4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_OSC
35 #include <linux/workqueue.h>
36 #include <lprocfs_status.h>
37 #include <lustre_debug.h>
38 #include <lustre_dlm.h>
39 #include <lustre_fid.h>
40 #include <lustre_ha.h>
41 #include <uapi/linux/lustre/lustre_ioctl.h>
42 #include <lustre_net.h>
43 #include <lustre_obdo.h>
45 #include <obd_cksum.h>
46 #include <obd_class.h>
47 #include <lustre_osc.h>
49 #include "osc_internal.h"
51 atomic_t osc_pool_req_count;
52 unsigned int osc_reqpool_maxreqcount;
53 struct ptlrpc_request_pool *osc_rq_pool;
55 /* max memory used for request pool, unit is MB */
56 static unsigned int osc_reqpool_mem_max = 5;
57 module_param(osc_reqpool_mem_max, uint, 0444);
59 static int osc_idle_timeout = 20;
60 module_param(osc_idle_timeout, uint, 0644);
62 #define osc_grant_args osc_brw_async_args
64 struct osc_setattr_args {
66 obd_enqueue_update_f sa_upcall;
70 struct osc_fsync_args {
71 struct osc_object *fa_obj;
73 obd_enqueue_update_f fa_upcall;
77 struct osc_ladvise_args {
79 obd_enqueue_update_f la_upcall;
83 static void osc_release_ppga(struct brw_page **ppga, size_t count);
84 static int brw_interpret(const struct lu_env *env, struct ptlrpc_request *req,
87 void osc_pack_req_body(struct ptlrpc_request *req, struct obdo *oa)
89 struct ost_body *body;
91 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
94 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
97 static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
100 struct ptlrpc_request *req;
101 struct ost_body *body;
105 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
109 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
111 ptlrpc_request_free(req);
115 osc_pack_req_body(req, oa);
117 ptlrpc_request_set_replen(req);
119 rc = ptlrpc_queue_wait(req);
123 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
125 GOTO(out, rc = -EPROTO);
127 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
128 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
130 oa->o_blksize = cli_brw_size(exp->exp_obd);
131 oa->o_valid |= OBD_MD_FLBLKSZ;
135 ptlrpc_req_finished(req);
140 static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
143 struct ptlrpc_request *req;
144 struct ost_body *body;
148 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
150 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
154 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
156 ptlrpc_request_free(req);
160 osc_pack_req_body(req, oa);
162 ptlrpc_request_set_replen(req);
164 rc = ptlrpc_queue_wait(req);
168 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
170 GOTO(out, rc = -EPROTO);
172 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
176 ptlrpc_req_finished(req);
181 static int osc_setattr_interpret(const struct lu_env *env,
182 struct ptlrpc_request *req, void *args, int rc)
184 struct osc_setattr_args *sa = args;
185 struct ost_body *body;
192 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
194 GOTO(out, rc = -EPROTO);
196 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
199 rc = sa->sa_upcall(sa->sa_cookie, rc);
203 int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
204 obd_enqueue_update_f upcall, void *cookie,
205 struct ptlrpc_request_set *rqset)
207 struct ptlrpc_request *req;
208 struct osc_setattr_args *sa;
213 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
217 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
219 ptlrpc_request_free(req);
223 osc_pack_req_body(req, oa);
225 ptlrpc_request_set_replen(req);
227 /* do mds to ost setattr asynchronously */
229 /* Do not wait for response. */
230 ptlrpcd_add_req(req);
232 req->rq_interpret_reply = osc_setattr_interpret;
234 sa = ptlrpc_req_async_args(sa, req);
236 sa->sa_upcall = upcall;
237 sa->sa_cookie = cookie;
239 if (rqset == PTLRPCD_SET)
240 ptlrpcd_add_req(req);
242 ptlrpc_set_add_req(rqset, req);
248 static int osc_ladvise_interpret(const struct lu_env *env,
249 struct ptlrpc_request *req,
252 struct osc_ladvise_args *la = arg;
253 struct ost_body *body;
259 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
261 GOTO(out, rc = -EPROTO);
263 *la->la_oa = body->oa;
265 rc = la->la_upcall(la->la_cookie, rc);
270 * If rqset is NULL, do not wait for response. Upcall and cookie could also
271 * be NULL in this case
273 int osc_ladvise_base(struct obd_export *exp, struct obdo *oa,
274 struct ladvise_hdr *ladvise_hdr,
275 obd_enqueue_update_f upcall, void *cookie,
276 struct ptlrpc_request_set *rqset)
278 struct ptlrpc_request *req;
279 struct ost_body *body;
280 struct osc_ladvise_args *la;
282 struct lu_ladvise *req_ladvise;
283 struct lu_ladvise *ladvise = ladvise_hdr->lah_advise;
284 int num_advise = ladvise_hdr->lah_count;
285 struct ladvise_hdr *req_ladvise_hdr;
288 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_LADVISE);
292 req_capsule_set_size(&req->rq_pill, &RMF_OST_LADVISE, RCL_CLIENT,
293 num_advise * sizeof(*ladvise));
294 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_LADVISE);
296 ptlrpc_request_free(req);
299 req->rq_request_portal = OST_IO_PORTAL;
300 ptlrpc_at_set_req_timeout(req);
302 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
304 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
307 req_ladvise_hdr = req_capsule_client_get(&req->rq_pill,
308 &RMF_OST_LADVISE_HDR);
309 memcpy(req_ladvise_hdr, ladvise_hdr, sizeof(*ladvise_hdr));
311 req_ladvise = req_capsule_client_get(&req->rq_pill, &RMF_OST_LADVISE);
312 memcpy(req_ladvise, ladvise, sizeof(*ladvise) * num_advise);
313 ptlrpc_request_set_replen(req);
316 /* Do not wait for response. */
317 ptlrpcd_add_req(req);
321 req->rq_interpret_reply = osc_ladvise_interpret;
322 la = ptlrpc_req_async_args(la, req);
324 la->la_upcall = upcall;
325 la->la_cookie = cookie;
327 if (rqset == PTLRPCD_SET)
328 ptlrpcd_add_req(req);
330 ptlrpc_set_add_req(rqset, req);
335 static int osc_create(const struct lu_env *env, struct obd_export *exp,
338 struct ptlrpc_request *req;
339 struct ost_body *body;
344 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
345 LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
347 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
349 GOTO(out, rc = -ENOMEM);
351 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
353 ptlrpc_request_free(req);
357 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
360 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
362 ptlrpc_request_set_replen(req);
364 rc = ptlrpc_queue_wait(req);
368 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
370 GOTO(out_req, rc = -EPROTO);
372 CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
373 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
375 oa->o_blksize = cli_brw_size(exp->exp_obd);
376 oa->o_valid |= OBD_MD_FLBLKSZ;
378 CDEBUG(D_HA, "transno: %lld\n",
379 lustre_msg_get_transno(req->rq_repmsg));
381 ptlrpc_req_finished(req);
386 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
387 obd_enqueue_update_f upcall, void *cookie)
389 struct ptlrpc_request *req;
390 struct osc_setattr_args *sa;
391 struct obd_import *imp = class_exp2cliimp(exp);
392 struct ost_body *body;
397 req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
401 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
403 ptlrpc_request_free(req);
407 osc_set_io_portal(req);
409 ptlrpc_at_set_req_timeout(req);
411 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
413 lustre_set_wire_obdo(&imp->imp_connect_data, &body->oa, oa);
415 ptlrpc_request_set_replen(req);
417 req->rq_interpret_reply = osc_setattr_interpret;
418 sa = ptlrpc_req_async_args(sa, req);
420 sa->sa_upcall = upcall;
421 sa->sa_cookie = cookie;
423 ptlrpcd_add_req(req);
427 EXPORT_SYMBOL(osc_punch_send);
429 static int osc_sync_interpret(const struct lu_env *env,
430 struct ptlrpc_request *req, void *args, int rc)
432 struct osc_fsync_args *fa = args;
433 struct ost_body *body;
434 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
435 unsigned long valid = 0;
436 struct cl_object *obj;
442 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
444 CERROR("can't unpack ost_body\n");
445 GOTO(out, rc = -EPROTO);
448 *fa->fa_oa = body->oa;
449 obj = osc2cl(fa->fa_obj);
451 /* Update osc object's blocks attribute */
452 cl_object_attr_lock(obj);
453 if (body->oa.o_valid & OBD_MD_FLBLOCKS) {
454 attr->cat_blocks = body->oa.o_blocks;
459 cl_object_attr_update(env, obj, attr, valid);
460 cl_object_attr_unlock(obj);
463 rc = fa->fa_upcall(fa->fa_cookie, rc);
467 int osc_sync_base(struct osc_object *obj, struct obdo *oa,
468 obd_enqueue_update_f upcall, void *cookie,
469 struct ptlrpc_request_set *rqset)
471 struct obd_export *exp = osc_export(obj);
472 struct ptlrpc_request *req;
473 struct ost_body *body;
474 struct osc_fsync_args *fa;
478 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
482 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
484 ptlrpc_request_free(req);
488 /* overload the size and blocks fields in the oa with start/end */
489 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
491 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
493 ptlrpc_request_set_replen(req);
494 req->rq_interpret_reply = osc_sync_interpret;
496 fa = ptlrpc_req_async_args(fa, req);
499 fa->fa_upcall = upcall;
500 fa->fa_cookie = cookie;
502 if (rqset == PTLRPCD_SET)
503 ptlrpcd_add_req(req);
505 ptlrpc_set_add_req(rqset, req);
510 /* Find and cancel locally locks matched by @mode in the resource found by
511 * @objid. Found locks are added into @cancel list. Returns the amount of
512 * locks added to @cancels list. */
513 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
514 struct list_head *cancels,
515 enum ldlm_mode mode, __u64 lock_flags)
517 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
518 struct ldlm_res_id res_id;
519 struct ldlm_resource *res;
523 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
524 * export) but disabled through procfs (flag in NS).
