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, 2016, 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;
99 struct llog_cookie jra_lcookie;
103 static inline int osp_sync_running(struct osp_device *d)
105 return !!(d->opd_syn_thread.t_flags & SVC_RUNNING);
109 * Check status: whether OSP thread has stopped
111 * \param[in] d OSP device
113 * \retval 0 still running
116 static inline int osp_sync_stopped(struct osp_device *d)
118 return !!(d->opd_syn_thread.t_flags & SVC_STOPPED);
122 ** Check for new changes to sync
124 * \param[in] d OSP device
126 * \retval 1 there are changes
127 * \retval 0 there are no changes
129 static inline int osp_sync_has_new_job(struct osp_device *d)
131 return ((d->opd_syn_last_processed_id < d->opd_syn_last_used_id) &&
132 (d->opd_syn_last_processed_id < d->opd_syn_last_committed_id))
133 || (d->opd_syn_prev_done == 0);
136 static inline int osp_sync_inflight_conflict(struct osp_device *d,
137 struct llog_rec_hdr *h)
139 struct osp_job_req_args *jra;
143 if (h == NULL || h->lrh_type == LLOG_GEN_REC ||
144 list_empty(&d->opd_syn_inflight_list))
147 memset(&ostid, 0, sizeof(ostid));
148 switch (h->lrh_type) {
149 case MDS_UNLINK_REC: {
150 struct llog_unlink_rec *unlink = (struct llog_unlink_rec *)h;
152 ostid_set_seq(&ostid, unlink->lur_oseq);
153 if (ostid_set_id(&ostid, unlink->lur_oid)) {
154 CERROR("Bad %llu to set " DOSTID "\n",
155 (unsigned long long)(unlink->lur_oid),
161 case MDS_UNLINK64_REC:
162 fid_to_ostid(&((struct llog_unlink64_rec *)h)->lur_fid, &ostid);
164 case MDS_SETATTR64_REC:
165 ostid = ((struct llog_setattr64_rec *)h)->lsr_oi;
171 spin_lock(&d->opd_syn_lock);
172 list_for_each_entry(jra, &d->opd_syn_inflight_list, jra_inflight_link) {
173 struct ptlrpc_request *req;
174 struct ost_body *body;
176 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
178 req = container_of((void *)jra, struct ptlrpc_request,
180 body = req_capsule_client_get(&req->rq_pill,
184 if (memcmp(&ostid, &body->oa.o_oi, sizeof(ostid)) == 0) {
189 spin_unlock(&d->opd_syn_lock);
194 static inline int osp_sync_low_in_progress(struct osp_device *d)
196 return atomic_read(&d->opd_syn_rpc_in_progress) <
197 d->opd_syn_max_rpc_in_progress;
201 * Check for room in the network pipe to OST
203 * \param[in] d OSP device
205 * \retval 1 there is room
206 * \retval 0 no room, the pipe is full
208 static inline int osp_sync_low_in_flight(struct osp_device *d)
210 return atomic_read(&d->opd_syn_rpc_in_flight) <
211 d->opd_syn_max_rpc_in_flight;
215 * Wake up check for the main sync thread
217 * \param[in] d OSP device
219 * \retval 1 time to wake up
220 * \retval 0 no need to wake up
222 static inline int osp_sync_has_work(struct osp_device *d)
224 /* has new/old changes and low in-progress? */
225 if (osp_sync_has_new_job(d) && osp_sync_low_in_progress(d) &&
226 osp_sync_low_in_flight(d) && d->opd_imp_connected)
229 /* has remotely committed? */
230 if (!list_empty(&d->opd_syn_committed_there))
236 #define osp_sync_check_for_work(d) \
238 if (osp_sync_has_work(d)) { \
239 wake_up(&d->opd_syn_waitq); \
243 void __osp_sync_check_for_work(struct osp_device *d)
245 osp_sync_check_for_work(d);
248 static inline __u64 osp_sync_correct_id(struct osp_device *d,
249 struct llog_rec_hdr *rec)
252 * llog use cyclic store with 32 bit lrh_id
253 * so overflow lrh_id is possible. Range between
254 * last_processed and last_committed is less than
255 * 64745 ^ 2 and less than 2^32 - 1
257 __u64 correct_id = d->opd_syn_last_committed_id;
259 if ((correct_id & 0xffffffffULL) < rec->lrh_id)
260 correct_id -= 0x100000000ULL;
262 correct_id &= ~0xffffffffULL;
263 correct_id |= rec->lrh_id;
268 * Check and return ready-for-new status.
270 * The thread processing llog record uses this function to check whether
271 * it's time to take another record and process it. The number of conditions
272 * must be met: the connection should be ready, RPCs in flight not exceeding
273 * the limit, the record is committed locally, etc (see the lines below).
275 * \param[in] d OSP device
276 * \param[in] rec next llog record to process
278 * \retval 0 not ready
281 static inline int osp_sync_can_process_new(struct osp_device *d,
282 struct llog_rec_hdr *rec)
286 if (unlikely(atomic_read(&d->opd_syn_barrier) > 0))
288 if (unlikely(osp_sync_inflight_conflict(d, rec)))
290 if (!osp_sync_low_in_progress(d))
292 if (!osp_sync_low_in_flight(d))
294 if (!d->opd_imp_connected)
296 if (d->opd_syn_prev_done == 0)
298 if (atomic_read(&d->opd_syn_changes) == 0)
301 osp_sync_correct_id(d, rec) <= d->opd_syn_last_committed_id)
307 * Declare intention to add a new change.
309 * With regard to OSD API, we have to declare any changes ahead. In this
310 * case we declare an intention to add a llog record representing the
311 * change on the local storage.
313 * \param[in] env LU environment provided by the caller
314 * \param[in] o OSP object
315 * \param[in] type type of change: MDS_UNLINK64_REC or MDS_SETATTR64_REC
316 * \param[in] th transaction handle (local)
318 * \retval 0 on success
319 * \retval negative negated errno on error
321 int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
322 llog_op_type type, struct thandle *th)
324 struct osp_thread_info *osi = osp_env_info(env);
325 struct osp_device *d = lu2osp_dev(o->opo_obj.do_lu.lo_dev);
326 struct llog_ctxt *ctxt;
327 struct thandle *storage_th;
332 /* it's a layering violation, to access internals of th,
333 * but we can do this as a sanity check, for a while */
334 LASSERT(th->th_top != NULL);
335 storage_th = thandle_get_sub_by_dt(env, th->th_top, d->opd_storage);
336 if (IS_ERR(storage_th))
337 RETURN(PTR_ERR(storage_th));
340 case MDS_UNLINK64_REC:
341 osi->osi_hdr.lrh_len = sizeof(struct llog_unlink64_rec);
343 case MDS_SETATTR64_REC:
344 osi->osi_hdr.lrh_len = sizeof(struct llog_setattr64_rec_v2);
350 /* we want ->dt_trans_start() to allocate per-thandle structure */
351 storage_th->th_tags |= LCT_OSP_THREAD;
353 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
356 rc = llog_declare_add(env, ctxt->loc_handle, &osi->osi_hdr,
364 * Generate a llog record for a given change.
