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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_OSC
39 #include <libcfs/libcfs.h>
41 #include <lustre_dlm.h>
42 #include <lustre_net.h>
43 #include <lustre/lustre_user.h>
44 #include <obd_cksum.h>
45 #include <lustre_ha.h>
46 #include <lprocfs_status.h>
47 #include <lustre_ioctl.h>
48 #include <lustre_debug.h>
49 #include <lustre_param.h>
50 #include <lustre_fid.h>
51 #include <obd_class.h>
52 #include "osc_internal.h"
53 #include "osc_cl_internal.h"
55 struct osc_brw_async_args {
59 obd_count aa_page_count;
61 struct brw_page **aa_ppga;
62 struct client_obd *aa_cli;
63 struct list_head aa_oaps;
64 struct list_head aa_exts;
65 struct obd_capa *aa_ocapa;
66 struct cl_req *aa_clerq;
69 #define osc_grant_args osc_brw_async_args
71 struct osc_async_args {
72 struct obd_info *aa_oi;
75 struct osc_setattr_args {
77 obd_enqueue_update_f sa_upcall;
81 struct osc_fsync_args {
82 struct obd_info *fa_oi;
83 obd_enqueue_update_f fa_upcall;
87 struct osc_enqueue_args {
88 struct obd_export *oa_exp;
92 osc_enqueue_upcall_f oa_upcall;
94 struct ost_lvb *oa_lvb;
95 struct lustre_handle oa_lockh;
96 unsigned int oa_agl:1;
99 static void osc_release_ppga(struct brw_page **ppga, obd_count count);
100 static int brw_interpret(const struct lu_env *env, struct ptlrpc_request *req,
103 static inline void osc_pack_capa(struct ptlrpc_request *req,
104 struct ost_body *body, void *capa)
106 struct obd_capa *oc = (struct obd_capa *)capa;
107 struct lustre_capa *c;
112 c = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
115 body->oa.o_valid |= OBD_MD_FLOSSCAPA;
116 DEBUG_CAPA(D_SEC, c, "pack");
119 static inline void osc_pack_req_body(struct ptlrpc_request *req,
120 struct obd_info *oinfo)
122 struct ost_body *body;
124 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
127 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
129 osc_pack_capa(req, body, oinfo->oi_capa);
132 static inline void osc_set_capa_size(struct ptlrpc_request *req,
133 const struct req_msg_field *field,
137 req_capsule_set_size(&req->rq_pill, field, RCL_CLIENT, 0);
139 /* it is already calculated as sizeof struct obd_capa */
143 static int osc_getattr_interpret(const struct lu_env *env,
144 struct ptlrpc_request *req,
145 struct osc_async_args *aa, int rc)
147 struct ost_body *body;
153 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
155 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
156 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
157 aa->aa_oi->oi_oa, &body->oa);
159 /* This should really be sent by the OST */
160 aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE;
161 aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
163 CDEBUG(D_INFO, "can't unpack ost_body\n");
165 aa->aa_oi->oi_oa->o_valid = 0;
168 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
172 static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
173 struct ptlrpc_request_set *set)
175 struct ptlrpc_request *req;
176 struct osc_async_args *aa;
180 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
184 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
185 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
187 ptlrpc_request_free(req);
191 osc_pack_req_body(req, oinfo);
193 ptlrpc_request_set_replen(req);
194 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret;
196 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
197 aa = ptlrpc_req_async_args(req);
200 ptlrpc_set_add_req(set, req);
204 static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
205 struct obd_info *oinfo)
207 struct ptlrpc_request *req;
208 struct ost_body *body;
212 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
216 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
217 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
219 ptlrpc_request_free(req);
223 osc_pack_req_body(req, oinfo);
225 ptlrpc_request_set_replen(req);
227 rc = ptlrpc_queue_wait(req);
231 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
233 GOTO(out, rc = -EPROTO);
235 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
236 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
239 oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd);
240 oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
244 ptlrpc_req_finished(req);
248 static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
249 struct obd_info *oinfo, struct obd_trans_info *oti)
251 struct ptlrpc_request *req;
252 struct ost_body *body;
256 LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP);
258 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
262 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
263 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
265 ptlrpc_request_free(req);
269 osc_pack_req_body(req, oinfo);
271 ptlrpc_request_set_replen(req);
273 rc = ptlrpc_queue_wait(req);
277 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
279 GOTO(out, rc = -EPROTO);
281 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
286 ptlrpc_req_finished(req);
290 static int osc_setattr_interpret(const struct lu_env *env,
291 struct ptlrpc_request *req,
292 struct osc_setattr_args *sa, int rc)
294 struct ost_body *body;
300 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
302 GOTO(out, rc = -EPROTO);
304 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
307 rc = sa->sa_upcall(sa->sa_cookie, rc);
311 int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo,
312 struct obd_trans_info *oti,
313 obd_enqueue_update_f upcall, void *cookie,
314 struct ptlrpc_request_set *rqset)
316 struct ptlrpc_request *req;
317 struct osc_setattr_args *sa;
321 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
325 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
326 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
328 ptlrpc_request_free(req);
332 if (oti && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
333 oinfo->oi_oa->o_lcookie = *oti->oti_logcookies;
335 osc_pack_req_body(req, oinfo);
337 ptlrpc_request_set_replen(req);
339 /* do mds to ost setattr asynchronously */
341 /* Do not wait for response. */
342 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
344 req->rq_interpret_reply =
345 (ptlrpc_interpterer_t)osc_setattr_interpret;
347 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
348 sa = ptlrpc_req_async_args(req);
349 sa->sa_oa = oinfo->oi_oa;
350 sa->sa_upcall = upcall;
351 sa->sa_cookie = cookie;
353 if (rqset == PTLRPCD_SET)
354 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
356 ptlrpc_set_add_req(rqset, req);
362 static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
363 struct obd_trans_info *oti,
364 struct ptlrpc_request_set *rqset)
366 return osc_setattr_async_base(exp, oinfo, oti,
367 oinfo->oi_cb_up, oinfo, rqset);
370 static int osc_create(const struct lu_env *env, struct obd_export *exp,
371 struct obdo *oa, struct obd_trans_info *oti)
373 struct ptlrpc_request *req;
374 struct ost_body *body;
379 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
380 LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
382 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
384 GOTO(out, rc = -ENOMEM);
386 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
388 ptlrpc_request_free(req);
392 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
395 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
397 ptlrpc_request_set_replen(req);
399 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
400 oa->o_flags == OBD_FL_DELORPHAN) {
402 "delorphan from OST integration");
403 /* Don't resend the delorphan req */
404 req->rq_no_resend = req->rq_no_delay = 1;
407 rc = ptlrpc_queue_wait(req);
411 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
413 GOTO(out_req, rc = -EPROTO);
415 CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
416 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
418 oa->o_blksize = cli_brw_size(exp->exp_obd);
419 oa->o_valid |= OBD_MD_FLBLKSZ;
422 if (oa->o_valid & OBD_MD_FLCOOKIE) {
423 if (oti->oti_logcookies == NULL)
424 oti->oti_logcookies = &oti->oti_onecookie;
426 *oti->oti_logcookies = oa->o_lcookie;
430 CDEBUG(D_HA, "transno: "LPD64"\n",
431 lustre_msg_get_transno(req->rq_repmsg));
433 ptlrpc_req_finished(req);
438 int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo,
439 obd_enqueue_update_f upcall, void *cookie,
440 struct ptlrpc_request_set *rqset)
442 struct ptlrpc_request *req;
443 struct osc_setattr_args *sa;
444 struct ost_body *body;
448 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH);
452 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
453 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
455 ptlrpc_request_free(req);
458 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
459 ptlrpc_at_set_req_timeout(req);
461 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
463 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
465 osc_pack_capa(req, body, oinfo->oi_capa);
467 ptlrpc_request_set_replen(req);
469 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
470 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
471 sa = ptlrpc_req_async_args(req);
472 sa->sa_oa = oinfo->oi_oa;
473 sa->sa_upcall = upcall;
474 sa->sa_cookie = cookie;
475 if (rqset == PTLRPCD_SET)
476 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
478 ptlrpc_set_add_req(rqset, req);
483 static int osc_sync_interpret(const struct lu_env *env,
484 struct ptlrpc_request *req,
487 struct osc_fsync_args *fa = arg;
488 struct ost_body *body;
494 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
496 CERROR ("can't unpack ost_body\n");
497 GOTO(out, rc = -EPROTO);
500 *fa->fa_oi->oi_oa = body->oa;
502 rc = fa->fa_upcall(fa->fa_cookie, rc);
506 int osc_sync_base(struct obd_export *exp, struct obd_info *oinfo,
507 obd_enqueue_update_f upcall, void *cookie,
508 struct ptlrpc_request_set *rqset)
510 struct ptlrpc_request *req;
511 struct ost_body *body;
512 struct osc_fsync_args *fa;
516 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
520 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
521 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
523 ptlrpc_request_free(req);
527 /* overload the size and blocks fields in the oa with start/end */
528 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
530 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
532 osc_pack_capa(req, body, oinfo->oi_capa);
534 ptlrpc_request_set_replen(req);
535 req->rq_interpret_reply = osc_sync_interpret;
537 CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
538 fa = ptlrpc_req_async_args(req);
540 fa->fa_upcall = upcall;
541 fa->fa_cookie = cookie;
543 if (rqset == PTLRPCD_SET)
544 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
546 ptlrpc_set_add_req(rqset, req);
551 /* Find and cancel locally locks matched by @mode in the resource found by
552 * @objid. Found locks are added into @cancel list. Returns the amount of
553 * locks added to @cancels list. */
554 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
555 struct list_head *cancels,
556 ldlm_mode_t mode, __u64 lock_flags)
558 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
559 struct ldlm_res_id res_id;
560 struct ldlm_resource *res;
564 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
565 * export) but disabled through procfs (flag in NS).
