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_setattr_args {
73 obd_enqueue_update_f sa_upcall;
77 struct osc_fsync_args {
78 struct obd_info *fa_oi;
79 obd_enqueue_update_f fa_upcall;
83 struct osc_enqueue_args {
84 struct obd_export *oa_exp;
88 osc_enqueue_upcall_f oa_upcall;
90 struct ost_lvb *oa_lvb;
91 struct lustre_handle oa_lockh;
92 unsigned int oa_agl:1;
95 static void osc_release_ppga(struct brw_page **ppga, obd_count count);
96 static int brw_interpret(const struct lu_env *env, struct ptlrpc_request *req,
99 static inline void osc_pack_capa(struct ptlrpc_request *req,
100 struct ost_body *body, void *capa)
102 struct obd_capa *oc = (struct obd_capa *)capa;
103 struct lustre_capa *c;
108 c = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
111 body->oa.o_valid |= OBD_MD_FLOSSCAPA;
112 DEBUG_CAPA(D_SEC, c, "pack");
115 void osc_pack_req_body(struct ptlrpc_request *req, struct obd_info *oinfo)
117 struct ost_body *body;
119 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
122 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
124 osc_pack_capa(req, body, oinfo->oi_capa);
127 void osc_set_capa_size(struct ptlrpc_request *req,
128 const struct req_msg_field *field,
132 req_capsule_set_size(&req->rq_pill, field, RCL_CLIENT, 0);
134 /* it is already calculated as sizeof struct obd_capa */
138 int osc_getattr_interpret(const struct lu_env *env,
139 struct ptlrpc_request *req,
140 struct osc_async_args *aa, int rc)
142 struct ost_body *body;
148 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
150 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
151 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
152 aa->aa_oi->oi_oa, &body->oa);
154 /* This should really be sent by the OST */
155 aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE;
156 aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
158 CDEBUG(D_INFO, "can't unpack ost_body\n");
160 aa->aa_oi->oi_oa->o_valid = 0;
163 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
167 static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
168 struct ptlrpc_request_set *set)
170 struct ptlrpc_request *req;
171 struct osc_async_args *aa;
175 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
179 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
180 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
182 ptlrpc_request_free(req);
186 osc_pack_req_body(req, oinfo);
188 ptlrpc_request_set_replen(req);
189 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret;
191 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
192 aa = ptlrpc_req_async_args(req);
195 ptlrpc_set_add_req(set, req);
199 static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
200 struct obd_info *oinfo)
202 struct ptlrpc_request *req;
203 struct ost_body *body;
207 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
211 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
212 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
214 ptlrpc_request_free(req);
218 osc_pack_req_body(req, oinfo);
220 ptlrpc_request_set_replen(req);
222 rc = ptlrpc_queue_wait(req);
226 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
228 GOTO(out, rc = -EPROTO);
230 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
231 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
234 oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd);
235 oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
239 ptlrpc_req_finished(req);
243 static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
244 struct obd_info *oinfo, struct obd_trans_info *oti)
246 struct ptlrpc_request *req;
247 struct ost_body *body;
251 LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP);
253 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
257 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
258 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
260 ptlrpc_request_free(req);
264 osc_pack_req_body(req, oinfo);
266 ptlrpc_request_set_replen(req);
268 rc = ptlrpc_queue_wait(req);
272 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
274 GOTO(out, rc = -EPROTO);
276 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
281 ptlrpc_req_finished(req);
285 static int osc_setattr_interpret(const struct lu_env *env,
286 struct ptlrpc_request *req,
287 struct osc_setattr_args *sa, int rc)
289 struct ost_body *body;
295 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
297 GOTO(out, rc = -EPROTO);
299 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
302 rc = sa->sa_upcall(sa->sa_cookie, rc);
306 int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo,
307 struct obd_trans_info *oti,
308 obd_enqueue_update_f upcall, void *cookie,
309 struct ptlrpc_request_set *rqset)
311 struct ptlrpc_request *req;
312 struct osc_setattr_args *sa;
316 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
320 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
321 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
323 ptlrpc_request_free(req);
327 if (oti && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
328 oinfo->oi_oa->o_lcookie = *oti->oti_logcookies;
330 osc_pack_req_body(req, oinfo);
332 ptlrpc_request_set_replen(req);
334 /* do mds to ost setattr asynchronously */
336 /* Do not wait for response. */
337 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
339 req->rq_interpret_reply =
340 (ptlrpc_interpterer_t)osc_setattr_interpret;
342 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
343 sa = ptlrpc_req_async_args(req);
344 sa->sa_oa = oinfo->oi_oa;
345 sa->sa_upcall = upcall;
346 sa->sa_cookie = cookie;
348 if (rqset == PTLRPCD_SET)
349 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
351 ptlrpc_set_add_req(rqset, req);
357 static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
358 struct obd_trans_info *oti,
359 struct ptlrpc_request_set *rqset)
361 return osc_setattr_async_base(exp, oinfo, oti,
362 oinfo->oi_cb_up, oinfo, rqset);
365 static int osc_create(const struct lu_env *env, struct obd_export *exp,
366 struct obdo *oa, struct obd_trans_info *oti)
368 struct ptlrpc_request *req;
369 struct ost_body *body;
374 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
375 LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
377 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
379 GOTO(out, rc = -ENOMEM);
381 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
383 ptlrpc_request_free(req);
387 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
390 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
392 ptlrpc_request_set_replen(req);
394 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
395 oa->o_flags == OBD_FL_DELORPHAN) {
397 "delorphan from OST integration");
398 /* Don't resend the delorphan req */
399 req->rq_no_resend = req->rq_no_delay = 1;
402 rc = ptlrpc_queue_wait(req);
406 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
408 GOTO(out_req, rc = -EPROTO);
410 CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
411 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
413 oa->o_blksize = cli_brw_size(exp->exp_obd);
414 oa->o_valid |= OBD_MD_FLBLKSZ;
417 if (oa->o_valid & OBD_MD_FLCOOKIE) {
418 if (oti->oti_logcookies == NULL)
419 oti->oti_logcookies = &oti->oti_onecookie;
421 *oti->oti_logcookies = oa->o_lcookie;
425 CDEBUG(D_HA, "transno: "LPD64"\n",
426 lustre_msg_get_transno(req->rq_repmsg));
428 ptlrpc_req_finished(req);
433 int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo,
434 obd_enqueue_update_f upcall, void *cookie,
435 struct ptlrpc_request_set *rqset)
437 struct ptlrpc_request *req;
438 struct osc_setattr_args *sa;
439 struct ost_body *body;
443 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH);
447 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
448 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
450 ptlrpc_request_free(req);
453 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
454 ptlrpc_at_set_req_timeout(req);
456 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
458 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
460 osc_pack_capa(req, body, oinfo->oi_capa);
462 ptlrpc_request_set_replen(req);
464 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
465 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
466 sa = ptlrpc_req_async_args(req);
467 sa->sa_oa = oinfo->oi_oa;
468 sa->sa_upcall = upcall;
469 sa->sa_cookie = cookie;
470 if (rqset == PTLRPCD_SET)
471 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
473 ptlrpc_set_add_req(rqset, req);
478 static int osc_sync_interpret(const struct lu_env *env,
479 struct ptlrpc_request *req,
482 struct osc_fsync_args *fa = arg;
483 struct ost_body *body;
489 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
491 CERROR ("can't unpack ost_body\n");
492 GOTO(out, rc = -EPROTO);
495 *fa->fa_oi->oi_oa = body->oa;
497 rc = fa->fa_upcall(fa->fa_cookie, rc);
501 int osc_sync_base(struct obd_export *exp, struct obd_info *oinfo,
502 obd_enqueue_update_f upcall, void *cookie,
503 struct ptlrpc_request_set *rqset)
505 struct ptlrpc_request *req;
506 struct ost_body *body;
507 struct osc_fsync_args *fa;
511 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
515 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
516 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
518 ptlrpc_request_free(req);
522 /* overload the size and blocks fields in the oa with start/end */
523 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
525 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
527 osc_pack_capa(req, body, oinfo->oi_capa);
529 ptlrpc_request_set_replen(req);
530 req->rq_interpret_reply = osc_sync_interpret;
532 CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
533 fa = ptlrpc_req_async_args(req);
535 fa->fa_upcall = upcall;
536 fa->fa_cookie = cookie;
538 if (rqset == PTLRPCD_SET)
539 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
541 ptlrpc_set_add_req(rqset, req);
546 /* Find and cancel locally locks matched by @mode in the resource found by
547 * @objid. Found locks are added into @cancel list. Returns the amount of
548 * locks added to @cancels list. */
549 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
550 struct list_head *cancels,
551 ldlm_mode_t mode, __u64 lock_flags)
553 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
554 struct ldlm_res_id res_id;
555 struct ldlm_resource *res;
559 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
560 * export) but disabled through procfs (flag in NS).
