lustre/osc/osc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_OSC
  34
  35 #include <lprocfs_status.h>
  36 #include <lustre_debug.h>
  37 #include <lustre_dlm.h>
  38 #include <lustre_fid.h>
  39 #include <lustre_ha.h>
  40 #include <uapi/linux/lustre/lustre_ioctl.h>
  41 #include <lustre_net.h>
  42 #include <lustre_obdo.h>
  43 #include <uapi/linux/lustre/lustre_param.h>
  44 #include <obd.h>
  45 #include <obd_cksum.h>
  46 #include <obd_class.h>
  47 #include <lustre_osc.h>
  48
  49 #include "osc_internal.h"
  50
  51 atomic_t osc_pool_req_count;
  52 unsigned int osc_reqpool_maxreqcount;
  53 struct ptlrpc_request_pool *osc_rq_pool;
  54
  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);
  58
  59 static int osc_idle_timeout = 20;
  60 module_param(osc_idle_timeout, uint, 0644);
  61
  62 #define osc_grant_args osc_brw_async_args
  63
  64 struct osc_setattr_args {
  65         struct obdo             *sa_oa;
  66         obd_enqueue_update_f     sa_upcall;
  67         void                    *sa_cookie;
  68 };
  69
  70 struct osc_fsync_args {
  71         struct osc_object       *fa_obj;
  72         struct obdo             *fa_oa;
  73         obd_enqueue_update_f    fa_upcall;
  74         void                    *fa_cookie;
  75 };
  76
  77 struct osc_ladvise_args {
  78         struct obdo             *la_oa;
  79         obd_enqueue_update_f     la_upcall;
  80         void                    *la_cookie;
  81 };
  82
  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,
  85                          void *data, int rc);
  86
  87 void osc_pack_req_body(struct ptlrpc_request *req, struct obdo *oa)
  88 {
  89         struct ost_body *body;
  90
  91         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
  92         LASSERT(body);
  93
  94         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
  95 }
  96
  97 static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
  98                        struct obdo *oa)
  99 {
 100         struct ptlrpc_request   *req;
 101         struct ost_body         *body;
 102         int                      rc;
 103
 104         ENTRY;
 105         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
 106         if (req == NULL)
 107                 RETURN(-ENOMEM);
 108
 109         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
 110         if (rc) {
 111                 ptlrpc_request_free(req);
 112                 RETURN(rc);
 113         }
 114
 115         osc_pack_req_body(req, oa);
 116
 117         ptlrpc_request_set_replen(req);
 118
 119         rc = ptlrpc_queue_wait(req);
 120         if (rc)
 121                 GOTO(out, rc);
 122
 123         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 124         if (body == NULL)
 125                 GOTO(out, rc = -EPROTO);
 126
 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);
 129
 130         oa->o_blksize = cli_brw_size(exp->exp_obd);
 131         oa->o_valid |= OBD_MD_FLBLKSZ;
 132
 133         EXIT;
 134 out:
 135         ptlrpc_req_finished(req);
 136
 137         return rc;
 138 }
 139
 140 static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
 141                        struct obdo *oa)
 142 {
 143         struct ptlrpc_request   *req;
 144         struct ost_body         *body;
 145         int                      rc;
 146
 147         ENTRY;
 148         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
 149
 150         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
 151         if (req == NULL)
 152                 RETURN(-ENOMEM);
 153
 154         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
 155         if (rc) {
 156                 ptlrpc_request_free(req);
 157                 RETURN(rc);
 158         }
 159
 160         osc_pack_req_body(req, oa);
 161
 162         ptlrpc_request_set_replen(req);
 163
 164         rc = ptlrpc_queue_wait(req);
 165         if (rc)
 166                 GOTO(out, rc);
 167
 168         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 169         if (body == NULL)
 170                 GOTO(out, rc = -EPROTO);
 171
 172         lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
 173
 174         EXIT;
 175 out:
 176         ptlrpc_req_finished(req);
 177
 178         RETURN(rc);
 179 }
 180
 181 static int osc_setattr_interpret(const struct lu_env *env,
 182                                  struct ptlrpc_request *req,
 183                                  struct osc_setattr_args *sa, int rc)
 184 {
 185         struct ost_body *body;
 186         ENTRY;
 187
 188         if (rc != 0)
 189                 GOTO(out, rc);
 190
 191         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 192         if (body == NULL)
 193                 GOTO(out, rc = -EPROTO);
 194
 195         lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
 196                              &body->oa);
 197 out:
 198         rc = sa->sa_upcall(sa->sa_cookie, rc);
 199         RETURN(rc);
 200 }
 201
 202 int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
 203                       obd_enqueue_update_f upcall, void *cookie,
 204                       struct ptlrpc_request_set *rqset)
 205 {
 206         struct ptlrpc_request   *req;
 207         struct osc_setattr_args *sa;
 208         int                      rc;
 209
 210         ENTRY;
 211
 212         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
 213         if (req == NULL)
 214                 RETURN(-ENOMEM);
 215
 216         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
 217         if (rc) {
 218                 ptlrpc_request_free(req);
 219                 RETURN(rc);
 220         }
 221
 222         osc_pack_req_body(req, oa);
 223
 224         ptlrpc_request_set_replen(req);
 225
 226         /* do mds to ost setattr asynchronously */
 227         if (!rqset) {
 228                 /* Do not wait for response. */
 229                 ptlrpcd_add_req(req);
 230         } else {
 231                 req->rq_interpret_reply =
 232                         (ptlrpc_interpterer_t)osc_setattr_interpret;
 233
 234                 CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
 235                 sa = ptlrpc_req_async_args(req);
 236                 sa->sa_oa = oa;
 237                 sa->sa_upcall = upcall;
 238                 sa->sa_cookie = cookie;
 239
 240                 if (rqset == PTLRPCD_SET)
 241                         ptlrpcd_add_req(req);
 242                 else
 243                         ptlrpc_set_add_req(rqset, req);
 244         }
 245
 246         RETURN(0);
 247 }
 248
 249 static int osc_ladvise_interpret(const struct lu_env *env,
 250                                  struct ptlrpc_request *req,
 251                                  void *arg, int rc)
 252 {
 253         struct osc_ladvise_args *la = arg;
 254         struct ost_body *body;
 255         ENTRY;
 256
 257         if (rc != 0)
 258                 GOTO(out, rc);
 259
 260         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 261         if (body == NULL)
 262                 GOTO(out, rc = -EPROTO);
 263
 264         *la->la_oa = body->oa;
 265 out:
 266         rc = la->la_upcall(la->la_cookie, rc);
 267         RETURN(rc);
 268 }
 269
 270 /**
 271  * If rqset is NULL, do not wait for response. Upcall and cookie could also
 272  * be NULL in this case
 273  */
 274 int osc_ladvise_base(struct obd_export *exp, struct obdo *oa,
 275                      struct ladvise_hdr *ladvise_hdr,
 276                      obd_enqueue_update_f upcall, void *cookie,
 277                      struct ptlrpc_request_set *rqset)
 278 {
 279         struct ptlrpc_request   *req;
 280         struct ost_body         *body;
 281         struct osc_ladvise_args *la;
 282         int                      rc;
 283         struct lu_ladvise       *req_ladvise;
 284         struct lu_ladvise       *ladvise = ladvise_hdr->lah_advise;
 285         int                      num_advise = ladvise_hdr->lah_count;
 286         struct ladvise_hdr      *req_ladvise_hdr;
 287         ENTRY;
 288
 289         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_LADVISE);
 290         if (req == NULL)
 291                 RETURN(-ENOMEM);
 292
 293         req_capsule_set_size(&req->rq_pill, &RMF_OST_LADVISE, RCL_CLIENT,
 294                              num_advise * sizeof(*ladvise));
 295         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_LADVISE);
 296         if (rc != 0) {
 297                 ptlrpc_request_free(req);
 298                 RETURN(rc);
 299         }
 300         req->rq_request_portal = OST_IO_PORTAL;
 301         ptlrpc_at_set_req_timeout(req);
 302
 303         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 304         LASSERT(body);
 305         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
 306                              oa);
 307
 308         req_ladvise_hdr = req_capsule_client_get(&req->rq_pill,
 309                                                  &RMF_OST_LADVISE_HDR);
 310         memcpy(req_ladvise_hdr, ladvise_hdr, sizeof(*ladvise_hdr));
 311
 312         req_ladvise = req_capsule_client_get(&req->rq_pill, &RMF_OST_LADVISE);
 313         memcpy(req_ladvise, ladvise, sizeof(*ladvise) * num_advise);
 314         ptlrpc_request_set_replen(req);
 315
 316         if (rqset == NULL) {
 317                 /* Do not wait for response. */
 318                 ptlrpcd_add_req(req);
 319                 RETURN(0);
 320         }
 321
 322         req->rq_interpret_reply = osc_ladvise_interpret;
 323         CLASSERT(sizeof(*la) <= sizeof(req->rq_async_args));
 324         la = ptlrpc_req_async_args(req);
 325         la->la_oa = oa;
 326         la->la_upcall = upcall;
 327         la->la_cookie = cookie;
 328
 329         if (rqset == PTLRPCD_SET)
 330                 ptlrpcd_add_req(req);
 331         else
 332                 ptlrpc_set_add_req(rqset, req);
 333
 334         RETURN(0);
 335 }
 336
 337 static int osc_create(const struct lu_env *env, struct obd_export *exp,
 338                       struct obdo *oa)
 339 {
 340         struct ptlrpc_request *req;
 341         struct ost_body       *body;
 342         int                    rc;
 343         ENTRY;
 344
 345         LASSERT(oa != NULL);
 346         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
 347         LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
 348
 349         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
 350         if (req == NULL)
 351                 GOTO(out, rc = -ENOMEM);
 352
 353         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
 354         if (rc) {
 355                 ptlrpc_request_free(req);
 356                 GOTO(out, rc);
 357         }
 358
 359         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 360         LASSERT(body);
 361
 362         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
 363
 364         ptlrpc_request_set_replen(req);
 365
 366         rc = ptlrpc_queue_wait(req);
 367         if (rc)
 368                 GOTO(out_req, rc);
 369
 370         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 371         if (body == NULL)
 372                 GOTO(out_req, rc = -EPROTO);
 373
 374         CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
 375         lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
 376
 377         oa->o_blksize = cli_brw_size(exp->exp_obd);
 378         oa->o_valid |= OBD_MD_FLBLKSZ;
 379
 380         CDEBUG(D_HA, "transno: %lld\n",
 381                lustre_msg_get_transno(req->rq_repmsg));
 382 out_req:
 383         ptlrpc_req_finished(req);
 384 out:
 385         RETURN(rc);
 386 }
 387
 388 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
 389                    obd_enqueue_update_f upcall, void *cookie)
 390 {
 391         struct ptlrpc_request *req;
 392         struct osc_setattr_args *sa;
 393         struct obd_import *imp = class_exp2cliimp(exp);
 394         struct ost_body *body;
 395         int rc;
 396
 397         ENTRY;
 398
 399         req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
 400         if (req == NULL)
 401                 RETURN(-ENOMEM);
 402
 403         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
 404         if (rc < 0) {
 405                 ptlrpc_request_free(req);
 406                 RETURN(rc);
 407         }
 408
 409         osc_set_io_portal(req);
 410
 411         ptlrpc_at_set_req_timeout(req);
 412
 413         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 414
 415         lustre_set_wire_obdo(&imp->imp_connect_data, &body->oa, oa);
 416
 417         ptlrpc_request_set_replen(req);
 418
 419         req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
 420         CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
 421         sa = ptlrpc_req_async_args(req);
 422         sa->sa_oa = oa;
 423         sa->sa_upcall = upcall;
 424         sa->sa_cookie = cookie;
 425
 426         ptlrpcd_add_req(req);
 427
 428         RETURN(0);
 429 }
 430 EXPORT_SYMBOL(osc_punch_send);
 431
 432 static int osc_sync_interpret(const struct lu_env *env,
 433                               struct ptlrpc_request *req,
 434                               void *arg, int rc)
 435 {
 436         struct osc_fsync_args   *fa = arg;
 437         struct ost_body         *body;
 438         struct cl_attr          *attr = &osc_env_info(env)->oti_attr;
 439         unsigned long           valid = 0;
 440         struct cl_object        *obj;
 441         ENTRY;
 442
 443         if (rc != 0)
 444                 GOTO(out, rc);
 445
 446         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 447         if (body == NULL) {
 448                 CERROR("can't unpack ost_body\n");
 449                 GOTO(out, rc = -EPROTO);
 450         }
 451
 452         *fa->fa_oa = body->oa;
 453         obj = osc2cl(fa->fa_obj);
 454
 455         /* Update osc object's blocks attribute */
 456         cl_object_attr_lock(obj);
 457         if (body->oa.o_valid & OBD_MD_FLBLOCKS) {
 458                 attr->cat_blocks = body->oa.o_blocks;
 459                 valid |= CAT_BLOCKS;
 460         }
 461
 462         if (valid != 0)
 463                 cl_object_attr_update(env, obj, attr, valid);
 464         cl_object_attr_unlock(obj);
 465
 466 out:
 467         rc = fa->fa_upcall(fa->fa_cookie, rc);
 468         RETURN(rc);
 469 }
 470
 471 int osc_sync_base(struct osc_object *obj, struct obdo *oa,
 472                   obd_enqueue_update_f upcall, void *cookie,
 473                   struct ptlrpc_request_set *rqset)
 474 {
 475         struct obd_export     *exp = osc_export(obj);
 476         struct ptlrpc_request *req;
 477         struct ost_body       *body;
 478         struct osc_fsync_args *fa;
 479         int                    rc;
 480         ENTRY;
 481
 482         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
 483         if (req == NULL)
 484                 RETURN(-ENOMEM);
 485
 486         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
 487         if (rc) {
 488                 ptlrpc_request_free(req);
 489                 RETURN(rc);
 490         }
 491
 492         /* overload the size and blocks fields in the oa with start/end */
 493         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 494         LASSERT(body);
 495         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
 496
 497         ptlrpc_request_set_replen(req);
 498         req->rq_interpret_reply = osc_sync_interpret;
 499
 500         CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
 501         fa = ptlrpc_req_async_args(req);
 502         fa->fa_obj = obj;
 503         fa->fa_oa = oa;
 504         fa->fa_upcall = upcall;
 505         fa->fa_cookie = cookie;
 506
 507         if (rqset == PTLRPCD_SET)
 508                 ptlrpcd_add_req(req);
 509         else
 510                 ptlrpc_set_add_req(rqset, req);
 511
 512         RETURN (0);
 513 }
 514
 515 /* Find and cancel locally locks matched by @mode in the resource found by
 516  * @objid. Found locks are added into @cancel list. Returns the amount of
 517  * locks added to @cancels list. */
 518 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
 519                                    struct list_head *cancels,
 520                                    enum ldlm_mode mode, __u64 lock_flags)
 521 {
 522         struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
 523         struct ldlm_res_id res_id;
 524         struct ldlm_resource *res;
 525         int count;
 526         ENTRY;
 527
 528         /* Return, i.e. cancel nothing, only if ELC is supported (flag in
 529          * export) but disabled through procfs (flag in NS).