366 * Generates a llog record for the change passed. The change can be of two
367 * types: unlink and setattr. The record gets an ID which later will be
368 * used to track commit status of the change. For unlink changes, the caller
369 * can supply a starting FID and the count of the objects to destroy. For
370 * setattr the caller should apply attributes to apply.
373 * \param[in] env LU environment provided by the caller
374 * \param[in] d OSP device
375 * \param[in] fid fid of the object the change should be applied to
376 * \param[in] type type of change: MDS_UNLINK64_REC or MDS_SETATTR64_REC
377 * \param[in] count count of objects to destroy
378 * \param[in] th transaction handle (local)
379 * \param[in] attr attributes for setattr
381 * \retval 0 on success
382 * \retval negative negated errno on error
384 static int osp_sync_add_rec(const struct lu_env *env, struct osp_device *d,
385 const struct lu_fid *fid, llog_op_type type,
386 int count, struct thandle *th,
387 const struct lu_attr *attr)
389 struct osp_thread_info *osi = osp_env_info(env);
390 struct llog_ctxt *ctxt;
391 struct osp_txn_info *txn;
392 struct thandle *storage_th;
397 /* it's a layering violation, to access internals of th,
398 * but we can do this as a sanity check, for a while */
399 LASSERT(th->th_top != NULL);
400 storage_th = thandle_get_sub_by_dt(env, th->th_top, d->opd_storage);
401 if (IS_ERR(storage_th))
402 RETURN(PTR_ERR(storage_th));
405 case MDS_UNLINK64_REC:
406 osi->osi_hdr.lrh_len = sizeof(osi->osi_unlink);
407 osi->osi_hdr.lrh_type = MDS_UNLINK64_REC;
408 osi->osi_unlink.lur_fid = *fid;
409 osi->osi_unlink.lur_count = count;
411 case MDS_SETATTR64_REC:
412 rc = fid_to_ostid(fid, &osi->osi_oi);
414 osi->osi_hdr.lrh_len = sizeof(osi->osi_setattr);
415 osi->osi_hdr.lrh_type = MDS_SETATTR64_REC;
416 osi->osi_setattr.lsr_oi = osi->osi_oi;
418 osi->osi_setattr.lsr_uid = attr->la_uid;
419 osi->osi_setattr.lsr_gid = attr->la_gid;
420 osi->osi_setattr.lsr_projid = attr->la_projid;
421 osi->osi_setattr.lsr_valid =
422 ((attr->la_valid & LA_UID) ? OBD_MD_FLUID : 0) |
423 ((attr->la_valid & LA_GID) ? OBD_MD_FLGID : 0) |
424 ((attr->la_valid & LA_PROJID) ? OBD_MD_FLPROJID : 0);
430 txn = osp_txn_info(&storage_th->th_ctx);
433 txn->oti_current_id = osp_sync_id_get(d, txn->oti_current_id);
434 osi->osi_hdr.lrh_id = (txn->oti_current_id & 0xffffffffULL);
435 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
439 rc = llog_add(env, ctxt->loc_handle, &osi->osi_hdr, &osi->osi_cookie,
443 if (likely(rc >= 0)) {
444 CDEBUG(D_OTHER, "%s: new record "DFID":%x.%u: rc = %d\n",
445 d->opd_obd->obd_name,
446 PFID(&osi->osi_cookie.lgc_lgl.lgl_oi.oi_fid),
447 osi->osi_cookie.lgc_lgl.lgl_ogen,
448 osi->osi_cookie.lgc_index, rc);
449 atomic_inc(&d->opd_syn_changes);
451 /* return 0 always here, error case just cause no llog record */
455 int osp_sync_add(const struct lu_env *env, struct osp_object *o,
456 llog_op_type type, struct thandle *th,
457 const struct lu_attr *attr)
459 return osp_sync_add_rec(env, lu2osp_dev(o->opo_obj.do_lu.lo_dev),
460 lu_object_fid(&o->opo_obj.do_lu), type, 1,
464 int osp_sync_gap(const struct lu_env *env, struct osp_device *d,
465 struct lu_fid *fid, int lost, struct thandle *th)
467 return osp_sync_add_rec(env, d, fid, MDS_UNLINK64_REC, lost, th, NULL);
471 * it's quite obvious we can't maintain all the structures in the memory:
472 * while OST is down, MDS can be processing thousands and thousands of unlinks
473 * filling persistent llogs and in-core respresentation
475 * this doesn't scale at all. so we need basically the following:
476 * a) destroy/setattr append llog records
477 * b) once llog has grown to X records, we process first Y committed records
479 * once record R is found via llog_process(), it becomes committed after any
480 * subsequent commit callback (at the most)
484 * ptlrpc commit callback.
486 * The callback is called by PTLRPC when a RPC is reported committed by the
487 * target (OST). We register the callback for the every RPC applying a change
488 * from the llog. This way we know then the llog records can be cancelled.
489 * Notice the callback can be called when OSP is finishing. We can detect this
490 * checking that actual transno in the request is less or equal of known
491 * committed transno (see osp_sync_process_committed() for the details).
492 * XXX: this is pretty expensive and can be improved later using batching.
494 * \param[in] req request
496 static void osp_sync_request_commit_cb(struct ptlrpc_request *req)
498 struct osp_device *d = req->rq_cb_data;
499 struct osp_job_req_args *jra;
501 CDEBUG(D_HA, "commit req %p, transno %llu\n", req, req->rq_transno);
503 if (unlikely(req->rq_transno == 0))
506 /* do not do any opd_dyn_rpc_* accounting here
507 * it's done in osp_sync_interpret sooner or later */
510 jra = ptlrpc_req_async_args(req);
511 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
512 LASSERT(list_empty(&jra->jra_committed_link));
514 ptlrpc_request_addref(req);
516 spin_lock(&d->opd_syn_lock);
517 list_add(&jra->jra_committed_link, &d->opd_syn_committed_there);
518 spin_unlock(&d->opd_syn_lock);
520 /* XXX: some batching wouldn't hurt */
521 wake_up(&d->opd_syn_waitq);
525 * RPC interpretation callback.