562 * This distinguishes from a case when ELC is not supported originally,
563 * when we still want to cancel locks in advance and just cancel them
564 * locally, without sending any RPC. */
565 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
568 ostid_build_res_name(&oa->o_oi, &res_id);
569 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
573 LDLM_RESOURCE_ADDREF(res);
574 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
575 lock_flags, 0, NULL);
576 LDLM_RESOURCE_DELREF(res);
577 ldlm_resource_putref(res);
581 static int osc_destroy_interpret(const struct lu_env *env,
582 struct ptlrpc_request *req, void *data,
585 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
587 atomic_dec(&cli->cl_destroy_in_flight);
588 wake_up(&cli->cl_destroy_waitq);
592 static int osc_can_send_destroy(struct client_obd *cli)
594 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
595 cli->cl_max_rpcs_in_flight) {
596 /* The destroy request can be sent */
599 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
600 cli->cl_max_rpcs_in_flight) {
602 * The counter has been modified between the two atomic
605 wake_up(&cli->cl_destroy_waitq);
610 /* Destroy requests can be async always on the client, and we don't even really
611 * care about the return code since the client cannot do anything at all about
613 * When the MDS is unlinking a filename, it saves the file objects into a
614 * recovery llog, and these object records are cancelled when the OST reports
615 * they were destroyed and sync'd to disk (i.e. transaction committed).
616 * If the client dies, or the OST is down when the object should be destroyed,
617 * the records are not cancelled, and when the OST reconnects to the MDS next,
618 * it will retrieve the llog unlink logs and then sends the log cancellation
619 * cookies to the MDS after committing destroy transactions. */
620 static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
621 struct obdo *oa, struct obd_trans_info *oti)
623 struct client_obd *cli = &exp->exp_obd->u.cli;
624 struct ptlrpc_request *req;
625 struct ost_body *body;
626 struct list_head cancels = LIST_HEAD_INIT(cancels);
631 CDEBUG(D_INFO, "oa NULL\n");
635 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
636 LDLM_FL_DISCARD_DATA);
638 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
640 ldlm_lock_list_put(&cancels, l_bl_ast, count);
644 osc_set_capa_size(req, &RMF_CAPA1, NULL);
645 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
648 ptlrpc_request_free(req);
652 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
653 ptlrpc_at_set_req_timeout(req);
655 if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
656 oa->o_lcookie = *oti->oti_logcookies;
657 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
659 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
661 ptlrpc_request_set_replen(req);
663 /* If osc_destory is for destroying the unlink orphan,
664 * sent from MDT to OST, which should not be blocked here,
665 * because the process might be triggered by ptlrpcd, and
666 * it is not good to block ptlrpcd thread (b=16006)*/
667 if (!(oa->o_flags & OBD_FL_DELORPHAN)) {
668 req->rq_interpret_reply = osc_destroy_interpret;
669 if (!osc_can_send_destroy(cli)) {
670 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP,
674 * Wait until the number of on-going destroy RPCs drops
675 * under max_rpc_in_flight
677 l_wait_event_exclusive(cli->cl_destroy_waitq,
678 osc_can_send_destroy(cli), &lwi);
682 /* Do not wait for response */
683 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
687 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
690 obd_flag bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT;
692 LASSERT(!(oa->o_valid & bits));
695 spin_lock(&cli->cl_loi_list_lock);
696 oa->o_dirty = cli->cl_dirty_pages << PAGE_CACHE_SHIFT;
697 if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
698 cli->cl_dirty_max_pages)) {
699 CERROR("dirty %lu - %lu > dirty_max %lu\n",
700 cli->cl_dirty_pages, cli->cl_dirty_transit,
701 cli->cl_dirty_max_pages);
703 } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
704 atomic_long_read(&obd_dirty_transit_pages) >
705 (obd_max_dirty_pages + 1))) {
706 /* The atomic_read() allowing the atomic_inc() are
707 * not covered by a lock thus they may safely race and trip
708 * this CERROR() unless we add in a small fudge factor (+1). */
709 CERROR("%s: dirty %ld - %ld > system dirty_max %lu\n",
710 cli->cl_import->imp_obd->obd_name,
711 atomic_long_read(&obd_dirty_pages),
712 atomic_long_read(&obd_dirty_transit_pages),
713 obd_max_dirty_pages);
715 } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
717 CERROR("dirty %lu - dirty_max %lu too big???\n",
718 cli->cl_dirty_pages, cli->cl_dirty_max_pages);
721 unsigned long max_in_flight = (cli->cl_max_pages_per_rpc <<
723 (cli->cl_max_rpcs_in_flight + 1);
724 oa->o_undirty = max(cli->cl_dirty_max_pages << PAGE_CACHE_SHIFT,
727 oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
728 oa->o_dropped = cli->cl_lost_grant;
729 cli->cl_lost_grant = 0;
730 spin_unlock(&cli->cl_loi_list_lock);
731 CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
732 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
736 void osc_update_next_shrink(struct client_obd *cli)
738 cli->cl_next_shrink_grant =
739 cfs_time_shift(cli->cl_grant_shrink_interval);
740 CDEBUG(D_CACHE, "next time %ld to shrink grant \n",
741 cli->cl_next_shrink_grant);
744 static void __osc_update_grant(struct client_obd *cli, obd_size grant)
746 spin_lock(&cli->cl_loi_list_lock);
747 cli->cl_avail_grant += grant;
748 spin_unlock(&cli->cl_loi_list_lock);
751 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
753 if (body->oa.o_valid & OBD_MD_FLGRANT) {
754 CDEBUG(D_CACHE, "got "LPU64" extra grant\n", body->oa.o_grant);
755 __osc_update_grant(cli, body->oa.o_grant);
759 static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
760 obd_count keylen, void *key, obd_count vallen,
761 void *val, struct ptlrpc_request_set *set);
763 static int osc_shrink_grant_interpret(const struct lu_env *env,
764 struct ptlrpc_request *req,
767 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
768 struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
769 struct ost_body *body;
772 __osc_update_grant(cli, oa->o_grant);
776 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
778 osc_update_grant(cli, body);
784 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
786 spin_lock(&cli->cl_loi_list_lock);
787 oa->o_grant = cli->cl_avail_grant / 4;
788 cli->cl_avail_grant -= oa->o_grant;
789 spin_unlock(&cli->cl_loi_list_lock);
790 if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
791 oa->o_valid |= OBD_MD_FLFLAGS;
794 oa->o_flags |= OBD_FL_SHRINK_GRANT;
795 osc_update_next_shrink(cli);
798 /* Shrink the current grant, either from some large amount to enough for a
799 * full set of in-flight RPCs, or if we have already shrunk to that limit
800 * then to enough for a single RPC. This avoids keeping more grant than
801 * needed, and avoids shrinking the grant piecemeal. */
802 static int osc_shrink_grant(struct client_obd *cli)
804 __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
805 (cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT);
807 spin_lock(&cli->cl_loi_list_lock);
808 if (cli->cl_avail_grant <= target_bytes)
809 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
810 spin_unlock(&cli->cl_loi_list_lock);
812 return osc_shrink_grant_to_target(cli, target_bytes);
815 int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
818 struct ost_body *body;
821 spin_lock(&cli->cl_loi_list_lock);
822 /* Don't shrink if we are already above or below the desired limit
823 * We don't want to shrink below a single RPC, as that will negatively
824 * impact block allocation and long-term performance. */
825 if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT)
826 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
828 if (target_bytes >= cli->cl_avail_grant) {
829 spin_unlock(&cli->cl_loi_list_lock);
832 spin_unlock(&cli->cl_loi_list_lock);
838 osc_announce_cached(cli, &body->oa, 0);
840 spin_lock(&cli->cl_loi_list_lock);
841 body->oa.o_grant = cli->cl_avail_grant - target_bytes;
842 cli->cl_avail_grant = target_bytes;
843 spin_unlock(&cli->cl_loi_list_lock);
844 if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
845 body->oa.o_valid |= OBD_MD_FLFLAGS;
846 body->oa.o_flags = 0;
848 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
849 osc_update_next_shrink(cli);
851 rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
852 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
853 sizeof(*body), body, NULL);
855 __osc_update_grant(cli, body->oa.o_grant);
860 static int osc_should_shrink_grant(struct client_obd *client)
862 cfs_time_t time = cfs_time_current();
863 cfs_time_t next_shrink = client->cl_next_shrink_grant;
865 if ((client->cl_import->imp_connect_data.ocd_connect_flags &
866 OBD_CONNECT_GRANT_SHRINK) == 0)
869 if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
870 /* Get the current RPC size directly, instead of going via:
871 * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
872 * Keep comment here so that it can be found by searching. */
873 int brw_size = client->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
875 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
876 client->cl_avail_grant > brw_size)
879 osc_update_next_shrink(client);
884 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
886 struct client_obd *client;
888 list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
889 if (osc_should_shrink_grant(client))
890 osc_shrink_grant(client);
895 static int osc_add_shrink_grant(struct client_obd *client)
899 rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
901 osc_grant_shrink_grant_cb, NULL,
902 &client->cl_grant_shrink_list);
904 CERROR("add grant client %s error %d\n",
905 client->cl_import->imp_obd->obd_name, rc);
908 CDEBUG(D_CACHE, "add grant client %s \n",
909 client->cl_import->imp_obd->obd_name);
910 osc_update_next_shrink(client);
914 static int osc_del_shrink_grant(struct client_obd *client)
916 return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
920 static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
923 * ocd_grant is the total grant amount we're expect to hold: if we've
924 * been evicted, it's the new avail_grant amount, cl_dirty_pages will
925 * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
928 * race is tolerable here: if we're evicted, but imp_state already
929 * left EVICTED state, then cl_dirty_pages must be 0 already.