 530          *
 531          * This distinguishes from a case when ELC is not supported originally,
 532          * when we still want to cancel locks in advance and just cancel them
 533          * locally, without sending any RPC. */
 534         if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
 535                 RETURN(0);
 536
 537         ostid_build_res_name(&oa->o_oi, &res_id);
 538         res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
 539         if (IS_ERR(res))
 540                 RETURN(0);
 541
 542         LDLM_RESOURCE_ADDREF(res);
 543         count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
 544                                            lock_flags, 0, NULL);
 545         LDLM_RESOURCE_DELREF(res);
 546         ldlm_resource_putref(res);
 547         RETURN(count);
 548 }
 549
 550 static int osc_destroy_interpret(const struct lu_env *env,
 551                                  struct ptlrpc_request *req, void *data,
 552                                  int rc)
 553 {
 554         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
 555
 556         atomic_dec(&cli->cl_destroy_in_flight);
 557         wake_up(&cli->cl_destroy_waitq);
 558         return 0;
 559 }
 560
 561 static int osc_can_send_destroy(struct client_obd *cli)
 562 {
 563         if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
 564             cli->cl_max_rpcs_in_flight) {
 565                 /* The destroy request can be sent */
 566                 return 1;
 567         }
 568         if (atomic_dec_return(&cli->cl_destroy_in_flight) <
 569             cli->cl_max_rpcs_in_flight) {
 570                 /*
 571                  * The counter has been modified between the two atomic
 572                  * operations.
 573                  */
 574                 wake_up(&cli->cl_destroy_waitq);
 575         }
 576         return 0;
 577 }
 578
 579 static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
 580                        struct obdo *oa)
 581 {
 582         struct client_obd     *cli = &exp->exp_obd->u.cli;
 583         struct ptlrpc_request *req;
 584         struct ost_body       *body;
 585         struct list_head       cancels = LIST_HEAD_INIT(cancels);
 586         int rc, count;
 587         ENTRY;
 588
 589         if (!oa) {
 590                 CDEBUG(D_INFO, "oa NULL\n");
 591                 RETURN(-EINVAL);
 592         }
 593
 594         count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
 595                                         LDLM_FL_DISCARD_DATA);
 596
 597         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
 598         if (req == NULL) {
 599                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 600                 RETURN(-ENOMEM);
 601         }
 602
 603         rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
 604                                0, &cancels, count);
 605         if (rc) {
 606                 ptlrpc_request_free(req);
 607                 RETURN(rc);
 608         }
 609
 610         req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
 611         ptlrpc_at_set_req_timeout(req);
 612
 613         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 614         LASSERT(body);
 615         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
 616
 617         ptlrpc_request_set_replen(req);
 618
 619         req->rq_interpret_reply = osc_destroy_interpret;
 620         if (!osc_can_send_destroy(cli)) {
 621                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
 622
 623                 /*
 624                  * Wait until the number of on-going destroy RPCs drops
 625                  * under max_rpc_in_flight
 626                  */
 627                 rc = l_wait_event_exclusive(cli->cl_destroy_waitq,
 628                                             osc_can_send_destroy(cli), &lwi);
 629                 if (rc) {
 630                         ptlrpc_req_finished(req);
 631                         RETURN(rc);
 632                 }
 633         }
 634
 635         /* Do not wait for response */
 636         ptlrpcd_add_req(req);
 637         RETURN(0);
 638 }
 639
 640 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
 641                                 long writing_bytes)
 642 {
 643         u64 bits = OBD_MD_FLBLOCKS | OBD_MD_FLGRANT;
 644
 645         LASSERT(!(oa->o_valid & bits));
 646
 647         oa->o_valid |= bits;
 648         spin_lock(&cli->cl_loi_list_lock);
 649         if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data, GRANT_PARAM))
 650                 oa->o_dirty = cli->cl_dirty_grant;
 651         else
 652                 oa->o_dirty = cli->cl_dirty_pages << PAGE_SHIFT;
 653         if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
 654                      cli->cl_dirty_max_pages)) {
 655                 CERROR("dirty %lu - %lu > dirty_max %lu\n",
 656                        cli->cl_dirty_pages, cli->cl_dirty_transit,
 657                        cli->cl_dirty_max_pages);
 658                 oa->o_undirty = 0;
 659         } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
 660                             atomic_long_read(&obd_dirty_transit_pages) >
 661                             (long)(obd_max_dirty_pages + 1))) {
 662                 /* The atomic_read() allowing the atomic_inc() are
 663                  * not covered by a lock thus they may safely race and trip
 664                  * this CERROR() unless we add in a small fudge factor (+1). */
 665                 CERROR("%s: dirty %ld - %ld > system dirty_max %ld\n",
 666                        cli_name(cli), atomic_long_read(&obd_dirty_pages),
 667                        atomic_long_read(&obd_dirty_transit_pages),
 668                        obd_max_dirty_pages);
 669                 oa->o_undirty = 0;
 670         } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
 671                             0x7fffffff)) {
 672                 CERROR("dirty %lu - dirty_max %lu too big???\n",
 673                        cli->cl_dirty_pages, cli->cl_dirty_max_pages);
 674                 oa->o_undirty = 0;
 675         } else {
 676                 unsigned long nrpages;
 677                 unsigned long undirty;
 678
 679                 nrpages = cli->cl_max_pages_per_rpc;
 680                 nrpages *= cli->cl_max_rpcs_in_flight + 1;
 681                 nrpages = max(nrpages, cli->cl_dirty_max_pages);
 682                 undirty = nrpages << PAGE_SHIFT;
 683                 if (OCD_HAS_FLAG(&cli->cl_import->imp_connect_data,
 684                                  GRANT_PARAM)) {
 685                         int nrextents;
 686
 687                         /* take extent tax into account when asking for more
 688                          * grant space */
 689                         nrextents = (nrpages + cli->cl_max_extent_pages - 1)  /
 690                                      cli->cl_max_extent_pages;
 691                         undirty += nrextents * cli->cl_grant_extent_tax;
 692                 }
 693                 /* Do not ask for more than OBD_MAX_GRANT - a margin for server
 694                  * to add extent tax, etc.
 695                  */
 696                 oa->o_undirty = min(undirty, OBD_MAX_GRANT -
 697                                     (PTLRPC_MAX_BRW_PAGES << PAGE_SHIFT)*4UL);
 698         }
 699         oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
 700         oa->o_dropped = cli->cl_lost_grant;
 701         cli->cl_lost_grant = 0;
 702         spin_unlock(&cli->cl_loi_list_lock);
 703         CDEBUG(D_CACHE, "dirty: %llu undirty: %u dropped %u grant: %llu\n",
 704                oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
 705 }
 706
 707 void osc_update_next_shrink(struct client_obd *cli)
 708 {
 709         cli->cl_next_shrink_grant = ktime_get_seconds() +
 710                                     cli->cl_grant_shrink_interval;
 711
 712         CDEBUG(D_CACHE, "next time %lld to shrink grant\n",
 713                cli->cl_next_shrink_grant);
 714 }
 715
 716 static void __osc_update_grant(struct client_obd *cli, u64 grant)
 717 {
 718         spin_lock(&cli->cl_loi_list_lock);
 719         cli->cl_avail_grant += grant;
 720         spin_unlock(&cli->cl_loi_list_lock);
 721 }
 722
 723 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
 724 {
 725         if (body->oa.o_valid & OBD_MD_FLGRANT) {
 726                 CDEBUG(D_CACHE, "got %llu extra grant\n", body->oa.o_grant);
 727                 __osc_update_grant(cli, body->oa.o_grant);
 728         }
 729 }
 730
 731 static int osc_shrink_grant_interpret(const struct lu_env *env,
 732                                       struct ptlrpc_request *req,
 733                                       void *aa, int rc)
 734 {
 735         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
 736         struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
 737         struct ost_body *body;
 738
 739         if (rc != 0) {
 740                 __osc_update_grant(cli, oa->o_grant);
 741                 GOTO(out, rc);
 742         }
 743
 744         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 745         LASSERT(body);
 746         osc_update_grant(cli, body);
 747 out:
 748         OBDO_FREE(oa);
 749         return rc;
 750 }
 751
 752 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
 753 {
 754         spin_lock(&cli->cl_loi_list_lock);
 755         oa->o_grant = cli->cl_avail_grant / 4;
 756         cli->cl_avail_grant -= oa->o_grant;
 757         spin_unlock(&cli->cl_loi_list_lock);
 758         if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
 759                 oa->o_valid |= OBD_MD_FLFLAGS;
 760                 oa->o_flags = 0;
 761         }
 762         oa->o_flags |= OBD_FL_SHRINK_GRANT;
 763         osc_update_next_shrink(cli);
 764 }
 765
 766 /* Shrink the current grant, either from some large amount to enough for a
 767  * full set of in-flight RPCs, or if we have already shrunk to that limit
 768  * then to enough for a single RPC.  This avoids keeping more grant than
 769  * needed, and avoids shrinking the grant piecemeal. */
 770 static int osc_shrink_grant(struct client_obd *cli)
 771 {
 772         __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
 773                              (cli->cl_max_pages_per_rpc << PAGE_SHIFT);
 774
 775         spin_lock(&cli->cl_loi_list_lock);
 776         if (cli->cl_avail_grant <= target_bytes)
 777                 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
 778         spin_unlock(&cli->cl_loi_list_lock);
 779
 780         return osc_shrink_grant_to_target(cli, target_bytes);
 781 }
 782
 783 int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
 784 {
 785         int                     rc = 0;
 786         struct ost_body        *body;
 787         ENTRY;
 788
 789         spin_lock(&cli->cl_loi_list_lock);
 790         /* Don't shrink if we are already above or below the desired limit
 791          * We don't want to shrink below a single RPC, as that will negatively
 792          * impact block allocation and long-term performance. */
 793         if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_SHIFT)
 794                 target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT;
 795
 796         if (target_bytes >= cli->cl_avail_grant) {
 797                 spin_unlock(&cli->cl_loi_list_lock);
 798                 RETURN(0);
 799         }
 800         spin_unlock(&cli->cl_loi_list_lock);
 801
 802         OBD_ALLOC_PTR(body);
 803         if (!body)
 804                 RETURN(-ENOMEM);
 805
 806         osc_announce_cached(cli, &body->oa, 0);
 807
 808         spin_lock(&cli->cl_loi_list_lock);
 809         body->oa.o_grant = cli->cl_avail_grant - target_bytes;
 810         cli->cl_avail_grant = target_bytes;
 811         spin_unlock(&cli->cl_loi_list_lock);
 812         if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
 813                 body->oa.o_valid |= OBD_MD_FLFLAGS;
 814                 body->oa.o_flags = 0;
 815         }
 816         body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
 817         osc_update_next_shrink(cli);
 818
 819         rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
 820                                 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
 821                                 sizeof(*body), body, NULL);
 822         if (rc != 0)
 823                 __osc_update_grant(cli, body->oa.o_grant);
 824         OBD_FREE_PTR(body);
 825         RETURN(rc);
 826 }
 827
 828 static int osc_should_shrink_grant(struct client_obd *client)
 829 {
 830         time64_t next_shrink = client->cl_next_shrink_grant;
 831
 832         if ((client->cl_import->imp_connect_data.ocd_connect_flags &
 833              OBD_CONNECT_GRANT_SHRINK) == 0)
 834                 return 0;
 835
 836         if (ktime_get_seconds() >= next_shrink - 5) {
 837                 /* Get the current RPC size directly, instead of going via:
 838                  * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
 839                  * Keep comment here so that it can be found by searching. */
 840                 int brw_size = client->cl_max_pages_per_rpc << PAGE_SHIFT;
 841
 842                 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
 843                     client->cl_avail_grant > brw_size)
 844                         return 1;
 845                 else
 846                         osc_update_next_shrink(client);
 847         }
 848         return 0;
 849 }
 850
 851 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
 852 {
 853         struct client_obd *client;
 854
 855         list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
 856                 if (osc_should_shrink_grant(client))
 857                         osc_shrink_grant(client);
 858         }
 859         return 0;
 860 }
 861
 862 static int osc_add_shrink_grant(struct client_obd *client)
 863 {
 864         int rc;
 865
 866         rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
 867                                        TIMEOUT_GRANT,
 868                                        osc_grant_shrink_grant_cb, NULL,
 869                                        &client->cl_grant_shrink_list);
 870         if (rc) {
 871                 CERROR("add grant client %s error %d\n", cli_name(client), rc);
 872                 return rc;
 873         }
 874         CDEBUG(D_CACHE, "add grant client %s\n", cli_name(client));
 875         osc_update_next_shrink(client);
 876         return 0;
 877 }
 878
 879 static int osc_del_shrink_grant(struct client_obd *client)
 880 {
 881         return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
 882                                          TIMEOUT_GRANT);
 883 }
 884
 885 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
 886 {
 887         /*
 888          * ocd_grant is the total grant amount we're expect to hold: if we've
 889          * been evicted, it's the new avail_grant amount, cl_dirty_pages will
 890          * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
 891          * dirty.
 892          *
 893          * race is tolerable here: if we're evicted, but imp_state already
 894          * left EVICTED state, then cl_dirty_pages must be 0 already.