567 * This distinguishes from a case when ELC is not supported originally,
568 * when we still want to cancel locks in advance and just cancel them
569 * locally, without sending any RPC. */
570 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
573 ostid_build_res_name(&oa->o_oi, &res_id);
574 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
578 LDLM_RESOURCE_ADDREF(res);
579 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
580 lock_flags, 0, NULL);
581 LDLM_RESOURCE_DELREF(res);
582 ldlm_resource_putref(res);
586 static int osc_destroy_interpret(const struct lu_env *env,
587 struct ptlrpc_request *req, void *data,
590 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
592 atomic_dec(&cli->cl_destroy_in_flight);
593 wake_up(&cli->cl_destroy_waitq);
597 static int osc_can_send_destroy(struct client_obd *cli)
599 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
600 cli->cl_max_rpcs_in_flight) {
601 /* The destroy request can be sent */
604 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
605 cli->cl_max_rpcs_in_flight) {
607 * The counter has been modified between the two atomic
610 wake_up(&cli->cl_destroy_waitq);
615 /* Destroy requests can be async always on the client, and we don't even really
616 * care about the return code since the client cannot do anything at all about
618 * When the MDS is unlinking a filename, it saves the file objects into a
619 * recovery llog, and these object records are cancelled when the OST reports
620 * they were destroyed and sync'd to disk (i.e. transaction committed).
621 * If the client dies, or the OST is down when the object should be destroyed,
622 * the records are not cancelled, and when the OST reconnects to the MDS next,
623 * it will retrieve the llog unlink logs and then sends the log cancellation
624 * cookies to the MDS after committing destroy transactions. */
625 static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
626 struct obdo *oa, struct obd_trans_info *oti)
628 struct client_obd *cli = &exp->exp_obd->u.cli;
629 struct ptlrpc_request *req;
630 struct ost_body *body;
631 struct list_head cancels = LIST_HEAD_INIT(cancels);
636 CDEBUG(D_INFO, "oa NULL\n");
640 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
641 LDLM_FL_DISCARD_DATA);
643 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
645 ldlm_lock_list_put(&cancels, l_bl_ast, count);
649 osc_set_capa_size(req, &RMF_CAPA1, NULL);
650 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
653 ptlrpc_request_free(req);
657 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
658 ptlrpc_at_set_req_timeout(req);
660 if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
661 oa->o_lcookie = *oti->oti_logcookies;
662 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
664 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
666 ptlrpc_request_set_replen(req);
668 /* If osc_destory is for destroying the unlink orphan,
669 * sent from MDT to OST, which should not be blocked here,
670 * because the process might be triggered by ptlrpcd, and
671 * it is not good to block ptlrpcd thread (b=16006)*/
672 if (!(oa->o_flags & OBD_FL_DELORPHAN)) {
673 req->rq_interpret_reply = osc_destroy_interpret;
674 if (!osc_can_send_destroy(cli)) {
675 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP,
679 * Wait until the number of on-going destroy RPCs drops
680 * under max_rpc_in_flight
682 l_wait_event_exclusive(cli->cl_destroy_waitq,
683 osc_can_send_destroy(cli), &lwi);
687 /* Do not wait for response */
688 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
692 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
695 obd_flag bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT;
697 LASSERT(!(oa->o_valid & bits));
700 spin_lock(&cli->cl_loi_list_lock);
701 oa->o_dirty = cli->cl_dirty_pages << PAGE_CACHE_SHIFT;
702 if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
703 cli->cl_dirty_max_pages)) {
704 CERROR("dirty %lu - %lu > dirty_max %lu\n",
705 cli->cl_dirty_pages, cli->cl_dirty_transit,
706 cli->cl_dirty_max_pages);
708 } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
709 atomic_long_read(&obd_dirty_transit_pages) >
710 (obd_max_dirty_pages + 1))) {
711 /* The atomic_read() allowing the atomic_inc() are
712 * not covered by a lock thus they may safely race and trip
713 * this CERROR() unless we add in a small fudge factor (+1). */
714 CERROR("%s: dirty %ld - %ld > system dirty_max %lu\n",
715 cli->cl_import->imp_obd->obd_name,
716 atomic_long_read(&obd_dirty_pages),
717 atomic_long_read(&obd_dirty_transit_pages),
718 obd_max_dirty_pages);
720 } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
722 CERROR("dirty %lu - dirty_max %lu too big???\n",
723 cli->cl_dirty_pages, cli->cl_dirty_max_pages);
726 unsigned long max_in_flight = (cli->cl_max_pages_per_rpc <<
728 (cli->cl_max_rpcs_in_flight + 1);
729 oa->o_undirty = max(cli->cl_dirty_max_pages << PAGE_CACHE_SHIFT,
732 oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
733 oa->o_dropped = cli->cl_lost_grant;
734 cli->cl_lost_grant = 0;
735 spin_unlock(&cli->cl_loi_list_lock);
736 CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
737 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
741 void osc_update_next_shrink(struct client_obd *cli)
743 cli->cl_next_shrink_grant =
744 cfs_time_shift(cli->cl_grant_shrink_interval);
745 CDEBUG(D_CACHE, "next time %ld to shrink grant \n",
746 cli->cl_next_shrink_grant);
749 static void __osc_update_grant(struct client_obd *cli, obd_size grant)
751 spin_lock(&cli->cl_loi_list_lock);
752 cli->cl_avail_grant += grant;
753 spin_unlock(&cli->cl_loi_list_lock);
756 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
758 if (body->oa.o_valid & OBD_MD_FLGRANT) {
759 CDEBUG(D_CACHE, "got "LPU64" extra grant\n", body->oa.o_grant);
760 __osc_update_grant(cli, body->oa.o_grant);
764 static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
765 obd_count keylen, void *key, obd_count vallen,
766 void *val, struct ptlrpc_request_set *set);
768 static int osc_shrink_grant_interpret(const struct lu_env *env,
769 struct ptlrpc_request *req,
772 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
773 struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
774 struct ost_body *body;
777 __osc_update_grant(cli, oa->o_grant);
781 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
783 osc_update_grant(cli, body);
789 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
791 spin_lock(&cli->cl_loi_list_lock);
792 oa->o_grant = cli->cl_avail_grant / 4;
793 cli->cl_avail_grant -= oa->o_grant;
794 spin_unlock(&cli->cl_loi_list_lock);
795 if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
796 oa->o_valid |= OBD_MD_FLFLAGS;
799 oa->o_flags |= OBD_FL_SHRINK_GRANT;
800 osc_update_next_shrink(cli);
803 /* Shrink the current grant, either from some large amount to enough for a
804 * full set of in-flight RPCs, or if we have already shrunk to that limit
805 * then to enough for a single RPC. This avoids keeping more grant than
806 * needed, and avoids shrinking the grant piecemeal. */
807 static int osc_shrink_grant(struct client_obd *cli)
809 __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
810 (cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT);
812 spin_lock(&cli->cl_loi_list_lock);
813 if (cli->cl_avail_grant <= target_bytes)
814 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
815 spin_unlock(&cli->cl_loi_list_lock);
817 return osc_shrink_grant_to_target(cli, target_bytes);
820 int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
823 struct ost_body *body;
826 spin_lock(&cli->cl_loi_list_lock);
827 /* Don't shrink if we are already above or below the desired limit
828 * We don't want to shrink below a single RPC, as that will negatively
829 * impact block allocation and long-term performance. */
830 if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT)
831 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
833 if (target_bytes >= cli->cl_avail_grant) {
834 spin_unlock(&cli->cl_loi_list_lock);
837 spin_unlock(&cli->cl_loi_list_lock);
843 osc_announce_cached(cli, &body->oa, 0);
845 spin_lock(&cli->cl_loi_list_lock);
846 body->oa.o_grant = cli->cl_avail_grant - target_bytes;
847 cli->cl_avail_grant = target_bytes;
848 spin_unlock(&cli->cl_loi_list_lock);
849 if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
850 body->oa.o_valid |= OBD_MD_FLFLAGS;
851 body->oa.o_flags = 0;
853 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
854 osc_update_next_shrink(cli);
856 rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
857 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
858 sizeof(*body), body, NULL);
860 __osc_update_grant(cli, body->oa.o_grant);
865 static int osc_should_shrink_grant(struct client_obd *client)
867 cfs_time_t time = cfs_time_current();
868 cfs_time_t next_shrink = client->cl_next_shrink_grant;
870 if ((client->cl_import->imp_connect_data.ocd_connect_flags &
871 OBD_CONNECT_GRANT_SHRINK) == 0)
874 if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
875 /* Get the current RPC size directly, instead of going via:
876 * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
877 * Keep comment here so that it can be found by searching. */
878 int brw_size = client->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
880 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
881 client->cl_avail_grant > brw_size)
884 osc_update_next_shrink(client);
889 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
891 struct client_obd *client;
893 list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
894 if (osc_should_shrink_grant(client))
895 osc_shrink_grant(client);
900 static int osc_add_shrink_grant(struct client_obd *client)
904 rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
906 osc_grant_shrink_grant_cb, NULL,
907 &client->cl_grant_shrink_list);
909 CERROR("add grant client %s error %d\n",
910 client->cl_import->imp_obd->obd_name, rc);
913 CDEBUG(D_CACHE, "add grant client %s \n",
914 client->cl_import->imp_obd->obd_name);
915 osc_update_next_shrink(client);
919 static int osc_del_shrink_grant(struct client_obd *client)
921 return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
925 static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
928 * ocd_grant is the total grant amount we're expect to hold: if we've
929 * been evicted, it's the new avail_grant amount, cl_dirty_pages will
930 * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
933 * race is tolerable here: if we're evicted, but imp_state already
934 * left EVICTED state, then cl_dirty_pages must be 0 already.