931 spin_lock(&cli->cl_loi_list_lock);
932 if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
933 cli->cl_avail_grant = ocd->ocd_grant;
935 cli->cl_avail_grant = ocd->ocd_grant -
936 (cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
938 if (cli->cl_avail_grant < 0) {
939 CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n",
940 cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant,
941 ocd->ocd_grant, cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
942 /* workaround for servers which do not have the patch from
944 cli->cl_avail_grant = ocd->ocd_grant;
947 /* determine the appropriate chunk size used by osc_extent. */
948 cli->cl_chunkbits = max_t(int, PAGE_CACHE_SHIFT, ocd->ocd_blocksize);
949 spin_unlock(&cli->cl_loi_list_lock);
951 CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
952 "chunk bits: %d.\n", cli->cl_import->imp_obd->obd_name,
953 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits);
955 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
956 list_empty(&cli->cl_grant_shrink_list))
957 osc_add_shrink_grant(cli);
960 /* We assume that the reason this OSC got a short read is because it read
961 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
962 * via the LOV, and it _knows_ it's reading inside the file, it's just that
963 * this stripe never got written at or beyond this stripe offset yet. */
964 static void handle_short_read(int nob_read, obd_count page_count,
965 struct brw_page **pga)
970 /* skip bytes read OK */
971 while (nob_read > 0) {
972 LASSERT (page_count > 0);
974 if (pga[i]->count > nob_read) {
975 /* EOF inside this page */
976 ptr = kmap(pga[i]->pg) +
977 (pga[i]->off & ~CFS_PAGE_MASK);
978 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
985 nob_read -= pga[i]->count;
990 /* zero remaining pages */
991 while (page_count-- > 0) {
992 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
993 memset(ptr, 0, pga[i]->count);
999 static int check_write_rcs(struct ptlrpc_request *req,
1000 int requested_nob, int niocount,
1001 obd_count page_count, struct brw_page **pga)
1006 remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1007 sizeof(*remote_rcs) *
1009 if (remote_rcs == NULL) {
1010 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1014 /* return error if any niobuf was in error */
1015 for (i = 0; i < niocount; i++) {
1016 if ((int)remote_rcs[i] < 0)
1017 return(remote_rcs[i]);
1019 if (remote_rcs[i] != 0) {
1020 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1021 i, remote_rcs[i], req);
1026 if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1027 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1028 req->rq_bulk->bd_nob_transferred, requested_nob);
1035 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1037 if (p1->flag != p2->flag) {
1038 unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE |
1039 OBD_BRW_SYNC | OBD_BRW_ASYNC |
1040 OBD_BRW_NOQUOTA | OBD_BRW_SOFT_SYNC);
1042 /* warn if we try to combine flags that we don't know to be
1043 * safe to combine */
1044 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1045 CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1046 "report this at https://jira.hpdd.intel.com/\n",
1047 p1->flag, p2->flag);
1052 return (p1->off + p1->count == p2->off);
1055 static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1056 struct brw_page **pga, int opc,
1057 cksum_type_t cksum_type)
1061 struct cfs_crypto_hash_desc *hdesc;
1062 unsigned int bufsize;
1064 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1066 LASSERT(pg_count > 0);
1068 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1069 if (IS_ERR(hdesc)) {
1070 CERROR("Unable to initialize checksum hash %s\n",
1071 cfs_crypto_hash_name(cfs_alg));
1072 return PTR_ERR(hdesc);
1075 while (nob > 0 && pg_count > 0) {
1076 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1078 /* corrupt the data before we compute the checksum, to
1079 * simulate an OST->client data error */
1080 if (i == 0 && opc == OST_READ &&
1081 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1082 unsigned char *ptr = kmap(pga[i]->pg);
1083 int off = pga[i]->off & ~CFS_PAGE_MASK;
1085 memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1088 cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
1089 pga[i]->off & ~CFS_PAGE_MASK,
1091 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
1092 (int)(pga[i]->off & ~CFS_PAGE_MASK));
1094 nob -= pga[i]->count;
1099 bufsize = sizeof(cksum);
1100 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1102 /* For sending we only compute the wrong checksum instead
1103 * of corrupting the data so it is still correct on a redo */
1104 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1110 static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1111 struct lov_stripe_md *lsm, obd_count page_count,
1112 struct brw_page **pga,
1113 struct ptlrpc_request **reqp,
1114 struct obd_capa *ocapa, int reserve,
1117 struct ptlrpc_request *req;
1118 struct ptlrpc_bulk_desc *desc;
1119 struct ost_body *body;
1120 struct obd_ioobj *ioobj;
1121 struct niobuf_remote *niobuf;
1122 int niocount, i, requested_nob, opc, rc;
1123 struct osc_brw_async_args *aa;
1124 struct req_capsule *pill;
1125 struct brw_page *pg_prev;
1128 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1129 RETURN(-ENOMEM); /* Recoverable */
1130 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1131 RETURN(-EINVAL); /* Fatal */
1133 if ((cmd & OBD_BRW_WRITE) != 0) {
1135 req = ptlrpc_request_alloc_pool(cli->cl_import,
1136 cli->cl_import->imp_rq_pool,
1137 &RQF_OST_BRW_WRITE);
1140 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1145 for (niocount = i = 1; i < page_count; i++) {
1146 if (!can_merge_pages(pga[i - 1], pga[i]))
1150 pill = &req->rq_pill;
1151 req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1153 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1154 niocount * sizeof(*niobuf));
1155 osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1157 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1159 ptlrpc_request_free(req);
1162 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1163 ptlrpc_at_set_req_timeout(req);
1164 /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1166 req->rq_no_retry_einprogress = 1;
1168 desc = ptlrpc_prep_bulk_imp(req, page_count,
1169 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1170 opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK,
1174 GOTO(out, rc = -ENOMEM);
1175 /* NB request now owns desc and will free it when it gets freed */
1177 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1178 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1179 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1180 LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1182 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
1184 obdo_to_ioobj(oa, ioobj);
1185 ioobj->ioo_bufcnt = niocount;
1186 /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1187 * that might be send for this request. The actual number is decided
1188 * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1189 * "max - 1" for old client compatibility sending "0", and also so the
1190 * the actual maximum is a power-of-two number, not one less. LU-1431 */
1191 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1192 osc_pack_capa(req, body, ocapa);
1193 LASSERT(page_count > 0);
1195 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1196 struct brw_page *pg = pga[i];
1197 int poff = pg->off & ~CFS_PAGE_MASK;
1199 LASSERT(pg->count > 0);
1200 /* make sure there is no gap in the middle of page array */
1201 LASSERTF(page_count == 1 ||
1202 (ergo(i == 0, poff + pg->count == PAGE_CACHE_SIZE) &&
1203 ergo(i > 0 && i < page_count - 1,
1204 poff == 0 && pg->count == PAGE_CACHE_SIZE) &&
1205 ergo(i == page_count - 1, poff == 0)),
1206 "i: %d/%d pg: %p off: "LPU64", count: %u\n",
1207 i, page_count, pg, pg->off, pg->count);
1208 LASSERTF(i == 0 || pg->off > pg_prev->off,
1209 "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1210 " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1212 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1213 pg_prev->pg, page_private(pg_prev->pg),
1214 pg_prev->pg->index, pg_prev->off);
1215 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1216 (pg->flag & OBD_BRW_SRVLOCK));
1218 ptlrpc_prep_bulk_page_pin(desc, pg->pg, poff, pg->count);
1219 requested_nob += pg->count;
1221 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1223 niobuf->rnb_len += pg->count;
1225 niobuf->rnb_offset = pg->off;
1226 niobuf->rnb_len = pg->count;
1227 niobuf->rnb_flags = pg->flag;
1232 LASSERTF((void *)(niobuf - niocount) ==
1233 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1234 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1235 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1237 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1239 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1240 body->oa.o_valid |= OBD_MD_FLFLAGS;
1241 body->oa.o_flags = 0;
1243 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1246 if (osc_should_shrink_grant(cli))
1247 osc_shrink_grant_local(cli, &body->oa);
1249 /* size[REQ_REC_OFF] still sizeof (*body) */
1250 if (opc == OST_WRITE) {
1251 if (cli->cl_checksum &&
1252 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1253 /* store cl_cksum_type in a local variable since
1254 * it can be changed via lprocfs */
1255 cksum_type_t cksum_type = cli->cl_cksum_type;
1257 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1258 oa->o_flags &= OBD_FL_LOCAL_MASK;
1259 body->oa.o_flags = 0;
1261 body->oa.o_flags |= cksum_type_pack(cksum_type);
1262 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1263 body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1267 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1269 /* save this in 'oa', too, for later checking */
