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, 2016, 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 <libcfs/libcfs.h>
37 #include <lprocfs_status.h>
38 #include <lustre_debug.h>
39 #include <lustre_dlm.h>
40 #include <lustre_fid.h>
41 #include <lustre_ha.h>
42 #include <uapi/linux/lustre/lustre_ioctl.h>
43 #include <lustre_net.h>
44 #include <lustre_obdo.h>
45 #include <uapi/linux/lustre/lustre_param.h>
47 #include <obd_cksum.h>
48 #include <obd_class.h>
49 #include <lustre_osc.h>
51 #include "osc_internal.h"
53 atomic_t osc_pool_req_count;
54 unsigned int osc_reqpool_maxreqcount;
55 struct ptlrpc_request_pool *osc_rq_pool;
57 /* max memory used for request pool, unit is MB */
58 static unsigned int osc_reqpool_mem_max = 5;
59 module_param(osc_reqpool_mem_max, uint, 0444);
61 #define osc_grant_args osc_brw_async_args
63 struct osc_setattr_args {
65 obd_enqueue_update_f sa_upcall;
69 struct osc_fsync_args {
70 struct osc_object *fa_obj;
72 obd_enqueue_update_f fa_upcall;
76 struct osc_ladvise_args {
78 obd_enqueue_update_f la_upcall;
82 static void osc_release_ppga(struct brw_page **ppga, size_t count);
83 static int brw_interpret(const struct lu_env *env, struct ptlrpc_request *req,
86 void osc_pack_req_body(struct ptlrpc_request *req, struct obdo *oa)
88 struct ost_body *body;
90 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
93 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
96 static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
99 struct ptlrpc_request *req;
100 struct ost_body *body;
104 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
108 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
110 ptlrpc_request_free(req);
114 osc_pack_req_body(req, oa);
116 ptlrpc_request_set_replen(req);
118 rc = ptlrpc_queue_wait(req);
122 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
124 GOTO(out, rc = -EPROTO);
126 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
127 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
129 oa->o_blksize = cli_brw_size(exp->exp_obd);
130 oa->o_valid |= OBD_MD_FLBLKSZ;
134 ptlrpc_req_finished(req);
139 static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
142 struct ptlrpc_request *req;
143 struct ost_body *body;
147 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
149 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
153 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
155 ptlrpc_request_free(req);
159 osc_pack_req_body(req, oa);
161 ptlrpc_request_set_replen(req);
163 rc = ptlrpc_queue_wait(req);
167 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
169 GOTO(out, rc = -EPROTO);
171 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
175 ptlrpc_req_finished(req);
180 static int osc_setattr_interpret(const struct lu_env *env,
181 struct ptlrpc_request *req,
182 struct osc_setattr_args *sa, int rc)
184 struct ost_body *body;
190 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
192 GOTO(out, rc = -EPROTO);
194 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
197 rc = sa->sa_upcall(sa->sa_cookie, rc);
201 int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
202 obd_enqueue_update_f upcall, void *cookie,
203 struct ptlrpc_request_set *rqset)
205 struct ptlrpc_request *req;
206 struct osc_setattr_args *sa;
211 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
215 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
217 ptlrpc_request_free(req);
221 osc_pack_req_body(req, oa);
223 ptlrpc_request_set_replen(req);
225 /* do mds to ost setattr asynchronously */
227 /* Do not wait for response. */
228 ptlrpcd_add_req(req);
230 req->rq_interpret_reply =
231 (ptlrpc_interpterer_t)osc_setattr_interpret;
233 CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
234 sa = ptlrpc_req_async_args(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 CLASSERT(sizeof(*la) <= sizeof(req->rq_async_args));
323 la = ptlrpc_req_async_args(req);
325 la->la_upcall = upcall;
326 la->la_cookie = cookie;
328 if (rqset == PTLRPCD_SET)
329 ptlrpcd_add_req(req);
331 ptlrpc_set_add_req(rqset, req);
336 static int osc_create(const struct lu_env *env, struct obd_export *exp,
339 struct ptlrpc_request *req;
340 struct ost_body *body;
345 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
346 LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
348 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
350 GOTO(out, rc = -ENOMEM);
352 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
354 ptlrpc_request_free(req);
358 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
361 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
363 ptlrpc_request_set_replen(req);
365 rc = ptlrpc_queue_wait(req);
369 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
371 GOTO(out_req, rc = -EPROTO);
373 CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
374 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
376 oa->o_blksize = cli_brw_size(exp->exp_obd);
377 oa->o_valid |= OBD_MD_FLBLKSZ;
379 CDEBUG(D_HA, "transno: %lld\n",
380 lustre_msg_get_transno(req->rq_repmsg));
382 ptlrpc_req_finished(req);
387 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
388 obd_enqueue_update_f upcall, void *cookie)
390 struct ptlrpc_request *req;
391 struct osc_setattr_args *sa;
392 struct obd_import *imp = class_exp2cliimp(exp);
393 struct ost_body *body;
398 req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
402 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
404 ptlrpc_request_free(req);
408 osc_set_io_portal(req);
410 ptlrpc_at_set_req_timeout(req);
412 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
414 lustre_set_wire_obdo(&imp->imp_connect_data, &body->oa, oa);
416 ptlrpc_request_set_replen(req);
418 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
419 CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
420 sa = ptlrpc_req_async_args(req);
422 sa->sa_upcall = upcall;
423 sa->sa_cookie = cookie;
425 ptlrpcd_add_req(req);
429 EXPORT_SYMBOL(osc_punch_send);
431 static int osc_sync_interpret(const struct lu_env *env,
432 struct ptlrpc_request *req,
435 struct osc_fsync_args *fa = arg;
436 struct ost_body *body;
437 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
438 unsigned long valid = 0;
439 struct cl_object *obj;
445 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
447 CERROR("can't unpack ost_body\n");
448 GOTO(out, rc = -EPROTO);
451 *fa->fa_oa = body->oa;
452 obj = osc2cl(fa->fa_obj);
454 /* Update osc object's blocks attribute */
455 cl_object_attr_lock(obj);
456 if (body->oa.o_valid & OBD_MD_FLBLOCKS) {
457 attr->cat_blocks = body->oa.o_blocks;
462 cl_object_attr_update(env, obj, attr, valid);
463 cl_object_attr_unlock(obj);
466 rc = fa->fa_upcall(fa->fa_cookie, rc);
470 int osc_sync_base(struct osc_object *obj, struct obdo *oa,
471 obd_enqueue_update_f upcall, void *cookie,
472 struct ptlrpc_request_set *rqset)
474 struct obd_export *exp = osc_export(obj);
475 struct ptlrpc_request *req;
476 struct ost_body *body;
477 struct osc_fsync_args *fa;
481 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
485 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
487 ptlrpc_request_free(req);
491 /* overload the size and blocks fields in the oa with start/end */
492 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
494 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
496 ptlrpc_request_set_replen(req);
497 req->rq_interpret_reply = osc_sync_interpret;
499 CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
500 fa = ptlrpc_req_async_args(req);
503 fa->fa_upcall = upcall;
504 fa->fa_cookie = cookie;
506 if (rqset == PTLRPCD_SET)
507 ptlrpcd_add_req(req);
509 ptlrpc_set_add_req(rqset, req);
514 /* Find and cancel locally locks matched by @mode in the resource found by
515 * @objid. Found locks are added into @cancel list. Returns the amount of
516 * locks added to @cancels list. */
517 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
518 struct list_head *cancels,
519 enum ldlm_mode mode, __u64 lock_flags)
521 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
522 struct ldlm_res_id res_id;
523 struct ldlm_resource *res;
527 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
528 * export) but disabled through procfs (flag in NS).
