1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 # define EXPORT_SYMTAB
40 #define DEBUG_SUBSYSTEM S_OSC
42 #include <libcfs/libcfs.h>
45 # include <liblustre.h>
48 #include <lustre_dlm.h>
49 #include <lustre_net.h>
50 #include <lustre/lustre_user.h>
51 #include <obd_cksum.h>
59 #include <lustre_ha.h>
60 #include <lprocfs_status.h>
61 #include <lustre_log.h>
62 #include <lustre_debug.h>
63 #include <lustre_param.h>
64 #include "osc_internal.h"
66 static quota_interface_t *quota_interface = NULL;
67 extern quota_interface_t osc_quota_interface;
69 static void osc_release_ppga(struct brw_page **ppga, obd_count count);
70 static int brw_interpret(const struct lu_env *env,
71 struct ptlrpc_request *req, void *data, int rc);
72 int osc_cleanup(struct obd_device *obd);
74 /* Pack OSC object metadata for disk storage (LE byte order). */
75 static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
76 struct lov_stripe_md *lsm)
81 lmm_size = sizeof(**lmmp);
86 OBD_FREE(*lmmp, lmm_size);
92 OBD_ALLOC(*lmmp, lmm_size);
98 LASSERT(lsm->lsm_object_id);
99 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
100 (*lmmp)->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
101 (*lmmp)->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
107 /* Unpack OSC object metadata from disk storage (LE byte order). */
108 static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
109 struct lov_mds_md *lmm, int lmm_bytes)
115 if (lmm_bytes < sizeof (*lmm)) {
116 CERROR("lov_mds_md too small: %d, need %d\n",
117 lmm_bytes, (int)sizeof(*lmm));
120 /* XXX LOV_MAGIC etc check? */
122 if (lmm->lmm_object_id == 0) {
123 CERROR("lov_mds_md: zero lmm_object_id\n");
128 lsm_size = lov_stripe_md_size(1);
132 if (*lsmp != NULL && lmm == NULL) {
133 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
134 OBD_FREE(*lsmp, lsm_size);
140 OBD_ALLOC(*lsmp, lsm_size);
143 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
144 if ((*lsmp)->lsm_oinfo[0] == NULL) {
145 OBD_FREE(*lsmp, lsm_size);
148 loi_init((*lsmp)->lsm_oinfo[0]);
152 /* XXX zero *lsmp? */
153 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
154 (*lsmp)->lsm_object_gr = le64_to_cpu (lmm->lmm_object_gr);
155 LASSERT((*lsmp)->lsm_object_id);
156 LASSERT_MDS_GROUP((*lsmp)->lsm_object_gr);
159 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
164 static inline void osc_pack_capa(struct ptlrpc_request *req,
165 struct ost_body *body, void *capa)
167 struct obd_capa *oc = (struct obd_capa *)capa;
168 struct lustre_capa *c;
173 c = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
176 body->oa.o_valid |= OBD_MD_FLOSSCAPA;
177 DEBUG_CAPA(D_SEC, c, "pack");
180 static inline void osc_pack_req_body(struct ptlrpc_request *req,
181 struct obd_info *oinfo)
183 struct ost_body *body;
185 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
188 lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
189 osc_pack_capa(req, body, oinfo->oi_capa);
192 static inline void osc_set_capa_size(struct ptlrpc_request *req,
193 const struct req_msg_field *field,
197 req_capsule_set_size(&req->rq_pill, field, RCL_CLIENT, 0);
199 /* it is already calculated as sizeof struct obd_capa */
203 static int osc_getattr_interpret(const struct lu_env *env,
204 struct ptlrpc_request *req,
205 struct osc_async_args *aa, int rc)
207 struct ost_body *body;
213 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
214 lustre_swab_ost_body);
216 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
217 lustre_get_wire_obdo(aa->aa_oi->oi_oa, &body->oa);
219 /* This should really be sent by the OST */
220 aa->aa_oi->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
221 aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
223 CDEBUG(D_INFO, "can't unpack ost_body\n");
225 aa->aa_oi->oi_oa->o_valid = 0;
228 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
232 static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
233 struct ptlrpc_request_set *set)
235 struct ptlrpc_request *req;
236 struct osc_async_args *aa;
240 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
244 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
245 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
247 ptlrpc_request_free(req);
251 osc_pack_req_body(req, oinfo);
253 ptlrpc_request_set_replen(req);
254 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret;
256 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
257 aa = ptlrpc_req_async_args(req);
260 ptlrpc_set_add_req(set, req);
264 static int osc_getattr(struct obd_export *exp, struct obd_info *oinfo)
266 struct ptlrpc_request *req;
267 struct ost_body *body;
271 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
275 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
276 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
278 ptlrpc_request_free(req);
282 osc_pack_req_body(req, oinfo);
284 ptlrpc_request_set_replen(req);
286 rc = ptlrpc_queue_wait(req);
290 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
292 GOTO(out, rc = -EPROTO);
294 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
295 lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
297 /* This should really be sent by the OST */
298 oinfo->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
299 oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
303 ptlrpc_req_finished(req);
307 static int osc_setattr(struct obd_export *exp, struct obd_info *oinfo,
308 struct obd_trans_info *oti)
310 struct ptlrpc_request *req;
311 struct ost_body *body;
315 LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP);
317 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
321 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
322 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
324 ptlrpc_request_free(req);
328 osc_pack_req_body(req, oinfo);
330 ptlrpc_request_set_replen(req);
332 rc = ptlrpc_queue_wait(req);
336 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
338 GOTO(out, rc = -EPROTO);
340 lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
344 ptlrpc_req_finished(req);
348 static int osc_setattr_interpret(const struct lu_env *env,
349 struct ptlrpc_request *req,
350 struct osc_async_args *aa, int rc)
352 struct ost_body *body;
358 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
360 GOTO(out, rc = -EPROTO);
362 lustre_get_wire_obdo(aa->aa_oi->oi_oa, &body->oa);
364 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
368 static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
369 struct obd_trans_info *oti,
370 struct ptlrpc_request_set *rqset)
372 struct ptlrpc_request *req;
373 struct osc_async_args *aa;
377 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
381 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
382 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
384 ptlrpc_request_free(req);
388 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
389 oinfo->oi_oa->o_lcookie = *oti->oti_logcookies;
391 osc_pack_req_body(req, oinfo);
393 ptlrpc_request_set_replen(req);
395 /* do mds to ost setattr asynchronously */
397 /* Do not wait for response. */
398 ptlrpcd_add_req(req, PSCOPE_OTHER);
400 req->rq_interpret_reply =
401 (ptlrpc_interpterer_t)osc_setattr_interpret;
403 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
404 aa = ptlrpc_req_async_args(req);
407 ptlrpc_set_add_req(rqset, req);
413 int osc_real_create(struct obd_export *exp, struct obdo *oa,
414 struct lov_stripe_md **ea, struct obd_trans_info *oti)
416 struct ptlrpc_request *req;
417 struct ost_body *body;
418 struct lov_stripe_md *lsm;
427 rc = obd_alloc_memmd(exp, &lsm);
432 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
434 GOTO(out, rc = -ENOMEM);
436 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
438 ptlrpc_request_free(req);
442 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
444 lustre_set_wire_obdo(&body->oa, oa);
446 ptlrpc_request_set_replen(req);
448 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
449 oa->o_flags == OBD_FL_DELORPHAN) {
451 "delorphan from OST integration");
452 /* Don't resend the delorphan req */
453 req->rq_no_resend = req->rq_no_delay = 1;
456 rc = ptlrpc_queue_wait(req);
460 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
462 GOTO(out_req, rc = -EPROTO);
464 lustre_get_wire_obdo(oa, &body->oa);
466 /* This should really be sent by the OST */
467 oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
468 oa->o_valid |= OBD_MD_FLBLKSZ;
470 /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
471 * have valid lsm_oinfo data structs, so don't go touching that.
472 * This needs to be fixed in a big way.
474 lsm->lsm_object_id = oa->o_id;
475 lsm->lsm_object_gr = oa->o_gr;
479 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
481 if (oa->o_valid & OBD_MD_FLCOOKIE) {
482 if (!oti->oti_logcookies)
483 oti_alloc_cookies(oti, 1);
484 *oti->oti_logcookies = oa->o_lcookie;
488 CDEBUG(D_HA, "transno: "LPD64"\n",
489 lustre_msg_get_transno(req->rq_repmsg));
491 ptlrpc_req_finished(req);
494 obd_free_memmd(exp, &lsm);
498 static int osc_punch_interpret(const struct lu_env *env,
499 struct ptlrpc_request *req,
500 struct osc_punch_args *aa, int rc)
502 struct ost_body *body;
508 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
510 GOTO(out, rc = -EPROTO);
512 lustre_get_wire_obdo(aa->pa_oa, &body->oa);
514 rc = aa->pa_upcall(aa->pa_cookie, rc);
518 int osc_punch_base(struct obd_export *exp, struct obdo *oa,
519 struct obd_capa *capa,
520 obd_enqueue_update_f upcall, void *cookie,
521 struct ptlrpc_request_set *rqset)
523 struct ptlrpc_request *req;
524 struct osc_punch_args *aa;
525 struct ost_body *body;
529 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH);
533 osc_set_capa_size(req, &RMF_CAPA1, capa);
534 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
536 ptlrpc_request_free(req);
539 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
540 ptlrpc_at_set_req_timeout(req);
542 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
544 lustre_set_wire_obdo(&body->oa, oa);
545 osc_pack_capa(req, body, capa);
547 ptlrpc_request_set_replen(req);
550 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_punch_interpret;
551 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
552 aa = ptlrpc_req_async_args(req);
554 aa->pa_upcall = upcall;
555 aa->pa_cookie = cookie;
556 if (rqset == PTLRPCD_SET)
557 ptlrpcd_add_req(req, PSCOPE_OTHER);
559 ptlrpc_set_add_req(rqset, req);
564 static int osc_punch(struct obd_export *exp, struct obd_info *oinfo,
565 struct obd_trans_info *oti,
566 struct ptlrpc_request_set *rqset)
568 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
569 oinfo->oi_oa->o_blocks = oinfo->oi_policy.l_extent.end;
570 oinfo->oi_oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
571 return osc_punch_base(exp, oinfo->oi_oa, oinfo->oi_capa,
572 oinfo->oi_cb_up, oinfo, rqset);
575 static int osc_sync(struct obd_export *exp, struct obdo *oa,
576 struct lov_stripe_md *md, obd_size start, obd_size end,
579 struct ptlrpc_request *req;
580 struct ost_body *body;
585 CDEBUG(D_INFO, "oa NULL\n");
589 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
593 osc_set_capa_size(req, &RMF_CAPA1, capa);
594 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
596 ptlrpc_request_free(req);
600 /* overload the size and blocks fields in the oa with start/end */
601 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
603 lustre_set_wire_obdo(&body->oa, oa);
604 body->oa.o_size = start;
605 body->oa.o_blocks = end;
606 body->oa.o_valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
607 osc_pack_capa(req, body, capa);
609 ptlrpc_request_set_replen(req);
611 rc = ptlrpc_queue_wait(req);
615 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
617 GOTO(out, rc = -EPROTO);
619 lustre_get_wire_obdo(oa, &body->oa);
623 ptlrpc_req_finished(req);
627 /* Find and cancel locally locks matched by @mode in the resource found by
628 * @objid. Found locks are added into @cancel list. Returns the amount of
629 * locks added to @cancels list. */
630 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
631 struct list_head *cancels, ldlm_mode_t mode,
634 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
635 struct ldlm_res_id res_id;
636 struct ldlm_resource *res;
640 osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
641 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
645 LDLM_RESOURCE_ADDREF(res);
646 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
647 lock_flags, 0, NULL);
648 LDLM_RESOURCE_DELREF(res);
649 ldlm_resource_putref(res);
653 static int osc_destroy_interpret(const struct lu_env *env,
654 struct ptlrpc_request *req, void *data,
657 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
659 atomic_dec(&cli->cl_destroy_in_flight);
660 cfs_waitq_signal(&cli->cl_destroy_waitq);
664 static int osc_can_send_destroy(struct client_obd *cli)
666 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
667 cli->cl_max_rpcs_in_flight) {
668 /* The destroy request can be sent */
671 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
672 cli->cl_max_rpcs_in_flight) {
674 * The counter has been modified between the two atomic
677 cfs_waitq_signal(&cli->cl_destroy_waitq);
682 /* Destroy requests can be async always on the client, and we don't even really
683 * care about the return code since the client cannot do anything at all about
685 * When the MDS is unlinking a filename, it saves the file objects into a
686 * recovery llog, and these object records are cancelled when the OST reports
687 * they were destroyed and sync'd to disk (i.e. transaction committed).
688 * If the client dies, or the OST is down when the object should be destroyed,
689 * the records are not cancelled, and when the OST reconnects to the MDS next,
690 * it will retrieve the llog unlink logs and then sends the log cancellation
691 * cookies to the MDS after committing destroy transactions. */
692 static int osc_destroy(struct obd_export *exp, struct obdo *oa,
693 struct lov_stripe_md *ea, struct obd_trans_info *oti,
694 struct obd_export *md_export, void *capa)
696 struct client_obd *cli = &exp->exp_obd->u.cli;
697 struct ptlrpc_request *req;
698 struct ost_body *body;
699 CFS_LIST_HEAD(cancels);
704 CDEBUG(D_INFO, "oa NULL\n");
708 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
709 LDLM_FL_DISCARD_DATA);
711 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
713 ldlm_lock_list_put(&cancels, l_bl_ast, count);
717 osc_set_capa_size(req, &RMF_CAPA1, (struct obd_capa *)capa);
718 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
721 ptlrpc_request_free(req);
725 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
726 ptlrpc_at_set_req_timeout(req);
728 if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
729 oa->o_lcookie = *oti->oti_logcookies;
730 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
732 lustre_set_wire_obdo(&body->oa, oa);
734 osc_pack_capa(req, body, (struct obd_capa *)capa);
735 ptlrpc_request_set_replen(req);
737 /* don't throttle destroy RPCs for the MDT */
738 if (!(cli->cl_import->imp_connect_flags_orig & OBD_CONNECT_MDS)) {
739 req->rq_interpret_reply = osc_destroy_interpret;
740 if (!osc_can_send_destroy(cli)) {
741 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP,
745 * Wait until the number of on-going destroy RPCs drops
746 * under max_rpc_in_flight
748 l_wait_event_exclusive(cli->cl_destroy_waitq,
749 osc_can_send_destroy(cli), &lwi);
753 /* Do not wait for response */
754 ptlrpcd_add_req(req, PSCOPE_OTHER);
758 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
761 obd_flag bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT;
763 LASSERT(!(oa->o_valid & bits));
766 client_obd_list_lock(&cli->cl_loi_list_lock);
767 oa->o_dirty = cli->cl_dirty;
768 if (cli->cl_dirty - cli->cl_dirty_transit > cli->cl_dirty_max) {
769 CERROR("dirty %lu - %lu > dirty_max %lu\n",
770 cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
772 } else if (atomic_read(&obd_dirty_pages) -
773 atomic_read(&obd_dirty_transit_pages) > obd_max_dirty_pages){
774 CERROR("dirty %d - %d > system dirty_max %d\n",
775 atomic_read(&obd_dirty_pages),
776 atomic_read(&obd_dirty_transit_pages),
777 obd_max_dirty_pages);
779 } else if (cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff) {
780 CERROR("dirty %lu - dirty_max %lu too big???\n",
781 cli->cl_dirty, cli->cl_dirty_max);
784 long max_in_flight = (cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT)*
785 (cli->cl_max_rpcs_in_flight + 1);
786 oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
788 oa->o_grant = cli->cl_avail_grant;
789 oa->o_dropped = cli->cl_lost_grant;
790 cli->cl_lost_grant = 0;
791 client_obd_list_unlock(&cli->cl_loi_list_lock);
792 CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
793 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
797 static void osc_update_next_shrink(struct client_obd *cli)
799 cli->cl_next_shrink_grant =
800 cfs_time_shift(cli->cl_grant_shrink_interval);
801 CDEBUG(D_CACHE, "next time %ld to shrink grant \n",
802 cli->cl_next_shrink_grant);
805 /* caller must hold loi_list_lock */
806 static void osc_consume_write_grant(struct client_obd *cli,
807 struct brw_page *pga)
809 LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
810 LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
811 atomic_inc(&obd_dirty_pages);
812 cli->cl_dirty += CFS_PAGE_SIZE;
813 cli->cl_avail_grant -= CFS_PAGE_SIZE;
814 pga->flag |= OBD_BRW_FROM_GRANT;
815 CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
816 CFS_PAGE_SIZE, pga, pga->pg);
817 LASSERT(cli->cl_avail_grant >= 0);
818 osc_update_next_shrink(cli);
821 /* the companion to osc_consume_write_grant, called when a brw has completed.