 895          */
 896         spin_lock(&cli->cl_loi_list_lock);
 897         cli->cl_avail_grant = ocd->ocd_grant;
 898         if (cli->cl_import->imp_state != LUSTRE_IMP_EVICTED) {
 899                 cli->cl_avail_grant -= cli->cl_reserved_grant;
 900                 if (OCD_HAS_FLAG(ocd, GRANT_PARAM))
 901                         cli->cl_avail_grant -= cli->cl_dirty_grant;
 902                 else
 903                         cli->cl_avail_grant -=
 904                                         cli->cl_dirty_pages << PAGE_SHIFT;
 905         }
 906
 907         if (OCD_HAS_FLAG(ocd, GRANT_PARAM)) {
 908                 u64 size;
 909                 int chunk_mask;
 910
 911                 /* overhead for each extent insertion */
 912                 cli->cl_grant_extent_tax = ocd->ocd_grant_tax_kb << 10;
 913                 /* determine the appropriate chunk size used by osc_extent. */
 914                 cli->cl_chunkbits = max_t(int, PAGE_SHIFT,
 915                                           ocd->ocd_grant_blkbits);
 916                 /* max_pages_per_rpc must be chunk aligned */
 917                 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
 918                 cli->cl_max_pages_per_rpc = (cli->cl_max_pages_per_rpc +
 919                                              ~chunk_mask) & chunk_mask;
 920                 /* determine maximum extent size, in #pages */
 921                 size = (u64)ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits;
 922                 cli->cl_max_extent_pages = size >> PAGE_SHIFT;
 923                 if (cli->cl_max_extent_pages == 0)
 924                         cli->cl_max_extent_pages = 1;
 925         } else {
 926                 cli->cl_grant_extent_tax = 0;
 927                 cli->cl_chunkbits = PAGE_SHIFT;
 928                 cli->cl_max_extent_pages = DT_MAX_BRW_PAGES;
 929         }
 930         spin_unlock(&cli->cl_loi_list_lock);
 931
 932         CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
 933                 "chunk bits: %d cl_max_extent_pages: %d\n",
 934                 cli_name(cli),
 935                 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits,
 936                 cli->cl_max_extent_pages);
 937
 938         if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
 939             list_empty(&cli->cl_grant_shrink_list))
 940                 osc_add_shrink_grant(cli);
 941 }
 942 EXPORT_SYMBOL(osc_init_grant);
 943
 944 /* We assume that the reason this OSC got a short read is because it read
 945  * beyond the end of a stripe file; i.e. lustre is reading a sparse file
 946  * via the LOV, and it _knows_ it's reading inside the file, it's just that
 947  * this stripe never got written at or beyond this stripe offset yet. */
 948 static void handle_short_read(int nob_read, size_t page_count,
 949                               struct brw_page **pga)
 950 {
 951         char *ptr;
 952         int i = 0;
 953
 954         /* skip bytes read OK */
 955         while (nob_read > 0) {
 956                 LASSERT (page_count > 0);
 957
 958                 if (pga[i]->count > nob_read) {
 959                         /* EOF inside this page */
 960                         ptr = kmap(pga[i]->pg) +
 961                                 (pga[i]->off & ~PAGE_MASK);
 962                         memset(ptr + nob_read, 0, pga[i]->count - nob_read);
 963                         kunmap(pga[i]->pg);
 964                         page_count--;
 965                         i++;
 966                         break;
 967                 }
 968
 969                 nob_read -= pga[i]->count;
 970                 page_count--;
 971                 i++;
 972         }
 973
 974         /* zero remaining pages */
 975         while (page_count-- > 0) {
 976                 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~PAGE_MASK);
 977                 memset(ptr, 0, pga[i]->count);
 978                 kunmap(pga[i]->pg);
 979                 i++;
 980         }
 981 }
 982
 983 static int check_write_rcs(struct ptlrpc_request *req,
 984                            int requested_nob, int niocount,
 985                            size_t page_count, struct brw_page **pga)
 986 {
 987         int     i;
 988         __u32   *remote_rcs;
 989
 990         remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
 991                                                   sizeof(*remote_rcs) *
 992                                                   niocount);
 993         if (remote_rcs == NULL) {
 994                 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
 995                 return(-EPROTO);
 996         }
 997
 998         /* return error if any niobuf was in error */
 999         for (i = 0; i < niocount; i++) {
1000                 if ((int)remote_rcs[i] < 0)
1001                         return(remote_rcs[i]);
1002
1003                 if (remote_rcs[i] != 0) {
1004                         CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1005                                 i, remote_rcs[i], req);
1006                         return(-EPROTO);
1007                 }
1008         }
1009         if (req->rq_bulk != NULL &&
1010             req->rq_bulk->bd_nob_transferred != requested_nob) {
1011                 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1012                        req->rq_bulk->bd_nob_transferred, requested_nob);
1013                 return(-EPROTO);
1014         }
1015
1016         return (0);
1017 }
1018
1019 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1020 {
1021         if (p1->flag != p2->flag) {
1022                 unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE |
1023                                   OBD_BRW_SYNC       | OBD_BRW_ASYNC   |
1024                                   OBD_BRW_NOQUOTA    | OBD_BRW_SOFT_SYNC);
1025
1026                 /* warn if we try to combine flags that we don't know to be
1027                  * safe to combine */
1028                 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1029                         CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1030                               "report this at https://jira.hpdd.intel.com/\n",
1031                               p1->flag, p2->flag);
1032                 }
1033                 return 0;
1034         }
1035
1036         return (p1->off + p1->count == p2->off);
1037 }
1038
1039 static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
1040                                    size_t pg_count, struct brw_page **pga,
1041                                    int opc, obd_dif_csum_fn *fn,
1042                                    int sector_size,
1043                                    u32 *check_sum)
1044 {
1045         struct cfs_crypto_hash_desc *hdesc;
1046         /* Used Adler as the default checksum type on top of DIF tags */
1047         unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP);
1048         struct page *__page;
1049         unsigned char *buffer;
1050         __u16 *guard_start;
1051         unsigned int bufsize;
1052         int guard_number;
1053         int used_number = 0;
1054         int used;
1055         u32 cksum;
1056         int rc = 0;
1057         int i = 0;
1058
1059         LASSERT(pg_count > 0);
1060
1061         __page = alloc_page(GFP_KERNEL);
1062         if (__page == NULL)
1063                 return -ENOMEM;
1064
1065         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1066         if (IS_ERR(hdesc)) {
1067                 rc = PTR_ERR(hdesc);
1068                 CERROR("%s: unable to initialize checksum hash %s: rc = %d\n",
1069                        obd_name, cfs_crypto_hash_name(cfs_alg), rc);
1070                 GOTO(out, rc);
1071         }
1072
1073         buffer = kmap(__page);
1074         guard_start = (__u16 *)buffer;
1075         guard_number = PAGE_SIZE / sizeof(*guard_start);
1076         while (nob > 0 && pg_count > 0) {
1077                 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1078
1079                 /* corrupt the data before we compute the checksum, to
1080                  * simulate an OST->client data error */
1081                 if (unlikely(i == 0 && opc == OST_READ &&
1082                              OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE))) {
1083                         unsigned char *ptr = kmap(pga[i]->pg);
1084                         int off = pga[i]->off & ~PAGE_MASK;
1085
1086                         memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1087                         kunmap(pga[i]->pg);
1088                 }
1089
1090                 /*
1091                  * The left guard number should be able to hold checksums of a
1092                  * whole page
1093                  */
1094                 rc = obd_page_dif_generate_buffer(obd_name, pga[i]->pg, 0,
1095                                                   count,
1096                                                   guard_start + used_number,
1097                                                   guard_number - used_number,
1098                                                   &used, sector_size,
1099                                                   fn);
1100                 if (rc)
1101                         break;
1102
1103                 used_number += used;
1104                 if (used_number == guard_number) {
1105                         cfs_crypto_hash_update_page(hdesc, __page, 0,
1106                                 used_number * sizeof(*guard_start));
1107                         used_number = 0;
1108                 }
1109
1110                 nob -= pga[i]->count;
1111                 pg_count--;
1112                 i++;
1113         }
1114         kunmap(__page);
1115         if (rc)
1116                 GOTO(out, rc);
1117
1118         if (used_number != 0)
1119                 cfs_crypto_hash_update_page(hdesc, __page, 0,
1120                         used_number * sizeof(*guard_start));
1121
1122         bufsize = sizeof(cksum);
1123         cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1124
1125         /* For sending we only compute the wrong checksum instead
1126          * of corrupting the data so it is still correct on a redo */
1127         if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1128                 cksum++;
1129
1130         *check_sum = cksum;
1131 out:
1132         __free_page(__page);
1133         return rc;
1134 }
1135
1136 static int osc_checksum_bulk(int nob, size_t pg_count,
1137                              struct brw_page **pga, int opc,
1138                              enum cksum_types cksum_type,
1139                              u32 *cksum)
1140 {
1141         int                             i = 0;
1142         struct cfs_crypto_hash_desc     *hdesc;
1143         unsigned int                    bufsize;
1144         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1145
1146         LASSERT(pg_count > 0);
1147
1148         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1149         if (IS_ERR(hdesc)) {
1150                 CERROR("Unable to initialize checksum hash %s\n",
1151                        cfs_crypto_hash_name(cfs_alg));
1152                 return PTR_ERR(hdesc);
1153         }
1154
1155         while (nob > 0 && pg_count > 0) {
1156                 unsigned int count = pga[i]->count > nob ? nob : pga[i]->count;
1157
1158                 /* corrupt the data before we compute the checksum, to
1159                  * simulate an OST->client data error */
1160                 if (i == 0 && opc == OST_READ &&
1161                     OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1162                         unsigned char *ptr = kmap(pga[i]->pg);
1163                         int off = pga[i]->off & ~PAGE_MASK;
1164
1165                         memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
1166                         kunmap(pga[i]->pg);
1167                 }
1168                 cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
1169                                             pga[i]->off & ~PAGE_MASK,
1170                                             count);
1171                 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
1172                                (int)(pga[i]->off & ~PAGE_MASK));
1173
1174                 nob -= pga[i]->count;
1175                 pg_count--;
1176                 i++;
1177         }
1178
1179         bufsize = sizeof(*cksum);
1180         cfs_crypto_hash_final(hdesc, (unsigned char *)cksum, &bufsize);
1181
1182         /* For sending we only compute the wrong checksum instead
1183          * of corrupting the data so it is still correct on a redo */
1184         if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1185                 (*cksum)++;
1186
1187         return 0;
1188 }
1189
1190 static int osc_checksum_bulk_rw(const char *obd_name,
1191                                 enum cksum_types cksum_type,
1192                                 int nob, size_t pg_count,
1193                                 struct brw_page **pga, int opc,
1194                                 u32 *check_sum)
1195 {
1196         obd_dif_csum_fn *fn = NULL;
1197         int sector_size = 0;
1198         int rc;
1199
1200         ENTRY;
1201         obd_t10_cksum2dif(cksum_type, &fn, &sector_size);
1202
1203         if (fn)
1204                 rc = osc_checksum_bulk_t10pi(obd_name, nob, pg_count, pga,
1205                                              opc, fn, sector_size, check_sum);
1206         else
1207                 rc = osc_checksum_bulk(nob, pg_count, pga, opc, cksum_type,
1208                                        check_sum);
1209
1210         RETURN(rc);
1211 }
1212
1213 static int
1214 osc_brw_prep_request(int cmd, struct client_obd *cli, struct obdo *oa,
1215                      u32 page_count, struct brw_page **pga,
1216                      struct ptlrpc_request **reqp, int resend)
1217 {
1218         struct ptlrpc_request   *req;
1219         struct ptlrpc_bulk_desc *desc;
1220         struct ost_body         *body;
1221         struct obd_ioobj        *ioobj;
1222         struct niobuf_remote    *niobuf;
1223         int niocount, i, requested_nob, opc, rc, short_io_size = 0;
1224         struct osc_brw_async_args *aa;
1225         struct req_capsule      *pill;
1226         struct brw_page *pg_prev;
1227         void *short_io_buf;
1228         const char *obd_name = cli->cl_import->imp_obd->obd_name;
1229
1230         ENTRY;
1231         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1232                 RETURN(-ENOMEM); /* Recoverable */
1233         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1234                 RETURN(-EINVAL); /* Fatal */
1235
1236         if ((cmd & OBD_BRW_WRITE) != 0) {
1237                 opc = OST_WRITE;
1238                 req = ptlrpc_request_alloc_pool(cli->cl_import,
1239                                                 osc_rq_pool,
1240                                                 &RQF_OST_BRW_WRITE);
1241         } else {
1242                 opc = OST_READ;
1243                 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1244         }
1245         if (req == NULL)
1246                 RETURN(-ENOMEM);
1247
1248         for (niocount = i = 1; i < page_count; i++) {
1249                 if (!can_merge_pages(pga[i - 1], pga[i]))
1250                         niocount++;
1251         }
1252
1253         pill = &req->rq_pill;
1254         req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1255                              sizeof(*ioobj));
1256         req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1257                              niocount * sizeof(*niobuf));
1258
1259         for (i = 0; i < page_count; i++)
1260                 short_io_size += pga[i]->count;
1261
1262         /* Check if we can do a short io. */
1263         if (!(short_io_size <= cli->cl_short_io_bytes && niocount == 1 &&
1264             imp_connect_shortio(cli->cl_import)))
1265                 short_io_size = 0;
1266
1267         req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_CLIENT,
1268                              opc == OST_READ ? 0 : short_io_size);
1269         if (opc == OST_READ)
1270                 req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_SERVER,
1271                                      short_io_size);
1272
1273         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1274         if (rc) {
1275                 ptlrpc_request_free(req);
1276                 RETURN(rc);
1277         }
1278         osc_set_io_portal(req);
1279
1280         ptlrpc_at_set_req_timeout(req);
1281         /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1282          * retry logic */
1283         req->rq_no_retry_einprogress = 1;
1284
1285         if (short_io_size != 0) {
1286                 desc = NULL;
1287                 short_io_buf = NULL;
1288                 goto no_bulk;
1289         }
1290
1291         desc = ptlrpc_prep_bulk_imp(req, page_count,
1292                 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1293                 (opc == OST_WRITE ? PTLRPC_BULK_GET_SOURCE :
1294                         PTLRPC_BULK_PUT_SINK) |
1295                         PTLRPC_BULK_BUF_KIOV,
1296                 OST_BULK_PORTAL,
1297                 &ptlrpc_bulk_kiov_pin_ops);
1298
1299         if (desc == NULL)
1300                 GOTO(out, rc = -ENOMEM);
1301         /* NB request now owns desc and will free it when it gets freed */
1302 no_bulk:
1303         body = req_capsule_client_get(pill, &RMF_OST_BODY);
1304         ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1305         niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1306         LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1307
1308         lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
1309
1310         /* For READ and WRITE, we can't fill o_uid and o_gid using from_kuid()
1311          * and from_kgid(), because they are asynchronous. Fortunately, variable
1312          * oa contains valid o_uid and o_gid in these two operations.
1313          * Besides, filling o_uid and o_gid is enough for nrs-tbf, see LU-9658.