530 * This distinguishes from a case when ELC is not supported originally,
531 * when we still want to cancel locks in advance and just cancel them
532 * locally, without sending any RPC. */
533 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
536 ostid_build_res_name(&oa->o_oi, &res_id);
537 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
541 LDLM_RESOURCE_ADDREF(res);
542 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
543 lock_flags, 0, NULL);
544 LDLM_RESOURCE_DELREF(res);
545 ldlm_resource_putref(res);
549 static int osc_destroy_interpret(const struct lu_env *env,
550 struct ptlrpc_request *req, void *data,
553 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
555 atomic_dec(&cli->cl_destroy_in_flight);
556 wake_up(&cli->cl_destroy_waitq);
560 static int osc_can_send_destroy(struct client_obd *cli)
562 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
563 cli->cl_max_rpcs_in_flight) {
564 /* The destroy request can be sent */
567 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
568 cli->cl_max_rpcs_in_flight) {
570 * The counter has been modified between the two atomic
573 wake_up(&cli->cl_destroy_waitq);
578 static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
581 struct client_obd *cli = &exp->exp_obd->u.cli;
582 struct ptlrpc_request *req;
583 struct ost_body *body;
584 struct list_head cancels = LIST_HEAD_INIT(cancels);
589 CDEBUG(D_INFO, "oa NULL\n");
593 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
594 LDLM_FL_DISCARD_DATA);
596 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
598 ldlm_lock_list_put(&cancels, l_bl_ast, count);
602 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
605 ptlrpc_request_free(req);
609 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
610 ptlrpc_at_set_req_timeout(req);
612 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
614 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
616 ptlrpc_request_set_replen(req);
618 req->rq_interpret_reply = osc_destroy_interpret;
619 if (!osc_can_send_destroy(cli)) {
620 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
623 * Wait until the number of on-going destroy RPCs drops
624 * under max_rpc_in_flight
626 rc = l_wait_event_exclusive(cli->cl_destroy_waitq,
627 osc_can_send_destroy(cli), &lwi);
629 ptlrpc_req_finished(req);
634 /* Do not wait for response */
635 ptlrpcd_add_req(req);
639 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
642 u64 bits = OBD_MD_FLBLOCKS | OBD_MD_FLGRANT;
644 LASSERT(!(oa->o_valid & bits));
647 spin_lock(&cli->cl_loi_list_lock);
648 if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data, GRANT_PARAM))
649 oa->o_dirty = cli->cl_dirty_grant;
651 oa->o_dirty = cli->cl_dirty_pages << PAGE_SHIFT;
652 if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
653 cli->cl_dirty_max_pages)) {
654 CERROR("dirty %lu - %lu > dirty_max %lu\n",
655 cli->cl_dirty_pages, cli->cl_dirty_transit,
656 cli->cl_dirty_max_pages);
658 } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
659 atomic_long_read(&obd_dirty_transit_pages) >
660 (long)(obd_max_dirty_pages + 1))) {
661 /* The atomic_read() allowing the atomic_inc() are
662 * not covered by a lock thus they may safely race and trip
663 * this CERROR() unless we add in a small fudge factor (+1). */
664 CERROR("%s: dirty %ld - %ld > system dirty_max %ld\n",
665 cli_name(cli), atomic_long_read(&obd_dirty_pages),
666 atomic_long_read(&obd_dirty_transit_pages),
667 obd_max_dirty_pages);
669 } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
671 CERROR("dirty %lu - dirty_max %lu too big???\n",
672 cli->cl_dirty_pages, cli->cl_dirty_max_pages);
675 unsigned long nrpages;
677 nrpages = cli->cl_max_pages_per_rpc;
678 nrpages *= cli->cl_max_rpcs_in_flight + 1;
679 nrpages = max(nrpages, cli->cl_dirty_max_pages);
680 oa->o_undirty = nrpages << PAGE_SHIFT;
681 if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data,
685 /* take extent tax into account when asking for more
687 nrextents = (nrpages + cli->cl_max_extent_pages - 1) /
688 cli->cl_max_extent_pages;
689 oa->o_undirty += nrextents * cli->cl_grant_extent_tax;
692 oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
693 oa->o_dropped = cli->cl_lost_grant;
694 cli->cl_lost_grant = 0;
695 spin_unlock(&cli->cl_loi_list_lock);
696 CDEBUG(D_CACHE, "dirty: %llu undirty: %u dropped %u grant: %llu\n",
697 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
700 void osc_update_next_shrink(struct client_obd *cli)
702 cli->cl_next_shrink_grant = ktime_get_seconds() +
703 cli->cl_grant_shrink_interval;
705 CDEBUG(D_CACHE, "next time %lld to shrink grant\n",
706 cli->cl_next_shrink_grant);
709 static void __osc_update_grant(struct client_obd *cli, u64 grant)
711 spin_lock(&cli->cl_loi_list_lock);
712 cli->cl_avail_grant += grant;
713 spin_unlock(&cli->cl_loi_list_lock);
716 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
718 if (body->oa.o_valid & OBD_MD_FLGRANT) {
719 CDEBUG(D_CACHE, "got %llu extra grant\n", body->oa.o_grant);
720 __osc_update_grant(cli, body->oa.o_grant);
724 static int osc_shrink_grant_interpret(const struct lu_env *env,
725 struct ptlrpc_request *req,
728 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
729 struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
730 struct ost_body *body;
733 __osc_update_grant(cli, oa->o_grant);
737 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
739 osc_update_grant(cli, body);
745 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
747 spin_lock(&cli->cl_loi_list_lock);
748 oa->o_grant = cli->cl_avail_grant / 4;
749 cli->cl_avail_grant -= oa->o_grant;
750 spin_unlock(&cli->cl_loi_list_lock);
751 if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
752 oa->o_valid |= OBD_MD_FLFLAGS;
755 oa->o_flags |= OBD_FL_SHRINK_GRANT;
756 osc_update_next_shrink(cli);
759 /* Shrink the current grant, either from some large amount to enough for a
760 * full set of in-flight RPCs, or if we have already shrunk to that limit
761 * then to enough for a single RPC. This avoids keeping more grant than
762 * needed, and avoids shrinking the grant piecemeal. */
763 static int osc_shrink_grant(struct client_obd *cli)
765 __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
766 (cli->cl_max_pages_per_rpc << PAGE_SHIFT);
768 spin_lock(&cli->cl_loi_list_lock);
769 if (cli->cl_avail_grant <= target_bytes)
770 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
771 spin_unlock(&cli->cl_loi_list_lock);
773 return osc_shrink_grant_to_target(cli, target_bytes);
776 int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
779 struct ost_body *body;
782 spin_lock(&cli->cl_loi_list_lock);
783 /* Don't shrink if we are already above or below the desired limit
784 * We don't want to shrink below a single RPC, as that will negatively
785 * impact block allocation and long-term performance. */
786 if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_SHIFT)
787 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
789 if (target_bytes >= cli->cl_avail_grant) {
790 spin_unlock(&cli->cl_loi_list_lock);
793 spin_unlock(&cli->cl_loi_list_lock);
799 osc_announce_cached(cli, &body->oa, 0);
801 spin_lock(&cli->cl_loi_list_lock);
802 body->oa.o_grant = cli->cl_avail_grant - target_bytes;
803 cli->cl_avail_grant = target_bytes;
804 spin_unlock(&cli->cl_loi_list_lock);
805 if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
806 body->oa.o_valid |= OBD_MD_FLFLAGS;
807 body->oa.o_flags = 0;
809 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
810 osc_update_next_shrink(cli);
812 rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
813 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
814 sizeof(*body), body, NULL);
816 __osc_update_grant(cli, body->oa.o_grant);
821 static int osc_should_shrink_grant(struct client_obd *client)
823 time64_t next_shrink = client->cl_next_shrink_grant;
825 if ((client->cl_import->imp_connect_data.ocd_connect_flags &
826 OBD_CONNECT_GRANT_SHRINK) == 0)
829 if (ktime_get_seconds() >= next_shrink - 5) {
830 /* Get the current RPC size directly, instead of going via:
831 * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
832 * Keep comment here so that it can be found by searching. */
833 int brw_size = client->cl_max_pages_per_rpc << PAGE_SHIFT;
835 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
836 client->cl_avail_grant > brw_size)
839 osc_update_next_shrink(client);
844 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
846 struct client_obd *client;
848 list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
849 if (osc_should_shrink_grant(client))
850 osc_shrink_grant(client);
855 static int osc_add_shrink_grant(struct client_obd *client)
859 rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
861 osc_grant_shrink_grant_cb, NULL,
862 &client->cl_grant_shrink_list);
864 CERROR("add grant client %s error %d\n", cli_name(client), rc);
867 CDEBUG(D_CACHE, "add grant client %s\n", cli_name(client));
868 osc_update_next_shrink(client);
872 static int osc_del_shrink_grant(struct client_obd *client)
874 return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
878 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
881 * ocd_grant is the total grant amount we're expect to hold: if we've
882 * been evicted, it's the new avail_grant amount, cl_dirty_pages will
883 * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
886 * race is tolerable here: if we're evicted, but imp_state already
887 * left EVICTED state, then cl_dirty_pages must be 0 already.
889 spin_lock(&cli->cl_loi_list_lock);
890 cli->cl_avail_grant = ocd->ocd_grant;
891 if (cli->cl_import->imp_state != LUSTRE_IMP_EVICTED) {
892 cli->cl_avail_grant -= cli->cl_reserved_grant;
893 if (OCD_HAS_FLAG(ocd, GRANT_PARAM))
894 cli->cl_avail_grant -= cli->cl_dirty_grant;
896 cli->cl_avail_grant -=
897 cli->cl_dirty_pages << PAGE_SHIFT;
900 if (OCD_HAS_FLAG(ocd, GRANT_PARAM)) {
904 /* overhead for each extent insertion */
905 cli->cl_grant_extent_tax = ocd->ocd_grant_tax_kb << 10;
906 /* determine the appropriate chunk size used by osc_extent. */
907 cli->cl_chunkbits = max_t(int, PAGE_SHIFT,
908 ocd->ocd_grant_blkbits);
909 /* max_pages_per_rpc must be chunk aligned */
910 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
911 cli->cl_max_pages_per_rpc = (cli->cl_max_pages_per_rpc +
912 ~chunk_mask) & chunk_mask;
913 /* determine maximum extent size, in #pages */
914 size = (u64)ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits;
915 cli->cl_max_extent_pages = size >> PAGE_SHIFT;
916 if (cli->cl_max_extent_pages == 0)
917 cli->cl_max_extent_pages = 1;
919 cli->cl_grant_extent_tax = 0;
920 cli->cl_chunkbits = PAGE_SHIFT;
921 cli->cl_max_extent_pages = DT_MAX_BRW_PAGES;
923 spin_unlock(&cli->cl_loi_list_lock);
925 CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
926 "chunk bits: %d cl_max_extent_pages: %d\n",
928 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits,
929 cli->cl_max_extent_pages);
931 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
932 list_empty(&cli->cl_grant_shrink_list))
933 osc_add_shrink_grant(cli);
935 EXPORT_SYMBOL(osc_init_grant);
937 /* We assume that the reason this OSC got a short read is because it read
938 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
939 * via the LOV, and it _knows_ it's reading inside the file, it's just that
940 * this stripe never got written at or beyond this stripe offset yet. */