822 * must be called with the loi lock held. */
823 static void osc_release_write_grant(struct client_obd *cli,
824 struct brw_page *pga, int sent)
826 int blocksize = cli->cl_import->imp_obd->obd_osfs.os_bsize ? : 4096;
829 LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
830 if (!(pga->flag & OBD_BRW_FROM_GRANT)) {
835 pga->flag &= ~OBD_BRW_FROM_GRANT;
836 atomic_dec(&obd_dirty_pages);
837 cli->cl_dirty -= CFS_PAGE_SIZE;
838 if (pga->flag & OBD_BRW_NOCACHE) {
839 pga->flag &= ~OBD_BRW_NOCACHE;
840 atomic_dec(&obd_dirty_transit_pages);
841 cli->cl_dirty_transit -= CFS_PAGE_SIZE;
844 cli->cl_lost_grant += CFS_PAGE_SIZE;
845 CDEBUG(D_CACHE, "lost grant: %lu avail grant: %lu dirty: %lu\n",
846 cli->cl_lost_grant, cli->cl_avail_grant, cli->cl_dirty);
847 } else if (CFS_PAGE_SIZE != blocksize && pga->count != CFS_PAGE_SIZE) {
848 /* For short writes we shouldn't count parts of pages that
849 * span a whole block on the OST side, or our accounting goes
850 * wrong. Should match the code in filter_grant_check. */
851 int offset = pga->off & ~CFS_PAGE_MASK;
852 int count = pga->count + (offset & (blocksize - 1));
853 int end = (offset + pga->count) & (blocksize - 1);
855 count += blocksize - end;
857 cli->cl_lost_grant += CFS_PAGE_SIZE - count;
858 CDEBUG(D_CACHE, "lost %lu grant: %lu avail: %lu dirty: %lu\n",
859 CFS_PAGE_SIZE - count, cli->cl_lost_grant,
860 cli->cl_avail_grant, cli->cl_dirty);
866 static unsigned long rpcs_in_flight(struct client_obd *cli)
868 return cli->cl_r_in_flight + cli->cl_w_in_flight;
871 /* caller must hold loi_list_lock */
872 void osc_wake_cache_waiters(struct client_obd *cli)
874 struct list_head *l, *tmp;
875 struct osc_cache_waiter *ocw;
878 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
879 /* if we can't dirty more, we must wait until some is written */
880 if ((cli->cl_dirty + CFS_PAGE_SIZE > cli->cl_dirty_max) ||
881 (atomic_read(&obd_dirty_pages) + 1 > obd_max_dirty_pages)) {
882 CDEBUG(D_CACHE, "no dirty room: dirty: %ld "
883 "osc max %ld, sys max %d\n", cli->cl_dirty,
884 cli->cl_dirty_max, obd_max_dirty_pages);
888 /* if still dirty cache but no grant wait for pending RPCs that
889 * may yet return us some grant before doing sync writes */
890 if (cli->cl_w_in_flight && cli->cl_avail_grant < CFS_PAGE_SIZE) {
891 CDEBUG(D_CACHE, "%u BRW writes in flight, no grant\n",
892 cli->cl_w_in_flight);
896 ocw = list_entry(l, struct osc_cache_waiter, ocw_entry);
897 list_del_init(&ocw->ocw_entry);
898 if (cli->cl_avail_grant < CFS_PAGE_SIZE) {
899 /* no more RPCs in flight to return grant, do sync IO */
900 ocw->ocw_rc = -EDQUOT;
901 CDEBUG(D_INODE, "wake oap %p for sync\n", ocw->ocw_oap);
903 osc_consume_write_grant(cli,
904 &ocw->ocw_oap->oap_brw_page);
907 cfs_waitq_signal(&ocw->ocw_waitq);
913 static void __osc_update_grant(struct client_obd *cli, obd_size grant)
915 client_obd_list_lock(&cli->cl_loi_list_lock);
916 cli->cl_avail_grant += grant;
917 client_obd_list_unlock(&cli->cl_loi_list_lock);
920 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
922 if (body->oa.o_valid & OBD_MD_FLGRANT) {
923 CDEBUG(D_CACHE, "got "LPU64" extra grant\n", body->oa.o_grant);
924 __osc_update_grant(cli, body->oa.o_grant);
928 static int osc_set_info_async(struct obd_export *exp, obd_count keylen,
929 void *key, obd_count vallen, void *val,
930 struct ptlrpc_request_set *set);
932 static int osc_shrink_grant_interpret(const struct lu_env *env,
933 struct ptlrpc_request *req,
936 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
937 struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
938 struct ost_body *body;
941 __osc_update_grant(cli, oa->o_grant);
945 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
947 osc_update_grant(cli, body);
953 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
955 client_obd_list_lock(&cli->cl_loi_list_lock);
956 oa->o_grant = cli->cl_avail_grant / 4;
957 cli->cl_avail_grant -= oa->o_grant;
958 client_obd_list_unlock(&cli->cl_loi_list_lock);
959 oa->o_flags |= OBD_FL_SHRINK_GRANT;
960 osc_update_next_shrink(cli);
963 /* Shrink the current grant, either from some large amount to enough for a
964 * full set of in-flight RPCs, or if we have already shrunk to that limit
965 * then to enough for a single RPC. This avoids keeping more grant than
966 * needed, and avoids shrinking the grant piecemeal. */
967 static int osc_shrink_grant(struct client_obd *cli)
969 long target = (cli->cl_max_rpcs_in_flight + 1) *
970 cli->cl_max_pages_per_rpc;
972 client_obd_list_lock(&cli->cl_loi_list_lock);
973 if (cli->cl_avail_grant <= target)
974 target = cli->cl_max_pages_per_rpc;
975 client_obd_list_unlock(&cli->cl_loi_list_lock);
977 return osc_shrink_grant_to_target(cli, target);
980 int osc_shrink_grant_to_target(struct client_obd *cli, long target)
983 struct ost_body *body;
986 client_obd_list_lock(&cli->cl_loi_list_lock);
987 /* Don't shrink if we are already above or below the desired limit
988 * We don't want to shrink below a single RPC, as that will negatively
989 * impact block allocation and long-term performance. */
990 if (target < cli->cl_max_pages_per_rpc)
991 target = cli->cl_max_pages_per_rpc;
993 if (target >= cli->cl_avail_grant) {
994 client_obd_list_unlock(&cli->cl_loi_list_lock);
997 client_obd_list_unlock(&cli->cl_loi_list_lock);
1003 osc_announce_cached(cli, &body->oa, 0);
1005 client_obd_list_lock(&cli->cl_loi_list_lock);
1006 body->oa.o_grant = cli->cl_avail_grant - target;
1007 cli->cl_avail_grant = target;
1008 client_obd_list_unlock(&cli->cl_loi_list_lock);
1009 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
1010 osc_update_next_shrink(cli);
1012 rc = osc_set_info_async(cli->cl_import->imp_obd->obd_self_export,
1013 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
1014 sizeof(*body), body, NULL);
1016 __osc_update_grant(cli, body->oa.o_grant);
1021 #define GRANT_SHRINK_LIMIT PTLRPC_MAX_BRW_SIZE
1022 static int osc_should_shrink_grant(struct client_obd *client)
1024 cfs_time_t time = cfs_time_current();
1025 cfs_time_t next_shrink = client->cl_next_shrink_grant;
1026 if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
1027 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
1028 client->cl_avail_grant > GRANT_SHRINK_LIMIT)
1031 osc_update_next_shrink(client);
1036 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
1038 struct client_obd *client;
1040 list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
1041 if (osc_should_shrink_grant(client))
1042 osc_shrink_grant(client);
1047 static int osc_add_shrink_grant(struct client_obd *client)
1051 rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
1053 osc_grant_shrink_grant_cb, NULL,
1054 &client->cl_grant_shrink_list);
1056 CERROR("add grant client %s error %d\n",
1057 client->cl_import->imp_obd->obd_name, rc);
1060 CDEBUG(D_CACHE, "add grant client %s \n",
1061 client->cl_import->imp_obd->obd_name);
1062 osc_update_next_shrink(client);
1066 static int osc_del_shrink_grant(struct client_obd *client)
1068 return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
1072 static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
1074 client_obd_list_lock(&cli->cl_loi_list_lock);
1075 cli->cl_avail_grant = ocd->ocd_grant;
1076 client_obd_list_unlock(&cli->cl_loi_list_lock);
1078 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
1079 list_empty(&cli->cl_grant_shrink_list))
1080 osc_add_shrink_grant(cli);
1082 CDEBUG(D_CACHE, "setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
1083 cli->cl_avail_grant, cli->cl_lost_grant);
1084 LASSERT(cli->cl_avail_grant >= 0);
1087 /* We assume that the reason this OSC got a short read is because it read
1088 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
1089 * via the LOV, and it _knows_ it's reading inside the file, it's just that
1090 * this stripe never got written at or beyond this stripe offset yet. */
1091 static void handle_short_read(int nob_read, obd_count page_count,
1092 struct brw_page **pga)
1097 /* skip bytes read OK */
1098 while (nob_read > 0) {
1099 LASSERT (page_count > 0);
1101 if (pga[i]->count > nob_read) {
1102 /* EOF inside this page */
1103 ptr = cfs_kmap(pga[i]->pg) +
1104 (pga[i]->off & ~CFS_PAGE_MASK);
1105 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
1106 cfs_kunmap(pga[i]->pg);
1112 nob_read -= pga[i]->count;
1117 /* zero remaining pages */
1118 while (page_count-- > 0) {
1119 ptr = cfs_kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
1120 memset(ptr, 0, pga[i]->count);
1121 cfs_kunmap(pga[i]->pg);
1126 static int check_write_rcs(struct ptlrpc_request *req,
1127 int requested_nob, int niocount,
1128 obd_count page_count, struct brw_page **pga)
1132 /* return error if any niobuf was in error */
1133 remote_rcs = lustre_swab_repbuf(req, REQ_REC_OFF + 1,
1134 sizeof(*remote_rcs) * niocount, NULL);
1135 if (remote_rcs == NULL) {
1136 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1139 if (ptlrpc_rep_need_swab(req))
1140 for (i = 0; i < niocount; i++)
1141 __swab32s(&remote_rcs[i]);
1143 for (i = 0; i < niocount; i++) {
1144 if (remote_rcs[i] < 0)
1145 return(remote_rcs[i]);
1147 if (remote_rcs[i] != 0) {
1148 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1149 i, remote_rcs[i], req);
1154 if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1155 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1156 req->rq_bulk->bd_nob_transferred, requested_nob);
1163 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1165 if (p1->flag != p2->flag) {
1166 unsigned mask = ~(OBD_BRW_FROM_GRANT|
1167 OBD_BRW_NOCACHE|OBD_BRW_SYNC);
1169 /* warn if we try to combine flags that we don't know to be
1170 * safe to combine */
1171 if ((p1->flag & mask) != (p2->flag & mask))
1172 CERROR("is it ok to have flags 0x%x and 0x%x in the "
1173 "same brw?\n", p1->flag, p2->flag);
1177 return (p1->off + p1->count == p2->off);
1180 static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1181 struct brw_page **pga, int opc,
1182 cksum_type_t cksum_type)
1187 LASSERT (pg_count > 0);
1188 cksum = init_checksum(cksum_type);
1189 while (nob > 0 && pg_count > 0) {
1190 unsigned char *ptr = cfs_kmap(pga[i]->pg);
1191 int off = pga[i]->off & ~CFS_PAGE_MASK;
1192 int count = pga[i]->count > nob ? nob : pga[i]->count;
1194 /* corrupt the data before we compute the checksum, to
1195 * simulate an OST->client data error */
1196 if (i == 0 && opc == OST_READ &&
1197 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE))
1198 memcpy(ptr + off, "bad1", min(4, nob));
1199 cksum = compute_checksum(cksum, ptr + off, count, cksum_type);
1200 cfs_kunmap(pga[i]->pg);
1201 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d checksum %x\n",
1204 nob -= pga[i]->count;
1208 /* For sending we only compute the wrong checksum instead
1209 * of corrupting the data so it is still correct on a redo */
1210 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1216 static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1217 struct lov_stripe_md *lsm, obd_count page_count,
1218 struct brw_page **pga,
1219 struct ptlrpc_request **reqp,
1220 struct obd_capa *ocapa, int reserve)
1222 struct ptlrpc_request *req;
1223 struct ptlrpc_bulk_desc *desc;
1224 struct ost_body *body;
1225 struct obd_ioobj *ioobj;
1226 struct niobuf_remote *niobuf;
1227 int niocount, i, requested_nob, opc, rc;
1228 struct osc_brw_async_args *aa;
1229 struct req_capsule *pill;
1230 struct brw_page *pg_prev;
1233 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1234 RETURN(-ENOMEM); /* Recoverable */
1235 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1236 RETURN(-EINVAL); /* Fatal */
1238 if ((cmd & OBD_BRW_WRITE) != 0) {
1240 req = ptlrpc_request_alloc_pool(cli->cl_import,
1241 cli->cl_import->imp_rq_pool,
1245 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW);
1250 for (niocount = i = 1; i < page_count; i++) {
1251 if (!can_merge_pages(pga[i - 1], pga[i]))
1255 pill = &req->rq_pill;
1256 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1257 niocount * sizeof(*niobuf));
1258 osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1260 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1262 ptlrpc_request_free(req);
1265 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1266 ptlrpc_at_set_req_timeout(req);
1268 if (opc == OST_WRITE)
1269 desc = ptlrpc_prep_bulk_imp(req, page_count,
1270 BULK_GET_SOURCE, OST_BULK_PORTAL);
1272 desc = ptlrpc_prep_bulk_imp(req, page_count,
1273 BULK_PUT_SINK, OST_BULK_PORTAL);
1276 GOTO(out, rc = -ENOMEM);
1277 /* NB request now owns desc and will free it when it gets freed */
1279 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1280 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1281 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1282 LASSERT(body && ioobj && niobuf);
1284 lustre_set_wire_obdo(&body->oa, oa);
1286 obdo_to_ioobj(oa, ioobj);
1287 ioobj->ioo_bufcnt = niocount;
1288 osc_pack_capa(req, body, ocapa);
1289 LASSERT (page_count > 0);
1291 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1292 struct brw_page *pg = pga[i];
1294 LASSERT(pg->count > 0);
1295 LASSERTF((pg->off & ~CFS_PAGE_MASK) + pg->count <= CFS_PAGE_SIZE,
1296 "i: %d pg: %p off: "LPU64", count: %u\n", i, pg,
1297 pg->off, pg->count);
1299 LASSERTF(i == 0 || pg->off > pg_prev->off,
1300 "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1301 " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1303 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1304 pg_prev->pg, page_private(pg_prev->pg),
1305 pg_prev->pg->index, pg_prev->off);
1307 LASSERTF(i == 0 || pg->off > pg_prev->off,
1308 "i %d p_c %u\n", i, page_count);
1310 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1311 (pg->flag & OBD_BRW_SRVLOCK));
1313 ptlrpc_prep_bulk_page(desc, pg->pg, pg->off & ~CFS_PAGE_MASK,
1315 requested_nob += pg->count;
1317 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1319 niobuf->len += pg->count;
1321 niobuf->offset = pg->off;
1322 niobuf->len = pg->count;
1323 niobuf->flags = pg->flag;
1328 LASSERTF((void *)(niobuf - niocount) ==
1329 lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1330 niocount * sizeof(*niobuf)),
1331 "want %p - real %p\n", lustre_msg_buf(req->rq_reqmsg,
1332 REQ_REC_OFF + 2, niocount * sizeof(*niobuf)),
1333 (void *)(niobuf - niocount));
1335 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1336 if (osc_should_shrink_grant(cli))
1337 osc_shrink_grant_local(cli, &body->oa);
1339 /* size[REQ_REC_OFF] still sizeof (*body) */
1340 if (opc == OST_WRITE) {
1341 if (unlikely(cli->cl_checksum) &&
1342 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1343 /* store cl_cksum_type in a local variable since
1344 * it can be changed via lprocfs */
1345 cksum_type_t cksum_type = cli->cl_cksum_type;
1347 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1348 oa->o_flags &= OBD_FL_LOCAL_MASK;
1349 body->oa.o_flags = 0;
1351 body->oa.o_flags |= cksum_type_pack(cksum_type);
1352 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1353 body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1357 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1359 /* save this in 'oa', too, for later checking */
1360 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1361 oa->o_flags |= cksum_type_pack(cksum_type);
1363 /* clear out the checksum flag, in case this is a
1364 * resend but cl_checksum is no longer set. b=11238 */
1365 oa->o_valid &= ~OBD_MD_FLCKSUM;
1367 oa->o_cksum = body->oa.o_cksum;
1368 /* 1 RC per niobuf */
1369 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_SERVER,
1370 sizeof(__u32) * niocount);
1372 if (unlikely(cli->cl_checksum) &&
1373 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1374 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1375 body->oa.o_flags = 0;
1376 body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1377 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1379 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_SERVER, 0);
1380 /* 1 RC for the whole I/O */
1382 ptlrpc_request_set_replen(req);
1384 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1385 aa = ptlrpc_req_async_args(req);
1387 aa->aa_requested_nob = requested_nob;
1388 aa->aa_nio_count = niocount;
1389 aa->aa_page_count = page_count;
1393 CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1394 if (ocapa && reserve)
1395 aa->aa_ocapa = capa_get(ocapa);
1401 ptlrpc_req_finished(req);
1405 static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1406 __u32 client_cksum, __u32 server_cksum, int nob,
1407 obd_count page_count, struct brw_page **pga,
1408 cksum_type_t client_cksum_type)