526 * This distinguishes from a case when ELC is not supported originally,
527 * when we still want to cancel locks in advance and just cancel them
528 * locally, without sending any RPC. */
529 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
532 ostid_build_res_name(&oa->o_oi, &res_id);
533 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
537 LDLM_RESOURCE_ADDREF(res);
538 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
539 lock_flags, 0, NULL);
540 LDLM_RESOURCE_DELREF(res);
541 ldlm_resource_putref(res);
545 static int osc_destroy_interpret(const struct lu_env *env,
546 struct ptlrpc_request *req, void *args, int rc)
548 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
550 atomic_dec(&cli->cl_destroy_in_flight);
551 wake_up(&cli->cl_destroy_waitq);
556 static int osc_can_send_destroy(struct client_obd *cli)
558 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
559 cli->cl_max_rpcs_in_flight) {
560 /* The destroy request can be sent */
563 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
564 cli->cl_max_rpcs_in_flight) {
566 * The counter has been modified between the two atomic
569 wake_up(&cli->cl_destroy_waitq);
574 static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
577 struct client_obd *cli = &exp->exp_obd->u.cli;
578 struct ptlrpc_request *req;
579 struct ost_body *body;
585 CDEBUG(D_INFO, "oa NULL\n");
589 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
590 LDLM_FL_DISCARD_DATA);
592 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
594 ldlm_lock_list_put(&cancels, l_bl_ast, count);
598 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
601 ptlrpc_request_free(req);
605 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
606 ptlrpc_at_set_req_timeout(req);
608 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
610 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
612 ptlrpc_request_set_replen(req);
614 req->rq_interpret_reply = osc_destroy_interpret;
615 if (!osc_can_send_destroy(cli)) {
617 * Wait until the number of on-going destroy RPCs drops
618 * under max_rpc_in_flight
620 rc = l_wait_event_abortable_exclusive(
621 cli->cl_destroy_waitq,
622 osc_can_send_destroy(cli));
624 ptlrpc_req_finished(req);
629 /* Do not wait for response */
630 ptlrpcd_add_req(req);
634 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
637 u64 bits = OBD_MD_FLBLOCKS | OBD_MD_FLGRANT;
639 LASSERT(!(oa->o_valid & bits));
642 spin_lock(&cli->cl_loi_list_lock);
643 if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data, GRANT_PARAM))
644 oa->o_dirty = cli->cl_dirty_grant;
646 oa->o_dirty = cli->cl_dirty_pages << PAGE_SHIFT;
647 if (unlikely(cli->cl_dirty_pages > cli->cl_dirty_max_pages)) {
648 CERROR("dirty %lu > dirty_max %lu\n",
650 cli->cl_dirty_max_pages);
652 } else if (unlikely(atomic_long_read(&obd_dirty_pages) >
653 (long)(obd_max_dirty_pages + 1))) {
654 /* The atomic_read() allowing the atomic_inc() are
655 * not covered by a lock thus they may safely race and trip
656 * this CERROR() unless we add in a small fudge factor (+1). */
657 CERROR("%s: dirty %ld > system dirty_max %ld\n",
658 cli_name(cli), atomic_long_read(&obd_dirty_pages),
659 obd_max_dirty_pages);
661 } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
663 CERROR("dirty %lu - dirty_max %lu too big???\n",
664 cli->cl_dirty_pages, cli->cl_dirty_max_pages);
667 unsigned long nrpages;
668 unsigned long undirty;
670 nrpages = cli->cl_max_pages_per_rpc;
671 nrpages *= cli->cl_max_rpcs_in_flight + 1;
672 nrpages = max(nrpages, cli->cl_dirty_max_pages);
673 undirty = nrpages << PAGE_SHIFT;
674 if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data,
678 /* take extent tax into account when asking for more
680 nrextents = (nrpages + cli->cl_max_extent_pages - 1) /
681 cli->cl_max_extent_pages;
682 undirty += nrextents * cli->cl_grant_extent_tax;
684 /* Do not ask for more than OBD_MAX_GRANT - a margin for server
685 * to add extent tax, etc.
687 oa->o_undirty = min(undirty, OBD_MAX_GRANT &
688 ~(PTLRPC_MAX_BRW_SIZE * 4UL));
690 oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
691 oa->o_dropped = cli->cl_lost_grant;
692 cli->cl_lost_grant = 0;
693 spin_unlock(&cli->cl_loi_list_lock);
694 CDEBUG(D_CACHE, "dirty: %llu undirty: %u dropped %u grant: %llu\n",
695 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
698 void osc_update_next_shrink(struct client_obd *cli)
700 cli->cl_next_shrink_grant = ktime_get_seconds() +
701 cli->cl_grant_shrink_interval;
703 CDEBUG(D_CACHE, "next time %lld to shrink grant\n",
704 cli->cl_next_shrink_grant);
707 static void __osc_update_grant(struct client_obd *cli, u64 grant)
709 spin_lock(&cli->cl_loi_list_lock);
710 cli->cl_avail_grant += grant;
711 spin_unlock(&cli->cl_loi_list_lock);
714 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
716 if (body->oa.o_valid & OBD_MD_FLGRANT) {
717 CDEBUG(D_CACHE, "got %llu extra grant\n", body->oa.o_grant);
718 __osc_update_grant(cli, body->oa.o_grant);
723 * grant thread data for shrinking space.
725 struct grant_thread_data {
726 struct list_head gtd_clients;
727 struct mutex gtd_mutex;
728 unsigned long gtd_stopped:1;
730 static struct grant_thread_data client_gtd;
732 static int osc_shrink_grant_interpret(const struct lu_env *env,
733 struct ptlrpc_request *req,
736 struct osc_grant_args *aa = args;
737 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
738 struct ost_body *body;
741 __osc_update_grant(cli, aa->aa_oa->o_grant);
745 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
747 osc_update_grant(cli, body);
749 OBD_SLAB_FREE_PTR(aa->aa_oa, osc_obdo_kmem);
755 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
757 spin_lock(&cli->cl_loi_list_lock);
758 oa->o_grant = cli->cl_avail_grant / 4;
759 cli->cl_avail_grant -= oa->o_grant;
760 spin_unlock(&cli->cl_loi_list_lock);
761 if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
762 oa->o_valid |= OBD_MD_FLFLAGS;
765 oa->o_flags |= OBD_FL_SHRINK_GRANT;
766 osc_update_next_shrink(cli);
769 /* Shrink the current grant, either from some large amount to enough for a
770 * full set of in-flight RPCs, or if we have already shrunk to that limit
771 * then to enough for a single RPC. This avoids keeping more grant than
772 * needed, and avoids shrinking the grant piecemeal. */
773 static int osc_shrink_grant(struct client_obd *cli)
775 __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
776 (cli->cl_max_pages_per_rpc << PAGE_SHIFT);
778 spin_lock(&cli->cl_loi_list_lock);
779 if (cli->cl_avail_grant <= target_bytes)
780 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
781 spin_unlock(&cli->cl_loi_list_lock);
783 return osc_shrink_grant_to_target(cli, target_bytes);
786 int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
789 struct ost_body *body;
792 spin_lock(&cli->cl_loi_list_lock);
793 /* Don't shrink if we are already above or below the desired limit
794 * We don't want to shrink below a single RPC, as that will negatively
795 * impact block allocation and long-term performance. */
796 if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_SHIFT)
797 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
799 if (target_bytes >= cli->cl_avail_grant) {
800 spin_unlock(&cli->cl_loi_list_lock);
803 spin_unlock(&cli->cl_loi_list_lock);
809 osc_announce_cached(cli, &body->oa, 0);
811 spin_lock(&cli->cl_loi_list_lock);
812 if (target_bytes >= cli->cl_avail_grant) {
813 /* available grant has changed since target calculation */
814 spin_unlock(&cli->cl_loi_list_lock);
815 GOTO(out_free, rc = 0);
817 body->oa.o_grant = cli->cl_avail_grant - target_bytes;
818 cli->cl_avail_grant = target_bytes;
819 spin_unlock(&cli->cl_loi_list_lock);
820 if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
821 body->oa.o_valid |= OBD_MD_FLFLAGS;
822 body->oa.o_flags = 0;
824 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
825 osc_update_next_shrink(cli);
827 rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
828 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
829 sizeof(*body), body, NULL);
831 __osc_update_grant(cli, body->oa.o_grant);
837 static int osc_should_shrink_grant(struct client_obd *client)
839 time64_t next_shrink = client->cl_next_shrink_grant;
841 if (client->cl_import == NULL)
844 if (!OCD_HAS_FLAG(&client->cl_import->imp_connect_data, GRANT_SHRINK) ||
845 client->cl_import->imp_grant_shrink_disabled) {
846 osc_update_next_shrink(client);
850 if (ktime_get_seconds() >= next_shrink - 5) {
851 /* Get the current RPC size directly, instead of going via:
852 * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
853 * Keep comment here so that it can be found by searching. */
854 int brw_size = client->cl_max_pages_per_rpc << PAGE_SHIFT;
856 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
857 client->cl_avail_grant > brw_size)
860 osc_update_next_shrink(client);
865 #define GRANT_SHRINK_RPC_BATCH 100
867 static struct delayed_work work;
869 static void osc_grant_work_handler(struct work_struct *data)
871 struct client_obd *cli;
873 bool init_next_shrink = true;
874 time64_t next_shrink = ktime_get_seconds() + GRANT_SHRINK_INTERVAL;
877 mutex_lock(&client_gtd.gtd_mutex);
878 list_for_each_entry(cli, &client_gtd.gtd_clients,
880 if (rpc_sent < GRANT_SHRINK_RPC_BATCH &&
881 osc_should_shrink_grant(cli)) {
882 osc_shrink_grant(cli);
886 if (!init_next_shrink) {
887 if (cli->cl_next_shrink_grant < next_shrink &&
888 cli->cl_next_shrink_grant > ktime_get_seconds())
889 next_shrink = cli->cl_next_shrink_grant;
891 init_next_shrink = false;
892 next_shrink = cli->cl_next_shrink_grant;
895 mutex_unlock(&client_gtd.gtd_mutex);
897 if (client_gtd.gtd_stopped == 1)
900 if (next_shrink > ktime_get_seconds()) {
901 time64_t delay = next_shrink - ktime_get_seconds();
903 schedule_delayed_work(&work, cfs_time_seconds(delay));
905 schedule_work(&work.work);
909 void osc_schedule_grant_work(void)
911 cancel_delayed_work_sync(&work);
912 schedule_work(&work.work);
916 * Start grant thread for returing grant to server for idle clients.
918 static int osc_start_grant_work(void)
920 client_gtd.gtd_stopped = 0;
921 mutex_init(&client_gtd.gtd_mutex);
922 INIT_LIST_HEAD(&client_gtd.gtd_clients);
924 INIT_DELAYED_WORK(&work, osc_grant_work_handler);
925 schedule_work(&work.work);
930 static void osc_stop_grant_work(void)
932 client_gtd.gtd_stopped = 1;
933 cancel_delayed_work_sync(&work);
936 static void osc_add_grant_list(struct client_obd *client)
938 mutex_lock(&client_gtd.gtd_mutex);
939 list_add(&client->cl_grant_chain, &client_gtd.gtd_clients);
940 mutex_unlock(&client_gtd.gtd_mutex);
943 static void osc_del_grant_list(struct client_obd *client)
945 if (list_empty(&client->cl_grant_chain))
948 mutex_lock(&client_gtd.gtd_mutex);
949 list_del_init(&client->cl_grant_chain);
950 mutex_unlock(&client_gtd.gtd_mutex);
953 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
956 * ocd_grant is the total grant amount we're expect to hold: if we've
957 * been evicted, it's the new avail_grant amount, cl_dirty_pages will
958 * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
961 * race is tolerable here: if we're evicted, but imp_state already
962 * left EVICTED state, then cl_dirty_pages must be 0 already.