1314          * OBD_MD_FLUID and OBD_MD_FLUID is not set in order to avoid breaking
1315          * other process logic */
1316         body->oa.o_uid = oa->o_uid;
1317         body->oa.o_gid = oa->o_gid;
1318
1319         obdo_to_ioobj(oa, ioobj);
1320         ioobj->ioo_bufcnt = niocount;
1321         /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1322          * that might be send for this request.  The actual number is decided
1323          * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1324          * "max - 1" for old client compatibility sending "0", and also so the
1325          * the actual maximum is a power-of-two number, not one less. LU-1431 */
1326         if (desc != NULL)
1327                 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1328         else /* short io */
1329                 ioobj_max_brw_set(ioobj, 0);
1330
1331         if (short_io_size != 0) {
1332                 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1333                         body->oa.o_valid |= OBD_MD_FLFLAGS;
1334                         body->oa.o_flags = 0;
1335                 }
1336                 body->oa.o_flags |= OBD_FL_SHORT_IO;
1337                 CDEBUG(D_CACHE, "Using short io for data transfer, size = %d\n",
1338                        short_io_size);
1339                 if (opc == OST_WRITE) {
1340                         short_io_buf = req_capsule_client_get(pill,
1341                                                               &RMF_SHORT_IO);
1342                         LASSERT(short_io_buf != NULL);
1343                 }
1344         }
1345
1346         LASSERT(page_count > 0);
1347         pg_prev = pga[0];
1348         for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1349                 struct brw_page *pg = pga[i];
1350                 int poff = pg->off & ~PAGE_MASK;
1351
1352                 LASSERT(pg->count > 0);
1353                 /* make sure there is no gap in the middle of page array */
1354                 LASSERTF(page_count == 1 ||
1355                          (ergo(i == 0, poff + pg->count == PAGE_SIZE) &&
1356                           ergo(i > 0 && i < page_count - 1,
1357                                poff == 0 && pg->count == PAGE_SIZE)   &&
1358                           ergo(i == page_count - 1, poff == 0)),
1359                          "i: %d/%d pg: %p off: %llu, count: %u\n",
1360                          i, page_count, pg, pg->off, pg->count);
1361                 LASSERTF(i == 0 || pg->off > pg_prev->off,
1362                          "i %d p_c %u pg %p [pri %lu ind %lu] off %llu"
1363                          " prev_pg %p [pri %lu ind %lu] off %llu\n",
1364                          i, page_count,
1365                          pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1366                          pg_prev->pg, page_private(pg_prev->pg),
1367                          pg_prev->pg->index, pg_prev->off);
1368                 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1369                         (pg->flag & OBD_BRW_SRVLOCK));
1370                 if (short_io_size != 0 && opc == OST_WRITE) {
1371                         unsigned char *ptr = ll_kmap_atomic(pg->pg, KM_USER0);
1372
1373                         LASSERT(short_io_size >= requested_nob + pg->count);
1374                         memcpy(short_io_buf + requested_nob,
1375                                ptr + poff,
1376                                pg->count);
1377                         ll_kunmap_atomic(ptr, KM_USER0);
1378                 } else if (short_io_size == 0) {
1379                         desc->bd_frag_ops->add_kiov_frag(desc, pg->pg, poff,
1380                                                          pg->count);
1381                 }
1382                 requested_nob += pg->count;
1383
1384                 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1385                         niobuf--;
1386                         niobuf->rnb_len += pg->count;
1387                 } else {
1388                         niobuf->rnb_offset = pg->off;
1389                         niobuf->rnb_len    = pg->count;
1390                         niobuf->rnb_flags  = pg->flag;
1391                 }
1392                 pg_prev = pg;
1393         }
1394
1395         LASSERTF((void *)(niobuf - niocount) ==
1396                 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1397                 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1398                 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1399
1400         osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1401         if (resend) {
1402                 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1403                         body->oa.o_valid |= OBD_MD_FLFLAGS;
1404                         body->oa.o_flags = 0;
1405                 }
1406                 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1407         }
1408
1409         if (osc_should_shrink_grant(cli))
1410                 osc_shrink_grant_local(cli, &body->oa);
1411
1412         /* size[REQ_REC_OFF] still sizeof (*body) */
1413         if (opc == OST_WRITE) {
1414                 if (cli->cl_checksum &&
1415                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1416                         /* store cl_cksum_type in a local variable since
1417                          * it can be changed via lprocfs */
1418                         enum cksum_types cksum_type = cli->cl_cksum_type;
1419
1420                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1421                                 body->oa.o_flags = 0;
1422
1423                         body->oa.o_flags |= obd_cksum_type_pack(obd_name,
1424                                                                 cksum_type);
1425                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1426
1427                         rc = osc_checksum_bulk_rw(obd_name, cksum_type,
1428                                                   requested_nob, page_count,
1429                                                   pga, OST_WRITE,
1430                                                   &body->oa.o_cksum);
1431                         if (rc < 0) {
1432                                 CDEBUG(D_PAGE, "failed to checksum, rc = %d\n",
1433                                        rc);
1434                                 GOTO(out, rc);
1435                         }
1436                         CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1437                                body->oa.o_cksum);
1438
1439                         /* save this in 'oa', too, for later checking */
1440                         oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1441                         oa->o_flags |= obd_cksum_type_pack(obd_name,
1442                                                            cksum_type);
1443                 } else {
1444                         /* clear out the checksum flag, in case this is a
1445                          * resend but cl_checksum is no longer set. b=11238 */
1446                         oa->o_valid &= ~OBD_MD_FLCKSUM;
1447                 }
1448                 oa->o_cksum = body->oa.o_cksum;
1449                 /* 1 RC per niobuf */
1450                 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1451                                      sizeof(__u32) * niocount);
1452         } else {
1453                 if (cli->cl_checksum &&
1454                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1455                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1456                                 body->oa.o_flags = 0;
1457                         body->oa.o_flags |= obd_cksum_type_pack(obd_name,
1458                                 cli->cl_cksum_type);
1459                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1460                 }
1461
1462                 /* Client cksum has been already copied to wire obdo in previous
1463                  * lustre_set_wire_obdo(), and in the case a bulk-read is being
1464                  * resent due to cksum error, this will allow Server to
1465                  * check+dump pages on its side */
1466         }
1467         ptlrpc_request_set_replen(req);
1468
1469         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1470         aa = ptlrpc_req_async_args(req);
1471         aa->aa_oa = oa;
1472         aa->aa_requested_nob = requested_nob;
1473         aa->aa_nio_count = niocount;
1474         aa->aa_page_count = page_count;
1475         aa->aa_resends = 0;
1476         aa->aa_ppga = pga;
1477         aa->aa_cli = cli;
1478         INIT_LIST_HEAD(&aa->aa_oaps);
1479
1480         *reqp = req;
1481         niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1482         CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
1483                 req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
1484                 niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
1485         RETURN(0);
1486
1487  out:
1488         ptlrpc_req_finished(req);
1489         RETURN(rc);
1490 }
1491
1492 char dbgcksum_file_name[PATH_MAX];
1493
1494 static void dump_all_bulk_pages(struct obdo *oa, __u32 page_count,
1495                                 struct brw_page **pga, __u32 server_cksum,
1496                                 __u32 client_cksum)
1497 {
1498         struct file *filp;
1499         int rc, i;
1500         unsigned int len;
1501         char *buf;
1502
1503         /* will only keep dump of pages on first error for the same range in
1504          * file/fid, not during the resends/retries. */
1505         snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1506                  "%s-checksum_dump-osc-"DFID":[%llu-%llu]-%x-%x",
1507                  (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 ?
1508                   libcfs_debug_file_path_arr :
1509                   LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1510                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
1511                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1512                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1513                  pga[0]->off,
1514                  pga[page_count-1]->off + pga[page_count-1]->count - 1,
1515                  client_cksum, server_cksum);
1516         filp = filp_open(dbgcksum_file_name,
1517                          O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1518         if (IS_ERR(filp)) {
1519                 rc = PTR_ERR(filp);
1520                 if (rc == -EEXIST)
1521                         CDEBUG(D_INFO, "%s: can't open to dump pages with "
1522                                "checksum error: rc = %d\n", dbgcksum_file_name,
1523                                rc);
1524                 else
1525                         CERROR("%s: can't open to dump pages with checksum "
1526                                "error: rc = %d\n", dbgcksum_file_name, rc);
1527                 return;
1528         }
1529
1530         for (i = 0; i < page_count; i++) {
1531                 len = pga[i]->count;
1532                 buf = kmap(pga[i]->pg);
1533                 while (len != 0) {
1534                         rc = cfs_kernel_write(filp, buf, len, &filp->f_pos);
1535                         if (rc < 0) {
1536                                 CERROR("%s: wanted to write %u but got %d "
1537                                        "error\n", dbgcksum_file_name, len, rc);
1538                                 break;
1539                         }
1540                         len -= rc;
1541                         buf += rc;
1542                         CDEBUG(D_INFO, "%s: wrote %d bytes\n",
1543                                dbgcksum_file_name, rc);
1544                 }
1545                 kunmap(pga[i]->pg);
1546         }
1547
1548         rc = ll_vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1549         if (rc)
1550                 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1551         filp_close(filp, NULL);
1552         return;
1553 }
1554
1555 static int
1556 check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
1557                      __u32 client_cksum, __u32 server_cksum,
1558                      struct osc_brw_async_args *aa)
1559 {
1560         const char *obd_name = aa->aa_cli->cl_import->imp_obd->obd_name;
1561         enum cksum_types cksum_type;
1562         obd_dif_csum_fn *fn = NULL;
1563         int sector_size = 0;
1564         bool t10pi = false;
1565         __u32 new_cksum;
1566         char *msg;
1567         int rc;
1568
1569         if (server_cksum == client_cksum) {
1570                 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1571                 return 0;
1572         }
1573
1574         if (aa->aa_cli->cl_checksum_dump)
1575                 dump_all_bulk_pages(oa, aa->aa_page_count, aa->aa_ppga,
1576                                     server_cksum, client_cksum);
1577
1578         cksum_type = obd_cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1579                                            oa->o_flags : 0);
1580
1581         switch (cksum_type) {
1582         case OBD_CKSUM_T10IP512:
1583                 t10pi = true;
1584                 fn = obd_dif_ip_fn;
1585                 sector_size = 512;
1586                 break;
1587         case OBD_CKSUM_T10IP4K:
1588                 t10pi = true;
1589                 fn = obd_dif_ip_fn;
1590                 sector_size = 4096;
1591                 break;
1592         case OBD_CKSUM_T10CRC512:
1593                 t10pi = true;
1594                 fn = obd_dif_crc_fn;
1595                 sector_size = 512;
1596                 break;
1597         case OBD_CKSUM_T10CRC4K:
1598                 t10pi = true;
1599                 fn = obd_dif_crc_fn;
1600                 sector_size = 4096;
1601                 break;
1602         default:
1603                 break;
1604         }
1605
1606         if (t10pi)
1607                 rc = osc_checksum_bulk_t10pi(obd_name, aa->aa_requested_nob,
1608                                              aa->aa_page_count,
1609                                              aa->aa_ppga,
1610                                              OST_WRITE,
1611                                              fn,
1612                                              sector_size,
1613                                              &new_cksum);
1614         else
1615                 rc = osc_checksum_bulk(aa->aa_requested_nob, aa->aa_page_count,
1616                                        aa->aa_ppga, OST_WRITE, cksum_type,
1617                                        &new_cksum);
1618
1619         if (rc < 0)
1620                 msg = "failed to calculate the client write checksum";
1621         else if (cksum_type != obd_cksum_type_unpack(aa->aa_oa->o_flags))
1622                 msg = "the server did not use the checksum type specified in "
1623                       "the original request - likely a protocol problem";
1624         else if (new_cksum == server_cksum)
1625                 msg = "changed on the client after we checksummed it - "
1626                       "likely false positive due to mmap IO (bug 11742)";
1627         else if (new_cksum == client_cksum)
1628                 msg = "changed in transit before arrival at OST";
1629         else
1630                 msg = "changed in transit AND doesn't match the original - "
1631                       "likely false positive due to mmap IO (bug 11742)";
1632
1633         LCONSOLE_ERROR_MSG(0x132, "%s: BAD WRITE CHECKSUM: %s: from %s inode "
1634                            DFID " object "DOSTID" extent [%llu-%llu], original "
1635                            "client csum %x (type %x), server csum %x (type %x),"
1636                            " client csum now %x\n",
1637                            obd_name, msg, libcfs_nid2str(peer->nid),
1638                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1639                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1640                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1641                            POSTID(&oa->o_oi), aa->aa_ppga[0]->off,
1642                            aa->aa_ppga[aa->aa_page_count - 1]->off +
1643                                 aa->aa_ppga[aa->aa_page_count-1]->count - 1,
1644                            client_cksum,
1645                            obd_cksum_type_unpack(aa->aa_oa->o_flags),
1646                            server_cksum, cksum_type, new_cksum);
1647         return 1;
1648 }
1649
1650 /* Note rc enters this function as number of bytes transferred */
1651 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1652 {
1653         struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1654         struct client_obd *cli = aa->aa_cli;
1655         const char *obd_name = cli->cl_import->imp_obd->obd_name;
1656         const struct lnet_process_id *peer =
1657                 &req->rq_import->imp_connection->c_peer;
1658         struct ost_body *body;
1659         u32 client_cksum = 0;
1660         ENTRY;
1661
1662         if (rc < 0 && rc != -EDQUOT) {
1663                 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1664                 RETURN(rc);
1665         }
1666
1667         LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1668         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1669         if (body == NULL) {
1670                 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1671                 RETURN(-EPROTO);
1672         }
1673
1674         /* set/clear over quota flag for a uid/gid/projid */
1675         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1676             body->oa.o_valid & (OBD_MD_FLALLQUOTA)) {
1677                 unsigned qid[LL_MAXQUOTAS] = {
1678                                          body->oa.o_uid, body->oa.o_gid,
1679                                          body->oa.o_projid };
1680                 CDEBUG(D_QUOTA, "setdq for [%u %u %u] with valid %#llx, flags %x\n",
1681                        body->oa.o_uid, body->oa.o_gid, body->oa.o_projid,
1682                        body->oa.o_valid, body->oa.o_flags);
1683                        osc_quota_setdq(cli, qid, body->oa.o_valid,
1684                                        body->oa.o_flags);
1685         }
1686
1687         osc_update_grant(cli, body);
1688
1689         if (rc < 0)
1690                 RETURN(rc);
1691
1692         if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1693                 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1694
1695         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1696                 if (rc > 0) {
1697                         CERROR("Unexpected +ve rc %d\n", rc);
1698                         RETURN(-EPROTO);
1699                 }
1700
1701                 if (req->rq_bulk != NULL &&
1702                     sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1703                         RETURN(-EAGAIN);
1704
1705                 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1706                     check_write_checksum(&body->oa, peer, client_cksum,
1707                                          body->oa.o_cksum, aa))
1708                         RETURN(-EAGAIN);
1709
1710                 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1711                                      aa->aa_page_count, aa->aa_ppga);
1712                 GOTO(out, rc);
1713         }
1714
1715         /* The rest of this function executes only for OST_READs */
1716
1717         if (req->rq_bulk == NULL) {
1718                 rc = req_capsule_get_size(&req->rq_pill, &RMF_SHORT_IO,
1719                                           RCL_SERVER);
1720                 LASSERT(rc == req->rq_status);
1721         } else {
1722                 /* if unwrap_bulk failed, return -EAGAIN to retry */
1723                 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1724         }
1725         if (rc < 0)
1726                 GOTO(out, rc = -EAGAIN);
1727
1728         if (rc > aa->aa_requested_nob) {
1729                 CERROR("Unexpected rc %d (%d requested)\n", rc,
1730                        aa->aa_requested_nob);
1731                 RETURN(-EPROTO);
1732         }
1733
1734         if (req->rq_bulk != NULL && rc != req->rq_bulk->bd_nob_transferred) {
1735                 CERROR ("Unexpected rc %d (%d transferred)\n",
1736                         rc, req->rq_bulk->bd_nob_transferred);
1737                 return (-EPROTO);
1738         }
1739
1740         if (req->rq_bulk == NULL) {
1741                 /* short io */
1742                 int nob, pg_count, i = 0;
1743                 unsigned char *buf;
1744
1745                 CDEBUG(D_CACHE, "Using short io read, size %d\n", rc);
1746                 pg_count = aa->aa_page_count;
1747                 buf = req_capsule_server_sized_get(&req->rq_pill, &RMF_SHORT_IO,
1748                                                    rc);
1749                 nob = rc;
1750                 while (nob > 0 && pg_count > 0) {
1751                         unsigned char *ptr;
1752                         int count = aa->aa_ppga[i]->count > nob ?