1412 cksum_type_t cksum_type;
1414 if (server_cksum == client_cksum) {
1415 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1419 if (oa->o_valid & OBD_MD_FLFLAGS)
1420 cksum_type = cksum_type_unpack(oa->o_flags);
1422 cksum_type = OBD_CKSUM_CRC32;
1424 new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1427 if (cksum_type != client_cksum_type)
1428 msg = "the server did not use the checksum type specified in "
1429 "the original request - likely a protocol problem";
1430 else if (new_cksum == server_cksum)
1431 msg = "changed on the client after we checksummed it - "
1432 "likely false positive due to mmap IO (bug 11742)";
1433 else if (new_cksum == client_cksum)
1434 msg = "changed in transit before arrival at OST";
1436 msg = "changed in transit AND doesn't match the original - "
1437 "likely false positive due to mmap IO (bug 11742)";
1439 LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inum "
1440 LPU64"/"LPU64" object "LPU64"/"LPU64" extent "
1441 "["LPU64"-"LPU64"]\n",
1442 msg, libcfs_nid2str(peer->nid),
1443 oa->o_valid & OBD_MD_FLFID ? oa->o_fid : (__u64)0,
1444 oa->o_valid & OBD_MD_FLFID ? oa->o_generation :
1447 oa->o_valid & OBD_MD_FLGROUP ? oa->o_gr : (__u64)0,
1449 pga[page_count-1]->off + pga[page_count-1]->count - 1);
1450 CERROR("original client csum %x (type %x), server csum %x (type %x), "
1451 "client csum now %x\n", client_cksum, client_cksum_type,
1452 server_cksum, cksum_type, new_cksum);
1456 /* Note rc enters this function as number of bytes transferred */
1457 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1459 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1460 const lnet_process_id_t *peer =
1461 &req->rq_import->imp_connection->c_peer;
1462 struct client_obd *cli = aa->aa_cli;
1463 struct ost_body *body;
1464 __u32 client_cksum = 0;
1467 if (rc < 0 && rc != -EDQUOT)
1470 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1471 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1472 lustre_swab_ost_body);
1474 CDEBUG(D_INFO, "Can't unpack body\n");
1478 /* set/clear over quota flag for a uid/gid */
1479 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1480 body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1481 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1483 lquota_setdq(quota_interface, cli, qid, body->oa.o_valid,
1490 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1491 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1493 osc_update_grant(cli, body);
1495 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1497 CERROR("Unexpected +ve rc %d\n", rc);
1500 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1502 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1505 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1506 check_write_checksum(&body->oa, peer, client_cksum,
1507 body->oa.o_cksum, aa->aa_requested_nob,
1508 aa->aa_page_count, aa->aa_ppga,
1509 cksum_type_unpack(aa->aa_oa->o_flags)))
1512 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1513 aa->aa_page_count, aa->aa_ppga);
1517 /* The rest of this function executes only for OST_READs */
1519 /* if unwrap_bulk failed, return -EAGAIN to retry */
1520 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1522 GOTO(out, rc = -EAGAIN);
1524 if (rc > aa->aa_requested_nob) {
1525 CERROR("Unexpected rc %d (%d requested)\n", rc,
1526 aa->aa_requested_nob);
1530 if (rc != req->rq_bulk->bd_nob_transferred) {
1531 CERROR ("Unexpected rc %d (%d transferred)\n",
1532 rc, req->rq_bulk->bd_nob_transferred);
1536 if (rc < aa->aa_requested_nob)
1537 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1539 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1540 static int cksum_counter;
1541 __u32 server_cksum = body->oa.o_cksum;
1544 cksum_type_t cksum_type;
1546 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1547 cksum_type = cksum_type_unpack(body->oa.o_flags);
1549 cksum_type = OBD_CKSUM_CRC32;
1550 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1551 aa->aa_ppga, OST_READ,
1554 if (peer->nid == req->rq_bulk->bd_sender) {
1558 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1561 if (server_cksum == ~0 && rc > 0) {
1562 CERROR("Protocol error: server %s set the 'checksum' "
1563 "bit, but didn't send a checksum. Not fatal, "
1564 "but please notify on http://bugzilla.lustre.org/\n",
1565 libcfs_nid2str(peer->nid));
1566 } else if (server_cksum != client_cksum) {
1567 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1568 "%s%s%s inum "LPU64"/"LPU64" object "
1569 LPU64"/"LPU64" extent "
1570 "["LPU64"-"LPU64"]\n",
1571 req->rq_import->imp_obd->obd_name,
1572 libcfs_nid2str(peer->nid),
1574 body->oa.o_valid & OBD_MD_FLFID ?
1575 body->oa.o_fid : (__u64)0,
1576 body->oa.o_valid & OBD_MD_FLFID ?
1577 body->oa.o_generation :(__u64)0,
1579 body->oa.o_valid & OBD_MD_FLGROUP ?
1580 body->oa.o_gr : (__u64)0,
1581 aa->aa_ppga[0]->off,
1582 aa->aa_ppga[aa->aa_page_count-1]->off +
1583 aa->aa_ppga[aa->aa_page_count-1]->count -
1585 CERROR("client %x, server %x, cksum_type %x\n",
1586 client_cksum, server_cksum, cksum_type);
1588 aa->aa_oa->o_cksum = client_cksum;
1592 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1595 } else if (unlikely(client_cksum)) {
1596 static int cksum_missed;
1599 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1600 CERROR("Checksum %u requested from %s but not sent\n",
1601 cksum_missed, libcfs_nid2str(peer->nid));
1607 lustre_get_wire_obdo(aa->aa_oa, &body->oa);
1612 static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa,
1613 struct lov_stripe_md *lsm,
1614 obd_count page_count, struct brw_page **pga,
1615 struct obd_capa *ocapa)
1617 struct ptlrpc_request *req;
1621 struct l_wait_info lwi;
1625 cfs_waitq_init(&waitq);
1628 rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm,
1629 page_count, pga, &req, ocapa, 0);
1633 rc = ptlrpc_queue_wait(req);
1635 if (rc == -ETIMEDOUT && req->rq_resend) {
1636 DEBUG_REQ(D_HA, req, "BULK TIMEOUT");
1637 ptlrpc_req_finished(req);
1641 rc = osc_brw_fini_request(req, rc);
1643 ptlrpc_req_finished(req);
1644 if (osc_recoverable_error(rc)) {
1646 if (!osc_should_resend(resends, &exp->exp_obd->u.cli)) {
1647 CERROR("too many resend retries, returning error\n");
1651 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1652 l_wait_event(waitq, 0, &lwi);
1660 int osc_brw_redo_request(struct ptlrpc_request *request,
1661 struct osc_brw_async_args *aa)
1663 struct ptlrpc_request *new_req;
1664 struct ptlrpc_request_set *set = request->rq_set;
1665 struct osc_brw_async_args *new_aa;
1666 struct osc_async_page *oap;
1670 if (!osc_should_resend(aa->aa_resends, aa->aa_cli)) {
1671 CERROR("too many resend retries, returning error\n");
1675 DEBUG_REQ(D_ERROR, request, "redo for recoverable error");
1677 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1678 OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1679 aa->aa_cli, aa->aa_oa,
1680 NULL /* lsm unused by osc currently */,
1681 aa->aa_page_count, aa->aa_ppga,
1682 &new_req, aa->aa_ocapa, 0);
1686 client_obd_list_lock(&aa->aa_cli->cl_loi_list_lock);
1688 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1689 if (oap->oap_request != NULL) {
1690 LASSERTF(request == oap->oap_request,
1691 "request %p != oap_request %p\n",
1692 request, oap->oap_request);
1693 if (oap->oap_interrupted) {
1694 client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1695 ptlrpc_req_finished(new_req);
1700 /* New request takes over pga and oaps from old request.
1701 * Note that copying a list_head doesn't work, need to move it... */
1703 new_req->rq_interpret_reply = request->rq_interpret_reply;
1704 new_req->rq_async_args = request->rq_async_args;
1705 new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1707 new_aa = ptlrpc_req_async_args(new_req);
1709 CFS_INIT_LIST_HEAD(&new_aa->aa_oaps);
1710 list_splice(&aa->aa_oaps, &new_aa->aa_oaps);
1711 CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1713 list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1714 if (oap->oap_request) {
1715 ptlrpc_req_finished(oap->oap_request);
1716 oap->oap_request = ptlrpc_request_addref(new_req);
1720 new_aa->aa_ocapa = aa->aa_ocapa;
1721 aa->aa_ocapa = NULL;
1723 /* use ptlrpc_set_add_req is safe because interpret functions work
1724 * in check_set context. only one way exist with access to request
1725 * from different thread got -EINTR - this way protected with
1726 * cl_loi_list_lock */
1727 ptlrpc_set_add_req(set, new_req);
1729 client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1731 DEBUG_REQ(D_INFO, new_req, "new request");
1736 * ugh, we want disk allocation on the target to happen in offset order. we'll
1737 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1738 * fine for our small page arrays and doesn't require allocation. its an
1739 * insertion sort that swaps elements that are strides apart, shrinking the
1740 * stride down until its '1' and the array is sorted.
1742 static void sort_brw_pages(struct brw_page **array, int num)
1745 struct brw_page *tmp;
1749 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1754 for (i = stride ; i < num ; i++) {
1757 while (j >= stride && array[j - stride]->off > tmp->off) {
1758 array[j] = array[j - stride];
1763 } while (stride > 1);
1766 static obd_count max_unfragmented_pages(struct brw_page **pg, obd_count pages)
1772 LASSERT (pages > 0);
1773 offset = pg[i]->off & ~CFS_PAGE_MASK;
1777 if (pages == 0) /* that's all */
1780 if (offset + pg[i]->count < CFS_PAGE_SIZE)
1781 return count; /* doesn't end on page boundary */
1784 offset = pg[i]->off & ~CFS_PAGE_MASK;
1785 if (offset != 0) /* doesn't start on page boundary */
1792 static struct brw_page **osc_build_ppga(struct brw_page *pga, obd_count count)
1794 struct brw_page **ppga;
1797 OBD_ALLOC(ppga, sizeof(*ppga) * count);
1801 for (i = 0; i < count; i++)
1806 static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1808 LASSERT(ppga != NULL);
1809 OBD_FREE(ppga, sizeof(*ppga) * count);
1812 static int osc_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1813 obd_count page_count, struct brw_page *pga,
1814 struct obd_trans_info *oti)
1816 struct obdo *saved_oa = NULL;
1817 struct brw_page **ppga, **orig;
1818 struct obd_import *imp = class_exp2cliimp(exp);
1819 struct client_obd *cli;
1820 int rc, page_count_orig;
1823 LASSERT((imp != NULL) && (imp->imp_obd != NULL));
1824 cli = &imp->imp_obd->u.cli;
1826 if (cmd & OBD_BRW_CHECK) {
1827 /* The caller just wants to know if there's a chance that this
1828 * I/O can succeed */
1830 if (imp->imp_invalid)
1835 /* test_brw with a failed create can trip this, maybe others. */
1836 LASSERT(cli->cl_max_pages_per_rpc);
1840 orig = ppga = osc_build_ppga(pga, page_count);
1843 page_count_orig = page_count;
1845 sort_brw_pages(ppga, page_count);
1846 while (page_count) {
1847 obd_count pages_per_brw;
1849 if (page_count > cli->cl_max_pages_per_rpc)
1850 pages_per_brw = cli->cl_max_pages_per_rpc;
1852 pages_per_brw = page_count;
1854 pages_per_brw = max_unfragmented_pages(ppga, pages_per_brw);
1856 if (saved_oa != NULL) {
1857 /* restore previously saved oa */
1858 *oinfo->oi_oa = *saved_oa;
1859 } else if (page_count > pages_per_brw) {
1860 /* save a copy of oa (brw will clobber it) */
1861 OBDO_ALLOC(saved_oa);
1862 if (saved_oa == NULL)
1863 GOTO(out, rc = -ENOMEM);
1864 *saved_oa = *oinfo->oi_oa;
1867 rc = osc_brw_internal(cmd, exp, oinfo->oi_oa, oinfo->oi_md,
1868 pages_per_brw, ppga, oinfo->oi_capa);
1873 page_count -= pages_per_brw;
1874 ppga += pages_per_brw;
1878 osc_release_ppga(orig, page_count_orig);
1880 if (saved_oa != NULL)
1881 OBDO_FREE(saved_oa);
1886 /* The companion to osc_enter_cache(), called when @oap is no longer part of
1887 * the dirty accounting. Writeback completes or truncate happens before
1888 * writing starts. Must be called with the loi lock held. */
1889 static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap,
1892 osc_release_write_grant(cli, &oap->oap_brw_page, sent);
1896 /* This maintains the lists of pending pages to read/write for a given object
1897 * (lop). This is used by osc_check_rpcs->osc_next_loi() and loi_list_maint()
1898 * to quickly find objects that are ready to send an RPC. */
1899 static int lop_makes_rpc(struct client_obd *cli, struct loi_oap_pages *lop,
1905 if (lop->lop_num_pending == 0)
1908 /* if we have an invalid import we want to drain the queued pages
1909 * by forcing them through rpcs that immediately fail and complete
1910 * the pages. recovery relies on this to empty the queued pages
1911 * before canceling the locks and evicting down the llite pages */
1912 if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
1915 /* stream rpcs in queue order as long as as there is an urgent page
1916 * queued. this is our cheap solution for good batching in the case
1917 * where writepage marks some random page in the middle of the file
1918 * as urgent because of, say, memory pressure */
1919 if (!list_empty(&lop->lop_urgent)) {
1920 CDEBUG(D_CACHE, "urgent request forcing RPC\n");
1923 /* fire off rpcs when we have 'optimal' rpcs as tuned for the wire. */
1924 optimal = cli->cl_max_pages_per_rpc;
1925 if (cmd & OBD_BRW_WRITE) {
1926 /* trigger a write rpc stream as long as there are dirtiers
1927 * waiting for space. as they're waiting, they're not going to
1928 * create more pages to coallesce with what's waiting.. */
1929 if (!list_empty(&cli->cl_cache_waiters)) {
1930 CDEBUG(D_CACHE, "cache waiters forcing RPC\n");
1933 /* +16 to avoid triggering rpcs that would want to include pages
1934 * that are being queued but which can't be made ready until
1935 * the queuer finishes with the page. this is a wart for
1936 * llite::commit_write() */
1939 if (lop->lop_num_pending >= optimal)
1945 static int lop_makes_hprpc(struct loi_oap_pages *lop)
1947 struct osc_async_page *oap;
1950 if (list_empty(&lop->lop_urgent))
1953 oap = list_entry(lop->lop_urgent.next,
1954 struct osc_async_page, oap_urgent_item);
1956 if (oap->oap_async_flags & ASYNC_HP) {
1957 CDEBUG(D_CACHE, "hp request forcing RPC\n");
1964 static void on_list(struct list_head *item, struct list_head *list,
1967 if (list_empty(item) && should_be_on)
1968 list_add_tail(item, list);
1969 else if (!list_empty(item) && !should_be_on)
1970 list_del_init(item);
1973 /* maintain the loi's cli list membership invariants so that osc_send_oap_rpc
1974 * can find pages to build into rpcs quickly */
1975 void loi_list_maint(struct client_obd *cli, struct lov_oinfo *loi)
1977 if (lop_makes_hprpc(&loi->loi_write_lop) ||
1978 lop_makes_hprpc(&loi->loi_read_lop)) {
1980 on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list, 0);
1981 on_list(&loi->loi_hp_ready_item, &cli->cl_loi_hp_ready_list, 1);
1983 on_list(&loi->loi_hp_ready_item, &cli->cl_loi_hp_ready_list, 0);
1984 on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list,
1985 lop_makes_rpc(cli, &loi->loi_write_lop, OBD_BRW_WRITE)||
1986 lop_makes_rpc(cli, &loi->loi_read_lop, OBD_BRW_READ));
1989 on_list(&loi->loi_write_item, &cli->cl_loi_write_list,
1990 loi->loi_write_lop.lop_num_pending);
1992 on_list(&loi->loi_read_item, &cli->cl_loi_read_list,
1993 loi->loi_read_lop.lop_num_pending);
1996 static void lop_update_pending(struct client_obd *cli,
1997 struct loi_oap_pages *lop, int cmd, int delta)
1999 lop->lop_num_pending += delta;
2000 if (cmd & OBD_BRW_WRITE)
2001 cli->cl_pending_w_pages += delta;
2003 cli->cl_pending_r_pages += delta;
2007 * this is called when a sync waiter receives an interruption. Its job is to
2008 * get the caller woken as soon as possible. If its page hasn't been put in an
2009 * rpc yet it can dequeue immediately. Otherwise it has to mark the rpc as
2010 * desiring interruption which will forcefully complete the rpc once the rpc
2013 int osc_oap_interrupted(const struct lu_env *env, struct osc_async_page *oap)
2015 struct loi_oap_pages *lop;
2016 struct lov_oinfo *loi;
2020 LASSERT(!oap->oap_interrupted);
2021 oap->oap_interrupted = 1;
2023 /* ok, it's been put in an rpc. only one oap gets a request reference */
2024 if (oap->oap_request != NULL) {
2025 ptlrpc_mark_interrupted(oap->oap_request);
2026 ptlrpcd_wake(oap->oap_request);
2027 ptlrpc_req_finished(oap->oap_request);
2028 oap->oap_request = NULL;
2032 * page completion may be called only if ->cpo_prep() method was
2033 * executed by osc_io_submit(), that also adds page the to pending list
2035 if (!list_empty(&oap->oap_pending_item)) {
2036 list_del_init(&oap->oap_pending_item);
2037 list_del_init(&oap->oap_urgent_item);
2040 lop = (oap->oap_cmd & OBD_BRW_WRITE) ?