964 spin_lock(&cli->cl_loi_list_lock);
965 cli->cl_avail_grant = ocd->ocd_grant;
966 if (cli->cl_import->imp_state != LUSTRE_IMP_EVICTED) {
967 cli->cl_avail_grant -= cli->cl_reserved_grant;
968 if (OCD_HAS_FLAG(ocd, GRANT_PARAM))
969 cli->cl_avail_grant -= cli->cl_dirty_grant;
971 cli->cl_avail_grant -=
972 cli->cl_dirty_pages << PAGE_SHIFT;
975 if (OCD_HAS_FLAG(ocd, GRANT_PARAM)) {
979 /* overhead for each extent insertion */
980 cli->cl_grant_extent_tax = ocd->ocd_grant_tax_kb << 10;
981 /* determine the appropriate chunk size used by osc_extent. */
982 cli->cl_chunkbits = max_t(int, PAGE_SHIFT,
983 ocd->ocd_grant_blkbits);
984 /* max_pages_per_rpc must be chunk aligned */
985 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
986 cli->cl_max_pages_per_rpc = (cli->cl_max_pages_per_rpc +
987 ~chunk_mask) & chunk_mask;
988 /* determine maximum extent size, in #pages */
989 size = (u64)ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits;
990 cli->cl_max_extent_pages = size >> PAGE_SHIFT;
991 if (cli->cl_max_extent_pages == 0)
992 cli->cl_max_extent_pages = 1;
994 cli->cl_grant_extent_tax = 0;
995 cli->cl_chunkbits = PAGE_SHIFT;
996 cli->cl_max_extent_pages = DT_MAX_BRW_PAGES;
998 spin_unlock(&cli->cl_loi_list_lock);
1000 CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
1001 "chunk bits: %d cl_max_extent_pages: %d\n",
1003 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits,
1004 cli->cl_max_extent_pages);
1006 if (OCD_HAS_FLAG(ocd, GRANT_SHRINK) && list_empty(&cli->cl_grant_chain))
1007 osc_add_grant_list(cli);
1009 EXPORT_SYMBOL(osc_init_grant);
1011 /* We assume that the reason this OSC got a short read is because it read
1012 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
1013 * via the LOV, and it _knows_ it's reading inside the file, it's just that
1014 * this stripe never got written at or beyond this stripe offset yet. */
1015 static void handle_short_read(int nob_read, size_t page_count,
1016 struct brw_page **pga)
1021 /* skip bytes read OK */
1022 while (nob_read > 0) {
1023 LASSERT (page_count > 0);
1025 if (pga[i]->count > nob_read) {
1026 /* EOF inside this page */
1027 ptr = kmap(pga[i]->pg) +
1028 (pga[i]->off & ~PAGE_MASK);
1029 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
1036 nob_read -= pga[i]->count;
1041 /* zero remaining pages */
1042 while (page_count-- > 0) {
1043 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~PAGE_MASK);
1044 memset(ptr, 0, pga[i]->count);
1050 static int check_write_rcs(struct ptlrpc_request *req,
1051 int requested_nob, int niocount,
1052 size_t page_count, struct brw_page **pga)
1057 remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1058 sizeof(*remote_rcs) *
1060 if (remote_rcs == NULL) {
1061 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1065 /* return error if any niobuf was in error */
1066 for (i = 0; i < niocount; i++) {
1067 if ((int)remote_rcs[i] < 0) {
1068 CDEBUG(D_INFO, "rc[%d]: %d req %p\n",
1069 i, remote_rcs[i], req);
1070 return remote_rcs[i];
1073 if (remote_rcs[i] != 0) {
1074 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1075 i, remote_rcs[i], req);
1079 if (req->rq_bulk != NULL &&
1080 req->rq_bulk->bd_nob_transferred != requested_nob) {
1081 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1082 req->rq_bulk->bd_nob_transferred, requested_nob);
1089 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1091 if (p1->flag != p2->flag) {
1092 unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE |
1093 OBD_BRW_SYNC | OBD_BRW_ASYNC |
1094 OBD_BRW_NOQUOTA | OBD_BRW_SOFT_SYNC);
1096 /* warn if we try to combine flags that we don't know to be
1097 * safe to combine */
1098 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1099 CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1100 "report this at https://jira.whamcloud.com/\n",
1101 p1->flag, p2->flag);
1106 return (p1->off + p1->count == p2->off);
1109 #if IS_ENABLED(CONFIG_CRC_T10DIF)
1110 static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
1111 size_t pg_count, struct brw_page **pga,
1112 int opc, obd_dif_csum_fn *fn,
1116 struct ahash_request *req;
1117 /* Used Adler as the default checksum type on top of DIF tags */
1118 unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP);
1119 struct page *__page;
1120 unsigned char *buffer;
1122 unsigned int bufsize;
1124 int used_number = 0;
1130 LASSERT(pg_count > 0);
1132 __page = alloc_page(GFP_KERNEL);
1136 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1139 CERROR("%s: unable to initialize checksum hash %s: rc = %d\n",
1140 obd_name, cfs_crypto_hash_name(cfs_alg), rc);
1144 buffer = kmap(__page);
1145 guard_start = (__u16 *)buffer;
1146 guard_number = PAGE_SIZE / sizeof(*guard_start);
1147 while (nob > 0 && pg_count > 0) {
1148 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1150 /* corrupt the data before we compute the checksum, to
1151 * simulate an OST->client data error */
1152 if (unlikely(i == 0 && opc == OST_READ &&
1153 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE))) {
1154 unsigned char *ptr = kmap(pga[i]->pg);
1155 int off = pga[i]->off & ~PAGE_MASK;
1157 memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1162 * The left guard number should be able to hold checksums of a
1165 rc = obd_page_dif_generate_buffer(obd_name, pga[i]->pg,
1166 pga[i]->off & ~PAGE_MASK,
1168 guard_start + used_number,
1169 guard_number - used_number,
1175 used_number += used;
1176 if (used_number == guard_number) {
1177 cfs_crypto_hash_update_page(req, __page, 0,
1178 used_number * sizeof(*guard_start));
1182 nob -= pga[i]->count;
1190 if (used_number != 0)
1191 cfs_crypto_hash_update_page(req, __page, 0,
1192 used_number * sizeof(*guard_start));
1194 bufsize = sizeof(cksum);
1195 cfs_crypto_hash_final(req, (unsigned char *)&cksum, &bufsize);
1197 /* For sending we only compute the wrong checksum instead
1198 * of corrupting the data so it is still correct on a redo */
1199 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1204 __free_page(__page);
1207 #else /* !CONFIG_CRC_T10DIF */
1208 #define obd_dif_ip_fn NULL
1209 #define obd_dif_crc_fn NULL
1210 #define osc_checksum_bulk_t10pi(name, nob, pgc, pga, opc, fn, ssize, csum) \
1212 #endif /* CONFIG_CRC_T10DIF */
1214 static int osc_checksum_bulk(int nob, size_t pg_count,
1215 struct brw_page **pga, int opc,
1216 enum cksum_types cksum_type,
1220 struct ahash_request *req;
1221 unsigned int bufsize;
1222 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1224 LASSERT(pg_count > 0);
1226 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1228 CERROR("Unable to initialize checksum hash %s\n",
1229 cfs_crypto_hash_name(cfs_alg));
1230 return PTR_ERR(req);
1233 while (nob > 0 && pg_count > 0) {
1234 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1236 /* corrupt the data before we compute the checksum, to
1237 * simulate an OST->client data error */
1238 if (i == 0 && opc == OST_READ &&
1239 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1240 unsigned char *ptr = kmap(pga[i]->pg);
1241 int off = pga[i]->off & ~PAGE_MASK;
1243 memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1246 cfs_crypto_hash_update_page(req, pga[i]->pg,
1247 pga[i]->off & ~PAGE_MASK,
1249 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
1250 (int)(pga[i]->off & ~PAGE_MASK));
1252 nob -= pga[i]->count;
1257 bufsize = sizeof(*cksum);
1258 cfs_crypto_hash_final(req, (unsigned char *)cksum, &bufsize);
1260 /* For sending we only compute the wrong checksum instead
1261 * of corrupting the data so it is still correct on a redo */
1262 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1268 static int osc_checksum_bulk_rw(const char *obd_name,
1269 enum cksum_types cksum_type,
1270 int nob, size_t pg_count,
1271 struct brw_page **pga, int opc,
1274 obd_dif_csum_fn *fn = NULL;
1275 int sector_size = 0;
1279 obd_t10_cksum2dif(cksum_type, &fn, §or_size);
1282 rc = osc_checksum_bulk_t10pi(obd_name, nob, pg_count, pga,
1283 opc, fn, sector_size, check_sum);
1285 rc = osc_checksum_bulk(nob, pg_count, pga, opc, cksum_type,
1292 osc_brw_prep_request(int cmd, struct client_obd *cli, struct obdo *oa,
1293 u32 page_count, struct brw_page **pga,
1294 struct ptlrpc_request **reqp, int resend)
1296 struct ptlrpc_request *req;
1297 struct ptlrpc_bulk_desc *desc;
1298 struct ost_body *body;
1299 struct obd_ioobj *ioobj;
1300 struct niobuf_remote *niobuf;
1301 int niocount, i, requested_nob, opc, rc, short_io_size = 0;
1302 struct osc_brw_async_args *aa;
1303 struct req_capsule *pill;
1304 struct brw_page *pg_prev;
1306 const char *obd_name = cli->cl_import->imp_obd->obd_name;
1309 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1310 RETURN(-ENOMEM); /* Recoverable */
1311 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1312 RETURN(-EINVAL); /* Fatal */
1314 if ((cmd & OBD_BRW_WRITE) != 0) {
1316 req = ptlrpc_request_alloc_pool(cli->cl_import,
1318 &RQF_OST_BRW_WRITE);
1321 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1326 for (niocount = i = 1; i < page_count; i++) {
1327 if (!can_merge_pages(pga[i - 1], pga[i]))
1331 pill = &req->rq_pill;
1332 req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1334 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1335 niocount * sizeof(*niobuf));
1337 for (i = 0; i < page_count; i++)
1338 short_io_size += pga[i]->count;
1340 /* Check if read/write is small enough to be a short io. */
1341 if (short_io_size > cli->cl_max_short_io_bytes || niocount > 1 ||
1342 !imp_connect_shortio(cli->cl_import))
1345 req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_CLIENT,
1346 opc == OST_READ ? 0 : short_io_size);
1347 if (opc == OST_READ)
1348 req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_SERVER,
1351 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1353 ptlrpc_request_free(req);
1356 osc_set_io_portal(req);
1358 ptlrpc_at_set_req_timeout(req);
1359 /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1361 req->rq_no_retry_einprogress = 1;
1363 if (short_io_size != 0) {
1365 short_io_buf = NULL;
1369 desc = ptlrpc_prep_bulk_imp(req, page_count,
1370 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1371 (opc == OST_WRITE ? PTLRPC_BULK_GET_SOURCE :
1372 PTLRPC_BULK_PUT_SINK) |
1373 PTLRPC_BULK_BUF_KIOV,
1375 &ptlrpc_bulk_kiov_pin_ops);
1378 GOTO(out, rc = -ENOMEM);
1379 /* NB request now owns desc and will free it when it gets freed */
1381 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1382 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1383 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1384 LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1386 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
1388 /* For READ and WRITE, we can't fill o_uid and o_gid using from_kuid()
1389 * and from_kgid(), because they are asynchronous. Fortunately, variable
1390 * oa contains valid o_uid and o_gid in these two operations.
1391 * Besides, filling o_uid and o_gid is enough for nrs-tbf, see LU-9658.