941 static void handle_short_read(int nob_read, size_t page_count,
942 struct brw_page **pga)
947 /* skip bytes read OK */
948 while (nob_read > 0) {
949 LASSERT (page_count > 0);
951 if (pga[i]->count > nob_read) {
952 /* EOF inside this page */
953 ptr = kmap(pga[i]->pg) +
954 (pga[i]->off & ~PAGE_MASK);
955 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
962 nob_read -= pga[i]->count;
967 /* zero remaining pages */
968 while (page_count-- > 0) {
969 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~PAGE_MASK);
970 memset(ptr, 0, pga[i]->count);
976 static int check_write_rcs(struct ptlrpc_request *req,
977 int requested_nob, int niocount,
978 size_t page_count, struct brw_page **pga)
983 remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
984 sizeof(*remote_rcs) *
986 if (remote_rcs == NULL) {
987 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
991 /* return error if any niobuf was in error */
992 for (i = 0; i < niocount; i++) {
993 if ((int)remote_rcs[i] < 0)
994 return(remote_rcs[i]);
996 if (remote_rcs[i] != 0) {
997 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
998 i, remote_rcs[i], req);
1002 if (req->rq_bulk != NULL &&
1003 req->rq_bulk->bd_nob_transferred != requested_nob) {
1004 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1005 req->rq_bulk->bd_nob_transferred, requested_nob);
1012 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1014 if (p1->flag != p2->flag) {
1015 unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE |
1016 OBD_BRW_SYNC | OBD_BRW_ASYNC |
1017 OBD_BRW_NOQUOTA | OBD_BRW_SOFT_SYNC);
1019 /* warn if we try to combine flags that we don't know to be
1020 * safe to combine */
1021 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1022 CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1023 "report this at https://jira.hpdd.intel.com/\n",
1024 p1->flag, p2->flag);
1029 return (p1->off + p1->count == p2->off);
1032 static u32 osc_checksum_bulk(int nob, size_t pg_count,
1033 struct brw_page **pga, int opc,
1034 enum cksum_types cksum_type)
1038 struct cfs_crypto_hash_desc *hdesc;
1039 unsigned int bufsize;
1040 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1042 LASSERT(pg_count > 0);
1044 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1045 if (IS_ERR(hdesc)) {
1046 CERROR("Unable to initialize checksum hash %s\n",
1047 cfs_crypto_hash_name(cfs_alg));
1048 return PTR_ERR(hdesc);
1051 while (nob > 0 && pg_count > 0) {
1052 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1054 /* corrupt the data before we compute the checksum, to
1055 * simulate an OST->client data error */
1056 if (i == 0 && opc == OST_READ &&
1057 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1058 unsigned char *ptr = kmap(pga[i]->pg);
1059 int off = pga[i]->off & ~PAGE_MASK;
1061 memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1064 cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
1065 pga[i]->off & ~PAGE_MASK,
1067 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
1068 (int)(pga[i]->off & ~PAGE_MASK));
1070 nob -= pga[i]->count;
1075 bufsize = sizeof(cksum);
1076 cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1078 /* For sending we only compute the wrong checksum instead
1079 * of corrupting the data so it is still correct on a redo */
1080 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1087 osc_brw_prep_request(int cmd, struct client_obd *cli, struct obdo *oa,
1088 u32 page_count, struct brw_page **pga,
1089 struct ptlrpc_request **reqp, int resend)
1091 struct ptlrpc_request *req;
1092 struct ptlrpc_bulk_desc *desc;
1093 struct ost_body *body;
1094 struct obd_ioobj *ioobj;
1095 struct niobuf_remote *niobuf;
1096 int niocount, i, requested_nob, opc, rc, short_io_size;
1097 struct osc_brw_async_args *aa;
1098 struct req_capsule *pill;
1099 struct brw_page *pg_prev;
1103 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1104 RETURN(-ENOMEM); /* Recoverable */
1105 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1106 RETURN(-EINVAL); /* Fatal */
1108 if ((cmd & OBD_BRW_WRITE) != 0) {
1110 req = ptlrpc_request_alloc_pool(cli->cl_import,
1112 &RQF_OST_BRW_WRITE);
1115 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1120 for (niocount = i = 1; i < page_count; i++) {
1121 if (!can_merge_pages(pga[i - 1], pga[i]))
1125 pill = &req->rq_pill;
1126 req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1128 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1129 niocount * sizeof(*niobuf));
1131 for (i = 0; i < page_count; i++)
1132 short_io_size += pga[i]->count;
1134 /* Check if we can do a short io. */
1135 if (!(short_io_size <= cli->cl_short_io_bytes && niocount == 1 &&
1136 imp_connect_shortio(cli->cl_import)))
1139 req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_CLIENT,
1140 opc == OST_READ ? 0 : short_io_size);
1141 if (opc == OST_READ)
1142 req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_SERVER,
1145 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1147 ptlrpc_request_free(req);
1150 osc_set_io_portal(req);
1152 ptlrpc_at_set_req_timeout(req);
1153 /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1155 req->rq_no_retry_einprogress = 1;
1157 if (short_io_size != 0) {
1159 short_io_buf = NULL;
1163 desc = ptlrpc_prep_bulk_imp(req, page_count,
1164 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1165 (opc == OST_WRITE ? PTLRPC_BULK_GET_SOURCE :
1166 PTLRPC_BULK_PUT_SINK) |
1167 PTLRPC_BULK_BUF_KIOV,
1169 &ptlrpc_bulk_kiov_pin_ops);
1172 GOTO(out, rc = -ENOMEM);
1173 /* NB request now owns desc and will free it when it gets freed */
1175 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1176 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1177 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1178 LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1180 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
1182 obdo_to_ioobj(oa, ioobj);
1183 ioobj->ioo_bufcnt = niocount;
1184 /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1185 * that might be send for this request. The actual number is decided
1186 * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1187 * "max - 1" for old client compatibility sending "0", and also so the
1188 * the actual maximum is a power-of-two number, not one less. LU-1431 */
1190 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1192 ioobj_max_brw_set(ioobj, 0);
1194 if (short_io_size != 0) {
1195 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1196 body->oa.o_valid |= OBD_MD_FLFLAGS;
1197 body->oa.o_flags = 0;
1199 body->oa.o_flags |= OBD_FL_SHORT_IO;
1200 CDEBUG(D_CACHE, "Using short io for data transfer, size = %d\n",
1202 if (opc == OST_WRITE) {
1203 short_io_buf = req_capsule_client_get(pill,
1205 LASSERT(short_io_buf != NULL);
1209 LASSERT(page_count > 0);
1211 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1212 struct brw_page *pg = pga[i];
1213 int poff = pg->off & ~PAGE_MASK;
1215 LASSERT(pg->count > 0);
1216 /* make sure there is no gap in the middle of page array */
1217 LASSERTF(page_count == 1 ||
1218 (ergo(i == 0, poff + pg->count == PAGE_SIZE) &&
1219 ergo(i > 0 && i < page_count - 1,
1220 poff == 0 && pg->count == PAGE_SIZE) &&
1221 ergo(i == page_count - 1, poff == 0)),
1222 "i: %d/%d pg: %p off: %llu, count: %u\n",
1223 i, page_count, pg, pg->off, pg->count);
1224 LASSERTF(i == 0 || pg->off > pg_prev->off,
1225 "i %d p_c %u pg %p [pri %lu ind %lu] off %llu"
1226 " prev_pg %p [pri %lu ind %lu] off %llu\n",
1228 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1229 pg_prev->pg, page_private(pg_prev->pg),
1230 pg_prev->pg->index, pg_prev->off);
1231 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1232 (pg->flag & OBD_BRW_SRVLOCK));
1233 if (short_io_size != 0 && opc == OST_WRITE) {
1234 unsigned char *ptr = ll_kmap_atomic(pg->pg, KM_USER0);
1236 LASSERT(short_io_size >= requested_nob + pg->count);
1237 memcpy(short_io_buf + requested_nob,
1240 ll_kunmap_atomic(ptr, KM_USER0);
1241 } else if (short_io_size == 0) {
1242 desc->bd_frag_ops->add_kiov_frag(desc, pg->pg, poff,
1245 requested_nob += pg->count;
1247 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1249 niobuf->rnb_len += pg->count;
1251 niobuf->rnb_offset = pg->off;
1252 niobuf->rnb_len = pg->count;
1253 niobuf->rnb_flags = pg->flag;
1258 LASSERTF((void *)(niobuf - niocount) ==
1259 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1260 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1261 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1263 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1265 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1266 body->oa.o_valid |= OBD_MD_FLFLAGS;
1267 body->oa.o_flags = 0;
1269 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1272 if (osc_should_shrink_grant(cli))
1273 osc_shrink_grant_local(cli, &body->oa);
1275 /* size[REQ_REC_OFF] still sizeof (*body) */
1276 if (opc == OST_WRITE) {
1277 if (cli->cl_checksum &&
1278 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1279 /* store cl_cksum_type in a local variable since
1280 * it can be changed via lprocfs */
1281 enum cksum_types cksum_type = cli->cl_cksum_type;
1283 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1284 body->oa.o_flags = 0;
1286 body->oa.o_flags |= cksum_type_pack(cksum_type);
1287 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1288 body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1292 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1294 /* save this in 'oa', too, for later checking */
1295 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1296 oa->o_flags |= cksum_type_pack(cksum_type);
1298 /* clear out the checksum flag, in case this is a
1299 * resend but cl_checksum is no longer set. b=11238 */
1300 oa->o_valid &= ~OBD_MD_FLCKSUM;
1302 oa->o_cksum = body->oa.o_cksum;
1303 /* 1 RC per niobuf */
1304 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1305 sizeof(__u32) * niocount);
1307 if (cli->cl_checksum &&
1308 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1309 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1310 body->oa.o_flags = 0;
1311 body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1312 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1315 /* Client cksum has been already copied to wire obdo in previous
1316 * lustre_set_wire_obdo(), and in the case a bulk-read is being
1317 * resent due to cksum error, this will allow Server to
1318 * check+dump pages on its side */
1320 ptlrpc_request_set_replen(req);
1322 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1323 aa = ptlrpc_req_async_args(req);
1325 aa->aa_requested_nob = requested_nob;
1326 aa->aa_nio_count = niocount;
1327 aa->aa_page_count = page_count;
1331 INIT_LIST_HEAD(&aa->aa_oaps);
1334 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1335 CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
1336 req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
1337 niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
1341 ptlrpc_req_finished(req);
1345 char dbgcksum_file_name[PATH_MAX];
1347 static void dump_all_bulk_pages(struct obdo *oa, __u32 page_count,
1348 struct brw_page **pga, __u32 server_cksum,
1357 /* will only keep dump of pages on first error for the same range in
1358 * file/fid, not during the resends/retries. */
1359 snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1360 "%s-checksum_dump-osc-"DFID":[%llu-%llu]-%x-%x",
1361 (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 ?