1270 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1271 oa->o_flags |= cksum_type_pack(cksum_type);
1273 /* clear out the checksum flag, in case this is a
1274 * resend but cl_checksum is no longer set. b=11238 */
1275 oa->o_valid &= ~OBD_MD_FLCKSUM;
1277 oa->o_cksum = body->oa.o_cksum;
1278 /* 1 RC per niobuf */
1279 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1280 sizeof(__u32) * niocount);
1282 if (cli->cl_checksum &&
1283 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1284 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1285 body->oa.o_flags = 0;
1286 body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1287 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1290 ptlrpc_request_set_replen(req);
1292 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1293 aa = ptlrpc_req_async_args(req);
1295 aa->aa_requested_nob = requested_nob;
1296 aa->aa_nio_count = niocount;
1297 aa->aa_page_count = page_count;
1301 INIT_LIST_HEAD(&aa->aa_oaps);
1302 if (ocapa && reserve)
1303 aa->aa_ocapa = capa_get(ocapa);
1306 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1307 CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
1308 req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
1309 niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
1313 ptlrpc_req_finished(req);
1317 static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1318 __u32 client_cksum, __u32 server_cksum, int nob,
1319 obd_count page_count, struct brw_page **pga,
1320 cksum_type_t client_cksum_type)
1324 cksum_type_t cksum_type;
1326 if (server_cksum == client_cksum) {
1327 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1331 cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1333 new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1336 if (cksum_type != client_cksum_type)
1337 msg = "the server did not use the checksum type specified in "
1338 "the original request - likely a protocol problem";
1339 else if (new_cksum == server_cksum)
1340 msg = "changed on the client after we checksummed it - "
1341 "likely false positive due to mmap IO (bug 11742)";
1342 else if (new_cksum == client_cksum)
1343 msg = "changed in transit before arrival at OST";
1345 msg = "changed in transit AND doesn't match the original - "
1346 "likely false positive due to mmap IO (bug 11742)";
1348 LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
1349 " object "DOSTID" extent ["LPU64"-"LPU64"]\n",
1350 msg, libcfs_nid2str(peer->nid),
1351 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1352 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1353 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1354 POSTID(&oa->o_oi), pga[0]->off,
1355 pga[page_count-1]->off + pga[page_count-1]->count - 1);
1356 CERROR("original client csum %x (type %x), server csum %x (type %x), "
1357 "client csum now %x\n", client_cksum, client_cksum_type,
1358 server_cksum, cksum_type, new_cksum);
1362 /* Note rc enters this function as number of bytes transferred */
1363 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1365 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1366 const lnet_process_id_t *peer =
1367 &req->rq_import->imp_connection->c_peer;
1368 struct client_obd *cli = aa->aa_cli;
1369 struct ost_body *body;
1370 u32 client_cksum = 0;
1373 if (rc < 0 && rc != -EDQUOT) {
1374 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1378 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1379 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1381 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1385 /* set/clear over quota flag for a uid/gid */
1386 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1387 body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1388 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1390 CDEBUG(D_QUOTA, "setdq for [%u %u] with valid "LPX64", flags %x\n",
1391 body->oa.o_uid, body->oa.o_gid, body->oa.o_valid,
1393 osc_quota_setdq(cli, qid, body->oa.o_valid, body->oa.o_flags);
1396 osc_update_grant(cli, body);
1401 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1402 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1404 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1406 CERROR("Unexpected +ve rc %d\n", rc);
1409 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1411 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1414 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1415 check_write_checksum(&body->oa, peer, client_cksum,
1416 body->oa.o_cksum, aa->aa_requested_nob,
1417 aa->aa_page_count, aa->aa_ppga,
1418 cksum_type_unpack(aa->aa_oa->o_flags)))
1421 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1422 aa->aa_page_count, aa->aa_ppga);
1426 /* The rest of this function executes only for OST_READs */
1428 /* if unwrap_bulk failed, return -EAGAIN to retry */
1429 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1431 GOTO(out, rc = -EAGAIN);
1433 if (rc > aa->aa_requested_nob) {
1434 CERROR("Unexpected rc %d (%d requested)\n", rc,
1435 aa->aa_requested_nob);
1439 if (rc != req->rq_bulk->bd_nob_transferred) {
1440 CERROR ("Unexpected rc %d (%d transferred)\n",
1441 rc, req->rq_bulk->bd_nob_transferred);
1445 if (rc < aa->aa_requested_nob)
1446 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1448 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1449 static int cksum_counter;
1450 u32 server_cksum = body->oa.o_cksum;
1453 cksum_type_t cksum_type;
1455 cksum_type = cksum_type_unpack(body->oa.o_valid &OBD_MD_FLFLAGS?
1456 body->oa.o_flags : 0);
1457 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1458 aa->aa_ppga, OST_READ,
1461 if (peer->nid != req->rq_bulk->bd_sender) {
1463 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1466 if (server_cksum != client_cksum) {
1467 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1468 "%s%s%s inode "DFID" object "DOSTID
1469 " extent ["LPU64"-"LPU64"]\n",
1470 req->rq_import->imp_obd->obd_name,
1471 libcfs_nid2str(peer->nid),
1473 body->oa.o_valid & OBD_MD_FLFID ?
1474 body->oa.o_parent_seq : (__u64)0,
1475 body->oa.o_valid & OBD_MD_FLFID ?
1476 body->oa.o_parent_oid : 0,
1477 body->oa.o_valid & OBD_MD_FLFID ?
1478 body->oa.o_parent_ver : 0,
1479 POSTID(&body->oa.o_oi),
1480 aa->aa_ppga[0]->off,
1481 aa->aa_ppga[aa->aa_page_count-1]->off +
1482 aa->aa_ppga[aa->aa_page_count-1]->count -
1484 CERROR("client %x, server %x, cksum_type %x\n",
1485 client_cksum, server_cksum, cksum_type);
1487 aa->aa_oa->o_cksum = client_cksum;
1491 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1494 } else if (unlikely(client_cksum)) {
1495 static int cksum_missed;
1498 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1499 CERROR("Checksum %u requested from %s but not sent\n",
1500 cksum_missed, libcfs_nid2str(peer->nid));
1506 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1507 aa->aa_oa, &body->oa);
1512 static int osc_brw_redo_request(struct ptlrpc_request *request,
1513 struct osc_brw_async_args *aa, int rc)
1515 struct ptlrpc_request *new_req;
1516 struct osc_brw_async_args *new_aa;
1517 struct osc_async_page *oap;
1520 DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1521 "redo for recoverable error %d", rc);
1523 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1524 OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1525 aa->aa_cli, aa->aa_oa,
1526 NULL /* lsm unused by osc currently */,
1527 aa->aa_page_count, aa->aa_ppga,
1528 &new_req, aa->aa_ocapa, 0, 1);
1532 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1533 if (oap->oap_request != NULL) {
1534 LASSERTF(request == oap->oap_request,
1535 "request %p != oap_request %p\n",
1536 request, oap->oap_request);
1537 if (oap->oap_interrupted) {
1538 ptlrpc_req_finished(new_req);
1543 /* New request takes over pga and oaps from old request.
1544 * Note that copying a list_head doesn't work, need to move it... */
1546 new_req->rq_interpret_reply = request->rq_interpret_reply;
1547 new_req->rq_async_args = request->rq_async_args;
1548 new_req->rq_commit_cb = request->rq_commit_cb;
1549 /* cap resend delay to the current request timeout, this is similar to
1550 * what ptlrpc does (see after_reply()) */
1551 if (aa->aa_resends > new_req->rq_timeout)
1552 new_req->rq_sent = cfs_time_current_sec() + new_req->rq_timeout;
1554 new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1555 new_req->rq_generation_set = 1;
1556 new_req->rq_import_generation = request->rq_import_generation;
1558 new_aa = ptlrpc_req_async_args(new_req);
1560 INIT_LIST_HEAD(&new_aa->aa_oaps);
1561 list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1562 INIT_LIST_HEAD(&new_aa->aa_exts);
1563 list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1564 new_aa->aa_resends = aa->aa_resends;
1566 list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1567 if (oap->oap_request) {
1568 ptlrpc_req_finished(oap->oap_request);
1569 oap->oap_request = ptlrpc_request_addref(new_req);
1573 new_aa->aa_ocapa = aa->aa_ocapa;
1574 aa->aa_ocapa = NULL;
1576 /* XXX: This code will run into problem if we're going to support
1577 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1578 * and wait for all of them to be finished. We should inherit request
1579 * set from old request. */
1580 ptlrpcd_add_req(new_req, PDL_POLICY_SAME, -1);
1582 DEBUG_REQ(D_INFO, new_req, "new request");
1587 * ugh, we want disk allocation on the target to happen in offset order. we'll
1588 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1589 * fine for our small page arrays and doesn't require allocation. its an
1590 * insertion sort that swaps elements that are strides apart, shrinking the
1591 * stride down until its '1' and the array is sorted.