527 * The callback is called by ptlrpc when RPC is replied. Now we have to decide
529 * - put request on a special list to wait until it's committed by the target,
530 * if the request is successful
531 * - schedule llog record cancel if no target object is found
532 * - try later (essentially after reboot) in case of unexpected error
534 * \param[in] env LU environment provided by the caller
535 * \param[in] req request replied
536 * \param[in] aa callback data
537 * \param[in] rc result of RPC
541 static int osp_sync_interpret(const struct lu_env *env,
542 struct ptlrpc_request *req, void *aa, int rc)
544 struct osp_device *d = req->rq_cb_data;
545 struct osp_job_req_args *jra = aa;
547 if (jra->jra_magic != OSP_JOB_MAGIC) {
548 DEBUG_REQ(D_ERROR, req, "bad magic %u\n", jra->jra_magic);
553 CDEBUG(D_HA, "reply req %p/%d, rc %d, transno %u\n", req,
554 atomic_read(&req->rq_refcount),
555 rc, (unsigned) req->rq_transno);
556 LASSERT(rc || req->rq_transno);
560 * we tried to destroy object or update attributes,
561 * but object doesn't exist anymore - cancell llog record
563 LASSERT(req->rq_transno == 0);
564 LASSERT(list_empty(&jra->jra_committed_link));
566 ptlrpc_request_addref(req);
568 spin_lock(&d->opd_syn_lock);
569 list_add(&jra->jra_committed_link, &d->opd_syn_committed_there);
570 spin_unlock(&d->opd_syn_lock);
572 wake_up(&d->opd_syn_waitq);
574 struct obd_import *imp = req->rq_import;
576 * error happened, we'll try to repeat on next boot ?
578 LASSERTF(req->rq_transno == 0 ||
579 req->rq_import_generation < imp->imp_generation,
580 "transno %llu, rc %d, gen: req %d, imp %d\n",
581 req->rq_transno, rc, req->rq_import_generation,
582 imp->imp_generation);
583 if (req->rq_transno == 0) {
584 /* this is the last time we see the request
585 * if transno is not zero, then commit cb
586 * will be called at some point */
587 LASSERT(atomic_read(&d->opd_syn_rpc_in_progress) > 0);
588 atomic_dec(&d->opd_syn_rpc_in_progress);
591 wake_up(&d->opd_syn_waitq);
592 } else if (d->opd_pre != NULL &&
593 unlikely(d->opd_pre_status == -ENOSPC)) {
595 * if current status is -ENOSPC (lack of free space on OST)
596 * then we should poll OST immediately once object destroy
599 osp_statfs_need_now(d);
602 spin_lock(&d->opd_syn_lock);
603 list_del_init(&jra->jra_inflight_link);
604 spin_unlock(&d->opd_syn_lock);
605 LASSERT(atomic_read(&d->opd_syn_rpc_in_flight) > 0);
606 atomic_dec(&d->opd_syn_rpc_in_flight);
607 if (unlikely(atomic_read(&d->opd_syn_barrier) > 0))
608 wake_up(&d->opd_syn_barrier_waitq);
609 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
610 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
611 atomic_read(&d->opd_syn_rpc_in_progress));
613 osp_sync_check_for_work(d);
619 ** Add request to ptlrpc queue.
621 * This is just a tiny helper function to put the request on the sending list
623 * \param[in] d OSP device
624 * \param[in] llh llog handle where the record is stored
625 * \param[in] h llog record
626 * \param[in] req request
628 static void osp_sync_send_new_rpc(struct osp_device *d,
629 struct llog_handle *llh,
630 struct llog_rec_hdr *h,
631 struct ptlrpc_request *req)
633 struct osp_job_req_args *jra;
635 LASSERT(atomic_read(&d->opd_syn_rpc_in_flight) <=
636 d->opd_syn_max_rpc_in_flight);
638 jra = ptlrpc_req_async_args(req);
639 jra->jra_magic = OSP_JOB_MAGIC;
640 jra->jra_lcookie.lgc_lgl = llh->lgh_id;
641 jra->jra_lcookie.lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
642 jra->jra_lcookie.lgc_index = h->lrh_index;
643 INIT_LIST_HEAD(&jra->jra_committed_link);
644 spin_lock(&d->opd_syn_lock);
645 list_add_tail(&jra->jra_inflight_link, &d->opd_syn_inflight_list);
646 spin_unlock(&d->opd_syn_lock);
648 ptlrpcd_add_req(req);
653 * Allocate and prepare RPC for a new change.
655 * The function allocates and initializes an RPC which will be sent soon to
656 * apply the change to the target OST. The request is initialized from the
657 * llog record passed. Notice only the fields common to all type of changes
660 * \param[in] d OSP device
661 * \param[in] op type of the change
662 * \param[in] format request format to be used
664 * \retval pointer new request on success
665 * \retval ERR_PTR(errno) on error
667 static struct ptlrpc_request *osp_sync_new_job(struct osp_device *d,
669 const struct req_format *format)
671 struct ptlrpc_request *req;
672 struct obd_import *imp;
675 /* Prepare the request */
676 imp = d->opd_obd->u.cli.cl_import;
679 if (OBD_FAIL_CHECK(OBD_FAIL_OSP_CHECK_ENOMEM))
680 RETURN(ERR_PTR(-ENOMEM));
682 req = ptlrpc_request_alloc(imp, format);
684 RETURN(ERR_PTR(-ENOMEM));
686 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, op);
688 ptlrpc_req_finished(req);
692 req->rq_interpret_reply = osp_sync_interpret;
693 req->rq_commit_cb = osp_sync_request_commit_cb;
696 ptlrpc_request_set_replen(req);
702 * Generate a request for setattr change.
704 * The function prepares a new RPC, initializes it with setattr specific
705 * bits and send the RPC.