936 spin_lock(&cli->cl_loi_list_lock);
937 if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
938 cli->cl_avail_grant = ocd->ocd_grant;
940 cli->cl_avail_grant = ocd->ocd_grant -
941 (cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
943 if (cli->cl_avail_grant < 0) {
944 CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n",
945 cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant,
946 ocd->ocd_grant, cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
947 /* workaround for servers which do not have the patch from
949 cli->cl_avail_grant = ocd->ocd_grant;
952 /* determine the appropriate chunk size used by osc_extent. */
953 cli->cl_chunkbits = max_t(int, PAGE_CACHE_SHIFT, ocd->ocd_blocksize);
954 spin_unlock(&cli->cl_loi_list_lock);
956 CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
957 "chunk bits: %d.\n", cli->cl_import->imp_obd->obd_name,
958 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits);
960 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
961 list_empty(&cli->cl_grant_shrink_list))
962 osc_add_shrink_grant(cli);
965 /* We assume that the reason this OSC got a short read is because it read
966 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
967 * via the LOV, and it _knows_ it's reading inside the file, it's just that
968 * this stripe never got written at or beyond this stripe offset yet. */
969 static void handle_short_read(int nob_read, obd_count page_count,
970 struct brw_page **pga)
975 /* skip bytes read OK */
976 while (nob_read > 0) {
977 LASSERT (page_count > 0);
979 if (pga[i]->count > nob_read) {
980 /* EOF inside this page */
981 ptr = kmap(pga[i]->pg) +
982 (pga[i]->off & ~CFS_PAGE_MASK);
983 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
990 nob_read -= pga[i]->count;
995 /* zero remaining pages */
996 while (page_count-- > 0) {
997 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
998 memset(ptr, 0, pga[i]->count);
1004 static int check_write_rcs(struct ptlrpc_request *req,
1005 int requested_nob, int niocount,
1006 obd_count page_count, struct brw_page **pga)
1011 remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1012 sizeof(*remote_rcs) *
1014 if (remote_rcs == NULL) {
1015 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1019 /* return error if any niobuf was in error */
1020 for (i = 0; i < niocount; i++) {
1021 if ((int)remote_rcs[i] < 0)
1022 return(remote_rcs[i]);
1024 if (remote_rcs[i] != 0) {
1025 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1026 i, remote_rcs[i], req);
1031 if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1032 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1033 req->rq_bulk->bd_nob_transferred, requested_nob);
1040 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1042 if (p1->flag != p2->flag) {
1043 unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE |
1044 OBD_BRW_SYNC | OBD_BRW_ASYNC |
1045 OBD_BRW_NOQUOTA | OBD_BRW_SOFT_SYNC);
1047 /* warn if we try to combine flags that we don't know to be
1048 * safe to combine */
1049 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1050 CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1051 "report this at https://jira.hpdd.intel.com/\n",
1052 p1->flag, p2->flag);
1057 return (p1->off + p1->count == p2->off);
1060 static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1061 struct brw_page **pga, int opc,
1062 cksum_type_t cksum_type)
1066 struct cfs_crypto_hash_desc *hdesc;
1067 unsigned int bufsize;
1069 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1071 LASSERT(pg_count > 0);
1073 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1074 if (IS_ERR(hdesc)) {
1075 CERROR("Unable to initialize checksum hash %s\n",
1076 cfs_crypto_hash_name(cfs_alg));
1077 return PTR_ERR(hdesc);
1080 while (nob > 0 && pg_count > 0) {
1081 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1083 /* corrupt the data before we compute the checksum, to
1084 * simulate an OST->client data error */
1085 if (i == 0 && opc == OST_READ &&
1086 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1087 unsigned char *ptr = kmap(pga[i]->pg);
1088 int off = pga[i]->off & ~CFS_PAGE_MASK;
1090 memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1093 cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
1094 pga[i]->off & ~CFS_PAGE_MASK,
1096 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
1097 (int)(pga[i]->off & ~CFS_PAGE_MASK));
1099 nob -= pga[i]->count;
1104 bufsize = sizeof(cksum);
1105 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1107 /* For sending we only compute the wrong checksum instead
1108 * of corrupting the data so it is still correct on a redo */
1109 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1115 static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1116 struct lov_stripe_md *lsm, obd_count page_count,
1117 struct brw_page **pga,
1118 struct ptlrpc_request **reqp,
1119 struct obd_capa *ocapa, int reserve,
1122 struct ptlrpc_request *req;
1123 struct ptlrpc_bulk_desc *desc;
1124 struct ost_body *body;
1125 struct obd_ioobj *ioobj;
1126 struct niobuf_remote *niobuf;
1127 int niocount, i, requested_nob, opc, rc;
1128 struct osc_brw_async_args *aa;
1129 struct req_capsule *pill;
1130 struct brw_page *pg_prev;
1133 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1134 RETURN(-ENOMEM); /* Recoverable */
1135 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1136 RETURN(-EINVAL); /* Fatal */
1138 if ((cmd & OBD_BRW_WRITE) != 0) {
1140 req = ptlrpc_request_alloc_pool(cli->cl_import,
1141 cli->cl_import->imp_rq_pool,
1142 &RQF_OST_BRW_WRITE);
1145 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1150 for (niocount = i = 1; i < page_count; i++) {
1151 if (!can_merge_pages(pga[i - 1], pga[i]))
1155 pill = &req->rq_pill;
1156 req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1158 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1159 niocount * sizeof(*niobuf));
1160 osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1162 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1164 ptlrpc_request_free(req);
1167 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1168 ptlrpc_at_set_req_timeout(req);
1169 /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1171 req->rq_no_retry_einprogress = 1;
1173 desc = ptlrpc_prep_bulk_imp(req, page_count,
1174 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1175 opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK,
1179 GOTO(out, rc = -ENOMEM);
1180 /* NB request now owns desc and will free it when it gets freed */
1182 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1183 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1184 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1185 LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1187 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
1189 obdo_to_ioobj(oa, ioobj);
1190 ioobj->ioo_bufcnt = niocount;
1191 /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1192 * that might be send for this request. The actual number is decided
1193 * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1194 * "max - 1" for old client compatibility sending "0", and also so the
1195 * the actual maximum is a power-of-two number, not one less. LU-1431 */
1196 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1197 osc_pack_capa(req, body, ocapa);
1198 LASSERT(page_count > 0);
1200 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1201 struct brw_page *pg = pga[i];
1202 int poff = pg->off & ~CFS_PAGE_MASK;
1204 LASSERT(pg->count > 0);
1205 /* make sure there is no gap in the middle of page array */
1206 LASSERTF(page_count == 1 ||
1207 (ergo(i == 0, poff + pg->count == PAGE_CACHE_SIZE) &&
1208 ergo(i > 0 && i < page_count - 1,
1209 poff == 0 && pg->count == PAGE_CACHE_SIZE) &&
1210 ergo(i == page_count - 1, poff == 0)),
1211 "i: %d/%d pg: %p off: "LPU64", count: %u\n",
1212 i, page_count, pg, pg->off, pg->count);
1213 LASSERTF(i == 0 || pg->off > pg_prev->off,
1214 "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1215 " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1217 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1218 pg_prev->pg, page_private(pg_prev->pg),
1219 pg_prev->pg->index, pg_prev->off);
1220 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1221 (pg->flag & OBD_BRW_SRVLOCK));
1223 ptlrpc_prep_bulk_page_pin(desc, pg->pg, poff, pg->count);
1224 requested_nob += pg->count;
1226 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1228 niobuf->rnb_len += pg->count;
1230 niobuf->rnb_offset = pg->off;
1231 niobuf->rnb_len = pg->count;
1232 niobuf->rnb_flags = pg->flag;
1237 LASSERTF((void *)(niobuf - niocount) ==
1238 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1239 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1240 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1242 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1244 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1245 body->oa.o_valid |= OBD_MD_FLFLAGS;
1246 body->oa.o_flags = 0;
1248 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1251 if (osc_should_shrink_grant(cli))
1252 osc_shrink_grant_local(cli, &body->oa);
1254 /* size[REQ_REC_OFF] still sizeof (*body) */
1255 if (opc == OST_WRITE) {
1256 if (cli->cl_checksum &&
1257 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1258 /* store cl_cksum_type in a local variable since
1259 * it can be changed via lprocfs */
1260 cksum_type_t cksum_type = cli->cl_cksum_type;
1262 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1263 oa->o_flags &= OBD_FL_LOCAL_MASK;
1264 body->oa.o_flags = 0;
1266 body->oa.o_flags |= cksum_type_pack(cksum_type);
1267 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1268 body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1272 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1274 /* save this in 'oa', too, for later checking */