2041 &loi->loi_write_lop : &loi->loi_read_lop;
2042 lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, -1);
2043 loi_list_maint(oap->oap_cli, oap->oap_loi);
2044 rc = oap->oap_caller_ops->ap_completion(env,
2045 oap->oap_caller_data,
2046 oap->oap_cmd, NULL, -EINTR);
2052 /* this is trying to propogate async writeback errors back up to the
2053 * application. As an async write fails we record the error code for later if
2054 * the app does an fsync. As long as errors persist we force future rpcs to be
2055 * sync so that the app can get a sync error and break the cycle of queueing
2056 * pages for which writeback will fail. */
2057 static void osc_process_ar(struct osc_async_rc *ar, __u64 xid,
2064 ar->ar_force_sync = 1;
2065 ar->ar_min_xid = ptlrpc_sample_next_xid();
2070 if (ar->ar_force_sync && (xid >= ar->ar_min_xid))
2071 ar->ar_force_sync = 0;
2074 void osc_oap_to_pending(struct osc_async_page *oap)
2076 struct loi_oap_pages *lop;
2078 if (oap->oap_cmd & OBD_BRW_WRITE)
2079 lop = &oap->oap_loi->loi_write_lop;
2081 lop = &oap->oap_loi->loi_read_lop;
2083 if (oap->oap_async_flags & ASYNC_HP)
2084 list_add(&oap->oap_urgent_item, &lop->lop_urgent);
2085 else if (oap->oap_async_flags & ASYNC_URGENT)
2086 list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
2087 list_add_tail(&oap->oap_pending_item, &lop->lop_pending);
2088 lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, 1);
2091 /* this must be called holding the loi list lock to give coverage to exit_cache,
2092 * async_flag maintenance, and oap_request */
2093 static void osc_ap_completion(const struct lu_env *env,
2094 struct client_obd *cli, struct obdo *oa,
2095 struct osc_async_page *oap, int sent, int rc)
2100 if (oap->oap_request != NULL) {
2101 xid = ptlrpc_req_xid(oap->oap_request);
2102 ptlrpc_req_finished(oap->oap_request);
2103 oap->oap_request = NULL;
2106 spin_lock(&oap->oap_lock);
2107 oap->oap_async_flags = 0;
2108 spin_unlock(&oap->oap_lock);
2109 oap->oap_interrupted = 0;
2111 if (oap->oap_cmd & OBD_BRW_WRITE) {
2112 osc_process_ar(&cli->cl_ar, xid, rc);
2113 osc_process_ar(&oap->oap_loi->loi_ar, xid, rc);
2116 if (rc == 0 && oa != NULL) {
2117 if (oa->o_valid & OBD_MD_FLBLOCKS)
2118 oap->oap_loi->loi_lvb.lvb_blocks = oa->o_blocks;
2119 if (oa->o_valid & OBD_MD_FLMTIME)
2120 oap->oap_loi->loi_lvb.lvb_mtime = oa->o_mtime;
2121 if (oa->o_valid & OBD_MD_FLATIME)
2122 oap->oap_loi->loi_lvb.lvb_atime = oa->o_atime;
2123 if (oa->o_valid & OBD_MD_FLCTIME)
2124 oap->oap_loi->loi_lvb.lvb_ctime = oa->o_ctime;
2127 rc = oap->oap_caller_ops->ap_completion(env, oap->oap_caller_data,
2128 oap->oap_cmd, oa, rc);
2130 /* ll_ap_completion (from llite) drops PG_locked. so, a new
2131 * I/O on the page could start, but OSC calls it under lock
2132 * and thus we can add oap back to pending safely */
2134 /* upper layer wants to leave the page on pending queue */
2135 osc_oap_to_pending(oap);
2137 osc_exit_cache(cli, oap, sent);
2141 static int brw_interpret(const struct lu_env *env,
2142 struct ptlrpc_request *req, void *data, int rc)
2144 struct osc_brw_async_args *aa = data;
2145 struct client_obd *cli;
2149 rc = osc_brw_fini_request(req, rc);
2150 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
2151 if (osc_recoverable_error(rc)) {
2152 rc = osc_brw_redo_request(req, aa);
2158 capa_put(aa->aa_ocapa);
2159 aa->aa_ocapa = NULL;
2164 client_obd_list_lock(&cli->cl_loi_list_lock);
2166 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2167 * is called so we know whether to go to sync BRWs or wait for more
2168 * RPCs to complete */
2169 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2170 cli->cl_w_in_flight--;
2172 cli->cl_r_in_flight--;
2174 async = list_empty(&aa->aa_oaps);
2175 if (!async) { /* from osc_send_oap_rpc() */
2176 struct osc_async_page *oap, *tmp;
2177 /* the caller may re-use the oap after the completion call so
2178 * we need to clean it up a little */
2179 list_for_each_entry_safe(oap, tmp, &aa->aa_oaps, oap_rpc_item) {
2180 list_del_init(&oap->oap_rpc_item);
2181 osc_ap_completion(env, cli, aa->aa_oa, oap, 1, rc);
2183 OBDO_FREE(aa->aa_oa);
2184 } else { /* from async_internal() */
2186 for (i = 0; i < aa->aa_page_count; i++)
2187 osc_release_write_grant(aa->aa_cli, aa->aa_ppga[i], 1);
2189 if (aa->aa_oa->o_flags & OBD_FL_TEMPORARY)
2190 OBDO_FREE(aa->aa_oa);
2192 osc_wake_cache_waiters(cli);
2193 osc_check_rpcs(env, cli);
2194 client_obd_list_unlock(&cli->cl_loi_list_lock);
2196 cl_req_completion(env, aa->aa_clerq, rc);
2197 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
2201 static struct ptlrpc_request *osc_build_req(const struct lu_env *env,
2202 struct client_obd *cli,
2203 struct list_head *rpc_list,
2204 int page_count, int cmd)
2206 struct ptlrpc_request *req;
2207 struct brw_page **pga = NULL;
2208 struct osc_brw_async_args *aa;
2209 struct obdo *oa = NULL;
2210 const struct obd_async_page_ops *ops = NULL;
2211 void *caller_data = NULL;
2212 struct osc_async_page *oap;
2213 struct osc_async_page *tmp;
2214 struct ost_body *body;
2215 struct cl_req *clerq = NULL;
2216 enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2217 struct ldlm_lock *lock = NULL;
2218 struct cl_req_attr crattr;
2222 LASSERT(!list_empty(rpc_list));
2224 memset(&crattr, 0, sizeof crattr);
2225 OBD_ALLOC(pga, sizeof(*pga) * page_count);
2227 GOTO(out, req = ERR_PTR(-ENOMEM));
2231 GOTO(out, req = ERR_PTR(-ENOMEM));
2234 list_for_each_entry(oap, rpc_list, oap_rpc_item) {
2235 struct cl_page *page = osc_oap2cl_page(oap);
2237 ops = oap->oap_caller_ops;
2238 caller_data = oap->oap_caller_data;
2240 clerq = cl_req_alloc(env, page, crt,
2241 1 /* only 1-object rpcs for
2244 GOTO(out, req = (void *)clerq);
2245 lock = oap->oap_ldlm_lock;
2247 pga[i] = &oap->oap_brw_page;
2248 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2249 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
2250 pga[i]->pg, cfs_page_index(oap->oap_page), oap, pga[i]->flag);
2252 cl_req_page_add(env, clerq, page);
2255 /* always get the data for the obdo for the rpc */
2256 LASSERT(ops != NULL);
2258 crattr.cra_capa = NULL;
2259 cl_req_attr_set(env, clerq, &crattr, ~0ULL);
2261 oa->o_handle = lock->l_remote_handle;
2262 oa->o_valid |= OBD_MD_FLHANDLE;
2265 rc = cl_req_prep(env, clerq);
2267 CERROR("cl_req_prep failed: %d\n", rc);
2268 GOTO(out, req = ERR_PTR(rc));
2271 sort_brw_pages(pga, page_count);
2272 rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
2273 pga, &req, crattr.cra_capa, 1);
2275 CERROR("prep_req failed: %d\n", rc);
2276 GOTO(out, req = ERR_PTR(rc));
2279 /* Need to update the timestamps after the request is built in case
2280 * we race with setattr (locally or in queue at OST). If OST gets
2281 * later setattr before earlier BRW (as determined by the request xid),
2282 * the OST will not use BRW timestamps. Sadly, there is no obvious
2283 * way to do this in a single call. bug 10150 */
2284 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2285 cl_req_attr_set(env, clerq, &crattr,
2286 OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
2288 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2289 aa = ptlrpc_req_async_args(req);
2290 CFS_INIT_LIST_HEAD(&aa->aa_oaps);
2291 list_splice(rpc_list, &aa->aa_oaps);
2292 CFS_INIT_LIST_HEAD(rpc_list);
2293 aa->aa_clerq = clerq;
2295 capa_put(crattr.cra_capa);
2300 OBD_FREE(pga, sizeof(*pga) * page_count);
2301 /* this should happen rarely and is pretty bad, it makes the
2302 * pending list not follow the dirty order */
2303 client_obd_list_lock(&cli->cl_loi_list_lock);
2304 list_for_each_entry_safe(oap, tmp, rpc_list, oap_rpc_item) {
2305 list_del_init(&oap->oap_rpc_item);
2307 /* queued sync pages can be torn down while the pages
2308 * were between the pending list and the rpc */
2309 if (oap->oap_interrupted) {
2310 CDEBUG(D_INODE, "oap %p interrupted\n", oap);
2311 osc_ap_completion(env, cli, NULL, oap, 0,
2315 osc_ap_completion(env, cli, NULL, oap, 0, PTR_ERR(req));
2317 if (clerq && !IS_ERR(clerq))
2318 cl_req_completion(env, clerq, PTR_ERR(req));
2324 * prepare pages for ASYNC io and put pages in send queue.
2326 * \param cmd OBD_BRW_* macroses
2327 * \param lop pending pages
2329 * \return zero if pages successfully add to send queue.
2330 * \return not zere if error occurring.
2333 osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli,
2334 struct lov_oinfo *loi,
2335 int cmd, struct loi_oap_pages *lop)
2337 struct ptlrpc_request *req;
2338 obd_count page_count = 0;
2339 struct osc_async_page *oap = NULL, *tmp;
2340 struct osc_brw_async_args *aa;
2341 const struct obd_async_page_ops *ops;
2342 CFS_LIST_HEAD(rpc_list);
2343 CFS_LIST_HEAD(tmp_list);
2344 unsigned int ending_offset;
2345 unsigned starting_offset = 0;
2347 struct cl_object *clob = NULL;
2350 /* ASYNC_HP pages first. At present, when the lock the pages is
2351 * to be canceled, the pages covered by the lock will be sent out
2352 * with ASYNC_HP. We have to send out them as soon as possible. */
2353 list_for_each_entry_safe(oap, tmp, &lop->lop_urgent, oap_urgent_item) {
2354 if (oap->oap_async_flags & ASYNC_HP)
2355 list_move(&oap->oap_pending_item, &tmp_list);
2357 list_move_tail(&oap->oap_pending_item, &tmp_list);
2358 if (++page_count >= cli->cl_max_pages_per_rpc)
2362 list_splice(&tmp_list, &lop->lop_pending);
2365 /* first we find the pages we're allowed to work with */
2366 list_for_each_entry_safe(oap, tmp, &lop->lop_pending,
2368 ops = oap->oap_caller_ops;
2370 LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
2371 "magic 0x%x\n", oap, oap->oap_magic);
2374 /* pin object in memory, so that completion call-backs
2375 * can be safely called under client_obd_list lock. */
2376 clob = osc_oap2cl_page(oap)->cp_obj;
2377 cl_object_get(clob);
2380 if (page_count != 0 &&
2381 srvlock != !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK)) {
2382 CDEBUG(D_PAGE, "SRVLOCK flag mismatch,"
2383 " oap %p, page %p, srvlock %u\n",
2384 oap, oap->oap_brw_page.pg, (unsigned)!srvlock);
2388 /* If there is a gap at the start of this page, it can't merge
2389 * with any previous page, so we'll hand the network a
2390 * "fragmented" page array that it can't transfer in 1 RDMA */
2391 if (page_count != 0 && oap->oap_page_off != 0)
2394 /* in llite being 'ready' equates to the page being locked
2395 * until completion unlocks it. commit_write submits a page
2396 * as not ready because its unlock will happen unconditionally
2397 * as the call returns. if we race with commit_write giving
2398 * us that page we dont' want to create a hole in the page
2399 * stream, so we stop and leave the rpc to be fired by
2400 * another dirtier or kupdated interval (the not ready page
2401 * will still be on the dirty list). we could call in
2402 * at the end of ll_file_write to process the queue again. */
2403 if (!(oap->oap_async_flags & ASYNC_READY)) {
2404 int rc = ops->ap_make_ready(env, oap->oap_caller_data,
2407 CDEBUG(D_INODE, "oap %p page %p returned %d "
2408 "instead of ready\n", oap,
2412 /* llite is telling us that the page is still
2413 * in commit_write and that we should try
2414 * and put it in an rpc again later. we
2415 * break out of the loop so we don't create
2416 * a hole in the sequence of pages in the rpc
2421 /* the io isn't needed.. tell the checks
2422 * below to complete the rpc with EINTR */
2423 spin_lock(&oap->oap_lock);
2424 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
2425 spin_unlock(&oap->oap_lock);
2426 oap->oap_count = -EINTR;
2429 spin_lock(&oap->oap_lock);
2430 oap->oap_async_flags |= ASYNC_READY;
2431 spin_unlock(&oap->oap_lock);
2434 LASSERTF(0, "oap %p page %p returned %d "
2435 "from make_ready\n", oap,
2443 * Page submitted for IO has to be locked. Either by
2444 * ->ap_make_ready() or by higher layers.