707 * \param[in] d OSP device
708 * \param[in] llh llog handle where the record is stored
709 * \param[in] h llog record
711 * \retval 0 on success
712 * \retval 1 on invalid record
713 * \retval negative negated errno on error
715 static int osp_sync_new_setattr_job(struct osp_device *d,
716 struct llog_handle *llh,
717 struct llog_rec_hdr *h)
719 struct llog_setattr64_rec *rec = (struct llog_setattr64_rec *)h;
720 struct ptlrpc_request *req;
721 struct ost_body *body;
724 LASSERT(h->lrh_type == MDS_SETATTR64_REC);
726 if (OBD_FAIL_CHECK(OBD_FAIL_OSP_CHECK_INVALID_REC))
729 /* lsr_valid can only be 0 or HAVE OBD_MD_{FLUID, FLGID, FLPROJID} set,
730 * so no bits other than these should be set. */
731 if ((rec->lsr_valid & ~(OBD_MD_FLUID | OBD_MD_FLGID |
732 OBD_MD_FLPROJID)) != 0) {
733 CERROR("%s: invalid setattr record, lsr_valid:%llu\n",
734 d->opd_obd->obd_name, rec->lsr_valid);
735 /* return 1 on invalid record */
739 req = osp_sync_new_job(d, OST_SETATTR, &RQF_OST_SETATTR);
741 RETURN(PTR_ERR(req));
743 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
745 body->oa.o_oi = rec->lsr_oi;
746 body->oa.o_uid = rec->lsr_uid;
747 body->oa.o_gid = rec->lsr_gid;
748 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID;
749 if (h->lrh_len > sizeof(struct llog_setattr64_rec))
750 body->oa.o_projid = ((struct llog_setattr64_rec_v2 *)
753 /* old setattr record (prior 2.6.0) doesn't have 'valid' stored,
754 * we assume that both UID and GID are valid in that case. */
755 if (rec->lsr_valid == 0)
756 body->oa.o_valid |= (OBD_MD_FLUID | OBD_MD_FLGID);
758 body->oa.o_valid |= rec->lsr_valid;
760 osp_sync_send_new_rpc(d, llh, h, req);
765 * Generate a request for unlink change.
767 * The function prepares a new RPC, initializes it with unlink(destroy)
768 * specific bits and sends the RPC. The function is used to handle
769 * llog_unlink_rec which were used in the older versions of Lustre.
770 * Current version uses llog_unlink_rec64.
772 * \param[in] d OSP device
773 * \param[in] llh llog handle where the record is stored
774 * \param[in] h llog record
776 * \retval 0 on success
777 * \retval negative negated errno on error
779 static int osp_sync_new_unlink_job(struct osp_device *d,
780 struct llog_handle *llh,
781 struct llog_rec_hdr *h)
783 struct llog_unlink_rec *rec = (struct llog_unlink_rec *)h;
784 struct ptlrpc_request *req;
785 struct ost_body *body;
789 LASSERT(h->lrh_type == MDS_UNLINK_REC);
791 req = osp_sync_new_job(d, OST_DESTROY, &RQF_OST_DESTROY);
793 RETURN(PTR_ERR(req));
795 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
797 ostid_set_seq(&body->oa.o_oi, rec->lur_oseq);
798 rc = ostid_set_id(&body->oa.o_oi, rec->lur_oid);
801 body->oa.o_misc = rec->lur_count;
802 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID;
804 body->oa.o_valid |= OBD_MD_FLOBJCOUNT;
806 osp_sync_send_new_rpc(d, llh, h, req);
811 * Generate a request for unlink change.
813 * The function prepares a new RPC, initializes it with unlink(destroy)
814 * specific bits and sends the RPC. Depending on the target (MDT or OST)
815 * two different protocols are used. For MDT we use OUT (basically OSD API
816 * updates transferred via a network). For OST we still use the old
817 * protocol (OBD?), originally for compatibility. Later we can start to
818 * use OUT for OST as well, this will allow batching and better code
821 * \param[in] d OSP device
822 * \param[in] llh llog handle where the record is stored
823 * \param[in] h llog record
825 * \retval 0 on success
826 * \retval negative negated errno on error
828 static int osp_sync_new_unlink64_job(struct osp_device *d,
829 struct llog_handle *llh,
830 struct llog_rec_hdr *h)
832 struct llog_unlink64_rec *rec = (struct llog_unlink64_rec *)h;
833 struct ptlrpc_request *req = NULL;
834 struct ost_body *body;
838 LASSERT(h->lrh_type == MDS_UNLINK64_REC);
839 req = osp_sync_new_job(d, OST_DESTROY, &RQF_OST_DESTROY);
841 RETURN(PTR_ERR(req));
843 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
846 rc = fid_to_ostid(&rec->lur_fid, &body->oa.o_oi);
849 body->oa.o_misc = rec->lur_count;
850 body->oa.o_valid = OBD_MD_FLGROUP | OBD_MD_FLID |
852 osp_sync_send_new_rpc(d, llh, h, req);
857 * Process llog records.
859 * This function is called to process the llog records committed locally.
860 * In the recovery model used by OSP we can apply a change to a remote
861 * target once corresponding transaction (like posix unlink) is committed
862 * locally so can't revert.
863 * Depending on the llog record type, a given handler is called that is
864 * responsible for preparing and sending the RPC to apply the change.
865 * Special record type LLOG_GEN_REC marking a reboot is cancelled right away.
867 * \param[in] env LU environment provided by the caller
868 * \param[in] d OSP device
869 * \param[in] llh llog handle where the record is stored
870 * \param[in] rec llog record
872 static void osp_sync_process_record(const struct lu_env *env,
873 struct osp_device *d,
874 struct llog_handle *llh,
875 struct llog_rec_hdr *rec)
877 struct llog_handle *cathandle = llh->u.phd.phd_cat_handle;
878 struct llog_cookie cookie;
883 cookie.lgc_lgl = llh->lgh_id;
884 cookie.lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
885 cookie.lgc_index = rec->lrh_index;
887 if (unlikely(rec->lrh_type == LLOG_GEN_REC)) {
888 struct llog_gen_rec *gen = (struct llog_gen_rec *)rec;
890 /* we're waiting for the record generated by this instance */
891 LASSERT(d->opd_syn_prev_done == 0);
892 if (!memcmp(&d->opd_syn_generation, &gen->lgr_gen,
893 sizeof(gen->lgr_gen))) {
894 CDEBUG(D_HA, "processed all old entries\n");
895 d->opd_syn_prev_done = 1;
898 /* cancel any generation record */
899 rc = llog_cat_cancel_records(env, cathandle, 1, &cookie);
905 * now we prepare and fill requests to OST, put them on the queue
906 * and fire after next commit callback
909 /* notice we increment counters before sending RPC, to be consistent
910 * in RPC interpret callback which may happen very quickly */
911 atomic_inc(&d->opd_syn_rpc_in_flight);
912 atomic_inc(&d->opd_syn_rpc_in_progress);
914 switch (rec->lrh_type) {
915 /* case MDS_UNLINK_REC is kept for compatibility */
917 rc = osp_sync_new_unlink_job(d, llh, rec);
919 case MDS_UNLINK64_REC:
920 rc = osp_sync_new_unlink64_job(d, llh, rec);
922 case MDS_SETATTR64_REC:
923 rc = osp_sync_new_setattr_job(d, llh, rec);
926 CERROR("%s: unknown record type: %x\n", d->opd_obd->obd_name,
928 /* treat "unknown record type" as "invalid" */
933 /* For all kinds of records, not matter successful or not,
934 * we should decrease changes and bump last_processed_id.