1362 libcfs_debug_file_path_arr :
1363 LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1364 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
1365 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1366 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1368 pga[page_count-1]->off + pga[page_count-1]->count - 1,
1369 client_cksum, server_cksum);
1370 filp = filp_open(dbgcksum_file_name,
1371 O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1375 CDEBUG(D_INFO, "%s: can't open to dump pages with "
1376 "checksum error: rc = %d\n", dbgcksum_file_name,
1379 CERROR("%s: can't open to dump pages with checksum "
1380 "error: rc = %d\n", dbgcksum_file_name, rc);
1386 for (i = 0; i < page_count; i++) {
1387 len = pga[i]->count;
1388 buf = kmap(pga[i]->pg);
1390 rc = vfs_write(filp, (__force const char __user *)buf,
1393 CERROR("%s: wanted to write %u but got %d "
1394 "error\n", dbgcksum_file_name, len, rc);
1399 CDEBUG(D_INFO, "%s: wrote %d bytes\n",
1400 dbgcksum_file_name, rc);
1406 rc = ll_vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1408 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1409 filp_close(filp, NULL);
1414 check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
1415 __u32 client_cksum, __u32 server_cksum,
1416 struct osc_brw_async_args *aa)
1420 enum cksum_types cksum_type;
1422 if (server_cksum == client_cksum) {
1423 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1427 if (aa->aa_cli->cl_checksum_dump)
1428 dump_all_bulk_pages(oa, aa->aa_page_count, aa->aa_ppga,
1429 server_cksum, client_cksum);
1431 cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1433 new_cksum = osc_checksum_bulk(aa->aa_requested_nob, aa->aa_page_count,
1434 aa->aa_ppga, OST_WRITE, cksum_type);
1436 if (cksum_type != cksum_type_unpack(aa->aa_oa->o_flags))
1437 msg = "the server did not use the checksum type specified in "
1438 "the original request - likely a protocol problem";
1439 else if (new_cksum == server_cksum)
1440 msg = "changed on the client after we checksummed it - "
1441 "likely false positive due to mmap IO (bug 11742)";
1442 else if (new_cksum == client_cksum)
1443 msg = "changed in transit before arrival at OST";
1445 msg = "changed in transit AND doesn't match the original - "
1446 "likely false positive due to mmap IO (bug 11742)";
1448 LCONSOLE_ERROR_MSG(0x132, "%s: BAD WRITE CHECKSUM: %s: from %s inode "
1449 DFID " object "DOSTID" extent [%llu-%llu], original "
1450 "client csum %x (type %x), server csum %x (type %x),"
1451 " client csum now %x\n",
1452 aa->aa_cli->cl_import->imp_obd->obd_name,
1453 msg, libcfs_nid2str(peer->nid),
1454 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1455 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1456 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1457 POSTID(&oa->o_oi), aa->aa_ppga[0]->off,
1458 aa->aa_ppga[aa->aa_page_count - 1]->off +
1459 aa->aa_ppga[aa->aa_page_count-1]->count - 1,
1460 client_cksum, cksum_type_unpack(aa->aa_oa->o_flags),
1461 server_cksum, cksum_type, new_cksum);
1465 /* Note rc enters this function as number of bytes transferred */
1466 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1468 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1469 const struct lnet_process_id *peer =
1470 &req->rq_import->imp_connection->c_peer;
1471 struct client_obd *cli = aa->aa_cli;
1472 struct ost_body *body;
1473 u32 client_cksum = 0;
1476 if (rc < 0 && rc != -EDQUOT) {
1477 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1481 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1482 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1484 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1488 /* set/clear over quota flag for a uid/gid/projid */
1489 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1490 body->oa.o_valid & (OBD_MD_FLALLQUOTA)) {
1491 unsigned qid[LL_MAXQUOTAS] = {
1492 body->oa.o_uid, body->oa.o_gid,
1493 body->oa.o_projid };
1494 CDEBUG(D_QUOTA, "setdq for [%u %u %u] with valid %#llx, flags %x\n",
1495 body->oa.o_uid, body->oa.o_gid, body->oa.o_projid,
1496 body->oa.o_valid, body->oa.o_flags);
1497 osc_quota_setdq(cli, qid, body->oa.o_valid,
1501 osc_update_grant(cli, body);
1506 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1507 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1509 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1511 CERROR("Unexpected +ve rc %d\n", rc);
1515 if (req->rq_bulk != NULL &&
1516 sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1519 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1520 check_write_checksum(&body->oa, peer, client_cksum,
1521 body->oa.o_cksum, aa))
1524 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1525 aa->aa_page_count, aa->aa_ppga);
1529 /* The rest of this function executes only for OST_READs */
1531 if (req->rq_bulk == NULL) {
1532 rc = req_capsule_get_size(&req->rq_pill, &RMF_SHORT_IO,
1534 LASSERT(rc == req->rq_status);
1536 /* if unwrap_bulk failed, return -EAGAIN to retry */
1537 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1540 GOTO(out, rc = -EAGAIN);
1542 if (rc > aa->aa_requested_nob) {
1543 CERROR("Unexpected rc %d (%d requested)\n", rc,
1544 aa->aa_requested_nob);
1548 if (req->rq_bulk != NULL && rc != req->rq_bulk->bd_nob_transferred) {
1549 CERROR ("Unexpected rc %d (%d transferred)\n",
1550 rc, req->rq_bulk->bd_nob_transferred);
1554 if (req->rq_bulk == NULL) {
1556 int nob, pg_count, i = 0;
1559 CDEBUG(D_CACHE, "Using short io read, size %d\n", rc);
1560 pg_count = aa->aa_page_count;
1561 buf = req_capsule_server_sized_get(&req->rq_pill, &RMF_SHORT_IO,
1564 while (nob > 0 && pg_count > 0) {
1566 int count = aa->aa_ppga[i]->count > nob ?
1567 nob : aa->aa_ppga[i]->count;
1569 CDEBUG(D_CACHE, "page %p count %d\n",
1570 aa->aa_ppga[i]->pg, count);
1571 ptr = ll_kmap_atomic(aa->aa_ppga[i]->pg, KM_USER0);
1572 memcpy(ptr + (aa->aa_ppga[i]->off & ~PAGE_MASK), buf,
1574 ll_kunmap_atomic((void *) ptr, KM_USER0);
1583 if (rc < aa->aa_requested_nob)
1584 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1586 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1587 static int cksum_counter;
1588 u32 server_cksum = body->oa.o_cksum;
1591 enum cksum_types cksum_type;
1593 cksum_type = cksum_type_unpack(body->oa.o_valid &OBD_MD_FLFLAGS?
1594 body->oa.o_flags : 0);
1595 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1596 aa->aa_ppga, OST_READ,
1599 if (req->rq_bulk != NULL &&
1600 peer->nid != req->rq_bulk->bd_sender) {
1602 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1605 if (server_cksum != client_cksum) {
1606 struct ost_body *clbody;
1607 u32 page_count = aa->aa_page_count;
1609 clbody = req_capsule_client_get(&req->rq_pill,
1611 if (cli->cl_checksum_dump)
1612 dump_all_bulk_pages(&clbody->oa, page_count,
1613 aa->aa_ppga, server_cksum,
1616 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1617 "%s%s%s inode "DFID" object "DOSTID
1618 " extent [%llu-%llu], client %x, "
1619 "server %x, cksum_type %x\n",
1620 req->rq_import->imp_obd->obd_name,
1621 libcfs_nid2str(peer->nid),
1623 clbody->oa.o_valid & OBD_MD_FLFID ?
1624 clbody->oa.o_parent_seq : 0ULL,
1625 clbody->oa.o_valid & OBD_MD_FLFID ?
1626 clbody->oa.o_parent_oid : 0,
1627 clbody->oa.o_valid & OBD_MD_FLFID ?
1628 clbody->oa.o_parent_ver : 0,
1629 POSTID(&body->oa.o_oi),
1630 aa->aa_ppga[0]->off,
1631 aa->aa_ppga[page_count-1]->off +
1632 aa->aa_ppga[page_count-1]->count - 1,
1633 client_cksum, server_cksum,
1636 aa->aa_oa->o_cksum = client_cksum;
1640 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1643 } else if (unlikely(client_cksum)) {
1644 static int cksum_missed;
1647 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1648 CERROR("Checksum %u requested from %s but not sent\n",
1649 cksum_missed, libcfs_nid2str(peer->nid));
1655 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1656 aa->aa_oa, &body->oa);
1661 static int osc_brw_redo_request(struct ptlrpc_request *request,
1662 struct osc_brw_async_args *aa, int rc)
1664 struct ptlrpc_request *new_req;
1665 struct osc_brw_async_args *new_aa;
1666 struct osc_async_page *oap;
1669 DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1670 "redo for recoverable error %d", rc);
1672 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1673 OST_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
1674 aa->aa_cli, aa->aa_oa, aa->aa_page_count,
1675 aa->aa_ppga, &new_req, 1);
1679 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1680 if (oap->oap_request != NULL) {
1681 LASSERTF(request == oap->oap_request,
1682 "request %p != oap_request %p\n",
1683 request, oap->oap_request);
1684 if (oap->oap_interrupted) {
1685 ptlrpc_req_finished(new_req);
1690 /* New request takes over pga and oaps from old request.
1691 * Note that copying a list_head doesn't work, need to move it... */
1693 new_req->rq_interpret_reply = request->rq_interpret_reply;
1694 new_req->rq_async_args = request->rq_async_args;
1695 new_req->rq_commit_cb = request->rq_commit_cb;
1696 /* cap resend delay to the current request timeout, this is similar to
1697 * what ptlrpc does (see after_reply()) */
1698 if (aa->aa_resends > new_req->rq_timeout)
1699 new_req->rq_sent = ktime_get_real_seconds() + new_req->rq_timeout;
1701 new_req->rq_sent = ktime_get_real_seconds() + aa->aa_resends;
1702 new_req->rq_generation_set = 1;
1703 new_req->rq_import_generation = request->rq_import_generation;
1705 new_aa = ptlrpc_req_async_args(new_req);
1707 INIT_LIST_HEAD(&new_aa->aa_oaps);
1708 list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1709 INIT_LIST_HEAD(&new_aa->aa_exts);
1710 list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1711 new_aa->aa_resends = aa->aa_resends;
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 /* XXX: This code will run into problem if we're going to support
1721 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1722 * and wait for all of them to be finished. We should inherit request
1723 * set from old request. */
1724 ptlrpcd_add_req(new_req);
1726 DEBUG_REQ(D_INFO, new_req, "new request");
1731 * ugh, we want disk allocation on the target to happen in offset order. we'll
1732 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1733 * fine for our small page arrays and doesn't require allocation. its an
1734 * insertion sort that swaps elements that are strides apart, shrinking the
1735 * stride down until its '1' and the array is sorted.