2446 #if defined(__KERNEL__) && defined(__linux__)
2448 struct cl_page *page;
2450 page = osc_oap2cl_page(oap);
2452 if (page->cp_type == CPT_CACHEABLE &&
2453 !(PageLocked(oap->oap_page) &&
2454 (CheckWriteback(oap->oap_page, cmd)))) {
2455 CDEBUG(D_PAGE, "page %p lost wb %lx/%x\n",
2457 (long)oap->oap_page->flags,
2458 oap->oap_async_flags);
2464 /* take the page out of our book-keeping */
2465 list_del_init(&oap->oap_pending_item);
2466 lop_update_pending(cli, lop, cmd, -1);
2467 list_del_init(&oap->oap_urgent_item);
2469 if (page_count == 0)
2470 starting_offset = (oap->oap_obj_off+oap->oap_page_off) &
2471 (PTLRPC_MAX_BRW_SIZE - 1);
2473 /* ask the caller for the size of the io as the rpc leaves. */
2474 if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
2476 ops->ap_refresh_count(env, oap->oap_caller_data,
2478 LASSERT(oap->oap_page_off + oap->oap_count <= CFS_PAGE_SIZE);
2480 if (oap->oap_count <= 0) {
2481 CDEBUG(D_CACHE, "oap %p count %d, completing\n", oap,
2483 osc_ap_completion(env, cli, NULL,
2484 oap, 0, oap->oap_count);
2488 /* now put the page back in our accounting */
2489 list_add_tail(&oap->oap_rpc_item, &rpc_list);
2490 if (page_count == 0)
2491 srvlock = !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK);
2492 if (++page_count >= cli->cl_max_pages_per_rpc)
2495 /* End on a PTLRPC_MAX_BRW_SIZE boundary. We want full-sized
2496 * RPCs aligned on PTLRPC_MAX_BRW_SIZE boundaries to help reads
2497 * have the same alignment as the initial writes that allocated
2498 * extents on the server. */
2499 ending_offset = (oap->oap_obj_off + oap->oap_page_off +
2500 oap->oap_count) & (PTLRPC_MAX_BRW_SIZE - 1);
2501 if (ending_offset == 0)
2504 /* If there is a gap at the end of this page, it can't merge
2505 * with any subsequent pages, so we'll hand the network a
2506 * "fragmented" page array that it can't transfer in 1 RDMA */
2507 if (oap->oap_page_off + oap->oap_count < CFS_PAGE_SIZE)
2511 osc_wake_cache_waiters(cli);
2513 loi_list_maint(cli, loi);
2515 client_obd_list_unlock(&cli->cl_loi_list_lock);
2518 cl_object_put(env, clob);
2520 if (page_count == 0) {
2521 client_obd_list_lock(&cli->cl_loi_list_lock);
2525 req = osc_build_req(env, cli, &rpc_list, page_count, cmd);
2527 LASSERT(list_empty(&rpc_list));
2528 loi_list_maint(cli, loi);
2529 RETURN(PTR_ERR(req));
2532 aa = ptlrpc_req_async_args(req);
2534 if (cmd == OBD_BRW_READ) {
2535 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2536 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2537 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2538 (starting_offset >> CFS_PAGE_SHIFT) + 1);
2540 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2541 lprocfs_oh_tally(&cli->cl_write_rpc_hist,
2542 cli->cl_w_in_flight);
2543 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2544 (starting_offset >> CFS_PAGE_SHIFT) + 1);
2546 ptlrpc_lprocfs_brw(req, aa->aa_requested_nob);
2548 client_obd_list_lock(&cli->cl_loi_list_lock);
2550 if (cmd == OBD_BRW_READ)
2551 cli->cl_r_in_flight++;
2553 cli->cl_w_in_flight++;
2555 /* queued sync pages can be torn down while the pages
2556 * were between the pending list and the rpc */
2558 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
2559 /* only one oap gets a request reference */
2562 if (oap->oap_interrupted && !req->rq_intr) {
2563 CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
2565 ptlrpc_mark_interrupted(req);
2569 tmp->oap_request = ptlrpc_request_addref(req);
2571 DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
2572 page_count, aa, cli->cl_r_in_flight, cli->cl_w_in_flight);
2574 req->rq_interpret_reply = brw_interpret;
2575 ptlrpcd_add_req(req, PSCOPE_BRW);
2579 #define LOI_DEBUG(LOI, STR, args...) \
2580 CDEBUG(D_INODE, "loi ready %d wr %d:%d rd %d:%d " STR, \
2581 !list_empty(&(LOI)->loi_ready_item) || \
2582 !list_empty(&(LOI)->loi_hp_ready_item), \
2583 (LOI)->loi_write_lop.lop_num_pending, \
2584 !list_empty(&(LOI)->loi_write_lop.lop_urgent), \
2585 (LOI)->loi_read_lop.lop_num_pending, \
2586 !list_empty(&(LOI)->loi_read_lop.lop_urgent), \
2589 /* This is called by osc_check_rpcs() to find which objects have pages that
2590 * we could be sending. These lists are maintained by lop_makes_rpc(). */
2591 struct lov_oinfo *osc_next_loi(struct client_obd *cli)
2595 /* First return objects that have blocked locks so that they
2596 * will be flushed quickly and other clients can get the lock,
2597 * then objects which have pages ready to be stuffed into RPCs */
2598 if (!list_empty(&cli->cl_loi_hp_ready_list))
2599 RETURN(list_entry(cli->cl_loi_hp_ready_list.next,
2600 struct lov_oinfo, loi_hp_ready_item));
2601 if (!list_empty(&cli->cl_loi_ready_list))
2602 RETURN(list_entry(cli->cl_loi_ready_list.next,
2603 struct lov_oinfo, loi_ready_item));
2605 /* then if we have cache waiters, return all objects with queued
2606 * writes. This is especially important when many small files
2607 * have filled up the cache and not been fired into rpcs because
2608 * they don't pass the nr_pending/object threshhold */
2609 if (!list_empty(&cli->cl_cache_waiters) &&
2610 !list_empty(&cli->cl_loi_write_list))
2611 RETURN(list_entry(cli->cl_loi_write_list.next,
2612 struct lov_oinfo, loi_write_item));
2614 /* then return all queued objects when we have an invalid import
2615 * so that they get flushed */
2616 if (cli->cl_import == NULL || cli->cl_import->imp_invalid) {
2617 if (!list_empty(&cli->cl_loi_write_list))
2618 RETURN(list_entry(cli->cl_loi_write_list.next,
2619 struct lov_oinfo, loi_write_item));
2620 if (!list_empty(&cli->cl_loi_read_list))
2621 RETURN(list_entry(cli->cl_loi_read_list.next,
2622 struct lov_oinfo, loi_read_item));
2627 static int osc_max_rpc_in_flight(struct client_obd *cli, struct lov_oinfo *loi)
2629 struct osc_async_page *oap;
2632 if (!list_empty(&loi->loi_write_lop.lop_urgent)) {
2633 oap = list_entry(loi->loi_write_lop.lop_urgent.next,
2634 struct osc_async_page, oap_urgent_item);
2635 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2638 if (!hprpc && !list_empty(&loi->loi_read_lop.lop_urgent)) {
2639 oap = list_entry(loi->loi_read_lop.lop_urgent.next,
2640 struct osc_async_page, oap_urgent_item);
2641 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2644 return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc;
2647 /* called with the loi list lock held */
2648 void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli)
2650 struct lov_oinfo *loi;
2651 int rc = 0, race_counter = 0;
2654 while ((loi = osc_next_loi(cli)) != NULL) {
2655 LOI_DEBUG(loi, "%lu in flight\n", rpcs_in_flight(cli));
2657 if (osc_max_rpc_in_flight(cli, loi))
2660 /* attempt some read/write balancing by alternating between
2661 * reads and writes in an object. The makes_rpc checks here
2662 * would be redundant if we were getting read/write work items
2663 * instead of objects. we don't want send_oap_rpc to drain a
2664 * partial read pending queue when we're given this object to
2665 * do io on writes while there are cache waiters */
2666 if (lop_makes_rpc(cli, &loi->loi_write_lop, OBD_BRW_WRITE)) {
2667 rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_WRITE,
2668 &loi->loi_write_lop);
2670 CERROR("Write request failed with %d\n", rc);
2672 /* osc_send_oap_rpc failed, mostly because of
2675 * It can't break here, because if:
2676 * - a page was submitted by osc_io_submit, so
2678 * - no request in flight
2679 * - no subsequent request
2680 * The system will be in live-lock state,
2681 * because there is no chance to call
2682 * osc_io_unplug() and osc_check_rpcs() any
2683 * more. pdflush can't help in this case,
2684 * because it might be blocked at grabbing
2685 * the page lock as we mentioned.
2687 * Anyway, continue to drain pages. */
2696 if (lop_makes_rpc(cli, &loi->loi_read_lop, OBD_BRW_READ)) {
2697 rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_READ,
2698 &loi->loi_read_lop);
2700 CERROR("Read request failed with %d\n", rc);
2708 /* attempt some inter-object balancing by issueing rpcs
2709 * for each object in turn */
2710 if (!list_empty(&loi->loi_hp_ready_item))
2711 list_del_init(&loi->loi_hp_ready_item);
2712 if (!list_empty(&loi->loi_ready_item))
2713 list_del_init(&loi->loi_ready_item);
2714 if (!list_empty(&loi->loi_write_item))
2715 list_del_init(&loi->loi_write_item);
2716 if (!list_empty(&loi->loi_read_item))
2717 list_del_init(&loi->loi_read_item);
2719 loi_list_maint(cli, loi);
2721 /* send_oap_rpc fails with 0 when make_ready tells it to
2722 * back off. llite's make_ready does this when it tries
2723 * to lock a page queued for write that is already locked.
2724 * we want to try sending rpcs from many objects, but we
2725 * don't want to spin failing with 0. */
2726 if (race_counter == 10)
2732 /* we're trying to queue a page in the osc so we're subject to the
2733 * 'cl_dirty_max' limit on the number of pages that can be queued in the osc.
2734 * If the osc's queued pages are already at that limit, then we want to sleep
2735 * until there is space in the osc's queue for us. We also may be waiting for
2736 * write credits from the OST if there are RPCs in flight that may return some
2737 * before we fall back to sync writes.