1753                                     nob : aa->aa_ppga[i]->count;
1754
1755                         CDEBUG(D_CACHE, "page %p count %d\n",
1756                                aa->aa_ppga[i]->pg, count);
1757                         ptr = ll_kmap_atomic(aa->aa_ppga[i]->pg, KM_USER0);
1758                         memcpy(ptr + (aa->aa_ppga[i]->off & ~PAGE_MASK), buf,
1759                                count);
1760                         ll_kunmap_atomic((void *) ptr, KM_USER0);
1761
1762                         buf += count;
1763                         nob -= count;
1764                         i++;
1765                         pg_count--;
1766                 }
1767         }
1768
1769         if (rc < aa->aa_requested_nob)
1770                 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1771
1772         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1773                 static int cksum_counter;
1774                 u32        server_cksum = body->oa.o_cksum;
1775                 char      *via = "";
1776                 char      *router = "";
1777                 enum cksum_types cksum_type;
1778                 u32 o_flags = body->oa.o_valid & OBD_MD_FLFLAGS ?
1779                         body->oa.o_flags : 0;
1780
1781                 cksum_type = obd_cksum_type_unpack(o_flags);
1782                 rc = osc_checksum_bulk_rw(obd_name, cksum_type, rc,
1783                                           aa->aa_page_count, aa->aa_ppga,
1784                                           OST_READ, &client_cksum);
1785                 if (rc < 0)
1786                         GOTO(out, rc);
1787
1788                 if (req->rq_bulk != NULL &&
1789                     peer->nid != req->rq_bulk->bd_sender) {
1790                         via = " via ";
1791                         router = libcfs_nid2str(req->rq_bulk->bd_sender);
1792                 }
1793
1794                 if (server_cksum != client_cksum) {
1795                         struct ost_body *clbody;
1796                         u32 page_count = aa->aa_page_count;
1797
1798                         clbody = req_capsule_client_get(&req->rq_pill,
1799                                                         &RMF_OST_BODY);
1800                         if (cli->cl_checksum_dump)
1801                                 dump_all_bulk_pages(&clbody->oa, page_count,
1802                                                     aa->aa_ppga, server_cksum,
1803                                                     client_cksum);
1804
1805                         LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1806                                            "%s%s%s inode "DFID" object "DOSTID
1807                                            " extent [%llu-%llu], client %x, "
1808                                            "server %x, cksum_type %x\n",
1809                                            obd_name,
1810                                            libcfs_nid2str(peer->nid),
1811                                            via, router,
1812                                            clbody->oa.o_valid & OBD_MD_FLFID ?
1813                                                 clbody->oa.o_parent_seq : 0ULL,
1814                                            clbody->oa.o_valid & OBD_MD_FLFID ?
1815                                                 clbody->oa.o_parent_oid : 0,
1816                                            clbody->oa.o_valid & OBD_MD_FLFID ?
1817                                                 clbody->oa.o_parent_ver : 0,
1818                                            POSTID(&body->oa.o_oi),
1819                                            aa->aa_ppga[0]->off,
1820                                            aa->aa_ppga[page_count-1]->off +
1821                                            aa->aa_ppga[page_count-1]->count - 1,
1822                                            client_cksum, server_cksum,
1823                                            cksum_type);
1824                         cksum_counter = 0;
1825                         aa->aa_oa->o_cksum = client_cksum;
1826                         rc = -EAGAIN;
1827                 } else {
1828                         cksum_counter++;
1829                         CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1830                         rc = 0;
1831                 }
1832         } else if (unlikely(client_cksum)) {
1833                 static int cksum_missed;
1834
1835                 cksum_missed++;
1836                 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1837                         CERROR("Checksum %u requested from %s but not sent\n",
1838                                cksum_missed, libcfs_nid2str(peer->nid));
1839         } else {
1840                 rc = 0;
1841         }
1842 out:
1843         if (rc >= 0)
1844                 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1845                                      aa->aa_oa, &body->oa);
1846
1847         RETURN(rc);
1848 }
1849
1850 static int osc_brw_redo_request(struct ptlrpc_request *request,
1851                                 struct osc_brw_async_args *aa, int rc)
1852 {
1853         struct ptlrpc_request *new_req;
1854         struct osc_brw_async_args *new_aa;
1855         struct osc_async_page *oap;
1856         ENTRY;
1857
1858         DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1859                   "redo for recoverable error %d", rc);
1860
1861         rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1862                                 OST_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
1863                                   aa->aa_cli, aa->aa_oa, aa->aa_page_count,
1864                                   aa->aa_ppga, &new_req, 1);
1865         if (rc)
1866                 RETURN(rc);
1867
1868         list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1869                 if (oap->oap_request != NULL) {
1870                         LASSERTF(request == oap->oap_request,
1871                                  "request %p != oap_request %p\n",
1872                                  request, oap->oap_request);
1873                         if (oap->oap_interrupted) {
1874                                 ptlrpc_req_finished(new_req);
1875                                 RETURN(-EINTR);
1876                         }
1877                 }
1878         }
1879         /* New request takes over pga and oaps from old request.
1880          * Note that copying a list_head doesn't work, need to move it... */
1881         aa->aa_resends++;
1882         new_req->rq_interpret_reply = request->rq_interpret_reply;
1883         new_req->rq_async_args = request->rq_async_args;
1884         new_req->rq_commit_cb = request->rq_commit_cb;
1885         /* cap resend delay to the current request timeout, this is similar to
1886          * what ptlrpc does (see after_reply()) */
1887         if (aa->aa_resends > new_req->rq_timeout)
1888                 new_req->rq_sent = ktime_get_real_seconds() + new_req->rq_timeout;
1889         else
1890                 new_req->rq_sent = ktime_get_real_seconds() + aa->aa_resends;
1891         new_req->rq_generation_set = 1;
1892         new_req->rq_import_generation = request->rq_import_generation;
1893
1894         new_aa = ptlrpc_req_async_args(new_req);
1895
1896         INIT_LIST_HEAD(&new_aa->aa_oaps);
1897         list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1898         INIT_LIST_HEAD(&new_aa->aa_exts);
1899         list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1900         new_aa->aa_resends = aa->aa_resends;
1901
1902         list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1903                 if (oap->oap_request) {
1904                         ptlrpc_req_finished(oap->oap_request);
1905                         oap->oap_request = ptlrpc_request_addref(new_req);
1906                 }
1907         }
1908
1909         /* XXX: This code will run into problem if we're going to support
1910          * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1911          * and wait for all of them to be finished. We should inherit request
1912          * set from old request. */
1913         ptlrpcd_add_req(new_req);
1914
1915         DEBUG_REQ(D_INFO, new_req, "new request");
1916         RETURN(0);
1917 }
1918
1919 /*
1920  * ugh, we want disk allocation on the target to happen in offset order.  we'll
1921  * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1922  * fine for our small page arrays and doesn't require allocation.  its an
1923  * insertion sort that swaps elements that are strides apart, shrinking the
1924  * stride down until its '1' and the array is sorted.
1925  */
1926 static void sort_brw_pages(struct brw_page **array, int num)
1927 {
1928         int stride, i, j;
1929         struct brw_page *tmp;
1930
1931         if (num == 1)
1932                 return;
1933         for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1934                 ;
1935
1936         do {
1937                 stride /= 3;
1938                 for (i = stride ; i < num ; i++) {
1939                         tmp = array[i];
1940                         j = i;
1941                         while (j >= stride && array[j - stride]->off > tmp->off) {
1942                                 array[j] = array[j - stride];
1943                                 j -= stride;
1944                         }
1945                         array[j] = tmp;
1946                 }
1947         } while (stride > 1);
1948 }
1949
1950 static void osc_release_ppga(struct brw_page **ppga, size_t count)
1951 {
1952         LASSERT(ppga != NULL);
1953         OBD_FREE(ppga, sizeof(*ppga) * count);
1954 }
1955
1956 static int brw_interpret(const struct lu_env *env,
1957                          struct ptlrpc_request *req, void *data, int rc)
1958 {
1959         struct osc_brw_async_args *aa = data;
1960         struct osc_extent *ext;
1961         struct osc_extent *tmp;
1962         struct client_obd *cli = aa->aa_cli;
1963         unsigned long           transferred = 0;
1964         ENTRY;
1965
1966         rc = osc_brw_fini_request(req, rc);
1967         CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
1968         /* When server return -EINPROGRESS, client should always retry
1969          * regardless of the number of times the bulk was resent already. */
1970         if (osc_recoverable_error(rc) && !req->rq_no_delay) {
1971                 if (req->rq_import_generation !=
1972                     req->rq_import->imp_generation) {
1973                         CDEBUG(D_HA, "%s: resend cross eviction for object: "
1974                                ""DOSTID", rc = %d.\n",
1975                                req->rq_import->imp_obd->obd_name,
1976                                POSTID(&aa->aa_oa->o_oi), rc);
1977                 } else if (rc == -EINPROGRESS ||
1978                     client_should_resend(aa->aa_resends, aa->aa_cli)) {
1979                         rc = osc_brw_redo_request(req, aa, rc);
1980                 } else {
1981                         CERROR("%s: too many resent retries for object: "
1982                                "%llu:%llu, rc = %d.\n",
1983                                req->rq_import->imp_obd->obd_name,
1984                                POSTID(&aa->aa_oa->o_oi), rc);
1985                 }
1986
1987                 if (rc == 0)
1988                         RETURN(0);
1989                 else if (rc == -EAGAIN || rc == -EINPROGRESS)
1990                         rc = -EIO;
1991         }
1992
1993         if (rc == 0) {
1994                 struct obdo *oa = aa->aa_oa;
1995                 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1996                 unsigned long valid = 0;
1997                 struct cl_object *obj;
1998                 struct osc_async_page *last;
1999
2000                 last = brw_page2oap(aa->aa_ppga[aa->aa_page_count - 1]);
2001                 obj = osc2cl(last->oap_obj);
2002
2003                 cl_object_attr_lock(obj);
2004                 if (oa->o_valid & OBD_MD_FLBLOCKS) {
2005                         attr->cat_blocks = oa->o_blocks;
2006                         valid |= CAT_BLOCKS;
2007                 }
2008                 if (oa->o_valid & OBD_MD_FLMTIME) {
2009                         attr->cat_mtime = oa->o_mtime;
2010                         valid |= CAT_MTIME;
2011                 }
2012                 if (oa->o_valid & OBD_MD_FLATIME) {
2013                         attr->cat_atime = oa->o_atime;
2014                         valid |= CAT_ATIME;
2015                 }
2016                 if (oa->o_valid & OBD_MD_FLCTIME) {
2017                         attr->cat_ctime = oa->o_ctime;
2018                         valid |= CAT_CTIME;
2019                 }
2020
2021                 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
2022                         struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
2023                         loff_t last_off = last->oap_count + last->oap_obj_off +
2024                                 last->oap_page_off;
2025
2026                         /* Change file size if this is an out of quota or
2027                          * direct IO write and it extends the file size */
2028                         if (loi->loi_lvb.lvb_size < last_off) {
2029                                 attr->cat_size = last_off;
2030                                 valid |= CAT_SIZE;
2031                         }
2032                         /* Extend KMS if it's not a lockless write */
2033                         if (loi->loi_kms < last_off &&
2034                             oap2osc_page(last)->ops_srvlock == 0) {
2035                                 attr->cat_kms = last_off;
2036                                 valid |= CAT_KMS;
2037                         }
2038                 }
2039
2040                 if (valid != 0)
2041                         cl_object_attr_update(env, obj, attr, valid);
2042                 cl_object_attr_unlock(obj);
2043         }
2044         OBDO_FREE(aa->aa_oa);
2045
2046         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
2047                 osc_inc_unstable_pages(req);
2048
2049         list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
2050                 list_del_init(&ext->oe_link);
2051                 osc_extent_finish(env, ext, 1,
2052                                   rc && req->rq_no_delay ? -EWOULDBLOCK : rc);
2053         }
2054         LASSERT(list_empty(&aa->aa_exts));
2055         LASSERT(list_empty(&aa->aa_oaps));
2056
2057         transferred = (req->rq_bulk == NULL ? /* short io */
2058                        aa->aa_requested_nob :
2059                        req->rq_bulk->bd_nob_transferred);
2060
2061         osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
2062         ptlrpc_lprocfs_brw(req, transferred);
2063
2064         spin_lock(&cli->cl_loi_list_lock);
2065         /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2066          * is called so we know whether to go to sync BRWs or wait for more
2067          * RPCs to complete */
2068         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2069                 cli->cl_w_in_flight--;
2070         else
2071                 cli->cl_r_in_flight--;
2072         osc_wake_cache_waiters(cli);
2073         spin_unlock(&cli->cl_loi_list_lock);
2074
2075         osc_io_unplug(env, cli, NULL);
2076         RETURN(rc);
2077 }
2078
2079 static void brw_commit(struct ptlrpc_request *req)
2080 {
2081         /* If osc_inc_unstable_pages (via osc_extent_finish) races with
2082          * this called via the rq_commit_cb, I need to ensure
2083          * osc_dec_unstable_pages is still called. Otherwise unstable
2084          * pages may be leaked. */
2085         spin_lock(&req->rq_lock);
2086         if (likely(req->rq_unstable)) {
2087                 req->rq_unstable = 0;
2088                 spin_unlock(&req->rq_lock);
2089
2090                 osc_dec_unstable_pages(req);
2091         } else {
2092                 req->rq_committed = 1;
2093                 spin_unlock(&req->rq_lock);
2094         }
2095 }
2096
2097 /**
2098  * Build an RPC by the list of extent @ext_list. The caller must ensure
2099  * that the total pages in this list are NOT over max pages per RPC.