1593 static void sort_brw_pages(struct brw_page **array, int num)
1596 struct brw_page *tmp;
1600 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1605 for (i = stride ; i < num ; i++) {
1608 while (j >= stride && array[j - stride]->off > tmp->off) {
1609 array[j] = array[j - stride];
1614 } while (stride > 1);
1617 static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1619 LASSERT(ppga != NULL);
1620 OBD_FREE(ppga, sizeof(*ppga) * count);
1623 static int brw_interpret(const struct lu_env *env,
1624 struct ptlrpc_request *req, void *data, int rc)
1626 struct osc_brw_async_args *aa = data;
1627 struct osc_extent *ext;
1628 struct osc_extent *tmp;
1629 struct client_obd *cli = aa->aa_cli;
1632 rc = osc_brw_fini_request(req, rc);
1633 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
1634 /* When server return -EINPROGRESS, client should always retry
1635 * regardless of the number of times the bulk was resent already. */
1636 if (osc_recoverable_error(rc)) {
1637 if (req->rq_import_generation !=
1638 req->rq_import->imp_generation) {
1639 CDEBUG(D_HA, "%s: resend cross eviction for object: "
1640 ""DOSTID", rc = %d.\n",
1641 req->rq_import->imp_obd->obd_name,
1642 POSTID(&aa->aa_oa->o_oi), rc);
1643 } else if (rc == -EINPROGRESS ||
1644 client_should_resend(aa->aa_resends, aa->aa_cli)) {
1645 rc = osc_brw_redo_request(req, aa, rc);
1647 CERROR("%s: too many resent retries for object: "
1648 ""LPU64":"LPU64", rc = %d.\n",
1649 req->rq_import->imp_obd->obd_name,
1650 POSTID(&aa->aa_oa->o_oi), rc);
1655 else if (rc == -EAGAIN || rc == -EINPROGRESS)
1660 capa_put(aa->aa_ocapa);
1661 aa->aa_ocapa = NULL;
1665 struct obdo *oa = aa->aa_oa;
1666 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1667 unsigned long valid = 0;
1668 struct cl_object *obj;
1669 struct osc_async_page *last;
1671 last = brw_page2oap(aa->aa_ppga[aa->aa_page_count - 1]);
1672 obj = osc2cl(last->oap_obj);
1674 cl_object_attr_lock(obj);
1675 if (oa->o_valid & OBD_MD_FLBLOCKS) {
1676 attr->cat_blocks = oa->o_blocks;
1677 valid |= CAT_BLOCKS;
1679 if (oa->o_valid & OBD_MD_FLMTIME) {
1680 attr->cat_mtime = oa->o_mtime;
1683 if (oa->o_valid & OBD_MD_FLATIME) {
1684 attr->cat_atime = oa->o_atime;
1687 if (oa->o_valid & OBD_MD_FLCTIME) {
1688 attr->cat_ctime = oa->o_ctime;
1692 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1693 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1694 loff_t last_off = last->oap_count + last->oap_obj_off +
1697 /* Change file size if this is an out of quota or
1698 * direct IO write and it extends the file size */
1699 if (loi->loi_lvb.lvb_size < last_off) {
1700 attr->cat_size = last_off;
1703 /* Extend KMS if it's not a lockless write */
1704 if (loi->loi_kms < last_off &&
1705 oap2osc_page(last)->ops_srvlock == 0) {
1706 attr->cat_kms = last_off;
1712 cl_object_attr_update(env, obj, attr, valid);
1713 cl_object_attr_unlock(obj);
1715 OBDO_FREE(aa->aa_oa);
1717 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
1718 osc_inc_unstable_pages(req);
1720 list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
1721 list_del_init(&ext->oe_link);
1722 osc_extent_finish(env, ext, 1, rc);
1724 LASSERT(list_empty(&aa->aa_exts));
1725 LASSERT(list_empty(&aa->aa_oaps));
1727 cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc :
1728 req->rq_bulk->bd_nob_transferred);
1729 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
1730 ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred);
1732 spin_lock(&cli->cl_loi_list_lock);
1733 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
1734 * is called so we know whether to go to sync BRWs or wait for more
1735 * RPCs to complete */
1736 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
1737 cli->cl_w_in_flight--;
1739 cli->cl_r_in_flight--;
1740 osc_wake_cache_waiters(cli);
1741 spin_unlock(&cli->cl_loi_list_lock);
1743 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
1747 static void brw_commit(struct ptlrpc_request *req)
1749 /* If osc_inc_unstable_pages (via osc_extent_finish) races with
1750 * this called via the rq_commit_cb, I need to ensure
1751 * osc_dec_unstable_pages is still called. Otherwise unstable
1752 * pages may be leaked. */
1753 spin_lock(&req->rq_lock);
1754 if (likely(req->rq_unstable)) {
1755 req->rq_unstable = 0;
1756 spin_unlock(&req->rq_lock);
1758 osc_dec_unstable_pages(req);
1760 req->rq_committed = 1;
1761 spin_unlock(&req->rq_lock);
1766 * Build an RPC by the list of extent @ext_list. The caller must ensure
1767 * that the total pages in this list are NOT over max pages per RPC.
1768 * Extents in the list must be in OES_RPC state.
1770 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
1771 struct list_head *ext_list, int cmd, pdl_policy_t pol)
1773 struct ptlrpc_request *req = NULL;
1774 struct osc_extent *ext;
1775 struct brw_page **pga = NULL;
1776 struct osc_brw_async_args *aa = NULL;
1777 struct obdo *oa = NULL;
1778 struct osc_async_page *oap;
1779 struct osc_async_page *tmp;
1780 struct cl_req *clerq = NULL;
1781 enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE :
1783 struct cl_req_attr *crattr = NULL;
1784 obd_off starting_offset = OBD_OBJECT_EOF;
1785 obd_off ending_offset = 0;
1789 bool soft_sync = false;
1792 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1793 struct ost_body *body;
1795 LASSERT(!list_empty(ext_list));
1797 /* add pages into rpc_list to build BRW rpc */
1798 list_for_each_entry(ext, ext_list, oe_link) {
1799 LASSERT(ext->oe_state == OES_RPC);
1800 mem_tight |= ext->oe_memalloc;
1801 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
1803 list_add_tail(&oap->oap_rpc_item, &rpc_list);
1804 if (starting_offset > oap->oap_obj_off)
1805 starting_offset = oap->oap_obj_off;
1807 LASSERT(oap->oap_page_off == 0);
1808 if (ending_offset < oap->oap_obj_off + oap->oap_count)
1809 ending_offset = oap->oap_obj_off +
1812 LASSERT(oap->oap_page_off + oap->oap_count ==
1817 soft_sync = osc_over_unstable_soft_limit(cli);
1819 mpflag = cfs_memory_pressure_get_and_set();
1821 OBD_ALLOC(crattr, sizeof(*crattr));
1823 GOTO(out, rc = -ENOMEM);
1825 OBD_ALLOC(pga, sizeof(*pga) * page_count);
1827 GOTO(out, rc = -ENOMEM);
1831 GOTO(out, rc = -ENOMEM);
1834 list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
1835 struct cl_page *page = oap2cl_page(oap);
1836 if (clerq == NULL) {
1837 clerq = cl_req_alloc(env, page, crt,
1838 1 /* only 1-object rpcs for now */);
1840 GOTO(out, rc = PTR_ERR(clerq));
1843 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
1845 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
1846 pga[i] = &oap->oap_brw_page;
1847 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
1848 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
1849 pga[i]->pg, page_index(oap->oap_page), oap,
1852 cl_req_page_add(env, clerq, page);
1855 /* always get the data for the obdo for the rpc */
1856 LASSERT(clerq != NULL);
1857 crattr->cra_oa = oa;
1858 cl_req_attr_set(env, clerq, crattr, ~0ULL);
1860 rc = cl_req_prep(env, clerq);
1862 CERROR("cl_req_prep failed: %d\n", rc);
1866 sort_brw_pages(pga, page_count);
1867 rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
1868 pga, &req, crattr->cra_capa, 1, 0);
1870 CERROR("prep_req failed: %d\n", rc);
1874 req->rq_commit_cb = brw_commit;
1875 req->rq_interpret_reply = brw_interpret;
1878 req->rq_memalloc = 1;
1880 /* Need to update the timestamps after the request is built in case
1881 * we race with setattr (locally or in queue at OST). If OST gets
1882 * later setattr before earlier BRW (as determined by the request xid),
1883 * the OST will not use BRW timestamps. Sadly, there is no obvious
1884 * way to do this in a single call. bug 10150 */
1885 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1886 crattr->cra_oa = &body->oa;
1887 cl_req_attr_set(env, clerq, crattr,
1888 OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
1890 lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
1892 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1893 aa = ptlrpc_req_async_args(req);
1894 INIT_LIST_HEAD(&aa->aa_oaps);
1895 list_splice_init(&rpc_list, &aa->aa_oaps);
1896 INIT_LIST_HEAD(&aa->aa_exts);
1897 list_splice_init(ext_list, &aa->aa_exts);
1898 aa->aa_clerq = clerq;
1900 /* queued sync pages can be torn down while the pages
1901 * were between the pending list and the rpc */
1903 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1904 /* only one oap gets a request reference */
1907 if (oap->oap_interrupted && !req->rq_intr) {
1908 CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
1910 ptlrpc_mark_interrupted(req);
1914 tmp->oap_request = ptlrpc_request_addref(req);
1916 spin_lock(&cli->cl_loi_list_lock);
1917 starting_offset >>= PAGE_CACHE_SHIFT;
1918 if (cmd == OBD_BRW_READ) {
1919 cli->cl_r_in_flight++;
1920 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
1921 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
1922 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
1923 starting_offset + 1);
1925 cli->cl_w_in_flight++;
1926 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
1927 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
1928 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
1929 starting_offset + 1);
1931 spin_unlock(&cli->cl_loi_list_lock);
1933 DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
1934 page_count, aa, cli->cl_r_in_flight,
1935 cli->cl_w_in_flight);
1937 /* XXX: Maybe the caller can check the RPC bulk descriptor to
1938 * see which CPU/NUMA node the majority of pages were allocated
1939 * on, and try to assign the async RPC to the CPU core
1940 * (PDL_POLICY_PREFERRED) to reduce cross-CPU memory traffic.
1942 * But on the other hand, we expect that multiple ptlrpcd
1943 * threads and the initial write sponsor can run in parallel,
1944 * especially when data checksum is enabled, which is CPU-bound
1945 * operation and single ptlrpcd thread cannot process in time.
1946 * So more ptlrpcd threads sharing BRW load
1947 * (with PDL_POLICY_ROUND) seems better.