1275 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1276 oa->o_flags |= cksum_type_pack(cksum_type);
1278 /* clear out the checksum flag, in case this is a
1279 * resend but cl_checksum is no longer set. b=11238 */
1280 oa->o_valid &= ~OBD_MD_FLCKSUM;
1282 oa->o_cksum = body->oa.o_cksum;
1283 /* 1 RC per niobuf */
1284 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1285 sizeof(__u32) * niocount);
1287 if (cli->cl_checksum &&
1288 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1289 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1290 body->oa.o_flags = 0;
1291 body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1292 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1295 ptlrpc_request_set_replen(req);
1297 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1298 aa = ptlrpc_req_async_args(req);
1300 aa->aa_requested_nob = requested_nob;
1301 aa->aa_nio_count = niocount;
1302 aa->aa_page_count = page_count;
1306 INIT_LIST_HEAD(&aa->aa_oaps);
1307 if (ocapa && reserve)
1308 aa->aa_ocapa = capa_get(ocapa);
1311 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1312 CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
1313 req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
1314 niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
1318 ptlrpc_req_finished(req);
1322 static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1323 __u32 client_cksum, __u32 server_cksum, int nob,
1324 obd_count page_count, struct brw_page **pga,
1325 cksum_type_t client_cksum_type)
1329 cksum_type_t cksum_type;
1331 if (server_cksum == client_cksum) {
1332 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1336 cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1338 new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1341 if (cksum_type != client_cksum_type)
1342 msg = "the server did not use the checksum type specified in "
1343 "the original request - likely a protocol problem";
1344 else if (new_cksum == server_cksum)
1345 msg = "changed on the client after we checksummed it - "
1346 "likely false positive due to mmap IO (bug 11742)";
1347 else if (new_cksum == client_cksum)
1348 msg = "changed in transit before arrival at OST";
1350 msg = "changed in transit AND doesn't match the original - "
1351 "likely false positive due to mmap IO (bug 11742)";
1353 LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
1354 " object "DOSTID" extent ["LPU64"-"LPU64"]\n",
1355 msg, libcfs_nid2str(peer->nid),
1356 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1357 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1358 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1359 POSTID(&oa->o_oi), pga[0]->off,
1360 pga[page_count-1]->off + pga[page_count-1]->count - 1);
1361 CERROR("original client csum %x (type %x), server csum %x (type %x), "
1362 "client csum now %x\n", client_cksum, client_cksum_type,
1363 server_cksum, cksum_type, new_cksum);
1367 /* Note rc enters this function as number of bytes transferred */
1368 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1370 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1371 const lnet_process_id_t *peer =
1372 &req->rq_import->imp_connection->c_peer;
1373 struct client_obd *cli = aa->aa_cli;
1374 struct ost_body *body;
1375 u32 client_cksum = 0;
1378 if (rc < 0 && rc != -EDQUOT) {
1379 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1383 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1384 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1386 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1390 /* set/clear over quota flag for a uid/gid */
1391 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1392 body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1393 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1395 CDEBUG(D_QUOTA, "setdq for [%u %u] with valid "LPX64", flags %x\n",
1396 body->oa.o_uid, body->oa.o_gid, body->oa.o_valid,
1398 osc_quota_setdq(cli, qid, body->oa.o_valid, body->oa.o_flags);
1401 osc_update_grant(cli, body);
1406 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1407 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1409 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1411 CERROR("Unexpected +ve rc %d\n", rc);
1414 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1416 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1419 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1420 check_write_checksum(&body->oa, peer, client_cksum,
1421 body->oa.o_cksum, aa->aa_requested_nob,
1422 aa->aa_page_count, aa->aa_ppga,
1423 cksum_type_unpack(aa->aa_oa->o_flags)))
1426 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1427 aa->aa_page_count, aa->aa_ppga);
1431 /* The rest of this function executes only for OST_READs */
1433 /* if unwrap_bulk failed, return -EAGAIN to retry */
1434 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1436 GOTO(out, rc = -EAGAIN);
1438 if (rc > aa->aa_requested_nob) {
1439 CERROR("Unexpected rc %d (%d requested)\n", rc,
1440 aa->aa_requested_nob);
1444 if (rc != req->rq_bulk->bd_nob_transferred) {
1445 CERROR ("Unexpected rc %d (%d transferred)\n",
1446 rc, req->rq_bulk->bd_nob_transferred);
1450 if (rc < aa->aa_requested_nob)
1451 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1453 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1454 static int cksum_counter;
1455 u32 server_cksum = body->oa.o_cksum;
1458 cksum_type_t cksum_type;
1460 cksum_type = cksum_type_unpack(body->oa.o_valid &OBD_MD_FLFLAGS?
1461 body->oa.o_flags : 0);
1462 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1463 aa->aa_ppga, OST_READ,
1466 if (peer->nid != req->rq_bulk->bd_sender) {
1468 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1471 if (server_cksum != client_cksum) {
1472 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1473 "%s%s%s inode "DFID" object "DOSTID
1474 " extent ["LPU64"-"LPU64"]\n",
1475 req->rq_import->imp_obd->obd_name,
1476 libcfs_nid2str(peer->nid),
1478 body->oa.o_valid & OBD_MD_FLFID ?
1479 body->oa.o_parent_seq : (__u64)0,
1480 body->oa.o_valid & OBD_MD_FLFID ?
1481 body->oa.o_parent_oid : 0,
1482 body->oa.o_valid & OBD_MD_FLFID ?
1483 body->oa.o_parent_ver : 0,
1484 POSTID(&body->oa.o_oi),
1485 aa->aa_ppga[0]->off,
1486 aa->aa_ppga[aa->aa_page_count-1]->off +
1487 aa->aa_ppga[aa->aa_page_count-1]->count -
1489 CERROR("client %x, server %x, cksum_type %x\n",
1490 client_cksum, server_cksum, cksum_type);
1492 aa->aa_oa->o_cksum = client_cksum;
1496 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1499 } else if (unlikely(client_cksum)) {
1500 static int cksum_missed;
1503 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1504 CERROR("Checksum %u requested from %s but not sent\n",
1505 cksum_missed, libcfs_nid2str(peer->nid));
1511 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1512 aa->aa_oa, &body->oa);
1517 static int osc_brw_redo_request(struct ptlrpc_request *request,
1518 struct osc_brw_async_args *aa, int rc)
1520 struct ptlrpc_request *new_req;
1521 struct osc_brw_async_args *new_aa;
1522 struct osc_async_page *oap;
1525 DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1526 "redo for recoverable error %d", rc);
1528 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1529 OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1530 aa->aa_cli, aa->aa_oa,
1531 NULL /* lsm unused by osc currently */,
1532 aa->aa_page_count, aa->aa_ppga,
1533 &new_req, aa->aa_ocapa, 0, 1);
1537 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1538 if (oap->oap_request != NULL) {
1539 LASSERTF(request == oap->oap_request,
1540 "request %p != oap_request %p\n",
1541 request, oap->oap_request);
1542 if (oap->oap_interrupted) {
1543 ptlrpc_req_finished(new_req);
1548 /* New request takes over pga and oaps from old request.
1549 * Note that copying a list_head doesn't work, need to move it... */
1551 new_req->rq_interpret_reply = request->rq_interpret_reply;
1552 new_req->rq_async_args = request->rq_async_args;
1553 new_req->rq_commit_cb = request->rq_commit_cb;
1554 /* cap resend delay to the current request timeout, this is similar to
1555 * what ptlrpc does (see after_reply()) */
1556 if (aa->aa_resends > new_req->rq_timeout)
1557 new_req->rq_sent = cfs_time_current_sec() + new_req->rq_timeout;
1559 new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1560 new_req->rq_generation_set = 1;
1561 new_req->rq_import_generation = request->rq_import_generation;
1563 new_aa = ptlrpc_req_async_args(new_req);
1565 INIT_LIST_HEAD(&new_aa->aa_oaps);
1566 list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1567 INIT_LIST_HEAD(&new_aa->aa_exts);
1568 list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1569 new_aa->aa_resends = aa->aa_resends;
1571 list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1572 if (oap->oap_request) {
1573 ptlrpc_req_finished(oap->oap_request);
1574 oap->oap_request = ptlrpc_request_addref(new_req);
1578 new_aa->aa_ocapa = aa->aa_ocapa;
1579 aa->aa_ocapa = NULL;
1581 /* XXX: This code will run into problem if we're going to support
1582 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1583 * and wait for all of them to be finished. We should inherit request
1584 * set from old request. */
1585 ptlrpcd_add_req(new_req, PDL_POLICY_SAME, -1);
1587 DEBUG_REQ(D_INFO, new_req, "new request");
1592 * ugh, we want disk allocation on the target to happen in offset order. we'll
1593 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1594 * fine for our small page arrays and doesn't require allocation. its an
1595 * insertion sort that swaps elements that are strides apart, shrinking the
1596 * stride down until its '1' and the array is sorted.