2100  * Extents in the list must be in OES_RPC state.
2101  */
2102 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
2103                   struct list_head *ext_list, int cmd)
2104 {
2105         struct ptlrpc_request           *req = NULL;
2106         struct osc_extent               *ext;
2107         struct brw_page                 **pga = NULL;
2108         struct osc_brw_async_args       *aa = NULL;
2109         struct obdo                     *oa = NULL;
2110         struct osc_async_page           *oap;
2111         struct osc_object               *obj = NULL;
2112         struct cl_req_attr              *crattr = NULL;
2113         loff_t                          starting_offset = OBD_OBJECT_EOF;
2114         loff_t                          ending_offset = 0;
2115         int                             mpflag = 0;
2116         int                             mem_tight = 0;
2117         int                             page_count = 0;
2118         bool                            soft_sync = false;
2119         bool                            interrupted = false;
2120         bool                            ndelay = false;
2121         int                             i;
2122         int                             grant = 0;
2123         int                             rc;
2124         __u32                           layout_version = 0;
2125         struct list_head                rpc_list = LIST_HEAD_INIT(rpc_list);
2126         struct ost_body                 *body;
2127         ENTRY;
2128         LASSERT(!list_empty(ext_list));
2129
2130         /* add pages into rpc_list to build BRW rpc */
2131         list_for_each_entry(ext, ext_list, oe_link) {
2132                 LASSERT(ext->oe_state == OES_RPC);
2133                 mem_tight |= ext->oe_memalloc;
2134                 grant += ext->oe_grants;
2135                 page_count += ext->oe_nr_pages;
2136                 layout_version = MAX(layout_version, ext->oe_layout_version);
2137                 if (obj == NULL)
2138                         obj = ext->oe_obj;
2139         }
2140
2141         soft_sync = osc_over_unstable_soft_limit(cli);
2142         if (mem_tight)
2143                 mpflag = cfs_memory_pressure_get_and_set();
2144
2145         OBD_ALLOC(pga, sizeof(*pga) * page_count);
2146         if (pga == NULL)
2147                 GOTO(out, rc = -ENOMEM);
2148
2149         OBDO_ALLOC(oa);
2150         if (oa == NULL)
2151                 GOTO(out, rc = -ENOMEM);
2152
2153         i = 0;
2154         list_for_each_entry(ext, ext_list, oe_link) {
2155                 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
2156                         if (mem_tight)
2157                                 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
2158                         if (soft_sync)
2159                                 oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
2160                         pga[i] = &oap->oap_brw_page;
2161                         pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2162                         i++;
2163
2164                         list_add_tail(&oap->oap_rpc_item, &rpc_list);
2165                         if (starting_offset == OBD_OBJECT_EOF ||
2166                             starting_offset > oap->oap_obj_off)
2167                                 starting_offset = oap->oap_obj_off;
2168                         else
2169                                 LASSERT(oap->oap_page_off == 0);
2170                         if (ending_offset < oap->oap_obj_off + oap->oap_count)
2171                                 ending_offset = oap->oap_obj_off +
2172                                                 oap->oap_count;
2173                         else
2174                                 LASSERT(oap->oap_page_off + oap->oap_count ==
2175                                         PAGE_SIZE);
2176                         if (oap->oap_interrupted)
2177                                 interrupted = true;
2178                 }
2179                 if (ext->oe_ndelay)
2180                         ndelay = true;
2181         }
2182
2183         /* first page in the list */
2184         oap = list_entry(rpc_list.next, typeof(*oap), oap_rpc_item);
2185
2186         crattr = &osc_env_info(env)->oti_req_attr;
2187         memset(crattr, 0, sizeof(*crattr));
2188         crattr->cra_type = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2189         crattr->cra_flags = ~0ULL;
2190         crattr->cra_page = oap2cl_page(oap);
2191         crattr->cra_oa = oa;
2192         cl_req_attr_set(env, osc2cl(obj), crattr);
2193
2194         if (cmd == OBD_BRW_WRITE) {
2195                 oa->o_grant_used = grant;
2196                 if (layout_version > 0) {
2197                         CDEBUG(D_LAYOUT, DFID": write with layout version %u\n",
2198                                PFID(&oa->o_oi.oi_fid), layout_version);
2199
2200                         oa->o_layout_version = layout_version;
2201                         oa->o_valid |= OBD_MD_LAYOUT_VERSION;
2202                 }
2203         }
2204
2205         sort_brw_pages(pga, page_count);
2206         rc = osc_brw_prep_request(cmd, cli, oa, page_count, pga, &req, 0);
2207         if (rc != 0) {
2208                 CERROR("prep_req failed: %d\n", rc);
2209                 GOTO(out, rc);
2210         }
2211
2212         req->rq_commit_cb = brw_commit;
2213         req->rq_interpret_reply = brw_interpret;
2214         req->rq_memalloc = mem_tight != 0;
2215         oap->oap_request = ptlrpc_request_addref(req);
2216         if (interrupted && !req->rq_intr)
2217                 ptlrpc_mark_interrupted(req);
2218         if (ndelay) {
2219                 req->rq_no_resend = req->rq_no_delay = 1;
2220                 /* probably set a shorter timeout value.
2221                  * to handle ETIMEDOUT in brw_interpret() correctly. */
2222                 /* lustre_msg_set_timeout(req, req->rq_timeout / 2); */
2223         }
2224
2225         /* Need to update the timestamps after the request is built in case
2226          * we race with setattr (locally or in queue at OST).  If OST gets
2227          * later setattr before earlier BRW (as determined by the request xid),
2228          * the OST will not use BRW timestamps.  Sadly, there is no obvious
2229          * way to do this in a single call.  bug 10150 */
2230         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2231         crattr->cra_oa = &body->oa;
2232         crattr->cra_flags = OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLATIME;
2233         cl_req_attr_set(env, osc2cl(obj), crattr);
2234         lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
2235
2236         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2237         aa = ptlrpc_req_async_args(req);
2238         INIT_LIST_HEAD(&aa->aa_oaps);
2239         list_splice_init(&rpc_list, &aa->aa_oaps);
2240         INIT_LIST_HEAD(&aa->aa_exts);
2241         list_splice_init(ext_list, &aa->aa_exts);
2242
2243         spin_lock(&cli->cl_loi_list_lock);
2244         starting_offset >>= PAGE_SHIFT;
2245         if (cmd == OBD_BRW_READ) {
2246                 cli->cl_r_in_flight++;
2247                 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2248                 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2249                 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2250                                       starting_offset + 1);
2251         } else {
2252                 cli->cl_w_in_flight++;
2253                 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2254                 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
2255                 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2256                                       starting_offset + 1);
2257         }
2258         spin_unlock(&cli->cl_loi_list_lock);
2259
2260         DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
2261                   page_count, aa, cli->cl_r_in_flight,
2262                   cli->cl_w_in_flight);
2263         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_IO, cfs_fail_val);
2264
2265         ptlrpcd_add_req(req);
2266         rc = 0;
2267         EXIT;
2268
2269 out:
2270         if (mem_tight != 0)
2271                 cfs_memory_pressure_restore(mpflag);
2272
2273         if (rc != 0) {
2274                 LASSERT(req == NULL);
2275
2276                 if (oa)
2277                         OBDO_FREE(oa);
2278                 if (pga)
2279                         OBD_FREE(pga, sizeof(*pga) * page_count);
2280                 /* this should happen rarely and is pretty bad, it makes the
2281                  * pending list not follow the dirty order */
2282                 while (!list_empty(ext_list)) {
2283                         ext = list_entry(ext_list->next, struct osc_extent,
2284                                          oe_link);
2285                         list_del_init(&ext->oe_link);
2286                         osc_extent_finish(env, ext, 0, rc);
2287                 }
2288         }
2289         RETURN(rc);
2290 }
2291
2292 static int osc_set_lock_data(struct ldlm_lock *lock, void *data)
2293 {
2294         int set = 0;
2295
2296         LASSERT(lock != NULL);
2297
2298         lock_res_and_lock(lock);
2299
2300         if (lock->l_ast_data == NULL)
2301                 lock->l_ast_data = data;
2302         if (lock->l_ast_data == data)
2303                 set = 1;
2304
2305         unlock_res_and_lock(lock);
2306
2307         return set;
2308 }
2309
2310 int osc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
2311                      void *cookie, struct lustre_handle *lockh,
2312                      enum ldlm_mode mode, __u64 *flags, bool speculative,
2313                      int errcode)
2314 {
2315         bool intent = *flags & LDLM_FL_HAS_INTENT;
2316         int rc;
2317         ENTRY;
2318
2319         /* The request was created before ldlm_cli_enqueue call. */
2320         if (intent && errcode == ELDLM_LOCK_ABORTED) {
2321                 struct ldlm_reply *rep;
2322
2323                 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
2324                 LASSERT(rep != NULL);
2325
2326                 rep->lock_policy_res1 =
2327                         ptlrpc_status_ntoh(rep->lock_policy_res1);
2328                 if (rep->lock_policy_res1)
2329                         errcode = rep->lock_policy_res1;
2330                 if (!speculative)
2331                         *flags |= LDLM_FL_LVB_READY;
2332         } else if (errcode == ELDLM_OK) {
2333                 *flags |= LDLM_FL_LVB_READY;
2334         }
2335
2336         /* Call the update callback. */
2337         rc = (*upcall)(cookie, lockh, errcode);
2338
2339         /* release the reference taken in ldlm_cli_enqueue() */
2340         if (errcode == ELDLM_LOCK_MATCHED)
2341                 errcode = ELDLM_OK;
2342         if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
2343                 ldlm_lock_decref(lockh, mode);
2344
2345         RETURN(rc);
2346 }
2347
2348 int osc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2349                           struct osc_enqueue_args *aa, int rc)
2350 {
2351         struct ldlm_lock *lock;
2352         struct lustre_handle *lockh = &aa->oa_lockh;
2353         enum ldlm_mode mode = aa->oa_mode;
2354         struct ost_lvb *lvb = aa->oa_lvb;
2355         __u32 lvb_len = sizeof(*lvb);
2356         __u64 flags = 0;
2357
2358         ENTRY;
2359
2360         /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2361          * be valid. */
2362         lock = ldlm_handle2lock(lockh);
2363         LASSERTF(lock != NULL,
2364                  "lockh %#llx, req %p, aa %p - client evicted?\n",
2365                  lockh->cookie, req, aa);
2366
2367         /* Take an additional reference so that a blocking AST that
2368          * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2369          * to arrive after an upcall has been executed by
2370          * osc_enqueue_fini(). */
2371         ldlm_lock_addref(lockh, mode);
2372
2373         /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
2374         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
2375
2376         /* Let CP AST to grant the lock first. */
2377         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2378
2379         if (aa->oa_speculative) {
2380                 LASSERT(aa->oa_lvb == NULL);
2381                 LASSERT(aa->oa_flags == NULL);
2382                 aa->oa_flags = &flags;
2383         }
2384
2385         /* Complete obtaining the lock procedure. */
2386         rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
2387                                    aa->oa_mode, aa->oa_flags, lvb, lvb_len,
2388                                    lockh, rc);
2389         /* Complete osc stuff. */
2390         rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
2391                               aa->oa_flags, aa->oa_speculative, rc);
2392
2393         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2394
2395         ldlm_lock_decref(lockh, mode);
2396         LDLM_LOCK_PUT(lock);
2397         RETURN(rc);
2398 }
2399
2400 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2401
2402 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2403  * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2404  * other synchronous requests, however keeping some locks and trying to obtain
2405  * others may take a considerable amount of time in a case of ost failure; and
2406  * when other sync requests do not get released lock from a client, the client
2407  * is evicted from the cluster -- such scenarious make the life difficult, so
2408  * release locks just after they are obtained. */
2409 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2410                      __u64 *flags, union ldlm_policy_data *policy,
2411                      struct ost_lvb *lvb, int kms_valid,
2412                      osc_enqueue_upcall_f upcall, void *cookie,
2413                      struct ldlm_enqueue_info *einfo,
2414                      struct ptlrpc_request_set *rqset, int async,
2415                      bool speculative)
2416 {
2417         struct obd_device *obd = exp->exp_obd;
2418         struct lustre_handle lockh = { 0 };
2419         struct ptlrpc_request *req = NULL;
2420         int intent = *flags & LDLM_FL_HAS_INTENT;
2421         __u64 match_flags = *flags;
2422         enum ldlm_mode mode;
2423         int rc;
2424         ENTRY;
2425
2426         /* Filesystem lock extents are extended to page boundaries so that
2427          * dealing with the page cache is a little smoother.  */
2428         policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2429         policy->l_extent.end |= ~PAGE_MASK;
2430
2431         /*
2432          * kms is not valid when either object is completely fresh (so that no
2433          * locks are cached), or object was evicted. In the latter case cached
2434          * lock cannot be used, because it would prime inode state with
2435          * potentially stale LVB.
2436          */
2437         if (!kms_valid)
2438                 goto no_match;
2439
2440         /* Next, search for already existing extent locks that will cover us */
2441         /* If we're trying to read, we also search for an existing PW lock.  The
2442          * VFS and page cache already protect us locally, so lots of readers/
2443          * writers can share a single PW lock.