936 if (d->opd_syn_prev_done) {
937 __u64 correct_id = osp_sync_correct_id(d, rec);
938 LASSERT(atomic_read(&d->opd_syn_changes) > 0);
939 LASSERT(correct_id <= d->opd_syn_last_committed_id);
940 /* NOTE: it's possible to meet same id if
941 * OST stores few stripes of same file
944 /* another thread may be trying to set new value */
946 if (correct_id > d->opd_syn_last_processed_id) {
947 d->opd_syn_last_processed_id = correct_id;
948 wake_up(&d->opd_syn_barrier_waitq);
952 atomic_dec(&d->opd_syn_changes);
955 atomic_dec(&d->opd_syn_rpc_in_flight);
956 atomic_dec(&d->opd_syn_rpc_in_progress);
959 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
960 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
961 atomic_read(&d->opd_syn_rpc_in_progress));
963 /* Delete the invalid record */
965 rc = llog_cat_cancel_records(env, cathandle, 1, &cookie);
967 CERROR("%s: can't delete invalid record: "
968 "fid = "DFID", rec_id = %u, rc = %d\n",
969 d->opd_obd->obd_name,
970 PFID(lu_object_fid(&cathandle->lgh_obj->do_lu)),
974 CDEBUG(D_HA, "found record %x, %d, idx %u, id %u\n",
975 rec->lrh_type, rec->lrh_len, rec->lrh_index, rec->lrh_id);
981 * Cancel llog records for the committed changes.
983 * The function walks through the list of the committed RPCs and cancels
984 * corresponding llog records. see osp_sync_request_commit_cb() for the
987 * \param[in] env LU environment provided by the caller
988 * \param[in] d OSP device
990 static void osp_sync_process_committed(const struct lu_env *env,
991 struct osp_device *d)
993 struct obd_device *obd = d->opd_obd;
994 struct obd_import *imp = obd->u.cli.cl_import;
995 struct ost_body *body;
996 struct ptlrpc_request *req;
997 struct llog_ctxt *ctxt;
998 struct llog_handle *llh;
999 struct list_head list;
1004 if (list_empty(&d->opd_syn_committed_there))
1008 * if current status is -ENOSPC (lack of free space on OST)
1009 * then we should poll OST immediately once object destroy
1011 * notice: we do this upon commit as well because some backends
1012 * (like DMU) do not release space right away.
1014 if (d->opd_pre != NULL && unlikely(d->opd_pre_status == -ENOSPC))
1015 osp_statfs_need_now(d);
1018 * now cancel them all
1019 * XXX: can we improve this using some batching?
1020 * with batch RPC that'll happen automatically?
1021 * XXX: can we store ctxt in lod_device and save few cycles ?
1023 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1026 llh = ctxt->loc_handle;
1029 INIT_LIST_HEAD(&list);
1030 spin_lock(&d->opd_syn_lock);
1031 list_splice(&d->opd_syn_committed_there, &list);
1032 INIT_LIST_HEAD(&d->opd_syn_committed_there);
1033 spin_unlock(&d->opd_syn_lock);
1035 while (!list_empty(&list)) {
1036 struct osp_job_req_args *jra;
1038 jra = list_entry(list.next, struct osp_job_req_args,
1039 jra_committed_link);
1040 LASSERT(jra->jra_magic == OSP_JOB_MAGIC);
1041 list_del_init(&jra->jra_committed_link);
1043 req = container_of((void *)jra, struct ptlrpc_request,
1045 body = req_capsule_client_get(&req->rq_pill,
1048 /* import can be closing, thus all commit cb's are
1049 * called we can check committness directly */
1050 if (req->rq_import_generation == imp->imp_generation) {
1051 rc = llog_cat_cancel_records(env, llh, 1,
1054 CERROR("%s: can't cancel record: %d\n",
1057 DEBUG_REQ(D_OTHER, req, "imp_committed = %llu",
1058 imp->imp_peer_committed_transno);
1060 ptlrpc_req_finished(req);
1064 llog_ctxt_put(ctxt);
1066 LASSERT(atomic_read(&d->opd_syn_rpc_in_progress) >= done);
1067 atomic_sub(done, &d->opd_syn_rpc_in_progress);
1068 CDEBUG(D_OTHER, "%s: %d in flight, %d in progress\n",
1069 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_flight),
1070 atomic_read(&d->opd_syn_rpc_in_progress));
1072 osp_sync_check_for_work(d);
1074 /* wake up the thread if requested to stop:
1075 * it might be waiting for in-progress to complete */
1076 if (unlikely(osp_sync_running(d) == 0))
1077 wake_up(&d->opd_syn_waitq);
1083 * The core of the syncing mechanism.
1085 * This is a callback called by the llog processing function. Essentially it
1086 * suspends llog processing until there is a record to process (it's supposed
1087 * to be committed locally). The function handles RPCs committed by the target
1088 * and cancels corresponding llog records.