1949 ptlrpcd_add_req(req, pol, -1);
1955 cfs_memory_pressure_restore(mpflag);
1957 if (crattr != NULL) {
1958 capa_put(crattr->cra_capa);
1959 OBD_FREE(crattr, sizeof(*crattr));
1963 LASSERT(req == NULL);
1968 OBD_FREE(pga, sizeof(*pga) * page_count);
1969 /* this should happen rarely and is pretty bad, it makes the
1970 * pending list not follow the dirty order */
1971 while (!list_empty(ext_list)) {
1972 ext = list_entry(ext_list->next, struct osc_extent,
1974 list_del_init(&ext->oe_link);
1975 osc_extent_finish(env, ext, 0, rc);
1977 if (clerq && !IS_ERR(clerq))
1978 cl_req_completion(env, clerq, rc);
1983 static int osc_set_lock_data_with_check(struct ldlm_lock *lock,
1984 struct ldlm_enqueue_info *einfo)
1986 void *data = einfo->ei_cbdata;
1989 LASSERT(lock != NULL);
1990 LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl);
1991 LASSERT(lock->l_resource->lr_type == einfo->ei_type);
1992 LASSERT(lock->l_completion_ast == einfo->ei_cb_cp);
1993 LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
1995 lock_res_and_lock(lock);
1997 if (lock->l_ast_data == NULL)
1998 lock->l_ast_data = data;
1999 if (lock->l_ast_data == data)
2002 unlock_res_and_lock(lock);
2007 static int osc_set_data_with_check(struct lustre_handle *lockh,
2008 struct ldlm_enqueue_info *einfo)
2010 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2014 set = osc_set_lock_data_with_check(lock, einfo);
2015 LDLM_LOCK_PUT(lock);
2017 CERROR("lockh %p, data %p - client evicted?\n",
2018 lockh, einfo->ei_cbdata);
2022 static int osc_enqueue_fini(struct ptlrpc_request *req,
2023 osc_enqueue_upcall_f upcall, void *cookie,
2024 struct lustre_handle *lockh, ldlm_mode_t mode,
2025 __u64 *flags, int agl, int errcode)
2027 bool intent = *flags & LDLM_FL_HAS_INTENT;
2031 /* The request was created before ldlm_cli_enqueue call. */
2032 if (intent && errcode == ELDLM_LOCK_ABORTED) {
2033 struct ldlm_reply *rep;
2035 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
2036 LASSERT(rep != NULL);
2038 rep->lock_policy_res1 =
2039 ptlrpc_status_ntoh(rep->lock_policy_res1);
2040 if (rep->lock_policy_res1)
2041 errcode = rep->lock_policy_res1;
2043 *flags |= LDLM_FL_LVB_READY;
2044 } else if (errcode == ELDLM_OK) {
2045 *flags |= LDLM_FL_LVB_READY;
2048 /* Call the update callback. */
2049 rc = (*upcall)(cookie, lockh, errcode);
2051 /* release the reference taken in ldlm_cli_enqueue() */
2052 if (errcode == ELDLM_LOCK_MATCHED)
2054 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
2055 ldlm_lock_decref(lockh, mode);
2060 static int osc_enqueue_interpret(const struct lu_env *env,
2061 struct ptlrpc_request *req,
2062 struct osc_enqueue_args *aa, int rc)
2064 struct ldlm_lock *lock;
2065 struct lustre_handle *lockh = &aa->oa_lockh;
2066 ldlm_mode_t mode = aa->oa_mode;
2067 struct ost_lvb *lvb = aa->oa_lvb;
2068 __u32 lvb_len = sizeof(*lvb);
2073 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2075 lock = ldlm_handle2lock(lockh);
2076 LASSERTF(lock != NULL,
2077 "lockh "LPX64", req %p, aa %p - client evicted?\n",
2078 lockh->cookie, req, aa);
2080 /* Take an additional reference so that a blocking AST that
2081 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2082 * to arrive after an upcall has been executed by
2083 * osc_enqueue_fini(). */
2084 ldlm_lock_addref(lockh, mode);
2086 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
2087 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
2089 /* Let CP AST to grant the lock first. */
2090 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2093 LASSERT(aa->oa_lvb == NULL);
2094 LASSERT(aa->oa_flags == NULL);
2095 aa->oa_flags = &flags;
2098 /* Complete obtaining the lock procedure. */
2099 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
2100 aa->oa_mode, aa->oa_flags, lvb, lvb_len,
2102 /* Complete osc stuff. */
2103 rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
2104 aa->oa_flags, aa->oa_agl, rc);
2106 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2108 ldlm_lock_decref(lockh, mode);
2109 LDLM_LOCK_PUT(lock);
2113 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2115 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2116 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2117 * other synchronous requests, however keeping some locks and trying to obtain
2118 * others may take a considerable amount of time in a case of ost failure; and
2119 * when other sync requests do not get released lock from a client, the client
2120 * is evicted from the cluster -- such scenarious make the life difficult, so
2121 * release locks just after they are obtained. */
2122 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2123 __u64 *flags, ldlm_policy_data_t *policy,
2124 struct ost_lvb *lvb, int kms_valid,
2125 osc_enqueue_upcall_f upcall, void *cookie,
2126 struct ldlm_enqueue_info *einfo,
2127 struct ptlrpc_request_set *rqset, int async, int agl)
2129 struct obd_device *obd = exp->exp_obd;
2130 struct lustre_handle lockh = { 0 };
2131 struct ptlrpc_request *req = NULL;
2132 int intent = *flags & LDLM_FL_HAS_INTENT;
2133 __u64 match_lvb = agl ? 0 : LDLM_FL_LVB_READY;
2138 /* Filesystem lock extents are extended to page boundaries so that
2139 * dealing with the page cache is a little smoother. */
2140 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2141 policy->l_extent.end |= ~CFS_PAGE_MASK;
2144 * kms is not valid when either object is completely fresh (so that no
2145 * locks are cached), or object was evicted. In the latter case cached
2146 * lock cannot be used, because it would prime inode state with
2147 * potentially stale LVB.
2152 /* Next, search for already existing extent locks that will cover us */
2153 /* If we're trying to read, we also search for an existing PW lock. The
2154 * VFS and page cache already protect us locally, so lots of readers/
2155 * writers can share a single PW lock.
2157 * There are problems with conversion deadlocks, so instead of
2158 * converting a read lock to a write lock, we'll just enqueue a new
2161 * At some point we should cancel the read lock instead of making them
2162 * send us a blocking callback, but there are problems with canceling
2163 * locks out from other users right now, too. */
2164 mode = einfo->ei_mode;
2165 if (einfo->ei_mode == LCK_PR)
2167 mode = ldlm_lock_match(obd->obd_namespace, *flags | match_lvb, res_id,
2168 einfo->ei_type, policy, mode, &lockh, 0);
2170 struct ldlm_lock *matched;
2172 if (*flags & LDLM_FL_TEST_LOCK)
2175 matched = ldlm_handle2lock(&lockh);
2177 /* AGL enqueues DLM locks speculatively. Therefore if
2178 * it already exists a DLM lock, it wll just inform the
2179 * caller to cancel the AGL process for this stripe. */
2180 ldlm_lock_decref(&lockh, mode);
2181 LDLM_LOCK_PUT(matched);
2183 } else if (osc_set_lock_data_with_check(matched, einfo)) {
2184 *flags |= LDLM_FL_LVB_READY;
2186 /* We already have a lock, and it's referenced. */
2187 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
2189 ldlm_lock_decref(&lockh, mode);
2190 LDLM_LOCK_PUT(matched);
2193 ldlm_lock_decref(&lockh, mode);
2194 LDLM_LOCK_PUT(matched);
2199 if (*flags & LDLM_FL_TEST_LOCK)
2203 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2204 &RQF_LDLM_ENQUEUE_LVB);
2208 rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
2210 ptlrpc_request_free(req);
2214 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
2216 ptlrpc_request_set_replen(req);
2219 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2220 *flags &= ~LDLM_FL_BLOCK_GRANTED;
2222 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2223 sizeof(*lvb), LVB_T_OST, &lockh, async);
2226 struct osc_enqueue_args *aa;
2227 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2228 aa = ptlrpc_req_async_args(req);
2230 aa->oa_mode = einfo->ei_mode;
2231 aa->oa_type = einfo->ei_type;
2232 lustre_handle_copy(&aa->oa_lockh, &lockh);
2233 aa->oa_upcall = upcall;
2234 aa->oa_cookie = cookie;
2237 aa->oa_flags = flags;
2240 /* AGL is essentially to enqueue an DLM lock
2241 * in advance, so we don't care about the
2242 * result of AGL enqueue. */
2244 aa->oa_flags = NULL;
2247 req->rq_interpret_reply =
2248 (ptlrpc_interpterer_t)osc_enqueue_interpret;
2249 if (rqset == PTLRPCD_SET)
2250 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
2252 ptlrpc_set_add_req(rqset, req);
2253 } else if (intent) {
2254 ptlrpc_req_finished(req);
2259 rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
2262 ptlrpc_req_finished(req);
2267 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2268 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2269 __u64 *flags, void *data, struct lustre_handle *lockh,
2272 struct obd_device *obd = exp->exp_obd;
2273 __u64 lflags = *flags;
2277 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
2280 /* Filesystem lock extents are extended to page boundaries so that
2281 * dealing with the page cache is a little smoother */
2282 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2283 policy->l_extent.end |= ~CFS_PAGE_MASK;
2285 /* Next, search for already existing extent locks that will cover us */
2286 /* If we're trying to read, we also search for an existing PW lock. The
2287 * VFS and page cache already protect us locally, so lots of readers/
2288 * writers can share a single PW lock. */
2292 rc = ldlm_lock_match(obd->obd_namespace, lflags,
2293 res_id, type, policy, rc, lockh, unref);
2296 if (!osc_set_data_with_check(lockh, data)) {
2297 if (!(lflags & LDLM_FL_TEST_LOCK))
2298 ldlm_lock_decref(lockh, rc);
2302 if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) {
2303 ldlm_lock_addref(lockh, LCK_PR);
2304 ldlm_lock_decref(lockh, LCK_PW);
2311 int osc_cancel_base(struct lustre_handle *lockh, __u32 mode)
2315 if (unlikely(mode == LCK_GROUP))
2316 ldlm_lock_decref_and_cancel(lockh, mode);
2318 ldlm_lock_decref(lockh, mode);
2323 static int osc_statfs_interpret(const struct lu_env *env,
2324 struct ptlrpc_request *req,
2325 struct osc_async_args *aa, int rc)
2327 struct obd_statfs *msfs;
2331 /* The request has in fact never been sent
2332 * due to issues at a higher level (LOV).