2444          *
2445          * There are problems with conversion deadlocks, so instead of
2446          * converting a read lock to a write lock, we'll just enqueue a new
2447          * one.
2448          *
2449          * At some point we should cancel the read lock instead of making them
2450          * send us a blocking callback, but there are problems with canceling
2451          * locks out from other users right now, too. */
2452         mode = einfo->ei_mode;
2453         if (einfo->ei_mode == LCK_PR)
2454                 mode |= LCK_PW;
2455         /* Normal lock requests must wait for the LVB to be ready before
2456          * matching a lock; speculative lock requests do not need to,
2457          * because they will not actually use the lock. */
2458         if (!speculative)
2459                 match_flags |= LDLM_FL_LVB_READY;
2460         if (intent != 0)
2461                 match_flags |= LDLM_FL_BLOCK_GRANTED;
2462         mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
2463                                einfo->ei_type, policy, mode, &lockh, 0);
2464         if (mode) {
2465                 struct ldlm_lock *matched;
2466
2467                 if (*flags & LDLM_FL_TEST_LOCK)
2468                         RETURN(ELDLM_OK);
2469
2470                 matched = ldlm_handle2lock(&lockh);
2471                 if (speculative) {
2472                         /* This DLM lock request is speculative, and does not
2473                          * have an associated IO request. Therefore if there
2474                          * is already a DLM lock, it wll just inform the
2475                          * caller to cancel the request for this stripe.*/
2476                         lock_res_and_lock(matched);
2477                         if (ldlm_extent_equal(&policy->l_extent,
2478                             &matched->l_policy_data.l_extent))
2479                                 rc = -EEXIST;
2480                         else
2481                                 rc = -ECANCELED;
2482                         unlock_res_and_lock(matched);
2483
2484                         ldlm_lock_decref(&lockh, mode);
2485                         LDLM_LOCK_PUT(matched);
2486                         RETURN(rc);
2487                 } else if (osc_set_lock_data(matched, einfo->ei_cbdata)) {
2488                         *flags |= LDLM_FL_LVB_READY;
2489
2490                         /* We already have a lock, and it's referenced. */
2491                         (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
2492
2493                         ldlm_lock_decref(&lockh, mode);
2494                         LDLM_LOCK_PUT(matched);
2495                         RETURN(ELDLM_OK);
2496                 } else {
2497                         ldlm_lock_decref(&lockh, mode);
2498                         LDLM_LOCK_PUT(matched);
2499                 }
2500         }
2501
2502 no_match:
2503         if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
2504                 RETURN(-ENOLCK);
2505
2506         if (intent) {
2507                 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2508                                            &RQF_LDLM_ENQUEUE_LVB);
2509                 if (req == NULL)
2510                         RETURN(-ENOMEM);
2511
2512                 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
2513                 if (rc) {
2514                         ptlrpc_request_free(req);
2515                         RETURN(rc);
2516                 }
2517
2518                 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
2519                                      sizeof *lvb);
2520                 ptlrpc_request_set_replen(req);
2521         }
2522
2523         /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2524         *flags &= ~LDLM_FL_BLOCK_GRANTED;
2525
2526         rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2527                               sizeof(*lvb), LVB_T_OST, &lockh, async);
2528         if (async) {
2529                 if (!rc) {
2530                         struct osc_enqueue_args *aa;
2531                         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2532                         aa = ptlrpc_req_async_args(req);
2533                         aa->oa_exp         = exp;
2534                         aa->oa_mode        = einfo->ei_mode;
2535                         aa->oa_type        = einfo->ei_type;
2536                         lustre_handle_copy(&aa->oa_lockh, &lockh);
2537                         aa->oa_upcall      = upcall;
2538                         aa->oa_cookie      = cookie;
2539                         aa->oa_speculative = speculative;
2540                         if (!speculative) {
2541                                 aa->oa_flags  = flags;
2542                                 aa->oa_lvb    = lvb;
2543                         } else {
2544                                 /* speculative locks are essentially to enqueue
2545                                  * a DLM lock  in advance, so we don't care
2546                                  * about the result of the enqueue. */
2547                                 aa->oa_lvb    = NULL;
2548                                 aa->oa_flags  = NULL;
2549                         }
2550
2551                         req->rq_interpret_reply =
2552                                 (ptlrpc_interpterer_t)osc_enqueue_interpret;
2553                         if (rqset == PTLRPCD_SET)
2554                                 ptlrpcd_add_req(req);
2555                         else
2556                                 ptlrpc_set_add_req(rqset, req);
2557                 } else if (intent) {
2558                         ptlrpc_req_finished(req);
2559                 }
2560                 RETURN(rc);
2561         }
2562
2563         rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
2564                               flags, speculative, rc);
2565         if (intent)
2566                 ptlrpc_req_finished(req);
2567
2568         RETURN(rc);
2569 }
2570
2571 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2572                    enum ldlm_type type, union ldlm_policy_data *policy,
2573                    enum ldlm_mode mode, __u64 *flags, void *data,
2574                    struct lustre_handle *lockh, int unref)
2575 {
2576         struct obd_device *obd = exp->exp_obd;
2577         __u64 lflags = *flags;
2578         enum ldlm_mode rc;
2579         ENTRY;
2580
2581         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
2582                 RETURN(-EIO);
2583
2584         /* Filesystem lock extents are extended to page boundaries so that
2585          * dealing with the page cache is a little smoother */
2586         policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
2587         policy->l_extent.end |= ~PAGE_MASK;
2588
2589         /* Next, search for already existing extent locks that will cover us */
2590         /* If we're trying to read, we also search for an existing PW lock.  The
2591          * VFS and page cache already protect us locally, so lots of readers/
2592          * writers can share a single PW lock. */
2593         rc = mode;
2594         if (mode == LCK_PR)
2595                 rc |= LCK_PW;
2596         rc = ldlm_lock_match(obd->obd_namespace, lflags,
2597                              res_id, type, policy, rc, lockh, unref);
2598         if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
2599                 RETURN(rc);
2600
2601         if (data != NULL) {
2602                 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2603
2604                 LASSERT(lock != NULL);
2605                 if (!osc_set_lock_data(lock, data)) {
2606                         ldlm_lock_decref(lockh, rc);
2607                         rc = 0;
2608                 }
2609                 LDLM_LOCK_PUT(lock);
2610         }
2611         RETURN(rc);
2612 }
2613
2614 static int osc_statfs_interpret(const struct lu_env *env,
2615                                 struct ptlrpc_request *req,
2616                                 struct osc_async_args *aa, int rc)
2617 {
2618         struct obd_statfs *msfs;
2619         ENTRY;
2620
2621         if (rc == -EBADR)
2622                 /* The request has in fact never been sent
2623                  * due to issues at a higher level (LOV).
2624                  * Exit immediately since the caller is
2625                  * aware of the problem and takes care
2626                  * of the clean up */
2627                  RETURN(rc);
2628
2629         if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2630             (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2631                 GOTO(out, rc = 0);
2632
2633         if (rc != 0)
2634                 GOTO(out, rc);
2635
2636         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2637         if (msfs == NULL) {
2638                 GOTO(out, rc = -EPROTO);
2639         }
2640
2641         *aa->aa_oi->oi_osfs = *msfs;
2642 out:
2643         rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
2644         RETURN(rc);
2645 }
2646
2647 static int osc_statfs_async(struct obd_export *exp,
2648                             struct obd_info *oinfo, time64_t max_age,
2649                             struct ptlrpc_request_set *rqset)
2650 {
2651         struct obd_device     *obd = class_exp2obd(exp);
2652         struct ptlrpc_request *req;
2653         struct osc_async_args *aa;
2654         int rc;
2655         ENTRY;
2656
2657         /* We could possibly pass max_age in the request (as an absolute
2658          * timestamp or a "seconds.usec ago") so the target can avoid doing
2659          * extra calls into the filesystem if that isn't necessary (e.g.
2660          * during mount that would help a bit).  Having relative timestamps
2661          * is not so great if request processing is slow, while absolute
2662          * timestamps are not ideal because they need time synchronization. */
2663         req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2664         if (req == NULL)
2665                 RETURN(-ENOMEM);
2666
2667         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2668         if (rc) {
2669                 ptlrpc_request_free(req);
2670                 RETURN(rc);
2671         }
2672         ptlrpc_request_set_replen(req);
2673         req->rq_request_portal = OST_CREATE_PORTAL;
2674         ptlrpc_at_set_req_timeout(req);
2675
2676         if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2677                 /* procfs requests not want stat in wait for avoid deadlock */
2678                 req->rq_no_resend = 1;
2679                 req->rq_no_delay = 1;
2680         }
2681
2682         req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
2683         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2684         aa = ptlrpc_req_async_args(req);
2685         aa->aa_oi = oinfo;
2686
2687         ptlrpc_set_add_req(rqset, req);
2688         RETURN(0);
2689 }
2690
2691 static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2692                       struct obd_statfs *osfs, time64_t max_age, __u32 flags)
2693 {
2694         struct obd_device     *obd = class_exp2obd(exp);
2695         struct obd_statfs     *msfs;
2696         struct ptlrpc_request *req;
2697         struct obd_import     *imp = NULL;
2698         int rc;
2699         ENTRY;
2700
2701
2702         /*Since the request might also come from lprocfs, so we need
2703          *sync this with client_disconnect_export Bug15684*/
2704         down_read(&obd->u.cli.cl_sem);
2705         if (obd->u.cli.cl_import)
2706                 imp = class_import_get(obd->u.cli.cl_import);
2707         up_read(&obd->u.cli.cl_sem);
2708         if (!imp)
2709                 RETURN(-ENODEV);
2710
2711         /* We could possibly pass max_age in the request (as an absolute
2712          * timestamp or a "seconds.usec ago") so the target can avoid doing
2713          * extra calls into the filesystem if that isn't necessary (e.g.
2714          * during mount that would help a bit).  Having relative timestamps
2715          * is not so great if request processing is slow, while absolute
2716          * timestamps are not ideal because they need time synchronization. */
2717         req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2718
2719         class_import_put(imp);
2720
2721         if (req == NULL)
2722                 RETURN(-ENOMEM);
2723
2724         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2725         if (rc) {
2726                 ptlrpc_request_free(req);
2727                 RETURN(rc);
2728         }
2729         ptlrpc_request_set_replen(req);
2730         req->rq_request_portal = OST_CREATE_PORTAL;
2731         ptlrpc_at_set_req_timeout(req);
2732
2733         if (flags & OBD_STATFS_NODELAY) {
2734                 /* procfs requests not want stat in wait for avoid deadlock */
2735                 req->rq_no_resend = 1;
2736                 req->rq_no_delay = 1;
2737         }
2738
2739         rc = ptlrpc_queue_wait(req);
2740         if (rc)
2741                 GOTO(out, rc);
2742
2743         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2744         if (msfs == NULL)
2745                 GOTO(out, rc = -EPROTO);
2746
2747         *osfs = *msfs;
2748
2749         EXIT;
2750 out:
2751         ptlrpc_req_finished(req);
2752         return rc;
2753 }
2754
2755 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2756                          void *karg, void __user *uarg)
2757 {
2758         struct obd_device *obd = exp->exp_obd;
2759         struct obd_ioctl_data *data = karg;
2760         int err = 0;
2761         ENTRY;
2762
2763         if (!try_module_get(THIS_MODULE)) {
2764                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2765                        module_name(THIS_MODULE));
2766                 return -EINVAL;
2767         }
2768         switch (cmd) {
2769         case OBD_IOC_CLIENT_RECOVER:
2770                 err = ptlrpc_recover_import(obd->u.cli.cl_import,
2771                                             data->ioc_inlbuf1, 0);
2772                 if (err > 0)
2773                         err = 0;
2774                 GOTO(out, err);
2775         case IOC_OSC_SET_ACTIVE:
2776                 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
2777                                                data->ioc_offset);
2778                 GOTO(out, err);
2779         case OBD_IOC_PING_TARGET:
2780                 err = ptlrpc_obd_ping(obd);
2781                 GOTO(out, err);
2782         default:
2783                 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
2784                        cmd, current_comm());
2785                 GOTO(out, err = -ENOTTY);
2786         }
2787 out:
2788         module_put(THIS_MODULE);
2789         return err;
2790 }
2791
2792 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
2793                        u32 keylen, void *key, u32 vallen, void *val,
2794                        struct ptlrpc_request_set *set)
2795 {
2796         struct ptlrpc_request *req;
2797         struct obd_device     *obd = exp->exp_obd;
2798         struct obd_import     *imp = class_exp2cliimp(exp);
2799         char                  *tmp;
2800         int                    rc;
2801         ENTRY;
2802
2803         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
2804
2805         if (KEY_IS(KEY_CHECKSUM)) {
2806                 if (vallen != sizeof(int))
2807                         RETURN(-EINVAL);
2808                 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
2809                 RETURN(0);
2810         }
2811
2812         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2813                 sptlrpc_conf_client_adapt(obd);
2814                 RETURN(0);
2815         }
2816
2817         if (KEY_IS(KEY_FLUSH_CTX)) {
2818                 sptlrpc_import_flush_my_ctx(imp);
2819                 RETURN(0);
2820         }
2821
2822         if (KEY_IS(KEY_CACHE_SET)) {
2823                 struct client_obd *cli = &obd->u.cli;
2824
2825                 LASSERT(cli->cl_cache == NULL); /* only once */
2826                 cli->cl_cache = (struct cl_client_cache *)val;
2827                 cl_cache_incref(cli->cl_cache);
2828                 cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
2829
2830                 /* add this osc into entity list */
2831                 LASSERT(list_empty(&cli->cl_lru_osc));
2832                 spin_lock(&cli->cl_cache->ccc_lru_lock);
2833                 list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
2834                 spin_unlock(&cli->cl_cache->ccc_lru_lock);
2835
2836                 RETURN(0);
2837         }
2838
2839         if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
2840                 struct client_obd *cli = &obd->u.cli;
2841                 long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
2842                 long target = *(long *)val;
2843
2844                 nr = osc_lru_shrink(env, cli, min(nr, target), true);
2845                 *(long *)val -= nr;
2846                 RETURN(0);
2847         }
2848
2849         if (!set && !KEY_IS(KEY_GRANT_SHRINK))
2850                 RETURN(-EINVAL);
2851
2852         /* We pass all other commands directly to OST. Since nobody calls osc
2853            methods directly and everybody is supposed to go through LOV, we
2854            assume lov checked invalid values for us.