1598 static void sort_brw_pages(struct brw_page **array, int num)
1601 struct brw_page *tmp;
1605 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1610 for (i = stride ; i < num ; i++) {
1613 while (j >= stride && array[j - stride]->off > tmp->off) {
1614 array[j] = array[j - stride];
1619 } while (stride > 1);
1622 static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1624 LASSERT(ppga != NULL);
1625 OBD_FREE(ppga, sizeof(*ppga) * count);
1628 static int brw_interpret(const struct lu_env *env,
1629 struct ptlrpc_request *req, void *data, int rc)
1631 struct osc_brw_async_args *aa = data;
1632 struct osc_extent *ext;
1633 struct osc_extent *tmp;
1634 struct client_obd *cli = aa->aa_cli;
1637 rc = osc_brw_fini_request(req, rc);
1638 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
1639 /* When server return -EINPROGRESS, client should always retry
1640 * regardless of the number of times the bulk was resent already. */
1641 if (osc_recoverable_error(rc)) {
1642 if (req->rq_import_generation !=
1643 req->rq_import->imp_generation) {
1644 CDEBUG(D_HA, "%s: resend cross eviction for object: "
1645 ""DOSTID", rc = %d.\n",
1646 req->rq_import->imp_obd->obd_name,
1647 POSTID(&aa->aa_oa->o_oi), rc);
1648 } else if (rc == -EINPROGRESS ||
1649 client_should_resend(aa->aa_resends, aa->aa_cli)) {
1650 rc = osc_brw_redo_request(req, aa, rc);
1652 CERROR("%s: too many resent retries for object: "
1653 ""LPU64":"LPU64", rc = %d.\n",
1654 req->rq_import->imp_obd->obd_name,
1655 POSTID(&aa->aa_oa->o_oi), rc);
1660 else if (rc == -EAGAIN || rc == -EINPROGRESS)
1665 capa_put(aa->aa_ocapa);
1666 aa->aa_ocapa = NULL;
1670 struct obdo *oa = aa->aa_oa;
1671 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1672 unsigned long valid = 0;
1673 struct cl_object *obj;
1674 struct osc_async_page *last;
1676 last = brw_page2oap(aa->aa_ppga[aa->aa_page_count - 1]);
1677 obj = osc2cl(last->oap_obj);
1679 cl_object_attr_lock(obj);
1680 if (oa->o_valid & OBD_MD_FLBLOCKS) {
1681 attr->cat_blocks = oa->o_blocks;
1682 valid |= CAT_BLOCKS;
1684 if (oa->o_valid & OBD_MD_FLMTIME) {
1685 attr->cat_mtime = oa->o_mtime;
1688 if (oa->o_valid & OBD_MD_FLATIME) {
1689 attr->cat_atime = oa->o_atime;
1692 if (oa->o_valid & OBD_MD_FLCTIME) {
1693 attr->cat_ctime = oa->o_ctime;
1697 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1698 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1699 loff_t last_off = last->oap_count + last->oap_obj_off +
1702 /* Change file size if this is an out of quota or
1703 * direct IO write and it extends the file size */
1704 if (loi->loi_lvb.lvb_size < last_off) {
1705 attr->cat_size = last_off;
1708 /* Extend KMS if it's not a lockless write */
1709 if (loi->loi_kms < last_off &&
1710 oap2osc_page(last)->ops_srvlock == 0) {
1711 attr->cat_kms = last_off;
1717 cl_object_attr_update(env, obj, attr, valid);
1718 cl_object_attr_unlock(obj);
1720 OBDO_FREE(aa->aa_oa);
1722 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
1723 osc_inc_unstable_pages(req);
1725 list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
1726 list_del_init(&ext->oe_link);
1727 osc_extent_finish(env, ext, 1, rc);
1729 LASSERT(list_empty(&aa->aa_exts));
1730 LASSERT(list_empty(&aa->aa_oaps));
1732 cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc :
1733 req->rq_bulk->bd_nob_transferred);
1734 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
1735 ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred);
1737 spin_lock(&cli->cl_loi_list_lock);
1738 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
1739 * is called so we know whether to go to sync BRWs or wait for more
1740 * RPCs to complete */
1741 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
1742 cli->cl_w_in_flight--;
1744 cli->cl_r_in_flight--;
1745 osc_wake_cache_waiters(cli);
1746 spin_unlock(&cli->cl_loi_list_lock);
1748 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
1752 static void brw_commit(struct ptlrpc_request *req)
1754 /* If osc_inc_unstable_pages (via osc_extent_finish) races with
1755 * this called via the rq_commit_cb, I need to ensure
1756 * osc_dec_unstable_pages is still called. Otherwise unstable
1757 * pages may be leaked. */
1758 spin_lock(&req->rq_lock);
1759 if (likely(req->rq_unstable)) {
1760 req->rq_unstable = 0;
1761 spin_unlock(&req->rq_lock);
1763 osc_dec_unstable_pages(req);
1765 req->rq_committed = 1;
1766 spin_unlock(&req->rq_lock);
1771 * Build an RPC by the list of extent @ext_list. The caller must ensure
1772 * that the total pages in this list are NOT over max pages per RPC.
1773 * Extents in the list must be in OES_RPC state.
1775 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
1776 struct list_head *ext_list, int cmd, pdl_policy_t pol)
1778 struct ptlrpc_request *req = NULL;
1779 struct osc_extent *ext;
1780 struct brw_page **pga = NULL;
1781 struct osc_brw_async_args *aa = NULL;
1782 struct obdo *oa = NULL;
1783 struct osc_async_page *oap;
1784 struct osc_async_page *tmp;
1785 struct cl_req *clerq = NULL;
1786 enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE :
1788 struct cl_req_attr *crattr = NULL;
1789 obd_off starting_offset = OBD_OBJECT_EOF;
1790 obd_off ending_offset = 0;
1794 bool soft_sync = false;
1797 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1798 struct ost_body *body;
1800 LASSERT(!list_empty(ext_list));
1802 /* add pages into rpc_list to build BRW rpc */
1803 list_for_each_entry(ext, ext_list, oe_link) {
1804 LASSERT(ext->oe_state == OES_RPC);
1805 mem_tight |= ext->oe_memalloc;
1806 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
1808 list_add_tail(&oap->oap_rpc_item, &rpc_list);
1809 if (starting_offset > oap->oap_obj_off)
1810 starting_offset = oap->oap_obj_off;
1812 LASSERT(oap->oap_page_off == 0);
1813 if (ending_offset < oap->oap_obj_off + oap->oap_count)
1814 ending_offset = oap->oap_obj_off +
1817 LASSERT(oap->oap_page_off + oap->oap_count ==
1822 soft_sync = osc_over_unstable_soft_limit(cli);
1824 mpflag = cfs_memory_pressure_get_and_set();
1826 OBD_ALLOC(crattr, sizeof(*crattr));
1828 GOTO(out, rc = -ENOMEM);
1830 OBD_ALLOC(pga, sizeof(*pga) * page_count);
1832 GOTO(out, rc = -ENOMEM);
1836 GOTO(out, rc = -ENOMEM);
1839 list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
1840 struct cl_page *page = oap2cl_page(oap);
1841 if (clerq == NULL) {
1842 clerq = cl_req_alloc(env, page, crt,
1843 1 /* only 1-object rpcs for now */);
1845 GOTO(out, rc = PTR_ERR(clerq));
1848 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
1850 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
1851 pga[i] = &oap->oap_brw_page;
1852 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
1853 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
1854 pga[i]->pg, page_index(oap->oap_page), oap,
1857 cl_req_page_add(env, clerq, page);
1860 /* always get the data for the obdo for the rpc */
1861 LASSERT(clerq != NULL);
1862 crattr->cra_oa = oa;
1863 cl_req_attr_set(env, clerq, crattr, ~0ULL);
1865 rc = cl_req_prep(env, clerq);
1867 CERROR("cl_req_prep failed: %d\n", rc);
1871 sort_brw_pages(pga, page_count);
1872 rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
1873 pga, &req, crattr->cra_capa, 1, 0);
1875 CERROR("prep_req failed: %d\n", rc);
1879 req->rq_commit_cb = brw_commit;
1880 req->rq_interpret_reply = brw_interpret;
1883 req->rq_memalloc = 1;
1885 /* Need to update the timestamps after the request is built in case
1886 * we race with setattr (locally or in queue at OST). If OST gets
1887 * later setattr before earlier BRW (as determined by the request xid),
1888 * the OST will not use BRW timestamps. Sadly, there is no obvious
1889 * way to do this in a single call. bug 10150 */
1890 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1891 crattr->cra_oa = &body->oa;
1892 cl_req_attr_set(env, clerq, crattr,
1893 OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
1895 lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
1897 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1898 aa = ptlrpc_req_async_args(req);
1899 INIT_LIST_HEAD(&aa->aa_oaps);
1900 list_splice_init(&rpc_list, &aa->aa_oaps);
1901 INIT_LIST_HEAD(&aa->aa_exts);
1902 list_splice_init(ext_list, &aa->aa_exts);
1903 aa->aa_clerq = clerq;
1905 /* queued sync pages can be torn down while the pages
1906 * were between the pending list and the rpc */
1908 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1909 /* only one oap gets a request reference */
1912 if (oap->oap_interrupted && !req->rq_intr) {
1913 CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
1915 ptlrpc_mark_interrupted(req);
1919 tmp->oap_request = ptlrpc_request_addref(req);
1921 spin_lock(&cli->cl_loi_list_lock);
1922 starting_offset >>= PAGE_CACHE_SHIFT;
1923 if (cmd == OBD_BRW_READ) {
1924 cli->cl_r_in_flight++;
1925 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
1926 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
1927 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
1928 starting_offset + 1);
1930 cli->cl_w_in_flight++;
1931 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
1932 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
1933 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
1934 starting_offset + 1);
1936 spin_unlock(&cli->cl_loi_list_lock);
1938 DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
1939 page_count, aa, cli->cl_r_in_flight,
1940 cli->cl_w_in_flight);
1942 /* XXX: Maybe the caller can check the RPC bulk descriptor to
1943 * see which CPU/NUMA node the majority of pages were allocated
1944 * on, and try to assign the async RPC to the CPU core
1945 * (PDL_POLICY_PREFERRED) to reduce cross-CPU memory traffic.
1947 * But on the other hand, we expect that multiple ptlrpcd
1948 * threads and the initial write sponsor can run in parallel,
1949 * especially when data checksum is enabled, which is CPU-bound
1950 * operation and single ptlrpcd thread cannot process in time.
1951 * So more ptlrpcd threads sharing BRW load
1952 * (with PDL_POLICY_ROUND) seems better.