1737 static void sort_brw_pages(struct brw_page **array, int num)
1740 struct brw_page *tmp;
1744 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1749 for (i = stride ; i < num ; i++) {
1752 while (j >= stride && array[j - stride]->off > tmp->off) {
1753 array[j] = array[j - stride];
1758 } while (stride > 1);
1761 static void osc_release_ppga(struct brw_page **ppga, size_t count)
1763 LASSERT(ppga != NULL);
1764 OBD_FREE(ppga, sizeof(*ppga) * count);
1767 static int brw_interpret(const struct lu_env *env,
1768 struct ptlrpc_request *req, void *data, int rc)
1770 struct osc_brw_async_args *aa = data;
1771 struct osc_extent *ext;
1772 struct osc_extent *tmp;
1773 struct client_obd *cli = aa->aa_cli;
1774 unsigned long transferred = 0;
1777 rc = osc_brw_fini_request(req, rc);
1778 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
1779 /* When server return -EINPROGRESS, client should always retry
1780 * regardless of the number of times the bulk was resent already. */
1781 if (osc_recoverable_error(rc) && !req->rq_no_delay) {
1782 if (req->rq_import_generation !=
1783 req->rq_import->imp_generation) {
1784 CDEBUG(D_HA, "%s: resend cross eviction for object: "
1785 ""DOSTID", rc = %d.\n",
1786 req->rq_import->imp_obd->obd_name,
1787 POSTID(&aa->aa_oa->o_oi), rc);
1788 } else if (rc == -EINPROGRESS ||
1789 client_should_resend(aa->aa_resends, aa->aa_cli)) {
1790 rc = osc_brw_redo_request(req, aa, rc);
1792 CERROR("%s: too many resent retries for object: "
1793 "%llu:%llu, rc = %d.\n",
1794 req->rq_import->imp_obd->obd_name,
1795 POSTID(&aa->aa_oa->o_oi), rc);
1800 else if (rc == -EAGAIN || rc == -EINPROGRESS)
1805 struct obdo *oa = aa->aa_oa;
1806 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1807 unsigned long valid = 0;
1808 struct cl_object *obj;
1809 struct osc_async_page *last;
1811 last = brw_page2oap(aa->aa_ppga[aa->aa_page_count - 1]);
1812 obj = osc2cl(last->oap_obj);
1814 cl_object_attr_lock(obj);
1815 if (oa->o_valid & OBD_MD_FLBLOCKS) {
1816 attr->cat_blocks = oa->o_blocks;
1817 valid |= CAT_BLOCKS;
1819 if (oa->o_valid & OBD_MD_FLMTIME) {
1820 attr->cat_mtime = oa->o_mtime;
1823 if (oa->o_valid & OBD_MD_FLATIME) {
1824 attr->cat_atime = oa->o_atime;
1827 if (oa->o_valid & OBD_MD_FLCTIME) {
1828 attr->cat_ctime = oa->o_ctime;
1832 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1833 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1834 loff_t last_off = last->oap_count + last->oap_obj_off +
1837 /* Change file size if this is an out of quota or
1838 * direct IO write and it extends the file size */
1839 if (loi->loi_lvb.lvb_size < last_off) {
1840 attr->cat_size = last_off;
1843 /* Extend KMS if it's not a lockless write */
1844 if (loi->loi_kms < last_off &&
1845 oap2osc_page(last)->ops_srvlock == 0) {
1846 attr->cat_kms = last_off;
1852 cl_object_attr_update(env, obj, attr, valid);
1853 cl_object_attr_unlock(obj);
1855 OBDO_FREE(aa->aa_oa);
1857 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
1858 osc_inc_unstable_pages(req);
1860 list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
1861 list_del_init(&ext->oe_link);
1862 osc_extent_finish(env, ext, 1,
1863 rc && req->rq_no_delay ? -EWOULDBLOCK : rc);
1865 LASSERT(list_empty(&aa->aa_exts));
1866 LASSERT(list_empty(&aa->aa_oaps));
1868 transferred = (req->rq_bulk == NULL ? /* short io */
1869 aa->aa_requested_nob :
1870 req->rq_bulk->bd_nob_transferred);
1872 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
1873 ptlrpc_lprocfs_brw(req, transferred);
1875 spin_lock(&cli->cl_loi_list_lock);
1876 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
1877 * is called so we know whether to go to sync BRWs or wait for more
1878 * RPCs to complete */
1879 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
1880 cli->cl_w_in_flight--;
1882 cli->cl_r_in_flight--;
1883 osc_wake_cache_waiters(cli);
1884 spin_unlock(&cli->cl_loi_list_lock);
1886 osc_io_unplug(env, cli, NULL);
1890 static void brw_commit(struct ptlrpc_request *req)
1892 /* If osc_inc_unstable_pages (via osc_extent_finish) races with
1893 * this called via the rq_commit_cb, I need to ensure
1894 * osc_dec_unstable_pages is still called. Otherwise unstable
1895 * pages may be leaked. */
1896 spin_lock(&req->rq_lock);
1897 if (likely(req->rq_unstable)) {
1898 req->rq_unstable = 0;
1899 spin_unlock(&req->rq_lock);
1901 osc_dec_unstable_pages(req);
1903 req->rq_committed = 1;
1904 spin_unlock(&req->rq_lock);
1909 * Build an RPC by the list of extent @ext_list. The caller must ensure
1910 * that the total pages in this list are NOT over max pages per RPC.
1911 * Extents in the list must be in OES_RPC state.
1913 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
1914 struct list_head *ext_list, int cmd)
1916 struct ptlrpc_request *req = NULL;
1917 struct osc_extent *ext;
1918 struct brw_page **pga = NULL;
1919 struct osc_brw_async_args *aa = NULL;
1920 struct obdo *oa = NULL;
1921 struct osc_async_page *oap;
1922 struct osc_object *obj = NULL;
1923 struct cl_req_attr *crattr = NULL;
1924 loff_t starting_offset = OBD_OBJECT_EOF;
1925 loff_t ending_offset = 0;
1929 bool soft_sync = false;
1930 bool interrupted = false;
1931 bool ndelay = false;
1935 __u32 layout_version = 0;
1936 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1937 struct ost_body *body;
1939 LASSERT(!list_empty(ext_list));
1941 /* add pages into rpc_list to build BRW rpc */
1942 list_for_each_entry(ext, ext_list, oe_link) {
1943 LASSERT(ext->oe_state == OES_RPC);
1944 mem_tight |= ext->oe_memalloc;
1945 grant += ext->oe_grants;
1946 page_count += ext->oe_nr_pages;
1947 layout_version = MAX(layout_version, ext->oe_layout_version);
1952 soft_sync = osc_over_unstable_soft_limit(cli);
1954 mpflag = cfs_memory_pressure_get_and_set();
1956 OBD_ALLOC(pga, sizeof(*pga) * page_count);
1958 GOTO(out, rc = -ENOMEM);
1962 GOTO(out, rc = -ENOMEM);
1965 list_for_each_entry(ext, ext_list, oe_link) {
1966 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
1968 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
1970 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
1971 pga[i] = &oap->oap_brw_page;
1972 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
1975 list_add_tail(&oap->oap_rpc_item, &rpc_list);
1976 if (starting_offset == OBD_OBJECT_EOF ||
1977 starting_offset > oap->oap_obj_off)
1978 starting_offset = oap->oap_obj_off;
1980 LASSERT(oap->oap_page_off == 0);
1981 if (ending_offset < oap->oap_obj_off + oap->oap_count)
1982 ending_offset = oap->oap_obj_off +
1985 LASSERT(oap->oap_page_off + oap->oap_count ==
1987 if (oap->oap_interrupted)
1994 /* first page in the list */
1995 oap = list_entry(rpc_list.next, typeof(*oap), oap_rpc_item);
1997 crattr = &osc_env_info(env)->oti_req_attr;
1998 memset(crattr, 0, sizeof(*crattr));
1999 crattr->cra_type = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2000 crattr->cra_flags = ~0ULL;
2001 crattr->cra_page = oap2cl_page(oap);
2002 crattr->cra_oa = oa;
2003 cl_req_attr_set(env, osc2cl(obj), crattr);
2005 if (cmd == OBD_BRW_WRITE) {
2006 oa->o_grant_used = grant;
2007 if (layout_version > 0) {
2008 CDEBUG(D_LAYOUT, DFID": write with layout version %u\n",
2009 PFID(&oa->o_oi.oi_fid), layout_version);
2011 oa->o_layout_version = layout_version;
2012 oa->o_valid |= OBD_MD_LAYOUT_VERSION;
2016 sort_brw_pages(pga, page_count);
2017 rc = osc_brw_prep_request(cmd, cli, oa, page_count, pga, &req, 0);
2019 CERROR("prep_req failed: %d\n", rc);
2023 req->rq_commit_cb = brw_commit;
2024 req->rq_interpret_reply = brw_interpret;
2025 req->rq_memalloc = mem_tight != 0;
2026 oap->oap_request = ptlrpc_request_addref(req);
2027 if (interrupted && !req->rq_intr)
2028 ptlrpc_mark_interrupted(req);
2030 req->rq_no_resend = req->rq_no_delay = 1;
2031 /* probably set a shorter timeout value.