1392 * OBD_MD_FLUID and OBD_MD_FLUID is not set in order to avoid breaking
1393 * other process logic */
1394 body->oa.o_uid = oa->o_uid;
1395 body->oa.o_gid = oa->o_gid;
1397 obdo_to_ioobj(oa, ioobj);
1398 ioobj->ioo_bufcnt = niocount;
1399 /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1400 * that might be send for this request. The actual number is decided
1401 * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1402 * "max - 1" for old client compatibility sending "0", and also so the
1403 * the actual maximum is a power-of-two number, not one less. LU-1431 */
1405 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1407 ioobj_max_brw_set(ioobj, 0);
1409 if (short_io_size != 0) {
1410 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1411 body->oa.o_valid |= OBD_MD_FLFLAGS;
1412 body->oa.o_flags = 0;
1414 body->oa.o_flags |= OBD_FL_SHORT_IO;
1415 CDEBUG(D_CACHE, "Using short io for data transfer, size = %d\n",
1417 if (opc == OST_WRITE) {
1418 short_io_buf = req_capsule_client_get(pill,
1420 LASSERT(short_io_buf != NULL);
1424 LASSERT(page_count > 0);
1426 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1427 struct brw_page *pg = pga[i];
1428 int poff = pg->off & ~PAGE_MASK;
1430 LASSERT(pg->count > 0);
1431 /* make sure there is no gap in the middle of page array */
1432 LASSERTF(page_count == 1 ||
1433 (ergo(i == 0, poff + pg->count == PAGE_SIZE) &&
1434 ergo(i > 0 && i < page_count - 1,
1435 poff == 0 && pg->count == PAGE_SIZE) &&
1436 ergo(i == page_count - 1, poff == 0)),
1437 "i: %d/%d pg: %p off: %llu, count: %u\n",
1438 i, page_count, pg, pg->off, pg->count);
1439 LASSERTF(i == 0 || pg->off > pg_prev->off,
1440 "i %d p_c %u pg %p [pri %lu ind %lu] off %llu"
1441 " prev_pg %p [pri %lu ind %lu] off %llu\n",
1443 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1444 pg_prev->pg, page_private(pg_prev->pg),
1445 pg_prev->pg->index, pg_prev->off);
1446 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1447 (pg->flag & OBD_BRW_SRVLOCK));
1448 if (short_io_size != 0 && opc == OST_WRITE) {
1449 unsigned char *ptr = kmap_atomic(pg->pg);
1451 LASSERT(short_io_size >= requested_nob + pg->count);
1452 memcpy(short_io_buf + requested_nob,
1456 } else if (short_io_size == 0) {
1457 desc->bd_frag_ops->add_kiov_frag(desc, pg->pg, poff,
1460 requested_nob += pg->count;
1462 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1464 niobuf->rnb_len += pg->count;
1466 niobuf->rnb_offset = pg->off;
1467 niobuf->rnb_len = pg->count;
1468 niobuf->rnb_flags = pg->flag;
1473 LASSERTF((void *)(niobuf - niocount) ==
1474 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1475 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1476 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1478 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1480 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1481 body->oa.o_valid |= OBD_MD_FLFLAGS;
1482 body->oa.o_flags = 0;
1484 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1487 if (osc_should_shrink_grant(cli))
1488 osc_shrink_grant_local(cli, &body->oa);
1490 /* size[REQ_REC_OFF] still sizeof (*body) */
1491 if (opc == OST_WRITE) {
1492 if (cli->cl_checksum &&
1493 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1494 /* store cl_cksum_type in a local variable since
1495 * it can be changed via lprocfs */
1496 enum cksum_types cksum_type = cli->cl_cksum_type;
1498 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1499 body->oa.o_flags = 0;
1501 body->oa.o_flags |= obd_cksum_type_pack(obd_name,
1503 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1505 rc = osc_checksum_bulk_rw(obd_name, cksum_type,
1506 requested_nob, page_count,
1510 CDEBUG(D_PAGE, "failed to checksum, rc = %d\n",
1514 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1517 /* save this in 'oa', too, for later checking */
1518 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1519 oa->o_flags |= obd_cksum_type_pack(obd_name,
1522 /* clear out the checksum flag, in case this is a
1523 * resend but cl_checksum is no longer set. b=11238 */
1524 oa->o_valid &= ~OBD_MD_FLCKSUM;
1526 oa->o_cksum = body->oa.o_cksum;
1527 /* 1 RC per niobuf */
1528 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1529 sizeof(__u32) * niocount);
1531 if (cli->cl_checksum &&
1532 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1533 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1534 body->oa.o_flags = 0;
1535 body->oa.o_flags |= obd_cksum_type_pack(obd_name,
1536 cli->cl_cksum_type);
1537 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1540 /* Client cksum has been already copied to wire obdo in previous
1541 * lustre_set_wire_obdo(), and in the case a bulk-read is being
1542 * resent due to cksum error, this will allow Server to
1543 * check+dump pages on its side */
1545 ptlrpc_request_set_replen(req);
1547 aa = ptlrpc_req_async_args(aa, req);
1549 aa->aa_requested_nob = requested_nob;
1550 aa->aa_nio_count = niocount;
1551 aa->aa_page_count = page_count;
1555 INIT_LIST_HEAD(&aa->aa_oaps);
1558 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1559 CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
1560 req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
1561 niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
1565 ptlrpc_req_finished(req);
1569 char dbgcksum_file_name[PATH_MAX];
1571 static void dump_all_bulk_pages(struct obdo *oa, __u32 page_count,
1572 struct brw_page **pga, __u32 server_cksum,
1580 /* will only keep dump of pages on first error for the same range in
1581 * file/fid, not during the resends/retries. */
1582 snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1583 "%s-checksum_dump-osc-"DFID":[%llu-%llu]-%x-%x",
1584 (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 ?
1585 libcfs_debug_file_path_arr :
1586 LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1587 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
1588 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1589 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1591 pga[page_count-1]->off + pga[page_count-1]->count - 1,
1592 client_cksum, server_cksum);
1593 filp = filp_open(dbgcksum_file_name,
1594 O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1598 CDEBUG(D_INFO, "%s: can't open to dump pages with "
1599 "checksum error: rc = %d\n", dbgcksum_file_name,
1602 CERROR("%s: can't open to dump pages with checksum "
1603 "error: rc = %d\n", dbgcksum_file_name, rc);
1607 for (i = 0; i < page_count; i++) {
1608 len = pga[i]->count;
1609 buf = kmap(pga[i]->pg);
1611 rc = cfs_kernel_write(filp, buf, len, &filp->f_pos);
1613 CERROR("%s: wanted to write %u but got %d "
1614 "error\n", dbgcksum_file_name, len, rc);
1619 CDEBUG(D_INFO, "%s: wrote %d bytes\n",
1620 dbgcksum_file_name, rc);
1625 rc = vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1627 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1628 filp_close(filp, NULL);
1632 check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
1633 __u32 client_cksum, __u32 server_cksum,
1634 struct osc_brw_async_args *aa)
1636 const char *obd_name = aa->aa_cli->cl_import->imp_obd->obd_name;
1637 enum cksum_types cksum_type;
1638 obd_dif_csum_fn *fn = NULL;
1639 int sector_size = 0;
1644 if (server_cksum == client_cksum) {
1645 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1649 if (aa->aa_cli->cl_checksum_dump)
1650 dump_all_bulk_pages(oa, aa->aa_page_count, aa->aa_ppga,
1651 server_cksum, client_cksum);
1653 cksum_type = obd_cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1656 switch (cksum_type) {
1657 case OBD_CKSUM_T10IP512:
1661 case OBD_CKSUM_T10IP4K:
1665 case OBD_CKSUM_T10CRC512:
1666 fn = obd_dif_crc_fn;
1669 case OBD_CKSUM_T10CRC4K:
1670 fn = obd_dif_crc_fn;
1678 rc = osc_checksum_bulk_t10pi(obd_name, aa->aa_requested_nob,
1679 aa->aa_page_count, aa->aa_ppga,
1680 OST_WRITE, fn, sector_size,
1683 rc = osc_checksum_bulk(aa->aa_requested_nob, aa->aa_page_count,
1684 aa->aa_ppga, OST_WRITE, cksum_type,
1688 msg = "failed to calculate the client write checksum";
1689 else if (cksum_type != obd_cksum_type_unpack(aa->aa_oa->o_flags))
1690 msg = "the server did not use the checksum type specified in "
1691 "the original request - likely a protocol problem";
1692 else if (new_cksum == server_cksum)
1693 msg = "changed on the client after we checksummed it - "
1694 "likely false positive due to mmap IO (bug 11742)";
1695 else if (new_cksum == client_cksum)
1696 msg = "changed in transit before arrival at OST";
1698 msg = "changed in transit AND doesn't match the original - "
1699 "likely false positive due to mmap IO (bug 11742)";
1701 LCONSOLE_ERROR_MSG(0x132, "%s: BAD WRITE CHECKSUM: %s: from %s inode "
1702 DFID " object "DOSTID" extent [%llu-%llu], original "
1703 "client csum %x (type %x), server csum %x (type %x),"
1704 " client csum now %x\n",
1705 obd_name, msg, libcfs_nid2str(peer->nid),
1706 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1707 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1708 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1709 POSTID(&oa->o_oi), aa->aa_ppga[0]->off,
1710 aa->aa_ppga[aa->aa_page_count - 1]->off +
1711 aa->aa_ppga[aa->aa_page_count-1]->count - 1,
1713 obd_cksum_type_unpack(aa->aa_oa->o_flags),
1714 server_cksum, cksum_type, new_cksum);
1718 /* Note rc enters this function as number of bytes transferred */
1719 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1721 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1722 struct client_obd *cli = aa->aa_cli;
1723 const char *obd_name = cli->cl_import->imp_obd->obd_name;
1724 const struct lnet_process_id *peer =
1725 &req->rq_import->imp_connection->c_peer;
1726 struct ost_body *body;
1727 u32 client_cksum = 0;
1731 if (rc < 0 && rc != -EDQUOT) {
1732 DEBUG_REQ(D_INFO, req, "Failed request: rc = %d", rc);
1736 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1737 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1739 DEBUG_REQ(D_INFO, req, "cannot unpack body");
1743 /* set/clear over quota flag for a uid/gid/projid */
1744 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1745 body->oa.o_valid & (OBD_MD_FLALLQUOTA)) {
1746 unsigned qid[LL_MAXQUOTAS] = {
1747 body->oa.o_uid, body->oa.o_gid,
1748 body->oa.o_projid };
1750 "setdq for [%u %u %u] with valid %#llx, flags %x\n",
1751 body->oa.o_uid, body->oa.o_gid, body->oa.o_projid,
1752 body->oa.o_valid, body->oa.o_flags);
1753 osc_quota_setdq(cli, req->rq_xid, qid, body->oa.o_valid,
1757 osc_update_grant(cli, body);
1762 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1763 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1765 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1767 CERROR("%s: unexpected positive size %d\n",
1772 if (req->rq_bulk != NULL &&
1773 sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1776 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1777 check_write_checksum(&body->oa, peer, client_cksum,
1778 body->oa.o_cksum, aa))
1781 rc = check_write_rcs(req, aa->aa_requested_nob,
1782 aa->aa_nio_count, aa->aa_page_count,
1787 /* The rest of this function executes only for OST_READs */
1789 if (req->rq_bulk == NULL) {
1790 rc = req_capsule_get_size(&req->rq_pill, &RMF_SHORT_IO,
1792 LASSERT(rc == req->rq_status);
1794 /* if unwrap_bulk failed, return -EAGAIN to retry */
1795 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1798 GOTO(out, rc = -EAGAIN);
1800 if (rc > aa->aa_requested_nob) {
1801 CERROR("%s: unexpected size %d, requested %d\n", obd_name,
1802 rc, aa->aa_requested_nob);
1806 if (req->rq_bulk != NULL && rc != req->rq_bulk->bd_nob_transferred) {
1807 CERROR("%s: unexpected size %d, transferred %d\n", obd_name,
1808 rc, req->rq_bulk->bd_nob_transferred);
1812 if (req->rq_bulk == NULL) {
1814 int nob, pg_count, i = 0;
1817 CDEBUG(D_CACHE, "Using short io read, size %d\n", rc);
1818 pg_count = aa->aa_page_count;
1819 buf = req_capsule_server_sized_get(&req->rq_pill, &RMF_SHORT_IO,
1822 while (nob > 0 && pg_count > 0) {
1824 int count = aa->aa_ppga[i]->count > nob ?
1825 nob : aa->aa_ppga[i]->count;
1827 CDEBUG(D_CACHE, "page %p count %d\n",
1828 aa->aa_ppga[i]->pg, count);
1829 ptr = kmap_atomic(aa->aa_ppga[i]->pg);
1830 memcpy(ptr + (aa->aa_ppga[i]->off & ~PAGE_MASK), buf,
1832 kunmap_atomic((void *) ptr);
1841 if (rc < aa->aa_requested_nob)
1842 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1844 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1845 static int cksum_counter;
1846 u32 server_cksum = body->oa.o_cksum;
1849 enum cksum_types cksum_type;
1850 u32 o_flags = body->oa.o_valid & OBD_MD_FLFLAGS ?