2333 * Exit immediately since the caller is
2334 * aware of the problem and takes care
2335 * of the clean up */
2338 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2339 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2345 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2347 GOTO(out, rc = -EPROTO);
2350 *aa->aa_oi->oi_osfs = *msfs;
2352 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
2356 static int osc_statfs_async(struct obd_export *exp,
2357 struct obd_info *oinfo, __u64 max_age,
2358 struct ptlrpc_request_set *rqset)
2360 struct obd_device *obd = class_exp2obd(exp);
2361 struct ptlrpc_request *req;
2362 struct osc_async_args *aa;
2366 /* We could possibly pass max_age in the request (as an absolute
2367 * timestamp or a "seconds.usec ago") so the target can avoid doing
2368 * extra calls into the filesystem if that isn't necessary (e.g.
2369 * during mount that would help a bit). Having relative timestamps
2370 * is not so great if request processing is slow, while absolute
2371 * timestamps are not ideal because they need time synchronization. */
2372 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2376 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2378 ptlrpc_request_free(req);
2381 ptlrpc_request_set_replen(req);
2382 req->rq_request_portal = OST_CREATE_PORTAL;
2383 ptlrpc_at_set_req_timeout(req);
2385 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2386 /* procfs requests not want stat in wait for avoid deadlock */
2387 req->rq_no_resend = 1;
2388 req->rq_no_delay = 1;
2391 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
2392 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
2393 aa = ptlrpc_req_async_args(req);
2396 ptlrpc_set_add_req(rqset, req);
2400 static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2401 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
2403 struct obd_device *obd = class_exp2obd(exp);
2404 struct obd_statfs *msfs;
2405 struct ptlrpc_request *req;
2406 struct obd_import *imp = NULL;
2410 /*Since the request might also come from lprocfs, so we need
2411 *sync this with client_disconnect_export Bug15684*/
2412 down_read(&obd->u.cli.cl_sem);
2413 if (obd->u.cli.cl_import)
2414 imp = class_import_get(obd->u.cli.cl_import);
2415 up_read(&obd->u.cli.cl_sem);
2419 /* We could possibly pass max_age in the request (as an absolute
2420 * timestamp or a "seconds.usec ago") so the target can avoid doing
2421 * extra calls into the filesystem if that isn't necessary (e.g.
2422 * during mount that would help a bit). Having relative timestamps
2423 * is not so great if request processing is slow, while absolute
2424 * timestamps are not ideal because they need time synchronization. */
2425 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2427 class_import_put(imp);
2432 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2434 ptlrpc_request_free(req);
2437 ptlrpc_request_set_replen(req);
2438 req->rq_request_portal = OST_CREATE_PORTAL;
2439 ptlrpc_at_set_req_timeout(req);
2441 if (flags & OBD_STATFS_NODELAY) {
2442 /* procfs requests not want stat in wait for avoid deadlock */
2443 req->rq_no_resend = 1;
2444 req->rq_no_delay = 1;
2447 rc = ptlrpc_queue_wait(req);
2451 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2453 GOTO(out, rc = -EPROTO);
2460 ptlrpc_req_finished(req);
2464 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2465 void *karg, void *uarg)
2467 struct obd_device *obd = exp->exp_obd;
2468 struct obd_ioctl_data *data = karg;
2472 if (!try_module_get(THIS_MODULE)) {
2473 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2474 module_name(THIS_MODULE));
2478 case OBD_IOC_CLIENT_RECOVER:
2479 err = ptlrpc_recover_import(obd->u.cli.cl_import,
2480 data->ioc_inlbuf1, 0);
2484 case IOC_OSC_SET_ACTIVE:
2485 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
2488 case OBD_IOC_POLL_QUOTACHECK:
2489 err = osc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2491 case OBD_IOC_PING_TARGET:
2492 err = ptlrpc_obd_ping(obd);
2495 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
2496 cmd, current_comm());
2497 GOTO(out, err = -ENOTTY);
2500 module_put(THIS_MODULE);
2504 static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
2505 obd_count keylen, void *key, obd_count vallen,
2506 void *val, struct ptlrpc_request_set *set)
2508 struct ptlrpc_request *req;
2509 struct obd_device *obd = exp->exp_obd;
2510 struct obd_import *imp = class_exp2cliimp(exp);
2515 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
2517 if (KEY_IS(KEY_CHECKSUM)) {
2518 if (vallen != sizeof(int))
2520 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
2524 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2525 sptlrpc_conf_client_adapt(obd);
2529 if (KEY_IS(KEY_FLUSH_CTX)) {
2530 sptlrpc_import_flush_my_ctx(imp);
2534 if (KEY_IS(KEY_CACHE_SET)) {
2535 struct client_obd *cli = &obd->u.cli;
2537 LASSERT(cli->cl_cache == NULL); /* only once */
2538 cli->cl_cache = (struct cl_client_cache *)val;
2539 cl_cache_incref(cli->cl_cache);
2540 cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
2542 /* add this osc into entity list */
2543 LASSERT(list_empty(&cli->cl_lru_osc));
2544 spin_lock(&cli->cl_cache->ccc_lru_lock);
2545 list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
2546 spin_unlock(&cli->cl_cache->ccc_lru_lock);
2551 if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
2552 struct client_obd *cli = &obd->u.cli;
2553 long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
2554 long target = *(long *)val;
2556 nr = osc_lru_shrink(env, cli, min(nr, target), true);
2561 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
2564 /* We pass all other commands directly to OST. Since nobody calls osc
2565 methods directly and everybody is supposed to go through LOV, we
2566 assume lov checked invalid values for us.
2567 The only recognised values so far are evict_by_nid and mds_conn.
2568 Even if something bad goes through, we'd get a -EINVAL from OST
2571 req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
2572 &RQF_OST_SET_GRANT_INFO :
2577 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2578 RCL_CLIENT, keylen);
2579 if (!KEY_IS(KEY_GRANT_SHRINK))
2580 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
2581 RCL_CLIENT, vallen);
2582 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
2584 ptlrpc_request_free(req);
2588 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
2589 memcpy(tmp, key, keylen);
2590 tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
2593 memcpy(tmp, val, vallen);
2595 if (KEY_IS(KEY_GRANT_SHRINK)) {
2596 struct osc_grant_args *aa;
2599 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2600 aa = ptlrpc_req_async_args(req);
2603 ptlrpc_req_finished(req);
2606 *oa = ((struct ost_body *)val)->oa;
2608 req->rq_interpret_reply = osc_shrink_grant_interpret;
2611 ptlrpc_request_set_replen(req);
2612 if (!KEY_IS(KEY_GRANT_SHRINK)) {
2613 LASSERT(set != NULL);
2614 ptlrpc_set_add_req(set, req);
2615 ptlrpc_check_set(NULL, set);
2617 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
2622 static int osc_reconnect(const struct lu_env *env,
2623 struct obd_export *exp, struct obd_device *obd,
2624 struct obd_uuid *cluuid,
2625 struct obd_connect_data *data,
2628 struct client_obd *cli = &obd->u.cli;
2630 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
2633 spin_lock(&cli->cl_loi_list_lock);
2634 data->ocd_grant = (cli->cl_avail_grant +
2635 (cli->cl_dirty_pages << PAGE_CACHE_SHIFT)) ?:
2636 2 * cli_brw_size(obd);
2637 lost_grant = cli->cl_lost_grant;
2638 cli->cl_lost_grant = 0;
2639 spin_unlock(&cli->cl_loi_list_lock);
2641 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
2642 " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
2643 data->ocd_version, data->ocd_grant, lost_grant);
2649 static int osc_disconnect(struct obd_export *exp)
2651 struct obd_device *obd = class_exp2obd(exp);
2654 rc = client_disconnect_export(exp);
2656 * Initially we put del_shrink_grant before disconnect_export, but it
2657 * causes the following problem if setup (connect) and cleanup
2658 * (disconnect) are tangled together.
2659 * connect p1 disconnect p2
2660 * ptlrpc_connect_import
2661 * ............... class_manual_cleanup
2664 * ptlrpc_connect_interrupt
2666 * add this client to shrink list
2668 * Bang! pinger trigger the shrink.