2032 * to handle ETIMEDOUT in brw_interpret() correctly. */
2033 /* lustre_msg_set_timeout(req, req->rq_timeout / 2); */
2036 /* Need to update the timestamps after the request is built in case
2037 * we race with setattr (locally or in queue at OST). If OST gets
2038 * later setattr before earlier BRW (as determined by the request xid),
2039 * the OST will not use BRW timestamps. Sadly, there is no obvious
2040 * way to do this in a single call. bug 10150 */
2041 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2042 crattr->cra_oa = &body->oa;
2043 crattr->cra_flags = OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME;
2044 cl_req_attr_set(env, osc2cl(obj), crattr);
2045 lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
2047 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2048 aa = ptlrpc_req_async_args(req);
2049 INIT_LIST_HEAD(&aa->aa_oaps);
2050 list_splice_init(&rpc_list, &aa->aa_oaps);
2051 INIT_LIST_HEAD(&aa->aa_exts);
2052 list_splice_init(ext_list, &aa->aa_exts);
2054 spin_lock(&cli->cl_loi_list_lock);
2055 starting_offset >>= PAGE_SHIFT;
2056 if (cmd == OBD_BRW_READ) {
2057 cli->cl_r_in_flight++;
2058 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2059 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2060 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2061 starting_offset + 1);
2063 cli->cl_w_in_flight++;
2064 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2065 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
2066 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2067 starting_offset + 1);
2069 spin_unlock(&cli->cl_loi_list_lock);
2071 DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
2072 page_count, aa, cli->cl_r_in_flight,
2073 cli->cl_w_in_flight);
2074 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_IO, cfs_fail_val);
2076 ptlrpcd_add_req(req);
2082 cfs_memory_pressure_restore(mpflag);
2085 LASSERT(req == NULL);
2090 OBD_FREE(pga, sizeof(*pga) * page_count);
2091 /* this should happen rarely and is pretty bad, it makes the
2092 * pending list not follow the dirty order */
2093 while (!list_empty(ext_list)) {
2094 ext = list_entry(ext_list->next, struct osc_extent,
2096 list_del_init(&ext->oe_link);
2097 osc_extent_finish(env, ext, 0, rc);
2103 static int osc_set_lock_data(struct ldlm_lock *lock, void *data)
2107 LASSERT(lock != NULL);
2109 lock_res_and_lock(lock);
2111 if (lock->l_ast_data == NULL)
2112 lock->l_ast_data = data;
2113 if (lock->l_ast_data == data)
2116 unlock_res_and_lock(lock);
2121 int osc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
2122 void *cookie, struct lustre_handle *lockh,
2123 enum ldlm_mode mode, __u64 *flags, bool speculative,
2126 bool intent = *flags & LDLM_FL_HAS_INTENT;
2130 /* The request was created before ldlm_cli_enqueue call. */
2131 if (intent && errcode == ELDLM_LOCK_ABORTED) {
2132 struct ldlm_reply *rep;
2134 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
2135 LASSERT(rep != NULL);
2137 rep->lock_policy_res1 =
2138 ptlrpc_status_ntoh(rep->lock_policy_res1);
2139 if (rep->lock_policy_res1)
2140 errcode = rep->lock_policy_res1;
2142 *flags |= LDLM_FL_LVB_READY;
2143 } else if (errcode == ELDLM_OK) {
2144 *flags |= LDLM_FL_LVB_READY;
2147 /* Call the update callback. */
2148 rc = (*upcall)(cookie, lockh, errcode);
2150 /* release the reference taken in ldlm_cli_enqueue() */
2151 if (errcode == ELDLM_LOCK_MATCHED)
2153 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
2154 ldlm_lock_decref(lockh, mode);
2159 int osc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2160 struct osc_enqueue_args *aa, int rc)
2162 struct ldlm_lock *lock;
2163 struct lustre_handle *lockh = &aa->oa_lockh;
2164 enum ldlm_mode mode = aa->oa_mode;
2165 struct ost_lvb *lvb = aa->oa_lvb;
2166 __u32 lvb_len = sizeof(*lvb);
2171 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2173 lock = ldlm_handle2lock(lockh);
2174 LASSERTF(lock != NULL,
2175 "lockh %#llx, req %p, aa %p - client evicted?\n",
2176 lockh->cookie, req, aa);
2178 /* Take an additional reference so that a blocking AST that
2179 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2180 * to arrive after an upcall has been executed by
2181 * osc_enqueue_fini(). */
2182 ldlm_lock_addref(lockh, mode);
2184 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
2185 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
2187 /* Let CP AST to grant the lock first. */
2188 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2190 if (aa->oa_speculative) {
2191 LASSERT(aa->oa_lvb == NULL);
2192 LASSERT(aa->oa_flags == NULL);
2193 aa->oa_flags = &flags;
2196 /* Complete obtaining the lock procedure. */
2197 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
2198 aa->oa_mode, aa->oa_flags, lvb, lvb_len,
2200 /* Complete osc stuff. */
2201 rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
2202 aa->oa_flags, aa->oa_speculative, rc);
2204 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2206 ldlm_lock_decref(lockh, mode);
2207 LDLM_LOCK_PUT(lock);
2211 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2213 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2214 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2215 * other synchronous requests, however keeping some locks and trying to obtain
2216 * others may take a considerable amount of time in a case of ost failure; and
2217 * when other sync requests do not get released lock from a client, the client
2218 * is evicted from the cluster -- such scenarious make the life difficult, so
2219 * release locks just after they are obtained. */
2220 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2221 __u64 *flags, union ldlm_policy_data *policy,
2222 struct ost_lvb *lvb, int kms_valid,
2223 osc_enqueue_upcall_f upcall, void *cookie,
2224 struct ldlm_enqueue_info *einfo,
2225 struct ptlrpc_request_set *rqset, int async,
2228 struct obd_device *obd = exp->exp_obd;
2229 struct lustre_handle lockh = { 0 };
2230 struct ptlrpc_request *req = NULL;
2231 int intent = *flags & LDLM_FL_HAS_INTENT;
2232 __u64 match_flags = *flags;
2233 enum ldlm_mode mode;
2237 /* Filesystem lock extents are extended to page boundaries so that
2238 * dealing with the page cache is a little smoother. */
2239 policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2240 policy->l_extent.end |= ~PAGE_MASK;
2243 * kms is not valid when either object is completely fresh (so that no
2244 * locks are cached), or object was evicted. In the latter case cached
2245 * lock cannot be used, because it would prime inode state with
2246 * potentially stale LVB.
2251 /* Next, search for already existing extent locks that will cover us */
2252 /* If we're trying to read, we also search for an existing PW lock. The
2253 * VFS and page cache already protect us locally, so lots of readers/
2254 * writers can share a single PW lock.
2256 * There are problems with conversion deadlocks, so instead of
2257 * converting a read lock to a write lock, we'll just enqueue a new
2260 * At some point we should cancel the read lock instead of making them
2261 * send us a blocking callback, but there are problems with canceling
2262 * locks out from other users right now, too. */
2263 mode = einfo->ei_mode;
2264 if (einfo->ei_mode == LCK_PR)
2266 /* Normal lock requests must wait for the LVB to be ready before
2267 * matching a lock; speculative lock requests do not need to,
2268 * because they will not actually use the lock. */
2270 match_flags |= LDLM_FL_LVB_READY;
2272 match_flags |= LDLM_FL_BLOCK_GRANTED;
2273 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
2274 einfo->ei_type, policy, mode, &lockh, 0);
2276 struct ldlm_lock *matched;
2278 if (*flags & LDLM_FL_TEST_LOCK)
2281 matched = ldlm_handle2lock(&lockh);
2283 /* This DLM lock request is speculative, and does not
2284 * have an associated IO request. Therefore if there
2285 * is already a DLM lock, it wll just inform the
2286 * caller to cancel the request for this stripe.*/
2287 lock_res_and_lock(matched);
2288 if (ldlm_extent_equal(&policy->l_extent,
2289 &matched->l_policy_data.l_extent))
2293 unlock_res_and_lock(matched);
2295 ldlm_lock_decref(&lockh, mode);
2296 LDLM_LOCK_PUT(matched);
2298 } else if (osc_set_lock_data(matched, einfo->ei_cbdata)) {
2299 *flags |= LDLM_FL_LVB_READY;
2301 /* We already have a lock, and it's referenced. */
2302 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
2304 ldlm_lock_decref(&lockh, mode);
2305 LDLM_LOCK_PUT(matched);
2308 ldlm_lock_decref(&lockh, mode);
2309 LDLM_LOCK_PUT(matched);
2314 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
2318 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2319 &RQF_LDLM_ENQUEUE_LVB);
2323 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
2325 ptlrpc_request_free(req);
2329 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
2331 ptlrpc_request_set_replen(req);
2334 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2335 *flags &= ~LDLM_FL_BLOCK_GRANTED;
2337 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2338 sizeof(*lvb), LVB_T_OST, &lockh, async);
2341 struct osc_enqueue_args *aa;
2342 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2343 aa = ptlrpc_req_async_args(req);
2345 aa->oa_mode = einfo->ei_mode;
2346 aa->oa_type = einfo->ei_type;
2347 lustre_handle_copy(&aa->oa_lockh, &lockh);
2348 aa->oa_upcall = upcall;
2349 aa->oa_cookie = cookie;
2350 aa->oa_speculative = speculative;
2352 aa->oa_flags = flags;
2355 /* speculative locks are essentially to enqueue
2356 * a DLM lock in advance, so we don't care
2357 * about the result of the enqueue. */
2359 aa->oa_flags = NULL;
2362 req->rq_interpret_reply =
2363 (ptlrpc_interpterer_t)osc_enqueue_interpret;
2364 if (rqset == PTLRPCD_SET)
2365 ptlrpcd_add_req(req);
2367 ptlrpc_set_add_req(rqset, req);
2368 } else if (intent) {
2369 ptlrpc_req_finished(req);
2374 rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
2375 flags, speculative, rc);
2377 ptlrpc_req_finished(req);
2382 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2383 enum ldlm_type type, union ldlm_policy_data *policy,
2384 enum ldlm_mode mode, __u64 *flags, void *data,
2385 struct lustre_handle *lockh, int unref)
2387 struct obd_device *obd = exp->exp_obd;
2388 __u64 lflags = *flags;
2392 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
2395 /* Filesystem lock extents are extended to page boundaries so that
2396 * dealing with the page cache is a little smoother */
2397 policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2398 policy->l_extent.end |= ~PAGE_MASK;
2400 /* Next, search for already existing extent locks that will cover us */
2401 /* If we're trying to read, we also search for an existing PW lock. The
2402 * VFS and page cache already protect us locally, so lots of readers/
2403 * writers can share a single PW lock. */
2407 rc = ldlm_lock_match(obd->obd_namespace, lflags,
2408 res_id, type, policy, rc, lockh, unref);
2409 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
2413 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2415 LASSERT(lock != NULL);
2416 if (!osc_set_lock_data(lock, data)) {
2417 ldlm_lock_decref(lockh, rc);
2420 LDLM_LOCK_PUT(lock);
2425 static int osc_statfs_interpret(const struct lu_env *env,
2426 struct ptlrpc_request *req,
2427 struct osc_async_args *aa, int rc)
2429 struct obd_statfs *msfs;
2433 /* The request has in fact never been sent
2434 * due to issues at a higher level (LOV).
2435 * Exit immediately since the caller is
2436 * aware of the problem and takes care
2437 * of the clean up */
2440 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2441 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2447 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2449 GOTO(out, rc = -EPROTO);
2452 *aa->aa_oi->oi_osfs = *msfs;
2454 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
2458 static int osc_statfs_async(struct obd_export *exp,
2459 struct obd_info *oinfo, __u64 max_age,
2460 struct ptlrpc_request_set *rqset)
2462 struct obd_device *obd = class_exp2obd(exp);
2463 struct ptlrpc_request *req;
2464 struct osc_async_args *aa;
2468 /* We could possibly pass max_age in the request (as an absolute
2469 * timestamp or a "seconds.usec ago") so the target can avoid doing
2470 * extra calls into the filesystem if that isn't necessary (e.g.