1851 body->oa.o_flags : 0;
1853 cksum_type = obd_cksum_type_unpack(o_flags);
1854 rc = osc_checksum_bulk_rw(obd_name, cksum_type, rc,
1855 aa->aa_page_count, aa->aa_ppga,
1856 OST_READ, &client_cksum);
1860 if (req->rq_bulk != NULL &&
1861 peer->nid != req->rq_bulk->bd_sender) {
1863 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1866 if (server_cksum != client_cksum) {
1867 struct ost_body *clbody;
1868 u32 page_count = aa->aa_page_count;
1870 clbody = req_capsule_client_get(&req->rq_pill,
1872 if (cli->cl_checksum_dump)
1873 dump_all_bulk_pages(&clbody->oa, page_count,
1874 aa->aa_ppga, server_cksum,
1877 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1878 "%s%s%s inode "DFID" object "DOSTID
1879 " extent [%llu-%llu], client %x, "
1880 "server %x, cksum_type %x\n",
1882 libcfs_nid2str(peer->nid),
1884 clbody->oa.o_valid & OBD_MD_FLFID ?
1885 clbody->oa.o_parent_seq : 0ULL,
1886 clbody->oa.o_valid & OBD_MD_FLFID ?
1887 clbody->oa.o_parent_oid : 0,
1888 clbody->oa.o_valid & OBD_MD_FLFID ?
1889 clbody->oa.o_parent_ver : 0,
1890 POSTID(&body->oa.o_oi),
1891 aa->aa_ppga[0]->off,
1892 aa->aa_ppga[page_count-1]->off +
1893 aa->aa_ppga[page_count-1]->count - 1,
1894 client_cksum, server_cksum,
1897 aa->aa_oa->o_cksum = client_cksum;
1901 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1904 } else if (unlikely(client_cksum)) {
1905 static int cksum_missed;
1908 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1909 CERROR("%s: checksum %u requested from %s but not sent\n",
1910 obd_name, cksum_missed,
1911 libcfs_nid2str(peer->nid));
1917 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1918 aa->aa_oa, &body->oa);
1923 static int osc_brw_redo_request(struct ptlrpc_request *request,
1924 struct osc_brw_async_args *aa, int rc)
1926 struct ptlrpc_request *new_req;
1927 struct osc_brw_async_args *new_aa;
1928 struct osc_async_page *oap;
1931 /* The below message is checked in replay-ost-single.sh test_8ae*/
1932 DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1933 "redo for recoverable error %d", rc);
1935 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1936 OST_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
1937 aa->aa_cli, aa->aa_oa, aa->aa_page_count,
1938 aa->aa_ppga, &new_req, 1);
1942 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1943 if (oap->oap_request != NULL) {
1944 LASSERTF(request == oap->oap_request,
1945 "request %p != oap_request %p\n",
1946 request, oap->oap_request);
1947 if (oap->oap_interrupted) {
1948 ptlrpc_req_finished(new_req);
1954 * New request takes over pga and oaps from old request.
1955 * Note that copying a list_head doesn't work, need to move it...
1958 new_req->rq_interpret_reply = request->rq_interpret_reply;
1959 new_req->rq_async_args = request->rq_async_args;
1960 new_req->rq_commit_cb = request->rq_commit_cb;
1961 /* cap resend delay to the current request timeout, this is similar to
1962 * what ptlrpc does (see after_reply()) */
1963 if (aa->aa_resends > new_req->rq_timeout)
1964 new_req->rq_sent = ktime_get_real_seconds() + new_req->rq_timeout;
1966 new_req->rq_sent = ktime_get_real_seconds() + aa->aa_resends;
1967 new_req->rq_generation_set = 1;
1968 new_req->rq_import_generation = request->rq_import_generation;
1970 new_aa = ptlrpc_req_async_args(new_aa, new_req);
1972 INIT_LIST_HEAD(&new_aa->aa_oaps);
1973 list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1974 INIT_LIST_HEAD(&new_aa->aa_exts);
1975 list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1976 new_aa->aa_resends = aa->aa_resends;
1978 list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1979 if (oap->oap_request) {
1980 ptlrpc_req_finished(oap->oap_request);
1981 oap->oap_request = ptlrpc_request_addref(new_req);
1985 /* XXX: This code will run into problem if we're going to support
1986 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1987 * and wait for all of them to be finished. We should inherit request
1988 * set from old request. */
1989 ptlrpcd_add_req(new_req);
1991 DEBUG_REQ(D_INFO, new_req, "new request");
1996 * ugh, we want disk allocation on the target to happen in offset order. we'll
1997 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1998 * fine for our small page arrays and doesn't require allocation. its an
1999 * insertion sort that swaps elements that are strides apart, shrinking the
2000 * stride down until its '1' and the array is sorted.
2002 static void sort_brw_pages(struct brw_page **array, int num)
2005 struct brw_page *tmp;
2009 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
2014 for (i = stride ; i < num ; i++) {
2017 while (j >= stride && array[j - stride]->off > tmp->off) {
2018 array[j] = array[j - stride];
2023 } while (stride > 1);
2026 static void osc_release_ppga(struct brw_page **ppga, size_t count)
2028 LASSERT(ppga != NULL);
2029 OBD_FREE(ppga, sizeof(*ppga) * count);
2032 static int brw_interpret(const struct lu_env *env,
2033 struct ptlrpc_request *req, void *args, int rc)
2035 struct osc_brw_async_args *aa = args;
2036 struct osc_extent *ext;
2037 struct osc_extent *tmp;
2038 struct client_obd *cli = aa->aa_cli;
2039 unsigned long transferred = 0;
2043 rc = osc_brw_fini_request(req, rc);
2044 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
2046 * When server returns -EINPROGRESS, client should always retry
2047 * regardless of the number of times the bulk was resent already.
2049 if (osc_recoverable_error(rc) && !req->rq_no_delay) {
2050 if (req->rq_import_generation !=
2051 req->rq_import->imp_generation) {
2052 CDEBUG(D_HA, "%s: resend cross eviction for object: "
2053 ""DOSTID", rc = %d.\n",
2054 req->rq_import->imp_obd->obd_name,
2055 POSTID(&aa->aa_oa->o_oi), rc);
2056 } else if (rc == -EINPROGRESS ||
2057 client_should_resend(aa->aa_resends, aa->aa_cli)) {
2058 rc = osc_brw_redo_request(req, aa, rc);
2060 CERROR("%s: too many resent retries for object: "
2061 "%llu:%llu, rc = %d.\n",
2062 req->rq_import->imp_obd->obd_name,
2063 POSTID(&aa->aa_oa->o_oi), rc);
2068 else if (rc == -EAGAIN || rc == -EINPROGRESS)
2073 struct obdo *oa = aa->aa_oa;
2074 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
2075 unsigned long valid = 0;
2076 struct cl_object *obj;
2077 struct osc_async_page *last;
2079 last = brw_page2oap(aa->aa_ppga[aa->aa_page_count - 1]);
2080 obj = osc2cl(last->oap_obj);
2082 cl_object_attr_lock(obj);
2083 if (oa->o_valid & OBD_MD_FLBLOCKS) {
2084 attr->cat_blocks = oa->o_blocks;
2085 valid |= CAT_BLOCKS;
2087 if (oa->o_valid & OBD_MD_FLMTIME) {
2088 attr->cat_mtime = oa->o_mtime;
2091 if (oa->o_valid & OBD_MD_FLATIME) {
2092 attr->cat_atime = oa->o_atime;
2095 if (oa->o_valid & OBD_MD_FLCTIME) {
2096 attr->cat_ctime = oa->o_ctime;
2100 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
2101 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
2102 loff_t last_off = last->oap_count + last->oap_obj_off +
2105 /* Change file size if this is an out of quota or
2106 * direct IO write and it extends the file size */
2107 if (loi->loi_lvb.lvb_size < last_off) {
2108 attr->cat_size = last_off;
2111 /* Extend KMS if it's not a lockless write */
2112 if (loi->loi_kms < last_off &&
2113 oap2osc_page(last)->ops_srvlock == 0) {
2114 attr->cat_kms = last_off;
2120 cl_object_attr_update(env, obj, attr, valid);
2121 cl_object_attr_unlock(obj);
2123 OBD_SLAB_FREE_PTR(aa->aa_oa, osc_obdo_kmem);
2126 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
2127 osc_inc_unstable_pages(req);
2129 list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
2130 list_del_init(&ext->oe_link);
2131 osc_extent_finish(env, ext, 1,
2132 rc && req->rq_no_delay ? -EWOULDBLOCK : rc);
2134 LASSERT(list_empty(&aa->aa_exts));
2135 LASSERT(list_empty(&aa->aa_oaps));
2137 transferred = (req->rq_bulk == NULL ? /* short io */
2138 aa->aa_requested_nob :
2139 req->rq_bulk->bd_nob_transferred);
2141 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
2142 ptlrpc_lprocfs_brw(req, transferred);
2144 spin_lock(&cli->cl_loi_list_lock);
2145 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2146 * is called so we know whether to go to sync BRWs or wait for more
2147 * RPCs to complete */
2148 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2149 cli->cl_w_in_flight--;
2151 cli->cl_r_in_flight--;
2152 osc_wake_cache_waiters(cli);
2153 spin_unlock(&cli->cl_loi_list_lock);
2155 osc_io_unplug(env, cli, NULL);
2159 static void brw_commit(struct ptlrpc_request *req)
2161 /* If osc_inc_unstable_pages (via osc_extent_finish) races with
2162 * this called via the rq_commit_cb, I need to ensure
2163 * osc_dec_unstable_pages is still called. Otherwise unstable
2164 * pages may be leaked. */
2165 spin_lock(&req->rq_lock);
2166 if (likely(req->rq_unstable)) {
2167 req->rq_unstable = 0;
2168 spin_unlock(&req->rq_lock);
2170 osc_dec_unstable_pages(req);
2172 req->rq_committed = 1;
2173 spin_unlock(&req->rq_lock);
2178 * Build an RPC by the list of extent @ext_list. The caller must ensure
2179 * that the total pages in this list are NOT over max pages per RPC.
2180 * Extents in the list must be in OES_RPC state.