1954 ptlrpcd_add_req(req, pol, -1);
1960 cfs_memory_pressure_restore(mpflag);
1962 if (crattr != NULL) {
1963 capa_put(crattr->cra_capa);
1964 OBD_FREE(crattr, sizeof(*crattr));
1968 LASSERT(req == NULL);
1973 OBD_FREE(pga, sizeof(*pga) * page_count);
1974 /* this should happen rarely and is pretty bad, it makes the
1975 * pending list not follow the dirty order */
1976 while (!list_empty(ext_list)) {
1977 ext = list_entry(ext_list->next, struct osc_extent,
1979 list_del_init(&ext->oe_link);
1980 osc_extent_finish(env, ext, 0, rc);
1982 if (clerq && !IS_ERR(clerq))
1983 cl_req_completion(env, clerq, rc);
1988 static int osc_set_lock_data_with_check(struct ldlm_lock *lock,
1989 struct ldlm_enqueue_info *einfo)
1991 void *data = einfo->ei_cbdata;
1994 LASSERT(lock != NULL);
1995 LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl);
1996 LASSERT(lock->l_resource->lr_type == einfo->ei_type);
1997 LASSERT(lock->l_completion_ast == einfo->ei_cb_cp);
1998 LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
2000 lock_res_and_lock(lock);
2002 if (lock->l_ast_data == NULL)
2003 lock->l_ast_data = data;
2004 if (lock->l_ast_data == data)
2007 unlock_res_and_lock(lock);
2012 static int osc_set_data_with_check(struct lustre_handle *lockh,
2013 struct ldlm_enqueue_info *einfo)
2015 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2019 set = osc_set_lock_data_with_check(lock, einfo);
2020 LDLM_LOCK_PUT(lock);
2022 CERROR("lockh %p, data %p - client evicted?\n",
2023 lockh, einfo->ei_cbdata);
2027 static int osc_enqueue_fini(struct ptlrpc_request *req,
2028 osc_enqueue_upcall_f upcall, void *cookie,
2029 struct lustre_handle *lockh, ldlm_mode_t mode,
2030 __u64 *flags, int agl, int errcode)
2032 bool intent = *flags & LDLM_FL_HAS_INTENT;
2036 /* The request was created before ldlm_cli_enqueue call. */
2037 if (intent && errcode == ELDLM_LOCK_ABORTED) {
2038 struct ldlm_reply *rep;
2040 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
2041 LASSERT(rep != NULL);
2043 rep->lock_policy_res1 =
2044 ptlrpc_status_ntoh(rep->lock_policy_res1);
2045 if (rep->lock_policy_res1)
2046 errcode = rep->lock_policy_res1;
2048 *flags |= LDLM_FL_LVB_READY;
2049 } else if (errcode == ELDLM_OK) {
2050 *flags |= LDLM_FL_LVB_READY;
2053 /* Call the update callback. */
2054 rc = (*upcall)(cookie, lockh, errcode);
2056 /* release the reference taken in ldlm_cli_enqueue() */
2057 if (errcode == ELDLM_LOCK_MATCHED)
2059 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
2060 ldlm_lock_decref(lockh, mode);
2065 static int osc_enqueue_interpret(const struct lu_env *env,
2066 struct ptlrpc_request *req,
2067 struct osc_enqueue_args *aa, int rc)
2069 struct ldlm_lock *lock;
2070 struct lustre_handle *lockh = &aa->oa_lockh;
2071 ldlm_mode_t mode = aa->oa_mode;
2072 struct ost_lvb *lvb = aa->oa_lvb;
2073 __u32 lvb_len = sizeof(*lvb);
2078 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2080 lock = ldlm_handle2lock(lockh);
2081 LASSERTF(lock != NULL,
2082 "lockh "LPX64", req %p, aa %p - client evicted?\n",
2083 lockh->cookie, req, aa);
2085 /* Take an additional reference so that a blocking AST that
2086 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2087 * to arrive after an upcall has been executed by
2088 * osc_enqueue_fini(). */
2089 ldlm_lock_addref(lockh, mode);
2091 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
2092 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
2094 /* Let CP AST to grant the lock first. */
2095 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2098 LASSERT(aa->oa_lvb == NULL);
2099 LASSERT(aa->oa_flags == NULL);
2100 aa->oa_flags = &flags;
2103 /* Complete obtaining the lock procedure. */
2104 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
2105 aa->oa_mode, aa->oa_flags, lvb, lvb_len,
2107 /* Complete osc stuff. */
2108 rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
2109 aa->oa_flags, aa->oa_agl, rc);
2111 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2113 ldlm_lock_decref(lockh, mode);
2114 LDLM_LOCK_PUT(lock);
2118 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2120 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2121 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2122 * other synchronous requests, however keeping some locks and trying to obtain
2123 * others may take a considerable amount of time in a case of ost failure; and
2124 * when other sync requests do not get released lock from a client, the client
2125 * is evicted from the cluster -- such scenarious make the life difficult, so
2126 * release locks just after they are obtained. */
2127 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2128 __u64 *flags, ldlm_policy_data_t *policy,
2129 struct ost_lvb *lvb, int kms_valid,
2130 osc_enqueue_upcall_f upcall, void *cookie,
2131 struct ldlm_enqueue_info *einfo,
2132 struct ptlrpc_request_set *rqset, int async, int agl)
2134 struct obd_device *obd = exp->exp_obd;
2135 struct lustre_handle lockh = { 0 };
2136 struct ptlrpc_request *req = NULL;
2137 int intent = *flags & LDLM_FL_HAS_INTENT;
2138 __u64 match_lvb = agl ? 0 : LDLM_FL_LVB_READY;
2143 /* Filesystem lock extents are extended to page boundaries so that
2144 * dealing with the page cache is a little smoother. */
2145 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2146 policy->l_extent.end |= ~CFS_PAGE_MASK;
2149 * kms is not valid when either object is completely fresh (so that no
2150 * locks are cached), or object was evicted. In the latter case cached
2151 * lock cannot be used, because it would prime inode state with
2152 * potentially stale LVB.
2157 /* Next, search for already existing extent locks that will cover us */
2158 /* If we're trying to read, we also search for an existing PW lock. The
2159 * VFS and page cache already protect us locally, so lots of readers/
2160 * writers can share a single PW lock.
2162 * There are problems with conversion deadlocks, so instead of
2163 * converting a read lock to a write lock, we'll just enqueue a new
2166 * At some point we should cancel the read lock instead of making them
2167 * send us a blocking callback, but there are problems with canceling
2168 * locks out from other users right now, too. */
2169 mode = einfo->ei_mode;
2170 if (einfo->ei_mode == LCK_PR)
2172 mode = ldlm_lock_match(obd->obd_namespace, *flags | match_lvb, res_id,
2173 einfo->ei_type, policy, mode, &lockh, 0);
2175 struct ldlm_lock *matched;
2177 if (*flags & LDLM_FL_TEST_LOCK)
2180 matched = ldlm_handle2lock(&lockh);
2182 /* AGL enqueues DLM locks speculatively. Therefore if
2183 * it already exists a DLM lock, it wll just inform the
2184 * caller to cancel the AGL process for this stripe. */
2185 ldlm_lock_decref(&lockh, mode);
2186 LDLM_LOCK_PUT(matched);
2188 } else if (osc_set_lock_data_with_check(matched, einfo)) {
2189 *flags |= LDLM_FL_LVB_READY;
2191 /* We already have a lock, and it's referenced. */
2192 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
2194 ldlm_lock_decref(&lockh, mode);
2195 LDLM_LOCK_PUT(matched);
2198 ldlm_lock_decref(&lockh, mode);
2199 LDLM_LOCK_PUT(matched);
2204 if (*flags & LDLM_FL_TEST_LOCK)
2208 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2209 &RQF_LDLM_ENQUEUE_LVB);
2213 rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
2215 ptlrpc_request_free(req);
2219 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
2221 ptlrpc_request_set_replen(req);
2224 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2225 *flags &= ~LDLM_FL_BLOCK_GRANTED;
2227 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2228 sizeof(*lvb), LVB_T_OST, &lockh, async);
2231 struct osc_enqueue_args *aa;
2232 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2233 aa = ptlrpc_req_async_args(req);
2235 aa->oa_mode = einfo->ei_mode;
2236 aa->oa_type = einfo->ei_type;
2237 lustre_handle_copy(&aa->oa_lockh, &lockh);
2238 aa->oa_upcall = upcall;
2239 aa->oa_cookie = cookie;
2242 aa->oa_flags = flags;
2245 /* AGL is essentially to enqueue an DLM lock
2246 * in advance, so we don't care about the
2247 * result of AGL enqueue. */
2249 aa->oa_flags = NULL;
2252 req->rq_interpret_reply =
2253 (ptlrpc_interpterer_t)osc_enqueue_interpret;
2254 if (rqset == PTLRPCD_SET)
2255 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
2257 ptlrpc_set_add_req(rqset, req);
2258 } else if (intent) {
2259 ptlrpc_req_finished(req);
2264 rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
2267 ptlrpc_req_finished(req);
2272 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2273 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2274 __u64 *flags, void *data, struct lustre_handle *lockh,
2277 struct obd_device *obd = exp->exp_obd;
2278 __u64 lflags = *flags;
2282 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
2285 /* Filesystem lock extents are extended to page boundaries so that
2286 * dealing with the page cache is a little smoother */
2287 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2288 policy->l_extent.end |= ~CFS_PAGE_MASK;
2290 /* Next, search for already existing extent locks that will cover us */
2291 /* If we're trying to read, we also search for an existing PW lock. The
2292 * VFS and page cache already protect us locally, so lots of readers/
2293 * writers can share a single PW lock. */
2297 rc = ldlm_lock_match(obd->obd_namespace, lflags,
2298 res_id, type, policy, rc, lockh, unref);
2301 if (!osc_set_data_with_check(lockh, data)) {
2302 if (!(lflags & LDLM_FL_TEST_LOCK))
2303 ldlm_lock_decref(lockh, rc);
2307 if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) {
2308 ldlm_lock_addref(lockh, LCK_PR);
2309 ldlm_lock_decref(lockh, LCK_PW);
2316 int osc_cancel_base(struct lustre_handle *lockh, __u32 mode)
2320 if (unlikely(mode == LCK_GROUP))
2321 ldlm_lock_decref_and_cancel(lockh, mode);
2323 ldlm_lock_decref(lockh, mode);
2328 static int osc_statfs_interpret(const struct lu_env *env,
2329 struct ptlrpc_request *req,
2330 struct osc_async_args *aa, int rc)
2332 struct obd_statfs *msfs;
2336 /* The request has in fact never been sent
2337 * due to issues at a higher level (LOV).
2338 * Exit immediately since the caller is
2339 * aware of the problem and takes care
2340 * of the clean up */
2343 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2344 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2350 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2352 GOTO(out, rc = -EPROTO);
2355 *aa->aa_oi->oi_osfs = *msfs;
2357 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
2361 static int osc_statfs_async(struct obd_export *exp,
2362 struct obd_info *oinfo, __u64 max_age,
2363 struct ptlrpc_request_set *rqset)
2365 struct obd_device *obd = class_exp2obd(exp);
2366 struct ptlrpc_request *req;
2367 struct osc_async_args *aa;
2371 /* We could possibly pass max_age in the request (as an absolute
2372 * timestamp or a "seconds.usec ago") so the target can avoid doing
2373 * extra calls into the filesystem if that isn't necessary (e.g.