2739 * We need this know our allocation was granted in the presence of signals */
2740 static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw)
2744 client_obd_list_lock(&cli->cl_loi_list_lock);
2745 rc = list_empty(&ocw->ocw_entry) || rpcs_in_flight(cli) == 0;
2746 client_obd_list_unlock(&cli->cl_loi_list_lock);
2751 * Non-blocking version of osc_enter_cache() that consumes grant only when it
2754 int osc_enter_cache_try(const struct lu_env *env,
2755 struct client_obd *cli, struct lov_oinfo *loi,
2756 struct osc_async_page *oap, int transient)
2760 has_grant = cli->cl_avail_grant >= CFS_PAGE_SIZE;
2762 osc_consume_write_grant(cli, &oap->oap_brw_page);
2764 cli->cl_dirty_transit += CFS_PAGE_SIZE;
2765 atomic_inc(&obd_dirty_transit_pages);
2766 oap->oap_brw_flags |= OBD_BRW_NOCACHE;
2772 /* Caller must hold loi_list_lock - we drop/regain it if we need to wait for
2773 * grant or cache space. */
2774 static int osc_enter_cache(const struct lu_env *env,
2775 struct client_obd *cli, struct lov_oinfo *loi,
2776 struct osc_async_page *oap)
2778 struct osc_cache_waiter ocw;
2779 struct l_wait_info lwi = { 0 };
2783 CDEBUG(D_CACHE, "dirty: %ld/%d dirty_max: %ld/%d dropped: %lu "
2784 "grant: %lu\n", cli->cl_dirty, atomic_read(&obd_dirty_pages),
2785 cli->cl_dirty_max, obd_max_dirty_pages,
2786 cli->cl_lost_grant, cli->cl_avail_grant);
2788 /* force the caller to try sync io. this can jump the list
2789 * of queued writes and create a discontiguous rpc stream */
2790 if (cli->cl_dirty_max < CFS_PAGE_SIZE || cli->cl_ar.ar_force_sync ||
2791 loi->loi_ar.ar_force_sync)
2794 /* Hopefully normal case - cache space and write credits available */
2795 if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
2796 atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages &&
2797 osc_enter_cache_try(env, cli, loi, oap, 0))
2800 /* Make sure that there are write rpcs in flight to wait for. This
2801 * is a little silly as this object may not have any pending but
2802 * other objects sure might. */
2803 if (cli->cl_w_in_flight) {
2804 list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
2805 cfs_waitq_init(&ocw.ocw_waitq);
2809 loi_list_maint(cli, loi);
2810 osc_check_rpcs(env, cli);
2811 client_obd_list_unlock(&cli->cl_loi_list_lock);
2813 CDEBUG(D_CACHE, "sleeping for cache space\n");
2814 l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi);
2816 client_obd_list_lock(&cli->cl_loi_list_lock);
2817 if (!list_empty(&ocw.ocw_entry)) {
2818 list_del(&ocw.ocw_entry);
2828 int osc_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
2829 struct lov_oinfo *loi, cfs_page_t *page,
2830 obd_off offset, const struct obd_async_page_ops *ops,
2831 void *data, void **res, int nocache,
2832 struct lustre_handle *lockh)
2834 struct osc_async_page *oap;
2839 return size_round(sizeof(*oap));
2842 oap->oap_magic = OAP_MAGIC;
2843 oap->oap_cli = &exp->exp_obd->u.cli;
2846 oap->oap_caller_ops = ops;
2847 oap->oap_caller_data = data;
2849 oap->oap_page = page;
2850 oap->oap_obj_off = offset;
2851 if (!client_is_remote(exp) &&
2852 cfs_capable(CFS_CAP_SYS_RESOURCE))
2853 oap->oap_brw_flags = OBD_BRW_NOQUOTA;
2855 LASSERT(!(offset & ~CFS_PAGE_MASK));
2857 CFS_INIT_LIST_HEAD(&oap->oap_pending_item);
2858 CFS_INIT_LIST_HEAD(&oap->oap_urgent_item);
2859 CFS_INIT_LIST_HEAD(&oap->oap_rpc_item);
2860 CFS_INIT_LIST_HEAD(&oap->oap_page_list);
2862 spin_lock_init(&oap->oap_lock);
2863 CDEBUG(D_CACHE, "oap %p page %p obj off "LPU64"\n", oap, page, offset);
2867 struct osc_async_page *oap_from_cookie(void *cookie)
2869 struct osc_async_page *oap = cookie;
2870 if (oap->oap_magic != OAP_MAGIC)
2871 return ERR_PTR(-EINVAL);
2875 int osc_queue_async_io(const struct lu_env *env,
2876 struct obd_export *exp, struct lov_stripe_md *lsm,
2877 struct lov_oinfo *loi, void *cookie,
2878 int cmd, obd_off off, int count,
2879 obd_flag brw_flags, enum async_flags async_flags)
2881 struct client_obd *cli = &exp->exp_obd->u.cli;
2882 struct osc_async_page *oap;
2886 oap = oap_from_cookie(cookie);
2888 RETURN(PTR_ERR(oap));
2890 if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
2893 if (!list_empty(&oap->oap_pending_item) ||
2894 !list_empty(&oap->oap_urgent_item) ||
2895 !list_empty(&oap->oap_rpc_item))
2898 /* check if the file's owner/group is over quota */
2899 if ((cmd & OBD_BRW_WRITE) && !(cmd & OBD_BRW_NOQUOTA)) {
2900 struct cl_object *obj;
2901 struct cl_attr attr; /* XXX put attr into thread info */
2902 unsigned int qid[MAXQUOTAS];
2904 obj = cl_object_top(osc_oap2cl_page(oap)->cp_obj);
2906 cl_object_attr_lock(obj);
2907 rc = cl_object_attr_get(env, obj, &attr);
2908 cl_object_attr_unlock(obj);
2910 qid[USRQUOTA] = attr.cat_uid;
2911 qid[GRPQUOTA] = attr.cat_gid;
2913 lquota_chkdq(quota_interface, cli, qid) == NO_QUOTA)
2920 loi = lsm->lsm_oinfo[0];
2922 client_obd_list_lock(&cli->cl_loi_list_lock);
2924 LASSERT(off + count <= CFS_PAGE_SIZE);
2926 oap->oap_page_off = off;
2927 oap->oap_count = count;
2928 oap->oap_brw_flags = brw_flags;
2929 /* Give a hint to OST that requests are coming from kswapd - bug19529 */
2930 if (libcfs_memory_pressure_get())
2931 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
2932 spin_lock(&oap->oap_lock);
2933 oap->oap_async_flags = async_flags;
2934 spin_unlock(&oap->oap_lock);
2936 if (cmd & OBD_BRW_WRITE) {
2937 rc = osc_enter_cache(env, cli, loi, oap);
2939 client_obd_list_unlock(&cli->cl_loi_list_lock);
2944 osc_oap_to_pending(oap);
2945 loi_list_maint(cli, loi);
2947 LOI_DEBUG(loi, "oap %p page %p added for cmd %d\n", oap, oap->oap_page,
2950 osc_check_rpcs(env, cli);
2951 client_obd_list_unlock(&cli->cl_loi_list_lock);
2956 /* aka (~was & now & flag), but this is more clear :) */
2957 #define SETTING(was, now, flag) (!(was & flag) && (now & flag))
2959 int osc_set_async_flags_base(struct client_obd *cli,
2960 struct lov_oinfo *loi, struct osc_async_page *oap,
2961 obd_flag async_flags)
2963 struct loi_oap_pages *lop;
2967 LASSERT(!list_empty(&oap->oap_pending_item));
2969 if (oap->oap_cmd & OBD_BRW_WRITE) {
2970 lop = &loi->loi_write_lop;
2972 lop = &loi->loi_read_lop;
2975 if ((oap->oap_async_flags & async_flags) == async_flags)
2978 if (SETTING(oap->oap_async_flags, async_flags, ASYNC_READY))
2979 flags |= ASYNC_READY;
2981 if (SETTING(oap->oap_async_flags, async_flags, ASYNC_URGENT) &&
2982 list_empty(&oap->oap_rpc_item)) {
2983 if (oap->oap_async_flags & ASYNC_HP)
2984 list_add(&oap->oap_urgent_item, &lop->lop_urgent);
2986 list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
2987 flags |= ASYNC_URGENT;
2988 loi_list_maint(cli, loi);
2990 spin_lock(&oap->oap_lock);
2991 oap->oap_async_flags |= flags;
2992 spin_unlock(&oap->oap_lock);
2994 LOI_DEBUG(loi, "oap %p page %p has flags %x\n", oap, oap->oap_page,
2995 oap->oap_async_flags);
2999 int osc_teardown_async_page(struct obd_export *exp,
3000 struct lov_stripe_md *lsm,
3001 struct lov_oinfo *loi, void *cookie)
3003 struct client_obd *cli = &exp->exp_obd->u.cli;
3004 struct loi_oap_pages *lop;
3005 struct osc_async_page *oap;
3009 oap = oap_from_cookie(cookie);
3011 RETURN(PTR_ERR(oap));
3014 loi = lsm->lsm_oinfo[0];
3016 if (oap->oap_cmd & OBD_BRW_WRITE) {
3017 lop = &loi->loi_write_lop;
3019 lop = &loi->loi_read_lop;
3022 client_obd_list_lock(&cli->cl_loi_list_lock);
3024 if (!list_empty(&oap->oap_rpc_item))
3025 GOTO(out, rc = -EBUSY);
3027 osc_exit_cache(cli, oap, 0);
3028 osc_wake_cache_waiters(cli);
3030 if (!list_empty(&oap->oap_urgent_item)) {
3031 list_del_init(&oap->oap_urgent_item);
3032 spin_lock(&oap->oap_lock);
3033 oap->oap_async_flags &= ~(ASYNC_URGENT | ASYNC_HP);
3034 spin_unlock(&oap->oap_lock);
3036 if (!list_empty(&oap->oap_pending_item)) {
3037 list_del_init(&oap->oap_pending_item);
3038 lop_update_pending(cli, lop, oap->oap_cmd, -1);
3040 loi_list_maint(cli, loi);
3041 LOI_DEBUG(loi, "oap %p page %p torn down\n", oap, oap->oap_page);
3043 client_obd_list_unlock(&cli->cl_loi_list_lock);
3047 static void osc_set_lock_data_with_check(struct ldlm_lock *lock,
3048 struct ldlm_enqueue_info *einfo,
3051 void *data = einfo->ei_cbdata;
3053 LASSERT(lock != NULL);
3054 LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl);
3055 LASSERT(lock->l_resource->lr_type == einfo->ei_type);
3056 LASSERT(lock->l_completion_ast == einfo->ei_cb_cp);
3057 LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
3059 lock_res_and_lock(lock);
3060 spin_lock(&osc_ast_guard);
3061 LASSERT(lock->l_ast_data == NULL || lock->l_ast_data == data);
3062 lock->l_ast_data = data;
3063 spin_unlock(&osc_ast_guard);
3064 unlock_res_and_lock(lock);
3067 static void osc_set_data_with_check(struct lustre_handle *lockh,
3068 struct ldlm_enqueue_info *einfo,
3071 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
3074 osc_set_lock_data_with_check(lock, einfo, flags);
3075 LDLM_LOCK_PUT(lock);
3077 CERROR("lockh %p, data %p - client evicted?\n",
3078 lockh, einfo->ei_cbdata);
3081 static int osc_change_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
3082 ldlm_iterator_t replace, void *data)
3084 struct ldlm_res_id res_id;
3085 struct obd_device *obd = class_exp2obd(exp);
3087 osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_gr, &res_id);
3088 ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
3092 static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb,
3093 obd_enqueue_update_f upcall, void *cookie,
3096 int intent = *flags & LDLM_FL_HAS_INTENT;
3100 /* The request was created before ldlm_cli_enqueue call. */
3101 if (rc == ELDLM_LOCK_ABORTED) {
3102 struct ldlm_reply *rep;
3103 rep = req_capsule_server_get(&req->rq_pill,
3106 LASSERT(rep != NULL);
3107 if (rep->lock_policy_res1)
3108 rc = rep->lock_policy_res1;
3112 if ((intent && rc == ELDLM_LOCK_ABORTED) || !rc) {
3113 *flags |= LDLM_FL_LVB_READY;
3114 CDEBUG(D_INODE,"got kms "LPU64" blocks "LPU64" mtime "LPU64"\n",
3115 lvb->lvb_size, lvb->lvb_blocks, lvb->lvb_mtime);
3118 /* Call the update callback. */
3119 rc = (*upcall)(cookie, rc);
3123 static int osc_enqueue_interpret(const struct lu_env *env,
3124 struct ptlrpc_request *req,
3125 struct osc_enqueue_args *aa, int rc)
3127 struct ldlm_lock *lock;
3128 struct lustre_handle handle;
3131 /* Make a local copy of a lock handle and a mode, because aa->oa_*
3132 * might be freed anytime after lock upcall has been called. */
3133 lustre_handle_copy(&handle, aa->oa_lockh);
3134 mode = aa->oa_ei->ei_mode;
3136 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
3138 lock = ldlm_handle2lock(&handle);
3140 /* Take an additional reference so that a blocking AST that
3141 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
3142 * to arrive after an upcall has been executed by
3143 * osc_enqueue_fini(). */
3144 ldlm_lock_addref(&handle, mode);
3146 /* Complete obtaining the lock procedure. */
3147 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1,
3148 mode, aa->oa_flags, aa->oa_lvb,
3149 sizeof(*aa->oa_lvb), lustre_swab_ost_lvb,
3151 /* Complete osc stuff. */
3152 rc = osc_enqueue_fini(req, aa->oa_lvb,
3153 aa->oa_upcall, aa->oa_cookie, aa->oa_flags, rc);
3155 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
3157 /* Release the lock for async request. */
3158 if (lustre_handle_is_used(&handle) && rc == ELDLM_OK)
3160 * Releases a reference taken by ldlm_cli_enqueue(), if it is
3161 * not already released by
3162 * ldlm_cli_enqueue_fini()->failed_lock_cleanup()
3164 ldlm_lock_decref(&handle, mode);
3166 LASSERTF(lock != NULL, "lockh %p, req %p, aa %p - client evicted?\n",
3167 aa->oa_lockh, req, aa);
3168 ldlm_lock_decref(&handle, mode);
3169 LDLM_LOCK_PUT(lock);
3173 void osc_update_enqueue(struct lustre_handle *lov_lockhp,
3174 struct lov_oinfo *loi, int flags,
3175 struct ost_lvb *lvb, __u32 mode, int rc)
3177 if (rc == ELDLM_OK) {
3178 struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
3181 LASSERT(lock != NULL);
3182 loi->loi_lvb = *lvb;
3183 tmp = loi->loi_lvb.lvb_size;
3184 /* Extend KMS up to the end of this lock and no further
3185 * A lock on [x,y] means a KMS of up to y + 1 bytes! */
3186 if (tmp > lock->l_policy_data.l_extent.end)
3187 tmp = lock->l_policy_data.l_extent.end + 1;
3188 if (tmp >= loi->loi_kms) {
3189 LDLM_DEBUG(lock, "lock acquired, setting rss="LPU64
3190 ", kms="LPU64, loi->loi_lvb.lvb_size, tmp);
3191 loi_kms_set(loi, tmp);
3193 LDLM_DEBUG(lock, "lock acquired, setting rss="
3194 LPU64"; leaving kms="LPU64", end="LPU64,
3195 loi->loi_lvb.lvb_size, loi->loi_kms,
3196 lock->l_policy_data.l_extent.end);
3198 ldlm_lock_allow_match(lock);
3199 LDLM_LOCK_PUT(lock);
3200 } else if (rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT)) {
3201 loi->loi_lvb = *lvb;
3202 CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
3203 " kms="LPU64"\n", loi->loi_lvb.lvb_size, loi->loi_kms);
3207 EXPORT_SYMBOL(osc_update_enqueue);
3209 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
3211 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
3212 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
3213 * other synchronous requests, however keeping some locks and trying to obtain
3214 * others may take a considerable amount of time in a case of ost failure; and
3215 * when other sync requests do not get released lock from a client, the client
3216 * is excluded from the cluster -- such scenarious make the life difficult, so
3217 * release locks just after they are obtained. */
3218 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
3219 int *flags, ldlm_policy_data_t *policy,
3220 struct ost_lvb *lvb, int kms_valid,
3221 obd_enqueue_update_f upcall, void *cookie,
3222 struct ldlm_enqueue_info *einfo,
3223 struct lustre_handle *lockh,
3224 struct ptlrpc_request_set *rqset, int async)
3226 struct obd_device *obd = exp->exp_obd;
3227 struct ptlrpc_request *req = NULL;
3228 int intent = *flags & LDLM_FL_HAS_INTENT;
3233 /* Filesystem lock extents are extended to page boundaries so that
3234 * dealing with the page cache is a little smoother. */
3235 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
3236 policy->l_extent.end |= ~CFS_PAGE_MASK;
3239 * kms is not valid when either object is completely fresh (so that no
3240 * locks are cached), or object was evicted. In the latter case cached
3241 * lock cannot be used, because it would prime inode state with
3242 * potentially stale LVB.
3247 /* Next, search for already existing extent locks that will cover us */
3248 /* If we're trying to read, we also search for an existing PW lock. The
3249 * VFS and page cache already protect us locally, so lots of readers/
3250 * writers can share a single PW lock.
3252 * There are problems with conversion deadlocks, so instead of
3253 * converting a read lock to a write lock, we'll just enqueue a new
3256 * At some point we should cancel the read lock instead of making them
3257 * send us a blocking callback, but there are problems with canceling
3258 * locks out from other users right now, too. */
3259 mode = einfo->ei_mode;
3260 if (einfo->ei_mode == LCK_PR)
3262 mode = ldlm_lock_match(obd->obd_namespace,
3263 *flags | LDLM_FL_LVB_READY, res_id,
3264 einfo->ei_type, policy, mode, lockh, 0);
3266 struct ldlm_lock *matched = ldlm_handle2lock(lockh);
3268 if (matched->l_ast_data == NULL ||
3269 matched->l_ast_data == einfo->ei_cbdata) {
3270 /* addref the lock only if not async requests and PW
3271 * lock is matched whereas we asked for PR. */
3272 if (!rqset && einfo->ei_mode != mode)
3273 ldlm_lock_addref(lockh, LCK_PR);
3274 osc_set_lock_data_with_check(matched, einfo, *flags);
3276 /* I would like to be able to ASSERT here that
3277 * rss <= kms, but I can't, for reasons which
3278 * are explained in lov_enqueue() */
3281 /* We already have a lock, and it's referenced */
3282 (*upcall)(cookie, ELDLM_OK);
3284 /* For async requests, decref the lock. */
3285 if (einfo->ei_mode != mode)
3286 ldlm_lock_decref(lockh, LCK_PW);
3288 ldlm_lock_decref(lockh, einfo->ei_mode);
3289 LDLM_LOCK_PUT(matched);
3292 ldlm_lock_decref(lockh, mode);
3293 LDLM_LOCK_PUT(matched);
3298 CFS_LIST_HEAD(cancels);
3299 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
3300 &RQF_LDLM_ENQUEUE_LVB);
3304 rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0);
3308 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
3310 ptlrpc_request_set_replen(req);
3313 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
3314 *flags &= ~LDLM_FL_BLOCK_GRANTED;
3316 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
3317 sizeof(*lvb), lustre_swab_ost_lvb, lockh, async);
3320 struct osc_enqueue_args *aa;
3321 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
3322 aa = ptlrpc_req_async_args(req);
3325 aa->oa_flags = flags;
3326 aa->oa_upcall = upcall;
3327 aa->oa_cookie = cookie;
3329 aa->oa_lockh = lockh;
3331 req->rq_interpret_reply =
3332 (ptlrpc_interpterer_t)osc_enqueue_interpret;
3333 if (rqset == PTLRPCD_SET)
3334 ptlrpcd_add_req(req, PSCOPE_OTHER);
3336 ptlrpc_set_add_req(rqset, req);
3337 } else if (intent) {
3338 ptlrpc_req_finished(req);
3343 rc = osc_enqueue_fini(req, lvb, upcall, cookie, flags, rc);
3345 ptlrpc_req_finished(req);
3350 static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
3351 struct ldlm_enqueue_info *einfo,
3352 struct ptlrpc_request_set *rqset)
3354 struct ldlm_res_id res_id;
3358 osc_build_res_name(oinfo->oi_md->lsm_object_id,
3359 oinfo->oi_md->lsm_object_gr, &res_id);
3361 rc = osc_enqueue_base(exp, &res_id, &oinfo->oi_flags, &oinfo->oi_policy,
3362 &oinfo->oi_md->lsm_oinfo[0]->loi_lvb,
3363 oinfo->oi_md->lsm_oinfo[0]->loi_kms_valid,
3364 oinfo->oi_cb_up, oinfo, einfo, oinfo->oi_lockh,
3365 rqset, rqset != NULL);
3369 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
3370 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
3371 int *flags, void *data, struct lustre_handle *lockh,
3374 struct obd_device *obd = exp->exp_obd;
3375 int lflags = *flags;
3379 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
3382 /* Filesystem lock extents are extended to page boundaries so that
3383 * dealing with the page cache is a little smoother */
3384 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
3385 policy->l_extent.end |= ~CFS_PAGE_MASK;
3387 /* Next, search for already existing extent locks that will cover us */
3388 /* If we're trying to read, we also search for an existing PW lock. The
3389 * VFS and page cache already protect us locally, so lots of readers/
3390 * writers can share a single PW lock. */
3394 rc = ldlm_lock_match(obd->obd_namespace, lflags,
3395 res_id, type, policy, rc, lockh, unref);
3398 osc_set_data_with_check(lockh, data, lflags);
3399 if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) {
3400 ldlm_lock_addref(lockh, LCK_PR);
3401 ldlm_lock_decref(lockh, LCK_PW);
3408 int osc_cancel_base(struct lustre_handle *lockh, __u32 mode)
3412 if (unlikely(mode == LCK_GROUP))
3413 ldlm_lock_decref_and_cancel(lockh, mode);
3415 ldlm_lock_decref(lockh, mode);
3420 static int osc_cancel(struct obd_export *exp, struct lov_stripe_md *md,
3421 __u32 mode, struct lustre_handle *lockh)
3424 RETURN(osc_cancel_base(lockh, mode));
3427 static int osc_cancel_unused(struct obd_export *exp,
3428 struct lov_stripe_md *lsm, int flags,
3431 struct obd_device *obd = class_exp2obd(exp);
3432 struct ldlm_res_id res_id, *resp = NULL;
3435 resp = osc_build_res_name(lsm->lsm_object_id,
3436 lsm->lsm_object_gr, &res_id);
3439 return ldlm_cli_cancel_unused(obd->obd_namespace, resp, flags, opaque);
3442 static int osc_statfs_interpret(const struct lu_env *env,
3443 struct ptlrpc_request *req,
3444 struct osc_async_args *aa, int rc)
3446 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
3447 struct obd_statfs *msfs;
3452 /* The request has in fact never been sent
3453 * due to issues at a higher level (LOV).