2669 * So the osc should be disconnected from the shrink list, after we
2670 * are sure the import has been destroyed. BUG18662
2672 if (obd->u.cli.cl_import == NULL)
2673 osc_del_shrink_grant(&obd->u.cli);
2677 static int osc_import_event(struct obd_device *obd,
2678 struct obd_import *imp,
2679 enum obd_import_event event)
2681 struct client_obd *cli;
2685 LASSERT(imp->imp_obd == obd);
2688 case IMP_EVENT_DISCON: {
2690 spin_lock(&cli->cl_loi_list_lock);
2691 cli->cl_avail_grant = 0;
2692 cli->cl_lost_grant = 0;
2693 spin_unlock(&cli->cl_loi_list_lock);
2696 case IMP_EVENT_INACTIVE: {
2697 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2700 case IMP_EVENT_INVALIDATE: {
2701 struct ldlm_namespace *ns = obd->obd_namespace;
2705 env = cl_env_get(&refcheck);
2709 /* all pages go to failing rpcs due to the invalid
2711 osc_io_unplug(env, cli, NULL, PDL_POLICY_ROUND);
2713 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2714 cl_env_put(env, &refcheck);
2719 case IMP_EVENT_ACTIVE: {
2720 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2723 case IMP_EVENT_OCD: {
2724 struct obd_connect_data *ocd = &imp->imp_connect_data;
2726 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
2727 osc_init_grant(&obd->u.cli, ocd);
2730 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
2731 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
2733 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2736 case IMP_EVENT_DEACTIVATE: {
2737 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE, NULL);
2740 case IMP_EVENT_ACTIVATE: {
2741 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE, NULL);
2745 CERROR("Unknown import event %d\n", event);
2752 * Determine whether the lock can be canceled before replaying the lock
2753 * during recovery, see bug16774 for detailed information.
2755 * \retval zero the lock can't be canceled
2756 * \retval other ok to cancel
2758 static int osc_cancel_weight(struct ldlm_lock *lock)
2761 * Cancel all unused and granted extent lock.
2763 if (lock->l_resource->lr_type == LDLM_EXTENT &&
2764 lock->l_granted_mode == lock->l_req_mode &&
2765 osc_ldlm_weigh_ast(lock) == 0)
2771 static int brw_queue_work(const struct lu_env *env, void *data)
2773 struct client_obd *cli = data;
2775 CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
2777 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
2781 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
2783 struct client_obd *cli = &obd->u.cli;
2784 struct obd_type *type;
2789 rc = ptlrpcd_addref();
2793 rc = client_obd_setup(obd, lcfg);
2795 GOTO(out_ptlrpcd, rc);
2797 handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
2798 if (IS_ERR(handler))
2799 GOTO(out_client_setup, rc = PTR_ERR(handler));
2800 cli->cl_writeback_work = handler;
2802 handler = ptlrpcd_alloc_work(cli->cl_import, lru_queue_work, cli);
2803 if (IS_ERR(handler))
2804 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
2805 cli->cl_lru_work = handler;
2807 rc = osc_quota_setup(obd);
2809 GOTO(out_ptlrpcd_work, rc);
2811 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
2813 #ifdef CONFIG_PROC_FS
2814 obd->obd_vars = lprocfs_osc_obd_vars;
2816 /* If this is true then both client (osc) and server (osp) are on the
2817 * same node. The osp layer if loaded first will register the osc proc
2818 * directory. In that case this obd_device will be attached its proc
2819 * tree to type->typ_procsym instead of obd->obd_type->typ_procroot. */
2820 type = class_search_type(LUSTRE_OSP_NAME);
2821 if (type && type->typ_procsym) {
2822 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
2824 obd->obd_vars, obd);
2825 if (IS_ERR(obd->obd_proc_entry)) {
2826 rc = PTR_ERR(obd->obd_proc_entry);
2827 CERROR("error %d setting up lprocfs for %s\n", rc,
2829 obd->obd_proc_entry = NULL;
2832 rc = lprocfs_obd_setup(obd);
2835 /* If the basic OSC proc tree construction succeeded then
2836 * lets do the rest. */
2838 lproc_osc_attach_seqstat(obd);
2839 sptlrpc_lprocfs_cliobd_attach(obd);
2840 ptlrpc_lprocfs_register_obd(obd);
2843 /* We need to allocate a few requests more, because
2844 * brw_interpret tries to create new requests before freeing
2845 * previous ones, Ideally we want to have 2x max_rpcs_in_flight
2846 * reserved, but I'm afraid that might be too much wasted RAM
2847 * in fact, so 2 is just my guess and still should work. */
2848 cli->cl_import->imp_rq_pool =
2849 ptlrpc_init_rq_pool(cli->cl_max_rpcs_in_flight + 2,
2851 ptlrpc_add_rqs_to_pool);
2853 INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
2854 ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
2858 if (cli->cl_writeback_work != NULL) {
2859 ptlrpcd_destroy_work(cli->cl_writeback_work);
2860 cli->cl_writeback_work = NULL;
2862 if (cli->cl_lru_work != NULL) {
2863 ptlrpcd_destroy_work(cli->cl_lru_work);
2864 cli->cl_lru_work = NULL;
2867 client_obd_cleanup(obd);
2873 static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2879 case OBD_CLEANUP_EARLY: {
2880 struct obd_import *imp;
2881 imp = obd->u.cli.cl_import;
2882 CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name);
2883 /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */
2884 ptlrpc_deactivate_import(imp);
2885 spin_lock(&imp->imp_lock);
2886 imp->imp_pingable = 0;
2887 spin_unlock(&imp->imp_lock);
2890 case OBD_CLEANUP_EXPORTS: {
2891 struct client_obd *cli = &obd->u.cli;
2893 * for echo client, export may be on zombie list, wait for
2894 * zombie thread to cull it, because cli.cl_import will be
2895 * cleared in client_disconnect_export():
2896 * class_export_destroy() -> obd_cleanup() ->
2897 * echo_device_free() -> echo_client_cleanup() ->
2898 * obd_disconnect() -> osc_disconnect() ->
2899 * client_disconnect_export()
2901 obd_zombie_barrier();
2902 if (cli->cl_writeback_work) {
2903 ptlrpcd_destroy_work(cli->cl_writeback_work);
2904 cli->cl_writeback_work = NULL;
2906 if (cli->cl_lru_work) {
2907 ptlrpcd_destroy_work(cli->cl_lru_work);
2908 cli->cl_lru_work = NULL;
2910 obd_cleanup_client_import(obd);
2911 ptlrpc_lprocfs_unregister_obd(obd);
2912 lprocfs_obd_cleanup(obd);
2919 int osc_cleanup(struct obd_device *obd)
2921 struct client_obd *cli = &obd->u.cli;
2927 if (cli->cl_cache != NULL) {
2928 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
2929 spin_lock(&cli->cl_cache->ccc_lru_lock);
2930 list_del_init(&cli->cl_lru_osc);
2931 spin_unlock(&cli->cl_cache->ccc_lru_lock);
2932 cli->cl_lru_left = NULL;
2933 cl_cache_decref(cli->cl_cache);
2934 cli->cl_cache = NULL;
2937 /* free memory of osc quota cache */
2938 osc_quota_cleanup(obd);
2940 rc = client_obd_cleanup(obd);
2946 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
2948 int rc = class_process_proc_param(PARAM_OSC, obd->obd_vars, lcfg, obd);
2949 return rc > 0 ? 0: rc;
2952 static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
2954 return osc_process_config_base(obd, buf);
2957 static struct obd_ops osc_obd_ops = {
2958 .o_owner = THIS_MODULE,
2959 .o_setup = osc_setup,
2960 .o_precleanup = osc_precleanup,
2961 .o_cleanup = osc_cleanup,
2962 .o_add_conn = client_import_add_conn,
2963 .o_del_conn = client_import_del_conn,
2964 .o_connect = client_connect_import,
2965 .o_reconnect = osc_reconnect,
2966 .o_disconnect = osc_disconnect,
2967 .o_statfs = osc_statfs,
2968 .o_statfs_async = osc_statfs_async,
2969 .o_create = osc_create,
2970 .o_destroy = osc_destroy,
2971 .o_getattr = osc_getattr,
2972 .o_getattr_async = osc_getattr_async,
2973 .o_setattr = osc_setattr,
2974 .o_setattr_async = osc_setattr_async,
2975 .o_iocontrol = osc_iocontrol,
2976 .o_set_info_async = osc_set_info_async,
2977 .o_import_event = osc_import_event,
2978 .o_process_config = osc_process_config,
2979 .o_quotactl = osc_quotactl,
2980 .o_quotacheck = osc_quotacheck,
2983 static int __init osc_init(void)
2985 bool enable_proc = true;
2986 struct obd_type *type;
2990 /* print an address of _any_ initialized kernel symbol from this
2991 * module, to allow debugging with gdb that doesn't support data
2992 * symbols from modules.*/
2993 CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
2995 rc = lu_kmem_init(osc_caches);
2999 type = class_search_type(LUSTRE_OSP_NAME);
3000 if (type != NULL && type->typ_procsym != NULL)
3001 enable_proc = false;
3003 rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
3004 LUSTRE_OSC_NAME, &osc_device_type);
3006 lu_kmem_fini(osc_caches);
3013 static void /*__exit*/ osc_exit(void)
3015 class_unregister_type(LUSTRE_OSC_NAME);
3016 lu_kmem_fini(osc_caches);
3019 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3020 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3021 MODULE_LICENSE("GPL");
3023 cfs_module(osc, LUSTRE_VERSION_STRING, osc_init, osc_exit);
git://git.whamcloud.com / fs / lustre-release.git / blob