2182 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
2183 struct list_head *ext_list, int cmd)
2185 struct ptlrpc_request *req = NULL;
2186 struct osc_extent *ext;
2187 struct brw_page **pga = NULL;
2188 struct osc_brw_async_args *aa = NULL;
2189 struct obdo *oa = NULL;
2190 struct osc_async_page *oap;
2191 struct osc_object *obj = NULL;
2192 struct cl_req_attr *crattr = NULL;
2193 loff_t starting_offset = OBD_OBJECT_EOF;
2194 loff_t ending_offset = 0;
2198 bool soft_sync = false;
2199 bool interrupted = false;
2200 bool ndelay = false;
2204 __u32 layout_version = 0;
2205 LIST_HEAD(rpc_list);
2206 struct ost_body *body;
2208 LASSERT(!list_empty(ext_list));
2210 /* add pages into rpc_list to build BRW rpc */
2211 list_for_each_entry(ext, ext_list, oe_link) {
2212 LASSERT(ext->oe_state == OES_RPC);
2213 mem_tight |= ext->oe_memalloc;
2214 grant += ext->oe_grants;
2215 page_count += ext->oe_nr_pages;
2216 layout_version = max(layout_version, ext->oe_layout_version);
2221 soft_sync = osc_over_unstable_soft_limit(cli);
2223 mpflag = cfs_memory_pressure_get_and_set();
2225 OBD_ALLOC(pga, sizeof(*pga) * page_count);
2227 GOTO(out, rc = -ENOMEM);
2229 OBD_SLAB_ALLOC_PTR_GFP(oa, osc_obdo_kmem, GFP_NOFS);
2231 GOTO(out, rc = -ENOMEM);
2234 list_for_each_entry(ext, ext_list, oe_link) {
2235 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
2237 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
2239 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
2240 pga[i] = &oap->oap_brw_page;
2241 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2244 list_add_tail(&oap->oap_rpc_item, &rpc_list);
2245 if (starting_offset == OBD_OBJECT_EOF ||
2246 starting_offset > oap->oap_obj_off)
2247 starting_offset = oap->oap_obj_off;
2249 LASSERT(oap->oap_page_off == 0);
2250 if (ending_offset < oap->oap_obj_off + oap->oap_count)
2251 ending_offset = oap->oap_obj_off +
2254 LASSERT(oap->oap_page_off + oap->oap_count ==
2256 if (oap->oap_interrupted)
2263 /* first page in the list */
2264 oap = list_entry(rpc_list.next, typeof(*oap), oap_rpc_item);
2266 crattr = &osc_env_info(env)->oti_req_attr;
2267 memset(crattr, 0, sizeof(*crattr));
2268 crattr->cra_type = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2269 crattr->cra_flags = ~0ULL;
2270 crattr->cra_page = oap2cl_page(oap);
2271 crattr->cra_oa = oa;
2272 cl_req_attr_set(env, osc2cl(obj), crattr);
2274 if (cmd == OBD_BRW_WRITE) {
2275 oa->o_grant_used = grant;
2276 if (layout_version > 0) {
2277 CDEBUG(D_LAYOUT, DFID": write with layout version %u\n",
2278 PFID(&oa->o_oi.oi_fid), layout_version);
2280 oa->o_layout_version = layout_version;
2281 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
2285 sort_brw_pages(pga, page_count);
2286 rc = osc_brw_prep_request(cmd, cli, oa, page_count, pga, &req, 0);
2288 CERROR("prep_req failed: %d\n", rc);
2292 req->rq_commit_cb = brw_commit;
2293 req->rq_interpret_reply = brw_interpret;
2294 req->rq_memalloc = mem_tight != 0;
2295 oap->oap_request = ptlrpc_request_addref(req);
2296 if (interrupted && !req->rq_intr)
2297 ptlrpc_mark_interrupted(req);
2299 req->rq_no_resend = req->rq_no_delay = 1;
2300 /* probably set a shorter timeout value.
2301 * to handle ETIMEDOUT in brw_interpret() correctly. */
2302 /* lustre_msg_set_timeout(req, req->rq_timeout / 2); */
2305 /* Need to update the timestamps after the request is built in case
2306 * we race with setattr (locally or in queue at OST). If OST gets
2307 * later setattr before earlier BRW (as determined by the request xid),
2308 * the OST will not use BRW timestamps. Sadly, there is no obvious
2309 * way to do this in a single call. bug 10150 */
2310 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2311 crattr->cra_oa = &body->oa;
2312 crattr->cra_flags = OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLATIME;
2313 cl_req_attr_set(env, osc2cl(obj), crattr);
2314 lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
2316 aa = ptlrpc_req_async_args(aa, req);
2317 INIT_LIST_HEAD(&aa->aa_oaps);
2318 list_splice_init(&rpc_list, &aa->aa_oaps);
2319 INIT_LIST_HEAD(&aa->aa_exts);
2320 list_splice_init(ext_list, &aa->aa_exts);
2322 spin_lock(&cli->cl_loi_list_lock);
2323 starting_offset >>= PAGE_SHIFT;
2324 if (cmd == OBD_BRW_READ) {
2325 cli->cl_r_in_flight++;
2326 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2327 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2328 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2329 starting_offset + 1);
2331 cli->cl_w_in_flight++;
2332 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2333 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
2334 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2335 starting_offset + 1);
2337 spin_unlock(&cli->cl_loi_list_lock);
2339 DEBUG_REQ(D_INODE, req, "%d pages, aa %p, now %ur/%uw in flight",
2340 page_count, aa, cli->cl_r_in_flight,
2341 cli->cl_w_in_flight);
2342 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_IO, cfs_fail_val);
2344 ptlrpcd_add_req(req);
2350 cfs_memory_pressure_restore(mpflag);
2353 LASSERT(req == NULL);
2356 OBD_SLAB_FREE_PTR(oa, osc_obdo_kmem);
2358 OBD_FREE(pga, sizeof(*pga) * page_count);
2359 /* this should happen rarely and is pretty bad, it makes the
2360 * pending list not follow the dirty order */
2361 while (!list_empty(ext_list)) {
2362 ext = list_entry(ext_list->next, struct osc_extent,
2364 list_del_init(&ext->oe_link);
2365 osc_extent_finish(env, ext, 0, rc);
2371 static int osc_set_lock_data(struct ldlm_lock *lock, void *data)
2375 LASSERT(lock != NULL);
2377 lock_res_and_lock(lock);
2379 if (lock->l_ast_data == NULL)
2380 lock->l_ast_data = data;
2381 if (lock->l_ast_data == data)
2384 unlock_res_and_lock(lock);
2389 int osc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
2390 void *cookie, struct lustre_handle *lockh,
2391 enum ldlm_mode mode, __u64 *flags, bool speculative,
2394 bool intent = *flags & LDLM_FL_HAS_INTENT;
2398 /* The request was created before ldlm_cli_enqueue call. */
2399 if (intent && errcode == ELDLM_LOCK_ABORTED) {
2400 struct ldlm_reply *rep;
2402 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
2403 LASSERT(rep != NULL);
2405 rep->lock_policy_res1 =
2406 ptlrpc_status_ntoh(rep->lock_policy_res1);
2407 if (rep->lock_policy_res1)
2408 errcode = rep->lock_policy_res1;
2410 *flags |= LDLM_FL_LVB_READY;
2411 } else if (errcode == ELDLM_OK) {
2412 *flags |= LDLM_FL_LVB_READY;
2415 /* Call the update callback. */
2416 rc = (*upcall)(cookie, lockh, errcode);
2418 /* release the reference taken in ldlm_cli_enqueue() */
2419 if (errcode == ELDLM_LOCK_MATCHED)
2421 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
2422 ldlm_lock_decref(lockh, mode);
2427 int osc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2430 struct osc_enqueue_args *aa = args;
2431 struct ldlm_lock *lock;
2432 struct lustre_handle *lockh = &aa->oa_lockh;
2433 enum ldlm_mode mode = aa->oa_mode;
2434 struct ost_lvb *lvb = aa->oa_lvb;
2435 __u32 lvb_len = sizeof(*lvb);
2440 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2442 lock = ldlm_handle2lock(lockh);
2443 LASSERTF(lock != NULL,
2444 "lockh %#llx, req %p, aa %p - client evicted?\n",
2445 lockh->cookie, req, aa);
2447 /* Take an additional reference so that a blocking AST that
2448 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2449 * to arrive after an upcall has been executed by
2450 * osc_enqueue_fini(). */
2451 ldlm_lock_addref(lockh, mode);
2453 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
2454 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
2456 /* Let CP AST to grant the lock first. */
2457 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2459 if (aa->oa_speculative) {
2460 LASSERT(aa->oa_lvb == NULL);
2461 LASSERT(aa->oa_flags == NULL);
2462 aa->oa_flags = &flags;
2465 /* Complete obtaining the lock procedure. */
2466 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
2467 aa->oa_mode, aa->oa_flags, lvb, lvb_len,
2469 /* Complete osc stuff. */
2470 rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
2471 aa->oa_flags, aa->oa_speculative, rc);
2473 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2475 ldlm_lock_decref(lockh, mode);
2476 LDLM_LOCK_PUT(lock);
2480 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2482 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2483 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2484 * other synchronous requests, however keeping some locks and trying to obtain
2485 * others may take a considerable amount of time in a case of ost failure; and
2486 * when other sync requests do not get released lock from a client, the client
2487 * is evicted from the cluster -- such scenarious make the life difficult, so
2488 * release locks just after they are obtained. */
2489 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2490 __u64 *flags, union ldlm_policy_data *policy,
2491 struct ost_lvb *lvb, osc_enqueue_upcall_f upcall,
2492 void *cookie, struct ldlm_enqueue_info *einfo,
2493 struct ptlrpc_request_set *rqset, int async,
2496 struct obd_device *obd = exp->exp_obd;
2497 struct lustre_handle lockh = { 0 };
2498 struct ptlrpc_request *req = NULL;
2499 int intent = *flags & LDLM_FL_HAS_INTENT;
2500 __u64 match_flags = *flags;
2501 enum ldlm_mode mode;
2505 /* Filesystem lock extents are extended to page boundaries so that
2506 * dealing with the page cache is a little smoother. */
2507 policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2508 policy->l_extent.end |= ~PAGE_MASK;
2510 /* Next, search for already existing extent locks that will cover us */
2511 /* If we're trying to read, we also search for an existing PW lock. The
2512 * VFS and page cache already protect us locally, so lots of readers/
2513 * writers can share a single PW lock.