3454 * Exit immediately since the caller is
3455 * aware of the problem and takes care
3456 * of the clean up */
3459 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
3460 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
3466 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
3468 GOTO(out, rc = -EPROTO);
3471 /* Reinitialize the RDONLY and DEGRADED flags at the client
3472 * on each statfs, so they don't stay set permanently. */
3473 spin_lock(&cli->cl_oscc.oscc_lock);
3475 if (unlikely(msfs->os_state & OS_STATE_DEGRADED))
3476 cli->cl_oscc.oscc_flags |= OSCC_FLAG_DEGRADED;
3477 else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_DEGRADED))
3478 cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_DEGRADED;
3480 if (unlikely(msfs->os_state & OS_STATE_READONLY))
3481 cli->cl_oscc.oscc_flags |= OSCC_FLAG_RDONLY;
3482 else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_RDONLY))
3483 cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_RDONLY;
3485 /* Add a bit of hysteresis so this flag isn't continually flapping,
3486 * and ensure that new files don't get extremely fragmented due to
3487 * only a small amount of available space in the filesystem.
3488 * We want to set the NOSPC flag when there is less than ~0.1% free
3489 * and clear it when there is at least ~0.2% free space, so:
3490 * avail < ~0.1% max max = avail + used
3491 * 1025 * avail < avail + used used = blocks - free
3492 * 1024 * avail < used
3493 * 1024 * avail < blocks - free
3494 * avail < ((blocks - free) >> 10)
3496 * On very large disk, say 16TB 0.1% will be 16 GB. We don't want to
3497 * lose that amount of space so in those cases we report no space left
3498 * if their is less than 1 GB left. */
3499 used = min((msfs->os_blocks - msfs->os_bfree) >> 10, 1ULL << 30);
3500 if (unlikely(((cli->cl_oscc.oscc_flags & OSCC_FLAG_NOSPC) == 0) &&
3501 ((msfs->os_ffree < 32) || (msfs->os_bavail < used))))
3502 cli->cl_oscc.oscc_flags |= OSCC_FLAG_NOSPC;
3503 else if (unlikely(((cli->cl_oscc.oscc_flags & OSCC_FLAG_NOSPC) != 0) &&
3504 (msfs->os_ffree > 64) && (msfs->os_bavail > (used << 1))))
3505 cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_NOSPC;
3507 spin_unlock(&cli->cl_oscc.oscc_lock);
3509 *aa->aa_oi->oi_osfs = *msfs;
3511 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
3515 static int osc_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
3516 __u64 max_age, struct ptlrpc_request_set *rqset)
3518 struct ptlrpc_request *req;
3519 struct osc_async_args *aa;
3523 /* We could possibly pass max_age in the request (as an absolute
3524 * timestamp or a "seconds.usec ago") so the target can avoid doing
3525 * extra calls into the filesystem if that isn't necessary (e.g.
3526 * during mount that would help a bit). Having relative timestamps
3527 * is not so great if request processing is slow, while absolute
3528 * timestamps are not ideal because they need time synchronization. */
3529 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
3533 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
3535 ptlrpc_request_free(req);
3538 ptlrpc_request_set_replen(req);
3539 req->rq_request_portal = OST_CREATE_PORTAL;
3540 ptlrpc_at_set_req_timeout(req);
3542 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
3543 /* procfs requests not want stat in wait for avoid deadlock */
3544 req->rq_no_resend = 1;
3545 req->rq_no_delay = 1;
3548 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
3549 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
3550 aa = ptlrpc_req_async_args(req);
3553 ptlrpc_set_add_req(rqset, req);
3557 static int osc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
3558 __u64 max_age, __u32 flags)
3560 struct obd_statfs *msfs;
3561 struct ptlrpc_request *req;
3562 struct obd_import *imp = NULL;
3566 /*Since the request might also come from lprocfs, so we need
3567 *sync this with client_disconnect_export Bug15684*/
3568 down_read(&obd->u.cli.cl_sem);
3569 if (obd->u.cli.cl_import)
3570 imp = class_import_get(obd->u.cli.cl_import);
3571 up_read(&obd->u.cli.cl_sem);
3575 /* We could possibly pass max_age in the request (as an absolute
3576 * timestamp or a "seconds.usec ago") so the target can avoid doing
3577 * extra calls into the filesystem if that isn't necessary (e.g.
3578 * during mount that would help a bit). Having relative timestamps
3579 * is not so great if request processing is slow, while absolute
3580 * timestamps are not ideal because they need time synchronization. */
3581 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
3583 class_import_put(imp);
3588 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
3590 ptlrpc_request_free(req);
3593 ptlrpc_request_set_replen(req);
3594 req->rq_request_portal = OST_CREATE_PORTAL;
3595 ptlrpc_at_set_req_timeout(req);
3597 if (flags & OBD_STATFS_NODELAY) {
3598 /* procfs requests not want stat in wait for avoid deadlock */
3599 req->rq_no_resend = 1;
3600 req->rq_no_delay = 1;
3603 rc = ptlrpc_queue_wait(req);
3607 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
3609 GOTO(out, rc = -EPROTO);
3616 ptlrpc_req_finished(req);
3620 /* Retrieve object striping information.
3622 * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating
3623 * the maximum number of OST indices which will fit in the user buffer.
3624 * lmm_magic must be LOV_MAGIC (we only use 1 slot here).
3626 static int osc_getstripe(struct lov_stripe_md *lsm, struct lov_user_md *lump)
3628 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
3629 struct lov_user_md_v3 lum, *lumk;
3630 struct lov_user_ost_data_v1 *lmm_objects;
3631 int rc = 0, lum_size;
3637 /* we only need the header part from user space to get lmm_magic and
3638 * lmm_stripe_count, (the header part is common to v1 and v3) */
3639 lum_size = sizeof(struct lov_user_md_v1);
3640 if (copy_from_user(&lum, lump, lum_size))
3643 if ((lum.lmm_magic != LOV_USER_MAGIC_V1) &&
3644 (lum.lmm_magic != LOV_USER_MAGIC_V3))
3647 /* lov_user_md_vX and lov_mds_md_vX must have the same size */
3648 LASSERT(sizeof(struct lov_user_md_v1) == sizeof(struct lov_mds_md_v1));
3649 LASSERT(sizeof(struct lov_user_md_v3) == sizeof(struct lov_mds_md_v3));
3650 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lumk->lmm_objects[0]));
3652 /* we can use lov_mds_md_size() to compute lum_size
3653 * because lov_user_md_vX and lov_mds_md_vX have the same size */
3654 if (lum.lmm_stripe_count > 0) {
3655 lum_size = lov_mds_md_size(lum.lmm_stripe_count, lum.lmm_magic);
3656 OBD_ALLOC(lumk, lum_size);
3660 if (lum.lmm_magic == LOV_USER_MAGIC_V1)
3661 lmm_objects = &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
3663 lmm_objects = &(lumk->lmm_objects[0]);
3664 lmm_objects->l_object_id = lsm->lsm_object_id;
3666 lum_size = lov_mds_md_size(0, lum.lmm_magic);
3670 lumk->lmm_object_id = lsm->lsm_object_id;
3671 lumk->lmm_object_gr = lsm->lsm_object_gr;
3672 lumk->lmm_stripe_count = 1;
3674 if (copy_to_user(lump, lumk, lum_size))
3678 OBD_FREE(lumk, lum_size);
3684 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
3685 void *karg, void *uarg)
3687 struct obd_device *obd = exp->exp_obd;
3688 struct obd_ioctl_data *data = karg;
3692 if (!try_module_get(THIS_MODULE)) {
3693 CERROR("Can't get module. Is it alive?");
3697 case OBD_IOC_LOV_GET_CONFIG: {
3699 struct lov_desc *desc;
3700 struct obd_uuid uuid;
3704 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
3705 GOTO(out, err = -EINVAL);
3707 data = (struct obd_ioctl_data *)buf;
3709 if (sizeof(*desc) > data->ioc_inllen1) {
3710 obd_ioctl_freedata(buf, len);
3711 GOTO(out, err = -EINVAL);
3714 if (data->ioc_inllen2 < sizeof(uuid)) {
3715 obd_ioctl_freedata(buf, len);
3716 GOTO(out, err = -EINVAL);
3719 desc = (struct lov_desc *)data->ioc_inlbuf1;
3720 desc->ld_tgt_count = 1;
3721 desc->ld_active_tgt_count = 1;
3722 desc->ld_default_stripe_count = 1;
3723 desc->ld_default_stripe_size = 0;
3724 desc->ld_default_stripe_offset = 0;
3725 desc->ld_pattern = 0;
3726 memcpy(&desc->ld_uuid, &obd->obd_uuid, sizeof(uuid));
3728 memcpy(data->ioc_inlbuf2, &obd->obd_uuid, sizeof(uuid));
3730 err = copy_to_user((void *)uarg, buf, len);
3733 obd_ioctl_freedata(buf, len);
3736 case LL_IOC_LOV_SETSTRIPE:
3737 err = obd_alloc_memmd(exp, karg);
3741 case LL_IOC_LOV_GETSTRIPE:
3742 err = osc_getstripe(karg, uarg);
3744 case OBD_IOC_CLIENT_RECOVER:
3745 err = ptlrpc_recover_import(obd->u.cli.cl_import,
3750 case IOC_OSC_SET_ACTIVE:
3751 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
3754 case OBD_IOC_POLL_QUOTACHECK:
3755 err = lquota_poll_check(quota_interface, exp,
3756 (struct if_quotacheck *)karg);
3758 case OBD_IOC_PING_TARGET:
3759 err = ptlrpc_obd_ping(obd);
3762 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
3763 cmd, cfs_curproc_comm());
3764 GOTO(out, err = -ENOTTY);
3767 module_put(THIS_MODULE);
3771 static int osc_get_info(struct obd_export *exp, obd_count keylen,
3772 void *key, __u32 *vallen, void *val,
3773 struct lov_stripe_md *lsm)
3776 if (!vallen || !val)
3779 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
3780 __u32 *stripe = val;
3781 *vallen = sizeof(*stripe);
3784 } else if (KEY_IS(KEY_LAST_ID)) {
3785 struct ptlrpc_request *req;
3790 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
3791 &RQF_OST_GET_INFO_LAST_ID);
3795 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
3796 RCL_CLIENT, keylen);
3797 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
3799 ptlrpc_request_free(req);
3803 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
3804 memcpy(tmp, key, keylen);
3806 req->rq_no_delay = req->rq_no_resend = 1;
3807 ptlrpc_request_set_replen(req);
3808 rc = ptlrpc_queue_wait(req);
3812 reply = req_capsule_server_get(&req->rq_pill, &RMF_OBD_ID);
3814 GOTO(out, rc = -EPROTO);
3816 *((obd_id *)val) = *reply;
3818 ptlrpc_req_finished(req);
3820 } else if (KEY_IS(KEY_FIEMAP)) {
3821 struct ptlrpc_request *req;
3822 struct ll_user_fiemap *reply;
3826 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
3827 &RQF_OST_GET_INFO_FIEMAP);
3831 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY,
3832 RCL_CLIENT, keylen);
3833 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
3834 RCL_CLIENT, *vallen);
3835 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
3836 RCL_SERVER, *vallen);
3838 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
3840 ptlrpc_request_free(req);
3844 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
3845 memcpy(tmp, key, keylen);
3846 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
3847 memcpy(tmp, val, *vallen);
3849 ptlrpc_request_set_replen(req);
3850 rc = ptlrpc_queue_wait(req);
3854 reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
3856 GOTO(out1, rc = -EPROTO);
3858 memcpy(val, reply, *vallen);
3860 ptlrpc_req_finished(req);
3868 static int osc_setinfo_mds_connect_import(struct obd_import *imp)
3870 struct llog_ctxt *ctxt;
3874 ctxt = llog_get_context(imp->imp_obd, LLOG_MDS_OST_ORIG_CTXT);
3876 rc = llog_initiator_connect(ctxt);
3877 llog_ctxt_put(ctxt);
3879 /* XXX return an error? skip setting below flags? */
3882 spin_lock(&imp->imp_lock);
3883 imp->imp_server_timeout = 1;
3884 imp->imp_pingable = 1;
3885 spin_unlock(&imp->imp_lock);
3886 CDEBUG(D_RPCTRACE, "pinging OST %s\n", obd2cli_tgt(imp->imp_obd));
3891 static int osc_setinfo_mds_conn_interpret(const struct lu_env *env,
3892 struct ptlrpc_request *req,
3899 RETURN(osc_setinfo_mds_connect_import(req->rq_import));
3902 static int osc_set_info_async(struct obd_export *exp, obd_count keylen,
3903 void *key, obd_count vallen, void *val,
3904 struct ptlrpc_request_set *set)
3906 struct ptlrpc_request *req;
3907 struct obd_device *obd = exp->exp_obd;
3908 struct obd_import *imp = class_exp2cliimp(exp);
3913 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
3915 if (KEY_IS(KEY_NEXT_ID)) {
3917 struct osc_creator *oscc = &obd->u.cli.cl_oscc;
3919 if (vallen != sizeof(obd_id))
3924 if (vallen != sizeof(obd_id))
3927 /* avoid race between allocate new object and set next id
3928 * from ll_sync thread */
3929 spin_lock(&oscc->oscc_lock);
3930 new_val = *((obd_id*)val) + 1;
3931 if (new_val > oscc->oscc_next_id)
3932 oscc->oscc_next_id = new_val;
3933 spin_unlock(&oscc->oscc_lock);
3934 CDEBUG(D_HA, "%s: set oscc_next_id = "LPU64"\n",
3935 exp->exp_obd->obd_name,
3936 obd->u.cli.cl_oscc.oscc_next_id);
3941 if (KEY_IS(KEY_INIT_RECOV)) {
3942 if (vallen != sizeof(int))
3944 spin_lock(&imp->imp_lock);
3945 imp->imp_initial_recov = *(int *)val;
3946 spin_unlock(&imp->imp_lock);
3947 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
3948 exp->exp_obd->obd_name,
3949 imp->imp_initial_recov);
3953 if (KEY_IS(KEY_CHECKSUM)) {
3954 if (vallen != sizeof(int))
3956 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
3960 if (KEY_IS(KEY_SPTLRPC_CONF)) {
3961 sptlrpc_conf_client_adapt(obd);
3965 if (KEY_IS(KEY_FLUSH_CTX)) {
3966 sptlrpc_import_flush_my_ctx(imp);
3970 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
3973 /* We pass all other commands directly to OST. Since nobody calls osc
3974 methods directly and everybody is supposed to go through LOV, we
3975 assume lov checked invalid values for us.