2374 * during mount that would help a bit). Having relative timestamps
2375 * is not so great if request processing is slow, while absolute
2376 * timestamps are not ideal because they need time synchronization. */
2377 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2381 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2383 ptlrpc_request_free(req);
2386 ptlrpc_request_set_replen(req);
2387 req->rq_request_portal = OST_CREATE_PORTAL;
2388 ptlrpc_at_set_req_timeout(req);
2390 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2391 /* procfs requests not want stat in wait for avoid deadlock */
2392 req->rq_no_resend = 1;
2393 req->rq_no_delay = 1;
2396 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
2397 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
2398 aa = ptlrpc_req_async_args(req);
2401 ptlrpc_set_add_req(rqset, req);
2405 static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2406 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
2408 struct obd_device *obd = class_exp2obd(exp);
2409 struct obd_statfs *msfs;
2410 struct ptlrpc_request *req;
2411 struct obd_import *imp = NULL;
2415 /*Since the request might also come from lprocfs, so we need
2416 *sync this with client_disconnect_export Bug15684*/
2417 down_read(&obd->u.cli.cl_sem);
2418 if (obd->u.cli.cl_import)
2419 imp = class_import_get(obd->u.cli.cl_import);
2420 up_read(&obd->u.cli.cl_sem);
2424 /* We could possibly pass max_age in the request (as an absolute
2425 * timestamp or a "seconds.usec ago") so the target can avoid doing
2426 * extra calls into the filesystem if that isn't necessary (e.g.
2427 * during mount that would help a bit). Having relative timestamps
2428 * is not so great if request processing is slow, while absolute
2429 * timestamps are not ideal because they need time synchronization. */
2430 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2432 class_import_put(imp);
2437 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2439 ptlrpc_request_free(req);
2442 ptlrpc_request_set_replen(req);
2443 req->rq_request_portal = OST_CREATE_PORTAL;
2444 ptlrpc_at_set_req_timeout(req);
2446 if (flags & OBD_STATFS_NODELAY) {
2447 /* procfs requests not want stat in wait for avoid deadlock */
2448 req->rq_no_resend = 1;
2449 req->rq_no_delay = 1;
2452 rc = ptlrpc_queue_wait(req);
2456 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2458 GOTO(out, rc = -EPROTO);
2465 ptlrpc_req_finished(req);
2469 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2470 void *karg, void *uarg)
2472 struct obd_device *obd = exp->exp_obd;
2473 struct obd_ioctl_data *data = karg;
2477 if (!try_module_get(THIS_MODULE)) {
2478 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2479 module_name(THIS_MODULE));
2483 case OBD_IOC_CLIENT_RECOVER:
2484 err = ptlrpc_recover_import(obd->u.cli.cl_import,
2485 data->ioc_inlbuf1, 0);
2489 case IOC_OSC_SET_ACTIVE:
2490 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
2493 case OBD_IOC_POLL_QUOTACHECK:
2494 err = osc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2496 case OBD_IOC_PING_TARGET:
2497 err = ptlrpc_obd_ping(obd);
2500 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
2501 cmd, current_comm());
2502 GOTO(out, err = -ENOTTY);
2505 module_put(THIS_MODULE);
2509 static int osc_get_info(const struct lu_env *env, struct obd_export *exp,
2510 obd_count keylen, void *key, __u32 *vallen, void *val,
2511 struct lov_stripe_md *lsm)
2514 if (!vallen || !val)
2517 if (KEY_IS(KEY_FIEMAP)) {
2518 struct ll_fiemap_info_key *fm_key =
2519 (struct ll_fiemap_info_key *)key;
2520 struct ldlm_res_id res_id;
2521 ldlm_policy_data_t policy;
2522 struct lustre_handle lockh;
2523 ldlm_mode_t mode = 0;
2524 struct ptlrpc_request *req;
2525 struct ll_user_fiemap *reply;
2529 if (!(fm_key->fiemap.fm_flags & FIEMAP_FLAG_SYNC))
2532 policy.l_extent.start = fm_key->fiemap.fm_start &
2535 if (OBD_OBJECT_EOF - fm_key->fiemap.fm_length <=
2536 fm_key->fiemap.fm_start + PAGE_CACHE_SIZE - 1)
2537 policy.l_extent.end = OBD_OBJECT_EOF;
2539 policy.l_extent.end = (fm_key->fiemap.fm_start +
2540 fm_key->fiemap.fm_length +
2541 PAGE_CACHE_SIZE - 1) & CFS_PAGE_MASK;
2543 ostid_build_res_name(&fm_key->oa.o_oi, &res_id);
2544 mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
2545 LDLM_FL_BLOCK_GRANTED |
2547 &res_id, LDLM_EXTENT, &policy,
2548 LCK_PR | LCK_PW, &lockh, 0);
2549 if (mode) { /* lock is cached on client */
2550 if (mode != LCK_PR) {
2551 ldlm_lock_addref(&lockh, LCK_PR);
2552 ldlm_lock_decref(&lockh, LCK_PW);
2554 } else { /* no cached lock, needs acquire lock on server side */
2555 fm_key->oa.o_valid |= OBD_MD_FLFLAGS;
2556 fm_key->oa.o_flags |= OBD_FL_SRVLOCK;
2560 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2561 &RQF_OST_GET_INFO_FIEMAP);
2563 GOTO(drop_lock, rc = -ENOMEM);
2565 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY,
2566 RCL_CLIENT, keylen);
2567 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
2568 RCL_CLIENT, *vallen);
2569 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
2570 RCL_SERVER, *vallen);
2572 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
2574 ptlrpc_request_free(req);
2575 GOTO(drop_lock, rc);
2578 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
2579 memcpy(tmp, key, keylen);
2580 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
2581 memcpy(tmp, val, *vallen);
2583 ptlrpc_request_set_replen(req);
2584 rc = ptlrpc_queue_wait(req);
2588 reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
2590 GOTO(fini_req, rc = -EPROTO);
2592 memcpy(val, reply, *vallen);
2594 ptlrpc_req_finished(req);
2597 ldlm_lock_decref(&lockh, LCK_PR);
2604 static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
2605 obd_count keylen, void *key, obd_count vallen,
2606 void *val, struct ptlrpc_request_set *set)
2608 struct ptlrpc_request *req;
2609 struct obd_device *obd = exp->exp_obd;
2610 struct obd_import *imp = class_exp2cliimp(exp);
2615 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
2617 if (KEY_IS(KEY_CHECKSUM)) {
2618 if (vallen != sizeof(int))
2620 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
2624 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2625 sptlrpc_conf_client_adapt(obd);
2629 if (KEY_IS(KEY_FLUSH_CTX)) {
2630 sptlrpc_import_flush_my_ctx(imp);
2634 if (KEY_IS(KEY_CACHE_SET)) {
2635 struct client_obd *cli = &obd->u.cli;
2637 LASSERT(cli->cl_cache == NULL); /* only once */
2638 cli->cl_cache = (struct cl_client_cache *)val;
2639 cl_cache_incref(cli->cl_cache);
2640 cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
2642 /* add this osc into entity list */
2643 LASSERT(list_empty(&cli->cl_lru_osc));
2644 spin_lock(&cli->cl_cache->ccc_lru_lock);
2645 list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
2646 spin_unlock(&cli->cl_cache->ccc_lru_lock);
2651 if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
2652 struct client_obd *cli = &obd->u.cli;
2653 long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
2654 long target = *(long *)val;
2656 nr = osc_lru_shrink(env, cli, min(nr, target), true);
2661 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
2664 /* We pass all other commands directly to OST. Since nobody calls osc
2665 methods directly and everybody is supposed to go through LOV, we
2666 assume lov checked invalid values for us.
2667 The only recognised values so far are evict_by_nid and mds_conn.
2668 Even if something bad goes through, we'd get a -EINVAL from OST
2671 req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
2672 &RQF_OST_SET_GRANT_INFO :
2677 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2678 RCL_CLIENT, keylen);
2679 if (!KEY_IS(KEY_GRANT_SHRINK))
2680 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
2681 RCL_CLIENT, vallen);
2682 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
2684 ptlrpc_request_free(req);
2688 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
2689 memcpy(tmp, key, keylen);
2690 tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
2693 memcpy(tmp, val, vallen);
2695 if (KEY_IS(KEY_GRANT_SHRINK)) {
2696 struct osc_grant_args *aa;
2699 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2700 aa = ptlrpc_req_async_args(req);
2703 ptlrpc_req_finished(req);
2706 *oa = ((struct ost_body *)val)->oa;
2708 req->rq_interpret_reply = osc_shrink_grant_interpret;
2711 ptlrpc_request_set_replen(req);
2712 if (!KEY_IS(KEY_GRANT_SHRINK)) {
2713 LASSERT(set != NULL);
2714 ptlrpc_set_add_req(set, req);
2715 ptlrpc_check_set(NULL, set);
2717 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
2722 static int osc_reconnect(const struct lu_env *env,
2723 struct obd_export *exp, struct obd_device *obd,
2724 struct obd_uuid *cluuid,
2725 struct obd_connect_data *data,
2728 struct client_obd *cli = &obd->u.cli;
2730 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
2733 spin_lock(&cli->cl_loi_list_lock);
2734 data->ocd_grant = (cli->cl_avail_grant +
2735 (cli->cl_dirty_pages << PAGE_CACHE_SHIFT)) ?:
2736 2 * cli_brw_size(obd);
2737 lost_grant = cli->cl_lost_grant;
2738 cli->cl_lost_grant = 0;
2739 spin_unlock(&cli->cl_loi_list_lock);
2741 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
2742 " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
2743 data->ocd_version, data->ocd_grant, lost_grant);
2749 static int osc_disconnect(struct obd_export *exp)
2751 struct obd_device *obd = class_exp2obd(exp);
2754 rc = client_disconnect_export(exp);
2756 * Initially we put del_shrink_grant before disconnect_export, but it
2757 * causes the following problem if setup (connect) and cleanup
2758 * (disconnect) are tangled together.