2515 * There are problems with conversion deadlocks, so instead of
2516 * converting a read lock to a write lock, we'll just enqueue a new
2519 * At some point we should cancel the read lock instead of making them
2520 * send us a blocking callback, but there are problems with canceling
2521 * locks out from other users right now, too. */
2522 mode = einfo->ei_mode;
2523 if (einfo->ei_mode == LCK_PR)
2525 /* Normal lock requests must wait for the LVB to be ready before
2526 * matching a lock; speculative lock requests do not need to,
2527 * because they will not actually use the lock. */
2529 match_flags |= LDLM_FL_LVB_READY;
2531 match_flags |= LDLM_FL_BLOCK_GRANTED;
2532 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
2533 einfo->ei_type, policy, mode, &lockh, 0);
2535 struct ldlm_lock *matched;
2537 if (*flags & LDLM_FL_TEST_LOCK)
2540 matched = ldlm_handle2lock(&lockh);
2542 /* This DLM lock request is speculative, and does not
2543 * have an associated IO request. Therefore if there
2544 * is already a DLM lock, it wll just inform the
2545 * caller to cancel the request for this stripe.*/
2546 lock_res_and_lock(matched);
2547 if (ldlm_extent_equal(&policy->l_extent,
2548 &matched->l_policy_data.l_extent))
2552 unlock_res_and_lock(matched);
2554 ldlm_lock_decref(&lockh, mode);
2555 LDLM_LOCK_PUT(matched);
2557 } else if (osc_set_lock_data(matched, einfo->ei_cbdata)) {
2558 *flags |= LDLM_FL_LVB_READY;
2560 /* We already have a lock, and it's referenced. */
2561 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
2563 ldlm_lock_decref(&lockh, mode);
2564 LDLM_LOCK_PUT(matched);
2567 ldlm_lock_decref(&lockh, mode);
2568 LDLM_LOCK_PUT(matched);
2572 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
2576 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2577 &RQF_LDLM_ENQUEUE_LVB);
2581 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
2583 ptlrpc_request_free(req);
2587 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
2589 ptlrpc_request_set_replen(req);
2592 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2593 *flags &= ~LDLM_FL_BLOCK_GRANTED;
2595 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2596 sizeof(*lvb), LVB_T_OST, &lockh, async);
2599 struct osc_enqueue_args *aa;
2600 aa = ptlrpc_req_async_args(aa, req);
2602 aa->oa_mode = einfo->ei_mode;
2603 aa->oa_type = einfo->ei_type;
2604 lustre_handle_copy(&aa->oa_lockh, &lockh);
2605 aa->oa_upcall = upcall;
2606 aa->oa_cookie = cookie;
2607 aa->oa_speculative = speculative;
2609 aa->oa_flags = flags;
2612 /* speculative locks are essentially to enqueue
2613 * a DLM lock in advance, so we don't care
2614 * about the result of the enqueue. */
2616 aa->oa_flags = NULL;
2619 req->rq_interpret_reply = osc_enqueue_interpret;
2620 if (rqset == PTLRPCD_SET)
2621 ptlrpcd_add_req(req);
2623 ptlrpc_set_add_req(rqset, req);
2624 } else if (intent) {
2625 ptlrpc_req_finished(req);
2630 rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
2631 flags, speculative, rc);
2633 ptlrpc_req_finished(req);
2638 int osc_match_base(const struct lu_env *env, struct obd_export *exp,
2639 struct ldlm_res_id *res_id, enum ldlm_type type,
2640 union ldlm_policy_data *policy, enum ldlm_mode mode,
2641 __u64 *flags, struct osc_object *obj,
2642 struct lustre_handle *lockh, int unref)
2644 struct obd_device *obd = exp->exp_obd;
2645 __u64 lflags = *flags;
2649 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
2652 /* Filesystem lock extents are extended to page boundaries so that
2653 * dealing with the page cache is a little smoother */
2654 policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2655 policy->l_extent.end |= ~PAGE_MASK;
2657 /* Next, search for already existing extent locks that will cover us */
2658 /* If we're trying to read, we also search for an existing PW lock. The
2659 * VFS and page cache already protect us locally, so lots of readers/
2660 * writers can share a single PW lock. */
2664 rc = ldlm_lock_match(obd->obd_namespace, lflags,
2665 res_id, type, policy, rc, lockh, unref);
2666 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
2670 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2672 LASSERT(lock != NULL);
2673 if (osc_set_lock_data(lock, obj)) {
2674 lock_res_and_lock(lock);
2675 if (!ldlm_is_lvb_cached(lock)) {
2676 LASSERT(lock->l_ast_data == obj);
2677 osc_lock_lvb_update(env, obj, lock, NULL);
2678 ldlm_set_lvb_cached(lock);
2680 unlock_res_and_lock(lock);
2682 ldlm_lock_decref(lockh, rc);
2685 LDLM_LOCK_PUT(lock);
2690 static int osc_statfs_interpret(const struct lu_env *env,
2691 struct ptlrpc_request *req, void *args, int rc)
2693 struct osc_async_args *aa = args;
2694 struct obd_statfs *msfs;
2699 * The request has in fact never been sent due to issues at
2700 * a higher level (LOV). Exit immediately since the caller
2701 * is aware of the problem and takes care of the clean up.
2705 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2706 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2712 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2714 GOTO(out, rc = -EPROTO);
2716 *aa->aa_oi->oi_osfs = *msfs;
2718 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
2723 static int osc_statfs_async(struct obd_export *exp,
2724 struct obd_info *oinfo, time64_t max_age,
2725 struct ptlrpc_request_set *rqset)
2727 struct obd_device *obd = class_exp2obd(exp);
2728 struct ptlrpc_request *req;
2729 struct osc_async_args *aa;
2733 if (obd->obd_osfs_age >= max_age) {
2735 "%s: use %p cache blocks %llu/%llu objects %llu/%llu\n",
2736 obd->obd_name, &obd->obd_osfs,
2737 obd->obd_osfs.os_bavail, obd->obd_osfs.os_blocks,
2738 obd->obd_osfs.os_ffree, obd->obd_osfs.os_files);
2739 spin_lock(&obd->obd_osfs_lock);
2740 memcpy(oinfo->oi_osfs, &obd->obd_osfs, sizeof(*oinfo->oi_osfs));
2741 spin_unlock(&obd->obd_osfs_lock);
2742 oinfo->oi_flags |= OBD_STATFS_FROM_CACHE;
2743 if (oinfo->oi_cb_up)
2744 oinfo->oi_cb_up(oinfo, 0);
2749 /* We could possibly pass max_age in the request (as an absolute
2750 * timestamp or a "seconds.usec ago") so the target can avoid doing
2751 * extra calls into the filesystem if that isn't necessary (e.g.
2752 * during mount that would help a bit). Having relative timestamps
2753 * is not so great if request processing is slow, while absolute
2754 * timestamps are not ideal because they need time synchronization. */
2755 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2759 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2761 ptlrpc_request_free(req);
2764 ptlrpc_request_set_replen(req);
2765 req->rq_request_portal = OST_CREATE_PORTAL;
2766 ptlrpc_at_set_req_timeout(req);
2768 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2769 /* procfs requests not want stat in wait for avoid deadlock */
2770 req->rq_no_resend = 1;
2771 req->rq_no_delay = 1;
2774 req->rq_interpret_reply = osc_statfs_interpret;
2775 aa = ptlrpc_req_async_args(aa, req);
2778 ptlrpc_set_add_req(rqset, req);
2782 static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2783 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
2785 struct obd_device *obd = class_exp2obd(exp);
2786 struct obd_statfs *msfs;
2787 struct ptlrpc_request *req;
2788 struct obd_import *imp = NULL;
2793 /*Since the request might also come from lprocfs, so we need
2794 *sync this with client_disconnect_export Bug15684*/
2795 down_read(&obd->u.cli.cl_sem);
2796 if (obd->u.cli.cl_import)
2797 imp = class_import_get(obd->u.cli.cl_import);
2798 up_read(&obd->u.cli.cl_sem);
2802 /* We could possibly pass max_age in the request (as an absolute
2803 * timestamp or a "seconds.usec ago") so the target can avoid doing
2804 * extra calls into the filesystem if that isn't necessary (e.g.
2805 * during mount that would help a bit). Having relative timestamps
2806 * is not so great if request processing is slow, while absolute
2807 * timestamps are not ideal because they need time synchronization. */
2808 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2810 class_import_put(imp);
2815 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2817 ptlrpc_request_free(req);
2820 ptlrpc_request_set_replen(req);
2821 req->rq_request_portal = OST_CREATE_PORTAL;
2822 ptlrpc_at_set_req_timeout(req);
2824 if (flags & OBD_STATFS_NODELAY) {
2825 /* procfs requests not want stat in wait for avoid deadlock */
2826 req->rq_no_resend = 1;
2827 req->rq_no_delay = 1;
2830 rc = ptlrpc_queue_wait(req);
2834 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2836 GOTO(out, rc = -EPROTO);
2842 ptlrpc_req_finished(req);
2846 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2847 void *karg, void __user *uarg)
2849 struct obd_device *obd = exp->exp_obd;
2850 struct obd_ioctl_data *data = karg;
2854 if (!try_module_get(THIS_MODULE)) {
2855 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2856 module_name(THIS_MODULE));
2860 case OBD_IOC_CLIENT_RECOVER:
2861 rc = ptlrpc_recover_import(obd->u.cli.cl_import,
2862 data->ioc_inlbuf1, 0);
2866 case IOC_OSC_SET_ACTIVE:
2867 rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
2872 CDEBUG(D_INODE, "%s: unrecognised ioctl %#x by %s: rc = %d\n",
2873 obd->obd_name, cmd, current_comm(), rc);
2877 module_put(THIS_MODULE);
2881 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
2882 u32 keylen, void *key, u32 vallen, void *val,
2883 struct ptlrpc_request_set *set)
2885 struct ptlrpc_request *req;
2886 struct obd_device *obd = exp->exp_obd;
2887 struct obd_import *imp = class_exp2cliimp(exp);
2892 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
2894 if (KEY_IS(KEY_CHECKSUM)) {
2895 if (vallen != sizeof(int))
2897 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
2901 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2902 sptlrpc_conf_client_adapt(obd);
2906 if (KEY_IS(KEY_FLUSH_CTX)) {
2907 sptlrpc_import_flush_my_ctx(imp);
2911 if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
2912 struct client_obd *cli = &obd->u.cli;
2913 long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
2914 long target = *(long *)val;
2916 nr = osc_lru_shrink(env, cli, min(nr, target), true);
2921 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
2924 /* We pass all other commands directly to OST. Since nobody calls osc
2925 methods directly and everybody is supposed to go through LOV, we
2926 assume lov checked invalid values for us.
2927 The only recognised values so far are evict_by_nid and mds_conn.
2928 Even if something bad goes through, we'd get a -EINVAL from OST
2931 req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
2932 &RQF_OST_SET_GRANT_INFO :
2937 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2938 RCL_CLIENT, keylen);
2939 if (!KEY_IS(KEY_GRANT_SHRINK))
2940 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
2941 RCL_CLIENT, vallen);
2942 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
2944 ptlrpc_request_free(req);
2948 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
2949 memcpy(tmp, key, keylen);
2950 tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
2953 memcpy(tmp, val, vallen);
2955 if (KEY_IS(KEY_GRANT_SHRINK)) {
2956 struct osc_grant_args *aa;
2959 aa = ptlrpc_req_async_args(aa, req);
2960 OBD_SLAB_ALLOC_PTR_GFP(oa, osc_obdo_kmem, GFP_NOFS);
2962 ptlrpc_req_finished(req);
2965 *oa = ((struct ost_body *)val)->oa;
2967 req->rq_interpret_reply = osc_shrink_grant_interpret;
2970 ptlrpc_request_set_replen(req);
2971 if (!KEY_IS(KEY_GRANT_SHRINK)) {
2972 LASSERT(set != NULL);
2973 ptlrpc_set_add_req(set, req);
2974 ptlrpc_check_set(NULL, set);
2976 ptlrpcd_add_req(req);
2981 EXPORT_SYMBOL(osc_set_info_async);
2983 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
2984 struct obd_device *obd, struct obd_uuid *cluuid,
2985 struct obd_connect_data *data, void *localdata)
2987 struct client_obd *cli = &obd->u.cli;
2989 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
2993 spin_lock(&cli->cl_loi_list_lock);
2994 grant = cli->cl_avail_grant + cli->cl_reserved_grant;
2995 if (data->ocd_connect_flags & OBD_CONNECT_GRANT_PARAM) {
2996 /* restore ocd_grant_blkbits as client page bits */
2997 data->ocd_grant_blkbits = PAGE_SHIFT;
2998 grant += cli->cl_dirty_grant;
3000 grant += cli->cl_dirty_pages << PAGE_SHIFT;
3002 data->ocd_grant = grant ? : 2 * cli_brw_size(obd);
3003 lost_grant = cli->cl_lost_grant;
3004 cli->cl_lost_grant = 0;
3005 spin_unlock(&cli->cl_loi_list_lock);
3007 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d"
3008 " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
3009 data->ocd_version, data->ocd_grant, lost_grant);
3014 EXPORT_SYMBOL(osc_reconnect);
3016 int osc_disconnect(struct obd_export *exp)
3018 struct obd_device *obd = class_exp2obd(exp);
3021 rc = client_disconnect_export(exp);
3023 * Initially we put del_shrink_grant before disconnect_export, but it
3024 * causes the following problem if setup (connect) and cleanup
3025 * (disconnect) are tangled together.
