a mark in the llog till which we'll be processing
1349 * old records restless
1351 d->opd_syn_generation.mnt_cnt = cfs_time_current();
1352 d->opd_syn_generation.conn_cnt = cfs_time_current();
1354 osi->osi_hdr.lrh_type = LLOG_GEN_REC;
1355 osi->osi_hdr.lrh_len = sizeof(osi->osi_gen);
1357 memcpy(&osi->osi_gen.lgr_gen, &d->opd_syn_generation,
1358 sizeof(osi->osi_gen.lgr_gen));
1360 rc = llog_cat_add(env, lgh, &osi->osi_gen.lgr_hdr, &osi->osi_cookie);
1362 GOTO(out_close, rc);
1363 llog_ctxt_put(ctxt);
1366 llog_cat_close(env, lgh);
1368 llog_cleanup(env, ctxt);
1373 * Cleanup llog used for syncing.
1375 * Closes and cleanups the llog. The function is called when the device is
1378 * \param[in] env LU environment provided by the caller
1379 * \param[in] d OSP device
1381 static void osp_sync_llog_fini(const struct lu_env *env, struct osp_device *d)
1383 struct llog_ctxt *ctxt;
1385 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
1387 llog_cat_close(env, ctxt->loc_handle);
1388 llog_cleanup(env, ctxt);
1392 * Initialization of the sync component of OSP.
1394 * Initializes the llog and starts a new thread to handle the changes to
1395 * the remote target (OST or MDT).
1397 * \param[in] env LU environment provided by the caller
1398 * \param[in] d OSP device
1400 * \retval 0 on success
1401 * \retval negative negated errno on error
1403 int osp_sync_init(const struct lu_env *env, struct osp_device *d)
1405 struct l_wait_info lwi = { 0 };
1406 struct task_struct *task;
1411 rc = osp_sync_id_traction_init(d);
1416 * initialize llog storing changes
1418 rc = osp_sync_llog_init(env, d);
1420 CERROR("%s: can't initialize llog: rc = %d\n",
1421 d->opd_obd->obd_name, rc);
1426 * Start synchronization thread
1428 d->opd_syn_max_rpc_in_flight = OSP_MAX_IN_FLIGHT;
1429 d->opd_syn_max_rpc_in_progress = OSP_MAX_IN_PROGRESS;
1430 spin_lock_init(&d->opd_syn_lock);
1431 init_waitqueue_head(&d->opd_syn_waitq);
1432 init_waitqueue_head(&d->opd_syn_barrier_waitq);
1433 init_waitqueue_head(&d->opd_syn_thread.t_ctl_waitq);
1434 INIT_LIST_HEAD(&d->opd_syn_inflight_list);
1435 INIT_LIST_HEAD(&d->opd_syn_committed_there);
1437 task = kthread_run(osp_sync_thread, d, "osp-syn-%u-%u",
1438 d->opd_index, d->opd_group);
1441 CERROR("%s: cannot start sync thread: rc = %d\n",
1442 d->opd_obd->obd_name, rc);
1446 l_wait_event(d->opd_syn_thread.t_ctl_waitq,
1447 osp_sync_running(d) || osp_sync_stopped(d), &lwi);
1451 osp_sync_llog_fini(env, d);
1453 osp_sync_id_traction_fini(d);
1458 * Stop the syncing thread.
1460 * Asks the syncing thread to stop and wait until it's stopped.
1462 * \param[in] d OSP device
1466 int osp_sync_fini(struct osp_device *d)
1468 struct ptlrpc_thread *thread = &d->opd_syn_thread;
1472 thread->t_flags = SVC_STOPPING;
1473 wake_up(&d->opd_syn_waitq);
1474 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
1477 * unregister transaction callbacks only when sync thread
1478 * has finished operations with llog
1480 osp_sync_id_traction_fini(d);
1485 static DEFINE_MUTEX(osp_id_tracker_sem);
1486 static struct list_head osp_id_tracker_list =
1487 LIST_HEAD_INIT(osp_id_tracker_list);
1490 * OSD commit callback.
1492 * The function is used as a local OSD commit callback to track the highest
1493 * committed llog record id. see osp_sync_id_traction_init() for the details.
1495 * \param[in] th local transaction handle committed
1496 * \param[in] cookie commit callback data (our private structure)
1498 static void osp_sync_tracker_commit_cb(struct thandle *th, void *cookie)
1500 struct osp_id_tracker *tr = cookie;
1501 struct osp_device *d;
1502 struct osp_txn_info *txn;
1506 txn = osp_txn_info(&th->th_ctx);
1507 if (txn == NULL || txn->oti_current_id < tr->otr_committed_id)
1510 spin_lock(&tr->otr_lock);
1511 if (likely(txn->oti_current_id > tr->otr_committed_id)) {
1512 CDEBUG(D_OTHER, "committed: %llu -> %llu\n",
1513 tr->otr_committed_id, txn->oti_current_id);
1514 tr->otr_committed_id = txn->oti_current_id;
1516 list_for_each_entry(d, &tr->otr_wakeup_list,
1518 d->opd_syn_last_committed_id = tr->otr_committed_id;
1519 wake_up(&d->opd_syn_waitq);
1522 spin_unlock(&tr->otr_lock);
1526 * Initialize commit tracking mechanism.
1528 * Some setups may have thousands of OSTs and each will be represented by OSP.