2471 * during mount that would help a bit). Having relative timestamps
2472 * is not so great if request processing is slow, while absolute
2473 * timestamps are not ideal because they need time synchronization. */
2474 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2478 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2480 ptlrpc_request_free(req);
2483 ptlrpc_request_set_replen(req);
2484 req->rq_request_portal = OST_CREATE_PORTAL;
2485 ptlrpc_at_set_req_timeout(req);
2487 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2488 /* procfs requests not want stat in wait for avoid deadlock */
2489 req->rq_no_resend = 1;
2490 req->rq_no_delay = 1;
2493 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
2494 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2495 aa = ptlrpc_req_async_args(req);
2498 ptlrpc_set_add_req(rqset, req);
2502 static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2503 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
2505 struct obd_device *obd = class_exp2obd(exp);
2506 struct obd_statfs *msfs;
2507 struct ptlrpc_request *req;
2508 struct obd_import *imp = NULL;
2512 /*Since the request might also come from lprocfs, so we need
2513 *sync this with client_disconnect_export Bug15684*/
2514 down_read(&obd->u.cli.cl_sem);
2515 if (obd->u.cli.cl_import)
2516 imp = class_import_get(obd->u.cli.cl_import);
2517 up_read(&obd->u.cli.cl_sem);
2521 /* We could possibly pass max_age in the request (as an absolute
2522 * timestamp or a "seconds.usec ago") so the target can avoid doing
2523 * extra calls into the filesystem if that isn't necessary (e.g.
2524 * during mount that would help a bit). Having relative timestamps
2525 * is not so great if request processing is slow, while absolute
2526 * timestamps are not ideal because they need time synchronization. */
2527 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2529 class_import_put(imp);
2534 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2536 ptlrpc_request_free(req);
2539 ptlrpc_request_set_replen(req);
2540 req->rq_request_portal = OST_CREATE_PORTAL;
2541 ptlrpc_at_set_req_timeout(req);
2543 if (flags & OBD_STATFS_NODELAY) {
2544 /* procfs requests not want stat in wait for avoid deadlock */
2545 req->rq_no_resend = 1;
2546 req->rq_no_delay = 1;
2549 rc = ptlrpc_queue_wait(req);
2553 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2555 GOTO(out, rc = -EPROTO);
2562 ptlrpc_req_finished(req);
2566 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2567 void *karg, void __user *uarg)
2569 struct obd_device *obd = exp->exp_obd;
2570 struct obd_ioctl_data *data = karg;
2574 if (!try_module_get(THIS_MODULE)) {
2575 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2576 module_name(THIS_MODULE));
2580 case OBD_IOC_CLIENT_RECOVER:
2581 err = ptlrpc_recover_import(obd->u.cli.cl_import,
2582 data->ioc_inlbuf1, 0);
2586 case IOC_OSC_SET_ACTIVE:
2587 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
2590 case OBD_IOC_PING_TARGET:
2591 err = ptlrpc_obd_ping(obd);
2594 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
2595 cmd, current_comm());
2596 GOTO(out, err = -ENOTTY);
2599 module_put(THIS_MODULE);
2603 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
2604 u32 keylen, void *key, u32 vallen, void *val,
2605 struct ptlrpc_request_set *set)
2607 struct ptlrpc_request *req;
2608 struct obd_device *obd = exp->exp_obd;
2609 struct obd_import *imp = class_exp2cliimp(exp);
2614 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
2616 if (KEY_IS(KEY_CHECKSUM)) {
2617 if (vallen != sizeof(int))
2619 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
2623 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2624 sptlrpc_conf_client_adapt(obd);
2628 if (KEY_IS(KEY_FLUSH_CTX)) {
2629 sptlrpc_import_flush_my_ctx(imp);
2633 if (KEY_IS(KEY_CACHE_SET)) {
2634 struct client_obd *cli = &obd->u.cli;
2636 LASSERT(cli->cl_cache == NULL); /* only once */
2637 cli->cl_cache = (struct cl_client_cache *)val;
2638 cl_cache_incref(cli->cl_cache);
2639 cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
2641 /* add this osc into entity list */
2642 LASSERT(list_empty(&cli->cl_lru_osc));
2643 spin_lock(&cli->cl_cache->ccc_lru_lock);
2644 list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
2645 spin_unlock(&cli->cl_cache->ccc_lru_lock);
2650 if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
2651 struct client_obd *cli = &obd->u.cli;
2652 long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
2653 long target = *(long *)val;
2655 nr = osc_lru_shrink(env, cli, min(nr, target), true);
2660 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
2663 /* We pass all other commands directly to OST. Since nobody calls osc
2664 methods directly and everybody is supposed to go through LOV, we
2665 assume lov checked invalid values for us.
2666 The only recognised values so far are evict_by_nid and mds_conn.
2667 Even if something bad goes through, we'd get a -EINVAL from OST
2670 req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
2671 &RQF_OST_SET_GRANT_INFO :
2676 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2677 RCL_CLIENT, keylen);
2678 if (!KEY_IS(KEY_GRANT_SHRINK))
2679 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
2680 RCL_CLIENT, vallen);
2681 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
2683 ptlrpc_request_free(req);
2687 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
2688 memcpy(tmp, key, keylen);
2689 tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
2692 memcpy(tmp, val, vallen);
2694 if (KEY_IS(KEY_GRANT_SHRINK)) {
2695 struct osc_grant_args *aa;
2698 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2699 aa = ptlrpc_req_async_args(req);
2702 ptlrpc_req_finished(req);
2705 *oa = ((struct ost_body *)val)->oa;
2707 req->rq_interpret_reply = osc_shrink_grant_interpret;
2710 ptlrpc_request_set_replen(req);
2711 if (!KEY_IS(KEY_GRANT_SHRINK)) {
2712 LASSERT(set != NULL);
2713 ptlrpc_set_add_req(set, req);
2714 ptlrpc_check_set(NULL, set);
2716 ptlrpcd_add_req(req);
2721 EXPORT_SYMBOL(osc_set_info_async);
2723 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
2724 struct obd_device *obd, struct obd_uuid *cluuid,
2725 struct obd_connect_data *data, void *localdata)
2727 struct client_obd *cli = &obd->u.cli;
2729 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
2733 spin_lock(&cli->cl_loi_list_lock);
2734 grant = cli->cl_avail_grant + cli->cl_reserved_grant;
2735 if (data->ocd_connect_flags & OBD_CONNECT_GRANT_PARAM)
2736 grant += cli->cl_dirty_grant;
2738 grant += cli->cl_dirty_pages << PAGE_SHIFT;
2739 data->ocd_grant = grant ? : 2 * cli_brw_size(obd);
2740 lost_grant = cli->cl_lost_grant;
2741 cli->cl_lost_grant = 0;
2742 spin_unlock(&cli->cl_loi_list_lock);
2744 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d"
2745 " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
2746 data->ocd_version, data->ocd_grant, lost_grant);
2751 EXPORT_SYMBOL(osc_reconnect);
2753 int osc_disconnect(struct obd_export *exp)
2755 struct obd_device *obd = class_exp2obd(exp);
2758 rc = client_disconnect_export(exp);
2760 * Initially we put del_shrink_grant before disconnect_export, but it
2761 * causes the following problem if setup (connect) and cleanup
2762 * (disconnect) are tangled together.
2763 * connect p1 disconnect p2
2764 * ptlrpc_connect_import
2765 * ............... class_manual_cleanup
2768 * ptlrpc_connect_interrupt
2770 * add this client to shrink list
2772 * Bang! pinger trigger the shrink.
2773 * So the osc should be disconnected from the shrink list, after we
2774 * are sure the import has been destroyed. BUG18662
2776 if (obd->u.cli.cl_import == NULL)
2777 osc_del_shrink_grant(&obd->u.cli);
2780 EXPORT_SYMBOL(osc_disconnect);
2782 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2783 struct hlist_node *hnode, void *arg)
2785 struct lu_env *env = arg;
2786 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2787 struct ldlm_lock *lock;
2788 struct osc_object *osc = NULL;
2792 list_for_each_entry(lock, &res->lr_granted, l_res_link) {
2793 if (lock->l_ast_data != NULL && osc == NULL) {
2794 osc = lock->l_ast_data;
2795 cl_object_get(osc2cl(osc));
2798 /* clear LDLM_FL_CLEANED flag to make sure it will be canceled
2799 * by the 2nd round of ldlm_namespace_clean() call in
2800 * osc_import_event(). */
2801 ldlm_clear_cleaned(lock);
2806 osc_object_invalidate(env, osc);
2807 cl_object_put(env, osc2cl(osc));
2812 EXPORT_SYMBOL(osc_ldlm_resource_invalidate);
2814 static int osc_import_event(struct obd_device *obd,
2815 struct obd_import *imp,
2816 enum obd_import_event event)
2818 struct client_obd *cli;
2822 LASSERT(imp->imp_obd == obd);
2825 case IMP_EVENT_DISCON: {
2827 spin_lock(&cli->cl_loi_list_lock);
2828 cli->cl_avail_grant = 0;
2829 cli->cl_lost_grant = 0;
2830 spin_unlock(&cli->cl_loi_list_lock);
2833 case IMP_EVENT_INACTIVE: {
2834 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2837 case IMP_EVENT_INVALIDATE: {
2838 struct ldlm_namespace *ns = obd->obd_namespace;
2842 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2844 env = cl_env_get(&refcheck);
2846 osc_io_unplug(env, &obd->u.cli, NULL);
2848 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2849 osc_ldlm_resource_invalidate,
2851 cl_env_put(env, &refcheck);
2853 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2858 case IMP_EVENT_ACTIVE: {
2859 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2862 case IMP_EVENT_OCD: {
2863 struct obd_connect_data *ocd = &imp->imp_connect_data;
2865 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
2866 osc_init_grant(&obd->u.cli, ocd);
2869 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
2870 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
2872 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2875 case IMP_EVENT_DEACTIVATE: {
2876 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE);
2879 case IMP_EVENT_ACTIVATE: {
2880 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE);
2884 CERROR("Unknown import event %d\n", event);
2891 * Determine whether the lock can be canceled before replaying the lock
2892 * during recovery, see bug16774 for detailed information.