3976 The only recognised values so far are evict_by_nid and mds_conn.
3977 Even if something bad goes through, we'd get a -EINVAL from OST
3980 if (KEY_IS(KEY_GRANT_SHRINK))
3981 req = ptlrpc_request_alloc(imp, &RQF_OST_SET_GRANT_INFO);
3983 req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
3988 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
3989 RCL_CLIENT, keylen);
3990 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
3991 RCL_CLIENT, vallen);
3992 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
3994 ptlrpc_request_free(req);
3998 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
3999 memcpy(tmp, key, keylen);
4000 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
4001 memcpy(tmp, val, vallen);
4003 if (KEY_IS(KEY_MDS_CONN)) {
4004 struct osc_creator *oscc = &obd->u.cli.cl_oscc;
4006 oscc->oscc_oa.o_gr = (*(__u32 *)val);
4007 oscc->oscc_oa.o_valid |= OBD_MD_FLGROUP;
4008 LASSERT_MDS_GROUP(oscc->oscc_oa.o_gr);
4009 req->rq_no_delay = req->rq_no_resend = 1;
4010 req->rq_interpret_reply = osc_setinfo_mds_conn_interpret;
4011 } else if (KEY_IS(KEY_GRANT_SHRINK)) {
4012 struct osc_grant_args *aa;
4015 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
4016 aa = ptlrpc_req_async_args(req);
4019 ptlrpc_req_finished(req);
4022 *oa = ((struct ost_body *)val)->oa;
4024 req->rq_interpret_reply = osc_shrink_grant_interpret;
4027 ptlrpc_request_set_replen(req);
4028 if (!KEY_IS(KEY_GRANT_SHRINK)) {
4029 LASSERT(set != NULL);
4030 ptlrpc_set_add_req(set, req);
4031 ptlrpc_check_set(NULL, set);
4033 ptlrpcd_add_req(req, PSCOPE_OTHER);
4039 static struct llog_operations osc_size_repl_logops = {
4040 lop_cancel: llog_obd_repl_cancel
4043 static struct llog_operations osc_mds_ost_orig_logops;
4045 static int __osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
4046 struct obd_device *tgt, struct llog_catid *catid)
4051 rc = llog_setup(obd, &obd->obd_olg, LLOG_MDS_OST_ORIG_CTXT, tgt, 1,
4052 &catid->lci_logid, &osc_mds_ost_orig_logops);
4054 CERROR("failed LLOG_MDS_OST_ORIG_CTXT\n");
4058 rc = llog_setup(obd, &obd->obd_olg, LLOG_SIZE_REPL_CTXT, tgt, 1,
4059 NULL, &osc_size_repl_logops);
4061 struct llog_ctxt *ctxt =
4062 llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
4065 CERROR("failed LLOG_SIZE_REPL_CTXT\n");
4070 CERROR("osc '%s' tgt '%s' catid %p rc=%d\n",
4071 obd->obd_name, tgt->obd_name, catid, rc);
4072 CERROR("logid "LPX64":0x%x\n",
4073 catid->lci_logid.lgl_oid, catid->lci_logid.lgl_ogen);
4078 static int osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
4079 struct obd_device *disk_obd, int *index)
4081 struct llog_catid catid;
4082 static char name[32] = CATLIST;
4086 LASSERT(olg == &obd->obd_olg);
4088 mutex_down(&olg->olg_cat_processing);
4089 rc = llog_get_cat_list(disk_obd, name, *index, 1, &catid);
4091 CERROR("rc: %d\n", rc);
4095 CDEBUG(D_INFO, "%s: Init llog for %d - catid "LPX64"/"LPX64":%x\n",
4096 obd->obd_name, *index, catid.lci_logid.lgl_oid,
4097 catid.lci_logid.lgl_ogr, catid.lci_logid.lgl_ogen);
4099 rc = __osc_llog_init(obd, olg, disk_obd, &catid);
4101 CERROR("rc: %d\n", rc);
4105 rc = llog_put_cat_list(disk_obd, name, *index, 1, &catid);
4107 CERROR("rc: %d\n", rc);
4112 mutex_up(&olg->olg_cat_processing);
4117 static int osc_llog_finish(struct obd_device *obd, int count)
4119 struct llog_ctxt *ctxt;
4120 int rc = 0, rc2 = 0;
4123 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
4125 rc = llog_cleanup(ctxt);
4127 ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
4129 rc2 = llog_cleanup(ctxt);
4136 static int osc_reconnect(const struct lu_env *env,
4137 struct obd_export *exp, struct obd_device *obd,
4138 struct obd_uuid *cluuid,
4139 struct obd_connect_data *data,
4142 struct client_obd *cli = &obd->u.cli;
4144 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
4147 client_obd_list_lock(&cli->cl_loi_list_lock);
4148 data->ocd_grant = cli->cl_avail_grant ?:
4149 2 * cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
4150 lost_grant = cli->cl_lost_grant;
4151 cli->cl_lost_grant = 0;
4152 client_obd_list_unlock(&cli->cl_loi_list_lock);
4154 CDEBUG(D_CACHE, "request ocd_grant: %d cl_avail_grant: %ld "
4155 "cl_lost_grant: %ld\n", data->ocd_grant,
4156 cli->cl_avail_grant, lost_grant);
4157 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
4158 " ocd_grant: %d\n", data->ocd_connect_flags,
4159 data->ocd_version, data->ocd_grant);
4165 static int osc_disconnect(struct obd_export *exp)
4167 struct obd_device *obd = class_exp2obd(exp);
4168 struct llog_ctxt *ctxt;
4171 ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
4173 if (obd->u.cli.cl_conn_count == 1) {
4174 /* Flush any remaining cancel messages out to the
4176 llog_sync(ctxt, exp);
4178 llog_ctxt_put(ctxt);
4180 CDEBUG(D_HA, "No LLOG_SIZE_REPL_CTXT found in obd %p\n",
4184 rc = client_disconnect_export(exp);
4186 * Initially we put del_shrink_grant before disconnect_export, but it
4187 * causes the following problem if setup (connect) and cleanup
4188 * (disconnect) are tangled together.
4189 * connect p1 disconnect p2
4190 * ptlrpc_connect_import
4191 * ............... class_manual_cleanup
4194 * ptlrpc_connect_interrupt
4196 * add this client to shrink list
4198 * Bang! pinger trigger the shrink.
4199 * So the osc should be disconnected from the shrink list, after we
4200 * are sure the import has been destroyed. BUG18662
4202 if (obd->u.cli.cl_import == NULL)
4203 osc_del_shrink_grant(&obd->u.cli);
4207 static int osc_import_event(struct obd_device *obd,
4208 struct obd_import *imp,
4209 enum obd_import_event event)
4211 struct client_obd *cli;
4215 LASSERT(imp->imp_obd == obd);
4218 case IMP_EVENT_DISCON: {
4219 /* Only do this on the MDS OSC's */
4220 if (imp->imp_server_timeout) {
4221 struct osc_creator *oscc = &obd->u.cli.cl_oscc;
4223 spin_lock(&oscc->oscc_lock);
4224 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
4225 spin_unlock(&oscc->oscc_lock);
4228 client_obd_list_lock(&cli->cl_loi_list_lock);
4229 cli->cl_avail_grant = 0;
4230 cli->cl_lost_grant = 0;
4231 client_obd_list_unlock(&cli->cl_loi_list_lock);
4234 case IMP_EVENT_INACTIVE: {
4235 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
4238 case IMP_EVENT_INVALIDATE: {
4239 struct ldlm_namespace *ns = obd->obd_namespace;
4243 env = cl_env_get(&refcheck);
4247 client_obd_list_lock(&cli->cl_loi_list_lock);
4248 /* all pages go to failing rpcs due to the invalid
4250 osc_check_rpcs(env, cli);
4251 client_obd_list_unlock(&cli->cl_loi_list_lock);
4253 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
4254 cl_env_put(env, &refcheck);
4259 case IMP_EVENT_ACTIVE: {
4260 /* Only do this on the MDS OSC's */
4261 if (imp->imp_server_timeout) {
4262 struct osc_creator *oscc = &obd->u.cli.cl_oscc;
4264 spin_lock(&oscc->oscc_lock);
4265 oscc->oscc_flags &= ~OSCC_FLAG_NOSPC;
4266 spin_unlock(&oscc->oscc_lock);
4268 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
4271 case IMP_EVENT_OCD: {
4272 struct obd_connect_data *ocd = &imp->imp_connect_data;
4274 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
4275 osc_init_grant(&obd->u.cli, ocd);
4278 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
4279 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
4281 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
4285 CERROR("Unknown import event %d\n", event);
4291 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
4297 rc = ptlrpcd_addref();
4301 rc = client_obd_setup(obd, lcfg);
4305 struct lprocfs_static_vars lvars = { 0 };
4306 struct client_obd *cli = &obd->u.cli;
4308 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
4309 lprocfs_osc_init_vars(&lvars);
4310 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
4311 lproc_osc_attach_seqstat(obd);
4312 sptlrpc_lprocfs_cliobd_attach(obd);
4313 ptlrpc_lprocfs_register_obd(obd);
4317 /* We need to allocate a few requests more, because
4318 brw_interpret tries to create new requests before freeing
4319 previous ones. Ideally we want to have 2x max_rpcs_in_flight
4320 reserved, but I afraid that might be too much wasted RAM
4321 in fact, so 2 is just my guess and still should work. */
4322 cli->cl_import->imp_rq_pool =
4323 ptlrpc_init_rq_pool(cli->cl_max_rpcs_in_flight + 2,
4325 ptlrpc_add_rqs_to_pool);
4327 CFS_INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
4328 sema_init(&cli->cl_grant_sem, 1);
4334 static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
4340 case OBD_CLEANUP_EARLY: {
4341 struct obd_import *imp;
4342 imp = obd->u.cli.cl_import;
4343 CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name);
4344 /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */
4345 ptlrpc_deactivate_import(imp);
4346 spin_lock(&imp->imp_lock);
4347 imp->imp_pingable = 0;
4348 spin_unlock(&imp->imp_lock);
4351 case OBD_CLEANUP_EXPORTS: {
4352 /* If we set up but never connected, the
4353 client import will not have been cleaned. */
4354 if (obd->u.cli.cl_import) {
4355 struct obd_import *imp;
4356 down_write(&obd->u.cli.cl_sem);
4357 imp = obd->u.cli.cl_import;
4358 CDEBUG(D_CONFIG, "%s: client import never connected\n",
4360 ptlrpc_invalidate_import(imp);
4361 if (imp->imp_rq_pool) {
4362 ptlrpc_free_rq_pool(imp->imp_rq_pool);
4363 imp->imp_rq_pool = NULL;
4365 class_destroy_import(imp);
4366 up_write(&obd->u.cli.cl_sem);
4367 obd->u.cli.cl_import = NULL;
4369 rc = obd_llog_finish(obd, 0);
4371 CERROR("failed to cleanup llogging subsystems\n");
4378 int osc_cleanup(struct obd_device *obd)
4383 ptlrpc_lprocfs_unregister_obd(obd);
4384 lprocfs_obd_cleanup(obd);
4386 /* free memory of osc quota cache */
4387 lquota_cleanup(quota_interface, obd);
4389 rc = client_obd_cleanup(obd);
4395 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
4397 struct lprocfs_static_vars lvars = { 0 };
4400 lprocfs_osc_init_vars(&lvars);
4402 switch (lcfg->lcfg_command) {
4404 rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
4414 static int osc_process_config(struct obd_device *obd, obd_count len, void *buf)
4416 return osc_process_config_base(obd, buf);
4419 struct obd_ops osc_obd_ops = {
4420 .o_owner = THIS_MODULE,
4421 .o_setup = osc_setup,
4422 .o_precleanup = osc_precleanup,
4423 .o_cleanup = osc_cleanup,
4424 .o_add_conn = client_import_add_conn,
4425 .o_del_conn = client_import_del_conn,
4426 .o_connect = client_connect_import,
4427 .o_reconnect = osc_reconnect,
4428 .o_disconnect = osc_disconnect,
4429 .o_statfs = osc_statfs,
4430 .o_statfs_async = osc_statfs_async,
4431 .o_packmd = osc_packmd,
4432 .o_unpackmd = osc_unpackmd,
4433 .o_precreate = osc_precreate,
4434 .o_create = osc_create,
4435 .o_create_async = osc_create_async,
4436 .o_destroy = osc_destroy,
4437 .o_getattr = osc_getattr,
4438 .o_getattr_async = osc_getattr_async,
4439 .o_setattr = osc_setattr,
4440 .o_setattr_async = osc_setattr_async,
4442 .o_punch = osc_punch,
4444 .o_enqueue = osc_enqueue,
4445 .o_change_cbdata = osc_change_cbdata,
4446 .o_cancel = osc_cancel,
4447 .o_cancel_unused = osc_cancel_unused,
4448 .o_iocontrol = osc_iocontrol,
4449 .o_get_info = osc_get_info,
4450 .o_set_info_async = osc_set_info_async,
4451 .o_import_event = osc_import_event,
4452 .o_llog_init = osc_llog_init,
4453 .o_llog_finish = osc_llog_finish,
4454 .o_process_config = osc_process_config,
4457 extern struct lu_kmem_descr osc_caches[];
4458 extern spinlock_t osc_ast_guard;
4459 extern struct lock_class_key osc_ast_guard_class;
4461 int __init osc_init(void)
4463 struct lprocfs_static_vars lvars = { 0 };
4467 /* print an address of _any_ initialized kernel symbol from this
4468 * module, to allow debugging with gdb that doesn't support data
4469 * symbols from modules.*/
4470 CDEBUG(D_CONSOLE, "Lustre OSC module (%p).\n", &osc_caches);
4472 rc = lu_kmem_init(osc_caches);
4474 lprocfs_osc_init_vars(&lvars);
4476 request_module("lquota");
4477 quota_interface = PORTAL_SYMBOL_GET(osc_quota_interface);
4478 lquota_init(quota_interface);
4479 init_obd_quota_ops(quota_interface, &osc_obd_ops);
4481 rc = class_register_type(&osc_obd_ops, NULL, lvars.module_vars,
4482 LUSTRE_OSC_NAME, &osc_device_type);
4484 if (quota_interface)
4485 PORTAL_SYMBOL_PUT(osc_quota_interface);
4486 lu_kmem_fini(osc_caches);
4490 spin_lock_init(&osc_ast_guard);
4491 lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class);
4493 osc_mds_ost_orig_logops = llog_lvfs_ops;
4494 osc_mds_ost_orig_logops.lop_setup = llog_obd_origin_setup;
4495 osc_mds_ost_orig_logops.lop_cleanup = llog_obd_origin_cleanup;
4496 osc_mds_ost_orig_logops.lop_add = llog_obd_origin_add;
4497 osc_mds_ost_orig_logops.lop_connect = llog_origin_connect;
4503 static void /*__exit*/ osc_exit(void)
4505 lu_device_type_fini(&osc_device_type);
4507 lquota_exit(quota_interface);
4508 if (quota_interface)
4509 PORTAL_SYMBOL_PUT(osc_quota_interface);
4511 class_unregister_type(LUSTRE_OSC_NAME);
4512 lu_kmem_fini(osc_caches);
4515 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4516 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
4517 MODULE_LICENSE("GPL");
4519 cfs_module(osc, LUSTRE_VERSION_STRING, osc_init, osc_exit);
git://git.whamcloud.com / fs / lustre-release.git / blob