1090 * \param[in] env LU environment provided by the caller
1091 * \param[in] llh llog handle we're processing
1092 * \param[in] rec current llog record
1093 * \param[in] data callback data containing a pointer to the device
1095 * \retval 0 to ask the caller (llog_process()) to continue
1096 * \retval LLOG_PROC_BREAK to ask the caller to break
1098 static int osp_sync_process_queues(const struct lu_env *env,
1099 struct llog_handle *llh,
1100 struct llog_rec_hdr *rec,
1103 struct osp_device *d = data;
1106 struct l_wait_info lwi = { 0 };
1108 if (!osp_sync_running(d)) {
1109 CDEBUG(D_HA, "stop llog processing\n");
1110 return LLOG_PROC_BREAK;
1113 /* process requests committed by OST */
1114 osp_sync_process_committed(env, d);
1116 /* if we there are changes to be processed and we have
1117 * resources for this ... do now */
1118 if (osp_sync_can_process_new(d, rec)) {
1120 /* ask llog for another record */
1121 CDEBUG(D_HA, "%u changes, %u in progress,"
1123 atomic_read(&d->opd_syn_changes),
1124 atomic_read(&d->opd_syn_rpc_in_progress),
1125 atomic_read(&d->opd_syn_rpc_in_flight));
1128 osp_sync_process_record(env, d, llh, rec);
1133 if (d->opd_syn_last_processed_id == d->opd_syn_last_used_id)
1134 osp_sync_remove_from_tracker(d);
1136 l_wait_event(d->opd_syn_waitq,
1137 !osp_sync_running(d) ||
1138 osp_sync_can_process_new(d, rec) ||
1139 !list_empty(&d->opd_syn_committed_there),
1147 * This thread runs llog_cat_process() scanner calling our callback
1148 * to process llog records. in the callback we implement tricky
1149 * state machine as we don't want to start scanning of the llog again
1150 * and again, also we don't want to process too many records and send
1151 * too many RPCs a time. so, depending on current load (num of changes
1152 * being synced to OST) the callback can suspend awaiting for some
1153 * new conditions, like syncs completed.
1155 * In order to process llog records left by previous boots and to allow
1156 * llog_process_thread() to find something (otherwise it'd just exit
1157 * immediately) we add a special GENERATATION record on each boot.
1159 * \param[in] _arg a pointer to thread's arguments
1161 * \retval 0 on success
1162 * \retval negative negated errno on error
1164 static int osp_sync_thread(void *_arg)
1166 struct osp_device *d = _arg;
1167 struct ptlrpc_thread *thread = &d->opd_syn_thread;
1168 struct l_wait_info lwi = { 0 };
1169 struct llog_ctxt *ctxt;
1170 struct obd_device *obd = d->opd_obd;
1171 struct llog_handle *llh;
1177 rc = lu_env_init(&env, LCT_LOCAL);
1179 CERROR("%s: can't initialize env: rc = %d\n",
1182 spin_lock(&d->opd_syn_lock);
1183 thread->t_flags = SVC_STOPPED;
1184 spin_unlock(&d->opd_syn_lock);
1185 wake_up(&thread->t_ctl_waitq);
1190 spin_lock(&d->opd_syn_lock);
1191 thread->t_flags = SVC_RUNNING;
1192 spin_unlock(&d->opd_syn_lock);
1193 wake_up(&thread->t_ctl_waitq);
1195 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1197 CERROR("can't get appropriate context\n");
1198 GOTO(out, rc = -EINVAL);
1201 llh = ctxt->loc_handle;
1203 CERROR("can't get llh\n");
1204 llog_ctxt_put(ctxt);
1205 GOTO(out, rc = -EINVAL);
1208 rc = llog_cat_process(&env, llh, osp_sync_process_queues, d, 0, 0);
1210 CERROR("%s: llog process with osp_sync_process_queues "
1211 "failed: %d\n", d->opd_obd->obd_name, rc);
1214 LASSERTF(rc == 0 || rc == LLOG_PROC_BREAK,
1215 "%u changes, %u in progress, %u in flight: %d\n",
1216 atomic_read(&d->opd_syn_changes),
1217 atomic_read(&d->opd_syn_rpc_in_progress),
1218 atomic_read(&d->opd_syn_rpc_in_flight), rc);
1220 /* we don't expect llog_process_thread() to exit till umount */
1221 LASSERTF(thread->t_flags != SVC_RUNNING,
1222 "%u changes, %u in progress, %u in flight\n",
1223 atomic_read(&d->opd_syn_changes),
1224 atomic_read(&d->opd_syn_rpc_in_progress),
1225 atomic_read(&d->opd_syn_rpc_in_flight));
1227 /* wait till all the requests are completed */
1229 while (atomic_read(&d->opd_syn_rpc_in_progress) > 0) {
1230 osp_sync_process_committed(&env, d);
1232 lwi = LWI_TIMEOUT(cfs_time_seconds(5), NULL, NULL);
1233 rc = l_wait_event(d->opd_syn_waitq,
1234 atomic_read(&d->opd_syn_rpc_in_progress) == 0,
1236 if (rc == -ETIMEDOUT)
1238 LASSERTF(count < 10, "%s: %d %d %sempty\n",
1239 d->opd_obd->obd_name,
1240 atomic_read(&d->opd_syn_rpc_in_progress),
1241 atomic_read(&d->opd_syn_rpc_in_flight),
1242 list_empty(&d->opd_syn_committed_there) ? "" : "!");
1247 llog_cat_close(&env, llh);
1248 rc = llog_cleanup(&env, ctxt);
1250 CERROR("can't cleanup llog: %d\n", rc);
1252 LASSERTF(atomic_read(&d->opd_syn_rpc_in_progress) == 0,
1253 "%s: %d %d %sempty\n",
1254 d->opd_obd->obd_name, atomic_read(&d->opd_syn_rpc_in_progress),
1255 atomic_read(&d->opd_syn_rpc_in_flight),
1256 list_empty(&d->opd_syn_committed_there) ? "" : "!");
1258 thread->t_flags = SVC_STOPPED;
1260 wake_up(&thread->t_ctl_waitq);
1270 * Initializes the llog. Specific llog to be used depends on the type of the
1271 * target OSP represents (OST or MDT). The function adds appends a new llog
1272 * record to mark the place where the records associated with this boot
1275 * \param[in] env LU environment provided by the caller
1276 * \param[in] d OSP device
1278 * \retval 0 on success
1279 * \retval negative negated errno on error
1281 static int osp_sync_llog_init(const struct lu_env *env, struct osp_device *d)