3026 * connect p1 disconnect p2
3027 * ptlrpc_connect_import
3028 * ............... class_manual_cleanup
3031 * ptlrpc_connect_interrupt
3033 * add this client to shrink list
3035 * Bang! grant shrink thread trigger the shrink. BUG18662
3037 osc_del_grant_list(&obd->u.cli);
3040 EXPORT_SYMBOL(osc_disconnect);
3042 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
3043 struct hlist_node *hnode, void *arg)
3045 struct lu_env *env = arg;
3046 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
3047 struct ldlm_lock *lock;
3048 struct osc_object *osc = NULL;
3052 list_for_each_entry(lock, &res->lr_granted, l_res_link) {
3053 if (lock->l_ast_data != NULL && osc == NULL) {
3054 osc = lock->l_ast_data;
3055 cl_object_get(osc2cl(osc));
3058 /* clear LDLM_FL_CLEANED flag to make sure it will be canceled
3059 * by the 2nd round of ldlm_namespace_clean() call in
3060 * osc_import_event(). */
3061 ldlm_clear_cleaned(lock);
3066 osc_object_invalidate(env, osc);
3067 cl_object_put(env, osc2cl(osc));
3072 EXPORT_SYMBOL(osc_ldlm_resource_invalidate);
3074 static int osc_import_event(struct obd_device *obd,
3075 struct obd_import *imp,
3076 enum obd_import_event event)
3078 struct client_obd *cli;
3082 LASSERT(imp->imp_obd == obd);
3085 case IMP_EVENT_DISCON: {
3087 spin_lock(&cli->cl_loi_list_lock);
3088 cli->cl_avail_grant = 0;
3089 cli->cl_lost_grant = 0;
3090 spin_unlock(&cli->cl_loi_list_lock);
3093 case IMP_EVENT_INACTIVE: {
3094 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
3097 case IMP_EVENT_INVALIDATE: {
3098 struct ldlm_namespace *ns = obd->obd_namespace;
3102 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3104 env = cl_env_get(&refcheck);
3106 osc_io_unplug(env, &obd->u.cli, NULL);
3108 cfs_hash_for_each_nolock(ns->ns_rs_hash,
3109 osc_ldlm_resource_invalidate,
3111 cl_env_put(env, &refcheck);
3113 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3118 case IMP_EVENT_ACTIVE: {
3119 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
3122 case IMP_EVENT_OCD: {
3123 struct obd_connect_data *ocd = &imp->imp_connect_data;
3125 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
3126 osc_init_grant(&obd->u.cli, ocd);
3129 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
3130 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
3132 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
3135 case IMP_EVENT_DEACTIVATE: {
3136 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE);
3139 case IMP_EVENT_ACTIVATE: {
3140 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE);
3144 CERROR("Unknown import event %d\n", event);
3151 * Determine whether the lock can be canceled before replaying the lock
3152 * during recovery, see bug16774 for detailed information.
3154 * \retval zero the lock can't be canceled
3155 * \retval other ok to cancel
3157 static int osc_cancel_weight(struct ldlm_lock *lock)
3160 * Cancel all unused and granted extent lock.
3162 if (lock->l_resource->lr_type == LDLM_EXTENT &&
3163 ldlm_is_granted(lock) &&
3164 osc_ldlm_weigh_ast(lock) == 0)
3170 static int brw_queue_work(const struct lu_env *env, void *data)
3172 struct client_obd *cli = data;
3174 CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
3176 osc_io_unplug(env, cli, NULL);
3180 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg)
3182 struct client_obd *cli = &obd->u.cli;
3188 rc = ptlrpcd_addref();
3192 rc = client_obd_setup(obd, lcfg);
3194 GOTO(out_ptlrpcd, rc);
3197 handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
3198 if (IS_ERR(handler))
3199 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
3200 cli->cl_writeback_work = handler;
3202 handler = ptlrpcd_alloc_work(cli->cl_import, lru_queue_work, cli);
3203 if (IS_ERR(handler))
3204 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
3205 cli->cl_lru_work = handler;
3207 rc = osc_quota_setup(obd);
3209 GOTO(out_ptlrpcd_work, rc);
3211 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
3212 osc_update_next_shrink(cli);
3217 if (cli->cl_writeback_work != NULL) {
3218 ptlrpcd_destroy_work(cli->cl_writeback_work);
3219 cli->cl_writeback_work = NULL;
3221 if (cli->cl_lru_work != NULL) {
3222 ptlrpcd_destroy_work(cli->cl_lru_work);
3223 cli->cl_lru_work = NULL;
3225 client_obd_cleanup(obd);
3230 EXPORT_SYMBOL(osc_setup_common);
3232 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
3234 struct client_obd *cli = &obd->u.cli;
3242 rc = osc_setup_common(obd, lcfg);
3246 rc = osc_tunables_init(obd);
3251 * We try to control the total number of requests with a upper limit
3252 * osc_reqpool_maxreqcount. There might be some race which will cause
3253 * over-limit allocation, but it is fine.
3255 req_count = atomic_read(&osc_pool_req_count);
3256 if (req_count < osc_reqpool_maxreqcount) {
3257 adding = cli->cl_max_rpcs_in_flight + 2;
3258 if (req_count + adding > osc_reqpool_maxreqcount)
3259 adding = osc_reqpool_maxreqcount - req_count;
3261 added = ptlrpc_add_rqs_to_pool(osc_rq_pool, adding);
3262 atomic_add(added, &osc_pool_req_count);
3265 ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
3267 spin_lock(&osc_shrink_lock);
3268 list_add_tail(&cli->cl_shrink_list, &osc_shrink_list);
3269 spin_unlock(&osc_shrink_lock);
3270 cli->cl_import->imp_idle_timeout = osc_idle_timeout;
3271 cli->cl_import->imp_idle_debug = D_HA;
3276 int osc_precleanup_common(struct obd_device *obd)
3278 struct client_obd *cli = &obd->u.cli;
3282 * for echo client, export may be on zombie list, wait for
3283 * zombie thread to cull it, because cli.cl_import will be
3284 * cleared in client_disconnect_export():
3285 * class_export_destroy() -> obd_cleanup() ->
3286 * echo_device_free() -> echo_client_cleanup() ->
3287 * obd_disconnect() -> osc_disconnect() ->
3288 * client_disconnect_export()
3290 obd_zombie_barrier();
3291 if (cli->cl_writeback_work) {
3292 ptlrpcd_destroy_work(cli->cl_writeback_work);
3293 cli->cl_writeback_work = NULL;
3296 if (cli->cl_lru_work) {
3297 ptlrpcd_destroy_work(cli->cl_lru_work);
3298 cli->cl_lru_work = NULL;
3301 obd_cleanup_client_import(obd);
3304 EXPORT_SYMBOL(osc_precleanup_common);
3306 static int osc_precleanup(struct obd_device *obd)
3310 osc_precleanup_common(obd);
3312 ptlrpc_lprocfs_unregister_obd(obd);
3316 int osc_cleanup_common(struct obd_device *obd)
3318 struct client_obd *cli = &obd->u.cli;
3323 spin_lock(&osc_shrink_lock);
3324 list_del(&cli->cl_shrink_list);
3325 spin_unlock(&osc_shrink_lock);
3328 if (cli->cl_cache != NULL) {
3329 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
3330 spin_lock(&cli->cl_cache->ccc_lru_lock);
3331 list_del_init(&cli->cl_lru_osc);
3332 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3333 cli->cl_lru_left = NULL;
3334 cl_cache_decref(cli->cl_cache);
3335 cli->cl_cache = NULL;
3338 /* free memory of osc quota cache */
3339 osc_quota_cleanup(obd);
3341 rc = client_obd_cleanup(obd);
3346 EXPORT_SYMBOL(osc_cleanup_common);
3348 static const struct obd_ops osc_obd_ops = {
3349 .o_owner = THIS_MODULE,
3350 .o_setup = osc_setup,
3351 .o_precleanup = osc_precleanup,
3352 .o_cleanup = osc_cleanup_common,
3353 .o_add_conn = client_import_add_conn,
3354 .o_del_conn = client_import_del_conn,
3355 .o_connect = client_connect_import,
3356 .o_reconnect = osc_reconnect,
3357 .o_disconnect = osc_disconnect,
3358 .o_statfs = osc_statfs,
3359 .o_statfs_async = osc_statfs_async,
3360 .o_create = osc_create,
3361 .o_destroy = osc_destroy,
3362 .o_getattr = osc_getattr,
3363 .o_setattr = osc_setattr,
3364 .o_iocontrol = osc_iocontrol,
3365 .o_set_info_async = osc_set_info_async,
3366 .o_import_event = osc_import_event,
3367 .o_quotactl = osc_quotactl,
3370 static struct shrinker *osc_cache_shrinker;
3371 LIST_HEAD(osc_shrink_list);
3372 DEFINE_SPINLOCK(osc_shrink_lock);
3374 #ifndef HAVE_SHRINKER_COUNT
3375 static int osc_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
3377 struct shrink_control scv = {
3378 .nr_to_scan = shrink_param(sc, nr_to_scan),
3379 .gfp_mask = shrink_param(sc, gfp_mask)
3381 (void)osc_cache_shrink_scan(shrinker, &scv);
3383 return osc_cache_shrink_count(shrinker, &scv);
3387 static int __init osc_init(void)
3389 unsigned int reqpool_size;
3390 unsigned int reqsize;
3392 DEF_SHRINKER_VAR(osc_shvar, osc_cache_shrink,
3393 osc_cache_shrink_count, osc_cache_shrink_scan);
3396 /* print an address of _any_ initialized kernel symbol from this
3397 * module, to allow debugging with gdb that doesn't support data
3398 * symbols from modules.*/
3399 CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
3401 rc = lu_kmem_init(osc_caches);
3405 rc = class_register_type(&osc_obd_ops, NULL, true, NULL,
3406 LUSTRE_OSC_NAME, &osc_device_type);
3410 osc_cache_shrinker = set_shrinker(DEFAULT_SEEKS, &osc_shvar);
3412 /* This is obviously too much memory, only prevent overflow here */
3413 if (osc_reqpool_mem_max >= 1 << 12 || osc_reqpool_mem_max == 0)
3414 GOTO(out_type, rc = -EINVAL);
3416 reqpool_size = osc_reqpool_mem_max << 20;
3419 while (reqsize < OST_IO_MAXREQSIZE)
3420 reqsize = reqsize << 1;
3423 * We don't enlarge the request count in OSC pool according to
3424 * cl_max_rpcs_in_flight. The allocation from the pool will only be
3425 * tried after normal allocation failed. So a small OSC pool won't
3426 * cause much performance degression in most of cases.
3428 osc_reqpool_maxreqcount = reqpool_size / reqsize;
3430 atomic_set(&osc_pool_req_count, 0);
3431 osc_rq_pool = ptlrpc_init_rq_pool(0, OST_IO_MAXREQSIZE,
3432 ptlrpc_add_rqs_to_pool);
3434 if (osc_rq_pool == NULL)
3435 GOTO(out_type, rc = -ENOMEM);
3437 rc = osc_start_grant_work();
3439 GOTO(out_req_pool, rc);
3444 ptlrpc_free_rq_pool(osc_rq_pool);
3446 class_unregister_type(LUSTRE_OSC_NAME);
3448 lu_kmem_fini(osc_caches);
3453 static void __exit osc_exit(void)
3455 osc_stop_grant_work();
3456 remove_shrinker(osc_cache_shrinker);
3457 class_unregister_type(LUSTRE_OSC_NAME);
3458 lu_kmem_fini(osc_caches);
3459 ptlrpc_free_rq_pool(osc_rq_pool);
3462 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3463 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3464 MODULE_VERSION(LUSTRE_VERSION_STRING);
3465 MODULE_LICENSE("GPL");
3467 module_init(osc_init);
3468 module_exit(osc_exit);