2894 * \retval zero the lock can't be canceled
2895 * \retval other ok to cancel
2897 static int osc_cancel_weight(struct ldlm_lock *lock)
2900 * Cancel all unused and granted extent lock.
2902 if (lock->l_resource->lr_type == LDLM_EXTENT &&
2903 lock->l_granted_mode == lock->l_req_mode &&
2904 osc_ldlm_weigh_ast(lock) == 0)
2910 static int brw_queue_work(const struct lu_env *env, void *data)
2912 struct client_obd *cli = data;
2914 CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
2916 osc_io_unplug(env, cli, NULL);
2920 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg)
2922 struct client_obd *cli = &obd->u.cli;
2928 rc = ptlrpcd_addref();
2932 rc = client_obd_setup(obd, lcfg);
2934 GOTO(out_ptlrpcd, rc);
2937 handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
2938 if (IS_ERR(handler))
2939 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
2940 cli->cl_writeback_work = handler;
2942 handler = ptlrpcd_alloc_work(cli->cl_import, lru_queue_work, cli);
2943 if (IS_ERR(handler))
2944 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
2945 cli->cl_lru_work = handler;
2947 rc = osc_quota_setup(obd);
2949 GOTO(out_ptlrpcd_work, rc);
2951 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
2953 INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
2957 if (cli->cl_writeback_work != NULL) {
2958 ptlrpcd_destroy_work(cli->cl_writeback_work);
2959 cli->cl_writeback_work = NULL;
2961 if (cli->cl_lru_work != NULL) {
2962 ptlrpcd_destroy_work(cli->cl_lru_work);
2963 cli->cl_lru_work = NULL;
2965 client_obd_cleanup(obd);
2970 EXPORT_SYMBOL(osc_setup_common);
2972 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
2974 struct client_obd *cli = &obd->u.cli;
2975 struct obd_type *type;
2983 rc = osc_setup_common(obd, lcfg);
2987 #ifdef CONFIG_PROC_FS
2988 obd->obd_vars = lprocfs_osc_obd_vars;
2990 /* If this is true then both client (osc) and server (osp) are on the
2991 * same node. The osp layer if loaded first will register the osc proc
2992 * directory. In that case this obd_device will be attached its proc
2993 * tree to type->typ_procsym instead of obd->obd_type->typ_procroot.
2995 type = class_search_type(LUSTRE_OSP_NAME);
2996 if (type && type->typ_procsym) {
2997 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
2999 obd->obd_vars, obd);
3000 if (IS_ERR(obd->obd_proc_entry)) {
3001 rc = PTR_ERR(obd->obd_proc_entry);
3002 CERROR("error %d setting up lprocfs for %s\n", rc,
3004 obd->obd_proc_entry = NULL;
3008 rc = lprocfs_obd_setup(obd, false);
3010 /* If the basic OSC proc tree construction succeeded then
3013 lproc_osc_attach_seqstat(obd);
3014 sptlrpc_lprocfs_cliobd_attach(obd);
3015 ptlrpc_lprocfs_register_obd(obd);
3019 * We try to control the total number of requests with a upper limit
3020 * osc_reqpool_maxreqcount. There might be some race which will cause
3021 * over-limit allocation, but it is fine.
3023 req_count = atomic_read(&osc_pool_req_count);
3024 if (req_count < osc_reqpool_maxreqcount) {
3025 adding = cli->cl_max_rpcs_in_flight + 2;
3026 if (req_count + adding > osc_reqpool_maxreqcount)
3027 adding = osc_reqpool_maxreqcount - req_count;
3029 added = ptlrpc_add_rqs_to_pool(osc_rq_pool, adding);
3030 atomic_add(added, &osc_pool_req_count);
3033 INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
3034 ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
3036 spin_lock(&osc_shrink_lock);
3037 list_add_tail(&cli->cl_shrink_list, &osc_shrink_list);
3038 spin_unlock(&osc_shrink_lock);
3043 int osc_precleanup_common(struct obd_device *obd)
3045 struct client_obd *cli = &obd->u.cli;
3049 * for echo client, export may be on zombie list, wait for
3050 * zombie thread to cull it, because cli.cl_import will be
3051 * cleared in client_disconnect_export():
3052 * class_export_destroy() -> obd_cleanup() ->
3053 * echo_device_free() -> echo_client_cleanup() ->
3054 * obd_disconnect() -> osc_disconnect() ->
3055 * client_disconnect_export()
3057 obd_zombie_barrier();
3058 if (cli->cl_writeback_work) {
3059 ptlrpcd_destroy_work(cli->cl_writeback_work);
3060 cli->cl_writeback_work = NULL;
3063 if (cli->cl_lru_work) {
3064 ptlrpcd_destroy_work(cli->cl_lru_work);
3065 cli->cl_lru_work = NULL;
3068 obd_cleanup_client_import(obd);
3071 EXPORT_SYMBOL(osc_precleanup_common);
3073 static int osc_precleanup(struct obd_device *obd)
3077 osc_precleanup_common(obd);
3079 ptlrpc_lprocfs_unregister_obd(obd);
3083 int osc_cleanup_common(struct obd_device *obd)
3085 struct client_obd *cli = &obd->u.cli;
3090 spin_lock(&osc_shrink_lock);
3091 list_del(&cli->cl_shrink_list);
3092 spin_unlock(&osc_shrink_lock);
3095 if (cli->cl_cache != NULL) {
3096 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
3097 spin_lock(&cli->cl_cache->ccc_lru_lock);
3098 list_del_init(&cli->cl_lru_osc);
3099 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3100 cli->cl_lru_left = NULL;
3101 cl_cache_decref(cli->cl_cache);
3102 cli->cl_cache = NULL;
3105 /* free memory of osc quota cache */
3106 osc_quota_cleanup(obd);
3108 rc = client_obd_cleanup(obd);
3113 EXPORT_SYMBOL(osc_cleanup_common);
3115 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
3117 int rc = class_process_proc_param(PARAM_OSC, obd->obd_vars, lcfg, obd);
3118 return rc > 0 ? 0: rc;
3121 static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
3123 return osc_process_config_base(obd, buf);
3126 static struct obd_ops osc_obd_ops = {
3127 .o_owner = THIS_MODULE,
3128 .o_setup = osc_setup,
3129 .o_precleanup = osc_precleanup,
3130 .o_cleanup = osc_cleanup_common,
3131 .o_add_conn = client_import_add_conn,
3132 .o_del_conn = client_import_del_conn,
3133 .o_connect = client_connect_import,
3134 .o_reconnect = osc_reconnect,
3135 .o_disconnect = osc_disconnect,
3136 .o_statfs = osc_statfs,
3137 .o_statfs_async = osc_statfs_async,
3138 .o_create = osc_create,
3139 .o_destroy = osc_destroy,
3140 .o_getattr = osc_getattr,
3141 .o_setattr = osc_setattr,
3142 .o_iocontrol = osc_iocontrol,
3143 .o_set_info_async = osc_set_info_async,
3144 .o_import_event = osc_import_event,
3145 .o_process_config = osc_process_config,
3146 .o_quotactl = osc_quotactl,
3149 static struct shrinker *osc_cache_shrinker;
3150 struct list_head osc_shrink_list = LIST_HEAD_INIT(osc_shrink_list);
3151 DEFINE_SPINLOCK(osc_shrink_lock);
3153 #ifndef HAVE_SHRINKER_COUNT
3154 static int osc_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
3156 struct shrink_control scv = {
3157 .nr_to_scan = shrink_param(sc, nr_to_scan),
3158 .gfp_mask = shrink_param(sc, gfp_mask)
3160 #if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL)
3161 struct shrinker *shrinker = NULL;
3164 (void)osc_cache_shrink_scan(shrinker, &scv);
3166 return osc_cache_shrink_count(shrinker, &scv);
3170 static int __init osc_init(void)
3172 bool enable_proc = true;
3173 struct obd_type *type;
3174 unsigned int reqpool_size;
3175 unsigned int reqsize;
3177 DEF_SHRINKER_VAR(osc_shvar, osc_cache_shrink,
3178 osc_cache_shrink_count, osc_cache_shrink_scan);
3181 /* print an address of _any_ initialized kernel symbol from this
3182 * module, to allow debugging with gdb that doesn't support data
3183 * symbols from modules.*/
3184 CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
3186 rc = lu_kmem_init(osc_caches);
3190 type = class_search_type(LUSTRE_OSP_NAME);
3191 if (type != NULL && type->typ_procsym != NULL)
3192 enable_proc = false;
3194 rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
3195 LUSTRE_OSC_NAME, &osc_device_type);
3199 osc_cache_shrinker = set_shrinker(DEFAULT_SEEKS, &osc_shvar);
3201 /* This is obviously too much memory, only prevent overflow here */
3202 if (osc_reqpool_mem_max >= 1 << 12 || osc_reqpool_mem_max == 0)
3203 GOTO(out_type, rc = -EINVAL);
3205 reqpool_size = osc_reqpool_mem_max << 20;
3208 while (reqsize < OST_IO_MAXREQSIZE)
3209 reqsize = reqsize << 1;
3212 * We don't enlarge the request count in OSC pool according to
3213 * cl_max_rpcs_in_flight. The allocation from the pool will only be
3214 * tried after normal allocation failed. So a small OSC pool won't
3215 * cause much performance degression in most of cases.
3217 osc_reqpool_maxreqcount = reqpool_size / reqsize;
3219 atomic_set(&osc_pool_req_count, 0);
3220 osc_rq_pool = ptlrpc_init_rq_pool(0, OST_IO_MAXREQSIZE,
3221 ptlrpc_add_rqs_to_pool);
3223 if (osc_rq_pool != NULL)
3227 class_unregister_type(LUSTRE_OSC_NAME);
3229 lu_kmem_fini(osc_caches);
3234 static void __exit osc_exit(void)
3236 remove_shrinker(osc_cache_shrinker);
3237 class_unregister_type(LUSTRE_OSC_NAME);
3238 lu_kmem_fini(osc_caches);
3239 ptlrpc_free_rq_pool(osc_rq_pool);
3242 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3243 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3244 MODULE_VERSION(LUSTRE_VERSION_STRING);
3245 MODULE_LICENSE("GPL");
3247 module_init(osc_init);
3248 module_exit(osc_exit);