1283 struct osp_thread_info *osi = osp_env_info(env);
1284 struct lu_fid *fid = &osi->osi_fid;
1285 struct llog_handle *lgh = NULL;
1286 struct obd_device *obd = d->opd_obd;
1287 struct llog_ctxt *ctxt;
1295 * open llog corresponding to our OST
1297 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1298 obd->obd_lvfs_ctxt.dt = d->opd_storage;
1300 lu_local_obj_fid(fid, LLOG_CATALOGS_OID);
1302 rc = llog_osd_get_cat_list(env, d->opd_storage, d->opd_index, 1,
1303 &osi->osi_cid, fid);
1305 if (rc != -EFAULT) {
1306 CERROR("%s: can't get id from catalogs: rc = %d\n",
1311 /* After sparse OST indices is supported, the CATALOG file
1312 * may become a sparse file that results in failure on
1313 * reading. Skip this error as the llog will be created
1315 memset(&osi->osi_cid, 0, sizeof(osi->osi_cid));
1319 CDEBUG(D_INFO, "%s: Init llog for %d - catid "DFID":%x\n",
1320 obd->obd_name, d->opd_index,
1321 PFID(&osi->osi_cid.lci_logid.lgl_oi.oi_fid),
1322 osi->osi_cid.lci_logid.lgl_ogen);
1324 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_MDS_OST_ORIG_CTXT,
1325 d->opd_storage->dd_lu_dev.ld_obd,
1326 &osp_mds_ost_orig_logops);
1330 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
1333 if (likely(logid_id(&osi->osi_cid.lci_logid) != 0)) {
1334 rc = llog_open(env, ctxt, &lgh, &osi->osi_cid.lci_logid, NULL,
1336 /* re-create llog if it is missing */
1338 logid_set_id(&osi->osi_cid.lci_logid, 0);
1340 GOTO(out_cleanup, rc);
1343 if (unlikely(logid_id(&osi->osi_cid.lci_logid) == 0)) {
1344 rc = llog_open_create(env, ctxt, &lgh, NULL, NULL);
1346 GOTO(out_cleanup, rc);
1347 osi->osi_cid.lci_logid = lgh->lgh_id;
1350 LASSERT(lgh != NULL);
1351 ctxt->loc_handle = lgh;
1353 rc = llog_init_handle(env, lgh, LLOG_F_IS_CAT, NULL);
1355 GOTO(out_close, rc);
1357 rc = llog_osd_put_cat_list(env, d->opd_storage, d->opd_index, 1,
1358 &osi->osi_cid, fid);
1360 GOTO(out_close, rc);
1363 * put a mark in the llog till which we'll be processing
1364 * old records restless
1366 d->opd_syn_generation.mnt_cnt = cfs_time_current();
1367 d->opd_syn_generation.conn_cnt = cfs_time_current();
1369 osi->osi_hdr.lrh_type = LLOG_GEN_REC;
1370 osi->osi_hdr.lrh_len = sizeof(osi->osi_gen);
1372 memcpy(&osi->osi_gen.lgr_gen, &d->opd_syn_generation,
1373 sizeof(osi->osi_gen.lgr_gen));
1375 rc = llog_cat_add(env, lgh, &osi->osi_gen.lgr_hdr, &osi->osi_cookie);
1377 GOTO(out_close, rc);
1378 llog_ctxt_put(ctxt);
1381 llog_cat_close(env, lgh);
1383 llog_cleanup(env, ctxt);
1388 * Cleanup llog used for syncing.
1390 * Closes and cleanups the llog. The function is called when the device is
1393 * \param[in] env LU environment provided by the caller
1394 * \param[in] d OSP device
1396 static void osp_sync_llog_fini(const struct lu_env *env, struct osp_device *d)
1398 struct llog_ctxt *ctxt;
1400 ctxt = llog_get_context(d->opd_obd, LLOG_MDS_OST_ORIG_CTXT);
1402 llog_cat_close(env, ctxt->loc_handle);
1403 llog_cleanup(env, ctxt);
1408 * Initialization of the sync component of OSP.
1410 * Initializes the llog and starts a new thread to handle the changes to
1411 * the remote target (OST or MDT).
1413 * \param[in] env LU environment provided by the caller
1414 * \param[in] d OSP device
1416 * \retval 0 on success
1417 * \retval negative negated errno on error
1419 int osp_sync_init(const struct lu_env *env, struct osp_device *d)
1421 struct l_wait_info lwi = { 0 };
1422 struct task_struct *task;
1427 d->opd_syn_max_rpc_in_flight = OSP_MAX_IN_FLIGHT;
1428 d->opd_syn_max_rpc_in_progress = OSP_MAX_IN_PROGRESS;
1429 spin_lock_init(&d->opd_syn_lock);
1430 init_waitqueue_head(&d->opd_syn_waitq);
1431 init_waitqueue_head(&d->opd_syn_barrier_waitq);
1432 thread_set_flags(&d->opd_syn_thread, SVC_INIT);
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 if (d->opd_storage->dd_rdonly)
1440 rc = osp_sync_id_traction_init(d);
1445 * initialize llog storing changes
1447 rc = osp_sync_llog_init(env, d);
1449 CERROR("%s: can't initialize llog: rc = %d\n",
1450 d->opd_obd->obd_name, rc);
1455 * Start synchronization thread
1457 task = kthread_run(osp_sync_thread, d, "osp-syn-%u-%u",
1458 d->opd_index, d->opd_group);
1461 CERROR("%s: cannot start sync thread: rc = %d\n",
1462 d->opd_obd->obd_name, rc);
1466 l_wait_event(d->opd_syn_thread.t_ctl_waitq,
1467 osp_sync_running(d) || osp_sync_stopped(d), &lwi);
1471 osp_sync_llog_fini(env, d);
1473 osp_sync_id_traction_fini(d);
1478 * Stop the syncing thread.
1480 * Asks the syncing thread to stop and wait until it's stopped.
1482 * \param[in] d OSP device
1486 int osp_sync_fini(struct osp_device *d)
1488 struct ptlrpc_thread *thread = &d->opd_syn_thread;
1492 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1493 thread->t_flags = SVC_STOPPING;
1494 wake_up(&d->opd_syn_waitq);
1495 wait_event(thread->t_ctl_waitq, thread_is_stopped(thread));
1499 * unregister transaction callbacks only when sync thread
1500 * has finished operations with llog
1502 osp_sync_id_traction_fini(d);
1507 static DEFINE_MUTEX(osp_id_tracker_sem);
1508 static struct list_head osp_id_tracker_list =
1509 LIST_HEAD_INIT(osp_id_tracker_list);
1512 * OSD commit callback.
1514 * The function is used as a local OSD commit callback to track the highest
1515 * committed llog record id. see osp_sync_id_traction_init() for the details.