2759 * connect p1 disconnect p2
2760 * ptlrpc_connect_import
2761 * ............... class_manual_cleanup
2764 * ptlrpc_connect_interrupt
2766 * add this client to shrink list
2768 * Bang! pinger trigger the shrink.
2769 * So the osc should be disconnected from the shrink list, after we
2770 * are sure the import has been destroyed. BUG18662
2772 if (obd->u.cli.cl_import == NULL)
2773 osc_del_shrink_grant(&obd->u.cli);
2777 static int osc_import_event(struct obd_device *obd,
2778 struct obd_import *imp,
2779 enum obd_import_event event)
2781 struct client_obd *cli;
2785 LASSERT(imp->imp_obd == obd);
2788 case IMP_EVENT_DISCON: {
2790 spin_lock(&cli->cl_loi_list_lock);
2791 cli->cl_avail_grant = 0;
2792 cli->cl_lost_grant = 0;
2793 spin_unlock(&cli->cl_loi_list_lock);
2796 case IMP_EVENT_INACTIVE: {
2797 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2800 case IMP_EVENT_INVALIDATE: {
2801 struct ldlm_namespace *ns = obd->obd_namespace;
2805 env = cl_env_get(&refcheck);
2809 /* all pages go to failing rpcs due to the invalid
2811 osc_io_unplug(env, cli, NULL, PDL_POLICY_ROUND);
2813 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2814 cl_env_put(env, &refcheck);
2819 case IMP_EVENT_ACTIVE: {
2820 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2823 case IMP_EVENT_OCD: {
2824 struct obd_connect_data *ocd = &imp->imp_connect_data;
2826 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
2827 osc_init_grant(&obd->u.cli, ocd);
2830 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
2831 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
2833 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2836 case IMP_EVENT_DEACTIVATE: {
2837 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE, NULL);
2840 case IMP_EVENT_ACTIVATE: {
2841 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE, NULL);
2845 CERROR("Unknown import event %d\n", event);
2852 * Determine whether the lock can be canceled before replaying the lock
2853 * during recovery, see bug16774 for detailed information.
2855 * \retval zero the lock can't be canceled
2856 * \retval other ok to cancel
2858 static int osc_cancel_weight(struct ldlm_lock *lock)
2861 * Cancel all unused and granted extent lock.
2863 if (lock->l_resource->lr_type == LDLM_EXTENT &&
2864 lock->l_granted_mode == lock->l_req_mode &&
2865 osc_ldlm_weigh_ast(lock) == 0)
2871 static int brw_queue_work(const struct lu_env *env, void *data)
2873 struct client_obd *cli = data;
2875 CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
2877 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
2881 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
2883 struct client_obd *cli = &obd->u.cli;
2884 struct obd_type *type;
2889 rc = ptlrpcd_addref();
2893 rc = client_obd_setup(obd, lcfg);
2895 GOTO(out_ptlrpcd, rc);
2897 handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
2898 if (IS_ERR(handler))
2899 GOTO(out_client_setup, rc = PTR_ERR(handler));
2900 cli->cl_writeback_work = handler;
2902 handler = ptlrpcd_alloc_work(cli->cl_import, lru_queue_work, cli);
2903 if (IS_ERR(handler))
2904 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
2905 cli->cl_lru_work = handler;
2907 rc = osc_quota_setup(obd);
2909 GOTO(out_ptlrpcd_work, rc);
2911 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
2913 #ifdef CONFIG_PROC_FS
2914 obd->obd_vars = lprocfs_osc_obd_vars;
2916 /* If this is true then both client (osc) and server (osp) are on the
2917 * same node. The osp layer if loaded first will register the osc proc
2918 * directory. In that case this obd_device will be attached its proc
2919 * tree to type->typ_procsym instead of obd->obd_type->typ_procroot. */
2920 type = class_search_type(LUSTRE_OSP_NAME);
2921 if (type && type->typ_procsym) {
2922 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
2924 obd->obd_vars, obd);
2925 if (IS_ERR(obd->obd_proc_entry)) {
2926 rc = PTR_ERR(obd->obd_proc_entry);
2927 CERROR("error %d setting up lprocfs for %s\n", rc,
2929 obd->obd_proc_entry = NULL;
2932 rc = lprocfs_obd_setup(obd);
2935 /* If the basic OSC proc tree construction succeeded then
2936 * lets do the rest. */
2938 lproc_osc_attach_seqstat(obd);
2939 sptlrpc_lprocfs_cliobd_attach(obd);
2940 ptlrpc_lprocfs_register_obd(obd);
2943 /* We need to allocate a few requests more, because
2944 * brw_interpret tries to create new requests before freeing
2945 * previous ones, Ideally we want to have 2x max_rpcs_in_flight
2946 * reserved, but I'm afraid that might be too much wasted RAM
2947 * in fact, so 2 is just my guess and still should work. */
2948 cli->cl_import->imp_rq_pool =
2949 ptlrpc_init_rq_pool(cli->cl_max_rpcs_in_flight + 2,
2951 ptlrpc_add_rqs_to_pool);
2953 INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
2954 ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
2958 if (cli->cl_writeback_work != NULL) {
2959 ptlrpcd_destroy_work(cli->cl_writeback_work);
2960 cli->cl_writeback_work = NULL;
2962 if (cli->cl_lru_work != NULL) {
2963 ptlrpcd_destroy_work(cli->cl_lru_work);
2964 cli->cl_lru_work = NULL;
2967 client_obd_cleanup(obd);
2973 static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2979 case OBD_CLEANUP_EARLY: {
2980 struct obd_import *imp;
2981 imp = obd->u.cli.cl_import;
2982 CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name);
2983 /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */
2984 ptlrpc_deactivate_import(imp);
2985 spin_lock(&imp->imp_lock);
2986 imp->imp_pingable = 0;
2987 spin_unlock(&imp->imp_lock);
2990 case OBD_CLEANUP_EXPORTS: {
2991 struct client_obd *cli = &obd->u.cli;
2993 * for echo client, export may be on zombie list, wait for
2994 * zombie thread to cull it, because cli.cl_import will be
2995 * cleared in client_disconnect_export():
2996 * class_export_destroy() -> obd_cleanup() ->
2997 * echo_device_free() -> echo_client_cleanup() ->
2998 * obd_disconnect() -> osc_disconnect() ->
2999 * client_disconnect_export()
3001 obd_zombie_barrier();
3002 if (cli->cl_writeback_work) {
3003 ptlrpcd_destroy_work(cli->cl_writeback_work);
3004 cli->cl_writeback_work = NULL;
3006 if (cli->cl_lru_work) {
3007 ptlrpcd_destroy_work(cli->cl_lru_work);
3008 cli->cl_lru_work = NULL;
3010 obd_cleanup_client_import(obd);
3011 ptlrpc_lprocfs_unregister_obd(obd);
3012 lprocfs_obd_cleanup(obd);
3019 int osc_cleanup(struct obd_device *obd)
3021 struct client_obd *cli = &obd->u.cli;
3027 if (cli->cl_cache != NULL) {
3028 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
3029 spin_lock(&cli->cl_cache->ccc_lru_lock);
3030 list_del_init(&cli->cl_lru_osc);
3031 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3032 cli->cl_lru_left = NULL;
3033 cl_cache_decref(cli->cl_cache);
3034 cli->cl_cache = NULL;
3037 /* free memory of osc quota cache */
3038 osc_quota_cleanup(obd);
3040 rc = client_obd_cleanup(obd);
3046 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
3048 int rc = class_process_proc_param(PARAM_OSC, obd->obd_vars, lcfg, obd);
3049 return rc > 0 ? 0: rc;
3052 static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
3054 return osc_process_config_base(obd, buf);
3057 static struct obd_ops osc_obd_ops = {
3058 .o_owner = THIS_MODULE,
3059 .o_setup = osc_setup,
3060 .o_precleanup = osc_precleanup,
3061 .o_cleanup = osc_cleanup,
3062 .o_add_conn = client_import_add_conn,
3063 .o_del_conn = client_import_del_conn,
3064 .o_connect = client_connect_import,
3065 .o_reconnect = osc_reconnect,
3066 .o_disconnect = osc_disconnect,
3067 .o_statfs = osc_statfs,
3068 .o_statfs_async = osc_statfs_async,
3069 .o_create = osc_create,
3070 .o_destroy = osc_destroy,
3071 .o_getattr = osc_getattr,
3072 .o_getattr_async = osc_getattr_async,
3073 .o_setattr = osc_setattr,
3074 .o_setattr_async = osc_setattr_async,
3075 .o_iocontrol = osc_iocontrol,
3076 .o_get_info = osc_get_info,
3077 .o_set_info_async = osc_set_info_async,
3078 .o_import_event = osc_import_event,
3079 .o_process_config = osc_process_config,
3080 .o_quotactl = osc_quotactl,
3081 .o_quotacheck = osc_quotacheck,
3084 static int __init osc_init(void)
3086 bool enable_proc = true;
3087 struct obd_type *type;
3091 /* print an address of _any_ initialized kernel symbol from this
3092 * module, to allow debugging with gdb that doesn't support data
3093 * symbols from modules.*/
3094 CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
3096 rc = lu_kmem_init(osc_caches);
3100 type = class_search_type(LUSTRE_OSP_NAME);
3101 if (type != NULL && type->typ_procsym != NULL)
3102 enable_proc = false;
3104 rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
3105 LUSTRE_OSC_NAME, &osc_device_type);
3107 lu_kmem_fini(osc_caches);
3114 static void /*__exit*/ osc_exit(void)
3116 class_unregister_type(LUSTRE_OSC_NAME);
3117 lu_kmem_fini(osc_caches);
3120 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3121 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3122 MODULE_LICENSE("GPL");
3124 cfs_module(osc, LUSTRE_VERSION_STRING, osc_init, osc_exit);