1529 * Meaning order of magnitute many more changes to apply every second. In order
1530 * to keep the number of commit callbacks low this mechanism was introduced.
1531 * The mechanism is very similar to transno used by MDT service: it's an single
1532 * ID stream which can be assigned by any OSP to its llog records. The tricky
1533 * part is that ID is stored in per-transaction data and re-used by all the OSPs
1534 * involved in that transaction. Then all these OSPs are woken up utilizing a single OSD commit callback.
1536 * The function initializes the data used by the tracker described above.
1537 * A singler tracker per OSD device is created.
1539 * \param[in] d OSP device
1541 * \retval 0 on success
1542 * \retval negative negated errno on error
1544 static int osp_sync_id_traction_init(struct osp_device *d)
1546 struct osp_id_tracker *tr, *found = NULL;
1550 LASSERT(d->opd_storage);
1551 LASSERT(d->opd_syn_tracker == NULL);
1552 INIT_LIST_HEAD(&d->opd_syn_ontrack);
1554 mutex_lock(&osp_id_tracker_sem);
1555 list_for_each_entry(tr, &osp_id_tracker_list, otr_list) {
1556 if (tr->otr_dev == d->opd_storage) {
1557 LASSERT(atomic_read(&tr->otr_refcount));
1558 atomic_inc(&tr->otr_refcount);
1559 d->opd_syn_tracker = tr;
1565 if (found == NULL) {
1569 d->opd_syn_tracker = tr;
1570 spin_lock_init(&tr->otr_lock);
1571 tr->otr_dev = d->opd_storage;
1572 tr->otr_next_id = 1;
1573 tr->otr_committed_id = 0;
1574 atomic_set(&tr->otr_refcount, 1);
1575 INIT_LIST_HEAD(&tr->otr_wakeup_list);
1576 list_add(&tr->otr_list, &osp_id_tracker_list);
1577 tr->otr_tx_cb.dtc_txn_commit =
1578 osp_sync_tracker_commit_cb;
1579 tr->otr_tx_cb.dtc_cookie = tr;
1580 tr->otr_tx_cb.dtc_tag = LCT_MD_THREAD;
1581 dt_txn_callback_add(d->opd_storage, &tr->otr_tx_cb);
1585 mutex_unlock(&osp_id_tracker_sem);
1591 * Release commit tracker.
1593 * Decrease a refcounter on the tracker used by the given OSP device \a d.
1594 * If no more users left, then the tracker is released.
1596 * \param[in] d OSP device
1598 static void osp_sync_id_traction_fini(struct osp_device *d)
1600 struct osp_id_tracker *tr;
1605 tr = d->opd_syn_tracker;
1611 osp_sync_remove_from_tracker(d);
1613 mutex_lock(&osp_id_tracker_sem);
1614 if (atomic_dec_and_test(&tr->otr_refcount)) {
1615 dt_txn_callback_del(d->opd_storage, &tr->otr_tx_cb);
1616 LASSERT(list_empty(&tr->otr_wakeup_list));
1617 list_del(&tr->otr_list);
1619 d->opd_syn_tracker = NULL;
1621 mutex_unlock(&osp_id_tracker_sem);
1627 * Generate a new ID on a tracker.
1629 * Generates a new ID using the tracker associated with the given OSP device
1630 * \a d, if the given ID \a id is non-zero. Unconditially adds OSP device to
1631 * the wakeup list, so OSP won't miss when a transaction using the ID is
1634 * \param[in] d OSP device
1635 * \param[in] id 0 or ID generated previously
1637 * \retval ID the caller should use
1639 static __u64 osp_sync_id_get(struct osp_device *d, __u64 id)
1641 struct osp_id_tracker *tr;
1643 tr = d->opd_syn_tracker;
1646 /* XXX: we can improve this introducing per-cpu preallocated ids? */
1647 spin_lock(&tr->otr_lock);
1648 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_TRACK_OVERFLOW))
1649 tr->otr_next_id = 0xfffffff0;
1651 if (unlikely(tr->otr_next_id <= d->opd_syn_last_used_id)) {
1652 spin_unlock(&tr->otr_lock);
1653 CERROR("%s: next %llu, last synced %llu\n",
1654 d->opd_obd->obd_name, tr->otr_next_id,
1655 d->opd_syn_last_used_id);
1660 id = tr->otr_next_id++;
1661 if (id > d->opd_syn_last_used_id)
1662 d->opd_syn_last_used_id = id;
1663 if (list_empty(&d->opd_syn_ontrack))
1664 list_add(&d->opd_syn_ontrack, &tr->otr_wakeup_list);
1665 spin_unlock(&tr->otr_lock);
1666 CDEBUG(D_OTHER, "new id %llu\n", id);
1672 * Stop to propagate commit status to OSP.
1674 * If the OSP does not have any llog records she's waiting to commit, then
1675 * it is possible to unsubscribe from wakeups from the tracking using this
1678 * \param[in] d OSP device not willing to wakeup
1680 static void osp_sync_remove_from_tracker(struct osp_device *d)
1682 struct osp_id_tracker *tr;
1684 tr = d->opd_syn_tracker;
1687 if (list_empty(&d->opd_syn_ontrack))
1690 spin_lock(&tr->otr_lock);
1691 list_del_init(&d->opd_syn_ontrack);
1692 spin_unlock(&tr->otr_lock);