1517 * \param[in] th local transaction handle committed
1518 * \param[in] cookie commit callback data (our private structure)
1520 static void osp_sync_tracker_commit_cb(struct thandle *th, void *cookie)
1522 struct osp_id_tracker *tr = cookie;
1523 struct osp_device *d;
1524 struct osp_txn_info *txn;
1528 txn = osp_txn_info(&th->th_ctx);
1529 if (txn == NULL || txn->oti_current_id < tr->otr_committed_id)
1532 spin_lock(&tr->otr_lock);
1533 if (likely(txn->oti_current_id > tr->otr_committed_id)) {
1534 CDEBUG(D_OTHER, "committed: %llu -> %llu\n",
1535 tr->otr_committed_id, txn->oti_current_id);
1536 tr->otr_committed_id = txn->oti_current_id;
1538 list_for_each_entry(d, &tr->otr_wakeup_list,
1540 d->opd_syn_last_committed_id = tr->otr_committed_id;
1541 wake_up(&d->opd_syn_waitq);
1544 spin_unlock(&tr->otr_lock);
1548 * Initialize commit tracking mechanism.
1550 * Some setups may have thousands of OSTs and each will be represented by OSP.
1551 * Meaning order of magnitute many more changes to apply every second. In order
1552 * to keep the number of commit callbacks low this mechanism was introduced.
1553 * The mechanism is very similar to transno used by MDT service: it's an single
1554 * ID stream which can be assigned by any OSP to its llog records. The tricky
1555 * part is that ID is stored in per-transaction data and re-used by all the OSPs
1556 * involved in that transaction. Then all these OSPs are woken up utilizing a single OSD commit callback.
1558 * The function initializes the data used by the tracker described above.
1559 * A singler tracker per OSD device is created.
1561 * \param[in] d OSP device
1563 * \retval 0 on success
1564 * \retval negative negated errno on error
1566 static int osp_sync_id_traction_init(struct osp_device *d)
1568 struct osp_id_tracker *tr, *found = NULL;
1572 LASSERT(d->opd_storage);
1573 LASSERT(d->opd_syn_tracker == NULL);
1574 INIT_LIST_HEAD(&d->opd_syn_ontrack);
1576 mutex_lock(&osp_id_tracker_sem);
1577 list_for_each_entry(tr, &osp_id_tracker_list, otr_list) {
1578 if (tr->otr_dev == d->opd_storage) {
1579 LASSERT(atomic_read(&tr->otr_refcount));
1580 atomic_inc(&tr->otr_refcount);
1581 d->opd_syn_tracker = tr;
1587 if (found == NULL) {
1591 d->opd_syn_tracker = tr;
1592 spin_lock_init(&tr->otr_lock);
1593 tr->otr_dev = d->opd_storage;
1594 tr->otr_next_id = 1;
1595 tr->otr_committed_id = 0;
1596 atomic_set(&tr->otr_refcount, 1);
1597 INIT_LIST_HEAD(&tr->otr_wakeup_list);
1598 list_add(&tr->otr_list, &osp_id_tracker_list);
1599 tr->otr_tx_cb.dtc_txn_commit =
1600 osp_sync_tracker_commit_cb;
1601 tr->otr_tx_cb.dtc_cookie = tr;
1602 tr->otr_tx_cb.dtc_tag = LCT_MD_THREAD;
1603 dt_txn_callback_add(d->opd_storage, &tr->otr_tx_cb);
1607 mutex_unlock(&osp_id_tracker_sem);
1613 * Release commit tracker.
1615 * Decrease a refcounter on the tracker used by the given OSP device \a d.
1616 * If no more users left, then the tracker is released.
1618 * \param[in] d OSP device
1620 static void osp_sync_id_traction_fini(struct osp_device *d)
1622 struct osp_id_tracker *tr;
1627 tr = d->opd_syn_tracker;
1633 osp_sync_remove_from_tracker(d);
1635 mutex_lock(&osp_id_tracker_sem);
1636 if (atomic_dec_and_test(&tr->otr_refcount)) {
1637 dt_txn_callback_del(d->opd_storage, &tr->otr_tx_cb);
1638 LASSERT(list_empty(&tr->otr_wakeup_list));
1639 list_del(&tr->otr_list);
1641 d->opd_syn_tracker = NULL;
1643 mutex_unlock(&osp_id_tracker_sem);
1649 * Generate a new ID on a tracker.
1651 * Generates a new ID using the tracker associated with the given OSP device
1652 * \a d, if the given ID \a id is non-zero. Unconditially adds OSP device to
1653 * the wakeup list, so OSP won't miss when a transaction using the ID is
1656 * \param[in] d OSP device
1657 * \param[in] id 0 or ID generated previously
1659 * \retval ID the caller should use
1661 static __u64 osp_sync_id_get(struct osp_device *d, __u64 id)
1663 struct osp_id_tracker *tr;
1665 tr = d->opd_syn_tracker;
1668 /* XXX: we can improve this introducing per-cpu preallocated ids? */
1669 spin_lock(&tr->otr_lock);
1670 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_TRACK_OVERFLOW))
1671 tr->otr_next_id = 0xfffffff0;
1673 if (unlikely(tr->otr_next_id <= d->opd_syn_last_used_id)) {
1674 spin_unlock(&tr->otr_lock);
1675 CERROR("%s: next %llu, last synced %llu\n",
1676 d->opd_obd->obd_name, tr->otr_next_id,
1677 d->opd_syn_last_used_id);
1682 id = tr->otr_next_id++;
1683 if (id > d->opd_syn_last_used_id)
1684 d->opd_syn_last_used_id = id;
1685 if (list_empty(&d->opd_syn_ontrack))
1686 list_add(&d->opd_syn_ontrack, &tr->otr_wakeup_list);
1687 spin_unlock(&tr->otr_lock);
1688 CDEBUG(D_OTHER, "new id %llu\n", id);
1694 * Stop to propagate commit status to OSP.
1696 * If the OSP does not have any llog records she's waiting to commit, then
1697 * it is possible to unsubscribe from wakeups from the tracking using this
1700 * \param[in] d OSP device not willing to wakeup
1702 static void osp_sync_remove_from_tracker(struct osp_device *d)
1704 struct osp_id_tracker *tr;
1706 tr = d->opd_syn_tracker;
1709 if (list_empty(&d->opd_syn_ontrack))
1712 spin_lock(&tr->otr_lock);
1713 list_del_init(&d->opd_syn_ontrack);
1714 spin_unlock(&tr->otr_lock);