2855            The only recognised values so far are evict_by_nid and mds_conn.
2856            Even if something bad goes through, we'd get a -EINVAL from OST
2857            anyway. */
2858
2859         req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
2860                                                 &RQF_OST_SET_GRANT_INFO :
2861                                                 &RQF_OBD_SET_INFO);
2862         if (req == NULL)
2863                 RETURN(-ENOMEM);
2864
2865         req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2866                              RCL_CLIENT, keylen);
2867         if (!KEY_IS(KEY_GRANT_SHRINK))
2868                 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
2869                                      RCL_CLIENT, vallen);
2870         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
2871         if (rc) {
2872                 ptlrpc_request_free(req);
2873                 RETURN(rc);
2874         }
2875
2876         tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
2877         memcpy(tmp, key, keylen);
2878         tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
2879                                                         &RMF_OST_BODY :
2880                                                         &RMF_SETINFO_VAL);
2881         memcpy(tmp, val, vallen);
2882
2883         if (KEY_IS(KEY_GRANT_SHRINK)) {
2884                 struct osc_grant_args *aa;
2885                 struct obdo *oa;
2886
2887                 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2888                 aa = ptlrpc_req_async_args(req);
2889                 OBDO_ALLOC(oa);
2890                 if (!oa) {
2891                         ptlrpc_req_finished(req);
2892                         RETURN(-ENOMEM);
2893                 }
2894                 *oa = ((struct ost_body *)val)->oa;
2895                 aa->aa_oa = oa;
2896                 req->rq_interpret_reply = osc_shrink_grant_interpret;
2897         }
2898
2899         ptlrpc_request_set_replen(req);
2900         if (!KEY_IS(KEY_GRANT_SHRINK)) {
2901                 LASSERT(set != NULL);
2902                 ptlrpc_set_add_req(set, req);
2903                 ptlrpc_check_set(NULL, set);
2904         } else {
2905                 ptlrpcd_add_req(req);
2906         }
2907
2908         RETURN(0);
2909 }
2910 EXPORT_SYMBOL(osc_set_info_async);
2911
2912 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
2913                   struct obd_device *obd, struct obd_uuid *cluuid,
2914                   struct obd_connect_data *data, void *localdata)
2915 {
2916         struct client_obd *cli = &obd->u.cli;
2917
2918         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
2919                 long lost_grant;
2920                 long grant;
2921
2922                 spin_lock(&cli->cl_loi_list_lock);
2923                 grant = cli->cl_avail_grant + cli->cl_reserved_grant;
2924                 if (data->ocd_connect_flags & OBD_CONNECT_GRANT_PARAM)
2925                         grant += cli->cl_dirty_grant;
2926                 else
2927                         grant += cli->cl_dirty_pages << PAGE_SHIFT;
2928                 data->ocd_grant = grant ? : 2 * cli_brw_size(obd);
2929                 lost_grant = cli->cl_lost_grant;
2930                 cli->cl_lost_grant = 0;
2931                 spin_unlock(&cli->cl_loi_list_lock);
2932
2933                 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d"
2934                        " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
2935                        data->ocd_version, data->ocd_grant, lost_grant);
2936         }
2937
2938         RETURN(0);
2939 }
2940 EXPORT_SYMBOL(osc_reconnect);
2941
2942 int osc_disconnect(struct obd_export *exp)
2943 {
2944         struct obd_device *obd = class_exp2obd(exp);
2945         int rc;
2946
2947         rc = client_disconnect_export(exp);
2948         /**
2949          * Initially we put del_shrink_grant before disconnect_export, but it
2950          * causes the following problem if setup (connect) and cleanup
2951          * (disconnect) are tangled together.
2952          *      connect p1                     disconnect p2
2953          *   ptlrpc_connect_import
2954          *     ...............               class_manual_cleanup
2955          *                                     osc_disconnect
2956          *                                     del_shrink_grant
2957          *   ptlrpc_connect_interrupt
2958          *     init_grant_shrink
2959          *   add this client to shrink list
2960          *                                      cleanup_osc
2961          * Bang! pinger trigger the shrink.
2962          * So the osc should be disconnected from the shrink list, after we
2963          * are sure the import has been destroyed. BUG18662
2964          */
2965         if (obd->u.cli.cl_import == NULL)
2966                 osc_del_shrink_grant(&obd->u.cli);
2967         return rc;
2968 }
2969 EXPORT_SYMBOL(osc_disconnect);
2970
2971 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2972                                  struct hlist_node *hnode, void *arg)
2973 {
2974         struct lu_env *env = arg;
2975         struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2976         struct ldlm_lock *lock;
2977         struct osc_object *osc = NULL;
2978         ENTRY;
2979
2980         lock_res(res);
2981         list_for_each_entry(lock, &res->lr_granted, l_res_link) {
2982                 if (lock->l_ast_data != NULL && osc == NULL) {
2983                         osc = lock->l_ast_data;
2984                         cl_object_get(osc2cl(osc));
2985                 }
2986
2987                 /* clear LDLM_FL_CLEANED flag to make sure it will be canceled
2988                  * by the 2nd round of ldlm_namespace_clean() call in
2989                  * osc_import_event(). */
2990                 ldlm_clear_cleaned(lock);
2991         }
2992         unlock_res(res);
2993
2994         if (osc != NULL) {
2995                 osc_object_invalidate(env, osc);
2996                 cl_object_put(env, osc2cl(osc));
2997         }
2998
2999         RETURN(0);
3000 }
3001 EXPORT_SYMBOL(osc_ldlm_resource_invalidate);
3002
3003 static int osc_import_event(struct obd_device *obd,
3004                             struct obd_import *imp,
3005                             enum obd_import_event event)
3006 {
3007         struct client_obd *cli;
3008         int rc = 0;
3009
3010         ENTRY;
3011         LASSERT(imp->imp_obd == obd);
3012
3013         switch (event) {
3014         case IMP_EVENT_DISCON: {
3015                 cli = &obd->u.cli;
3016                 spin_lock(&cli->cl_loi_list_lock);
3017                 cli->cl_avail_grant = 0;
3018                 cli->cl_lost_grant = 0;
3019                 spin_unlock(&cli->cl_loi_list_lock);
3020                 break;
3021         }
3022         case IMP_EVENT_INACTIVE: {
3023                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
3024                 break;
3025         }
3026         case IMP_EVENT_INVALIDATE: {
3027                 struct ldlm_namespace *ns = obd->obd_namespace;
3028                 struct lu_env         *env;
3029                 __u16                  refcheck;
3030
3031                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3032
3033                 env = cl_env_get(&refcheck);
3034                 if (!IS_ERR(env)) {
3035                         osc_io_unplug(env, &obd->u.cli, NULL);
3036
3037                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
3038                                                  osc_ldlm_resource_invalidate,
3039                                                  env, 0);
3040                         cl_env_put(env, &refcheck);
3041
3042                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3043                 } else
3044                         rc = PTR_ERR(env);
3045                 break;
3046         }
3047         case IMP_EVENT_ACTIVE: {
3048                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
3049                 break;
3050         }
3051         case IMP_EVENT_OCD: {
3052                 struct obd_connect_data *ocd = &imp->imp_connect_data;
3053
3054                 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
3055                         osc_init_grant(&obd->u.cli, ocd);
3056
3057                 /* See bug 7198 */
3058                 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
3059                         imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
3060
3061                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
3062                 break;
3063         }
3064         case IMP_EVENT_DEACTIVATE: {
3065                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE);
3066                 break;
3067         }
3068         case IMP_EVENT_ACTIVATE: {
3069                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE);
3070                 break;
3071         }
3072         default:
3073                 CERROR("Unknown import event %d\n", event);
3074                 LBUG();
3075         }
3076         RETURN(rc);
3077 }
3078
3079 /**
3080  * Determine whether the lock can be canceled before replaying the lock
3081  * during recovery, see bug16774 for detailed information.
3082  *
3083  * \retval zero the lock can't be canceled
3084  * \retval other ok to cancel
3085  */
3086 static int osc_cancel_weight(struct ldlm_lock *lock)
3087 {
3088         /*
3089          * Cancel all unused and granted extent lock.
3090          */
3091         if (lock->l_resource->lr_type == LDLM_EXTENT &&
3092             lock->l_granted_mode == lock->l_req_mode &&
3093             osc_ldlm_weigh_ast(lock) == 0)
3094                 RETURN(1);
3095
3096         RETURN(0);
3097 }
3098
3099 static int brw_queue_work(const struct lu_env *env, void *data)
3100 {
3101         struct client_obd *cli = data;
3102
3103         CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
3104
3105         osc_io_unplug(env, cli, NULL);
3106         RETURN(0);
3107 }
3108
3109 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg)
3110 {
3111         struct client_obd *cli = &obd->u.cli;
3112         void *handler;
3113         int rc;
3114
3115         ENTRY;
3116
3117         rc = ptlrpcd_addref();
3118         if (rc)
3119                 RETURN(rc);
3120
3121         rc = client_obd_setup(obd, lcfg);
3122         if (rc)
3123                 GOTO(out_ptlrpcd, rc);
3124
3125
3126         handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
3127         if (IS_ERR(handler))
3128                 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
3129         cli->cl_writeback_work = handler;
3130
3131         handler = ptlrpcd_alloc_work(cli->cl_import, lru_queue_work, cli);
3132         if (IS_ERR(handler))
3133                 GOTO(out_ptlrpcd_work, rc = PTR_ERR(handler));
3134         cli->cl_lru_work = handler;
3135
3136         rc = osc_quota_setup(obd);
3137         if (rc)
3138                 GOTO(out_ptlrpcd_work, rc);
3139
3140         cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
3141
3142         INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
3143         RETURN(rc);
3144
3145 out_ptlrpcd_work:
3146         if (cli->cl_writeback_work != NULL) {
3147                 ptlrpcd_destroy_work(cli->cl_writeback_work);
3148                 cli->cl_writeback_work = NULL;
3149         }
3150         if (cli->cl_lru_work != NULL) {
3151                 ptlrpcd_destroy_work(cli->cl_lru_work);
3152                 cli->cl_lru_work = NULL;
3153         }
3154         client_obd_cleanup(obd);
3155 out_ptlrpcd:
3156         ptlrpcd_decref();
3157         RETURN(rc);
3158 }
3159 EXPORT_SYMBOL(osc_setup_common);
3160
3161 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
3162 {
3163         struct client_obd *cli = &obd->u.cli;
3164         int                adding;
3165         int                added;
3166         int                req_count;
3167         int                rc;
3168
3169         ENTRY;
3170
3171         rc = osc_setup_common(obd, lcfg);
3172         if (rc < 0)
3173                 RETURN(rc);
3174
3175         rc = osc_tunables_init(obd);
3176         if (rc)
3177                 RETURN(rc);
3178
3179         /*
3180          * We try to control the total number of requests with a upper limit
3181          * osc_reqpool_maxreqcount. There might be some race which will cause
3182          * over-limit allocation, but it is fine.
3183          */
3184         req_count = atomic_read(&osc_pool_req_count);
3185         if (req_count < osc_reqpool_maxreqcount) {
3186                 adding = cli->cl_max_rpcs_in_flight + 2;
3187                 if (req_count + adding > osc_reqpool_maxreqcount)
3188                         adding = osc_reqpool_maxreqcount - req_count;
3189
3190                 added = ptlrpc_add_rqs_to_pool(osc_rq_pool, adding);
3191                 atomic_add(added, &osc_pool_req_count);
3192         }
3193
3194         INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
3195         ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
3196
3197         spin_lock(&osc_shrink_lock);
3198         list_add_tail(&cli->cl_shrink_list, &osc_shrink_list);
3199         spin_unlock(&osc_shrink_lock);
3200         cli->cl_import->imp_idle_timeout = osc_idle_timeout;
3201
3202         RETURN(0);
3203 }
3204
3205 int osc_precleanup_common(struct obd_device *obd)
3206 {
3207         struct client_obd *cli = &obd->u.cli;
3208         ENTRY;
3209
3210         /* LU-464
3211          * for echo client, export may be on zombie list, wait for
3212          * zombie thread to cull it, because cli.cl_import will be
3213          * cleared in client_disconnect_export():
3214          *   class_export_destroy() -> obd_cleanup() ->
3215          *   echo_device_free() -> echo_client_cleanup() ->
3216          *   obd_disconnect() -> osc_disconnect() ->
3217          *   client_disconnect_export()
3218          */
3219         obd_zombie_barrier();
3220         if (cli->cl_writeback_work) {
3221                 ptlrpcd_destroy_work(cli->cl_writeback_work);
3222                 cli->cl_writeback_work = NULL;
3223         }
3224
3225         if (cli->cl_lru_work) {
3226                 ptlrpcd_destroy_work(cli->cl_lru_work);
3227                 cli->cl_lru_work = NULL;
3228         }
3229
3230         obd_cleanup_client_import(obd);
3231         RETURN(0);
3232 }
3233 EXPORT_SYMBOL(osc_precleanup_common);
3234
3235 static int osc_precleanup(struct obd_device *obd)
3236 {
3237         ENTRY;
3238
3239         osc_precleanup_common(obd);
3240
3241         ptlrpc_lprocfs_unregister_obd(obd);
3242         RETURN(0);
3243 }
3244
3245 int osc_cleanup_common(struct obd_device *obd)
3246 {
3247         struct client_obd *cli = &obd->u.cli;
3248         int rc;
3249
3250         ENTRY;
3251
3252         spin_lock(&osc_shrink_lock);
3253         list_del(&cli->cl_shrink_list);
3254         spin_unlock(&osc_shrink_lock);
3255
3256         /* lru cleanup */
3257         if (cli->cl_cache != NULL) {
3258                 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
3259                 spin_lock(&cli->cl_cache->ccc_lru_lock);
3260                 list_del_init(&cli->cl_lru_osc);
3261                 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3262                 cli->cl_lru_left = NULL;
3263                 cl_cache_decref(cli->cl_cache);
3264                 cli->cl_cache = NULL;
3265         }
3266
3267         /* free memory of osc quota cache */
3268         osc_quota_cleanup(obd);
3269
3270         rc = client_obd_cleanup(obd);
3271
3272         ptlrpcd_decref();
3273         RETURN(rc);
3274 }
3275 EXPORT_SYMBOL(osc_cleanup_common);
3276
3277 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
3278 {
3279         ssize_t count  = class_modify_config(lcfg, PARAM_OSC,
3280                                              &obd->obd_kset.kobj);
3281         return count > 0 ? 0 : count;
3282 }
3283
3284 static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
3285 {
3286         return osc_process_config_base(obd, buf);
3287 }
3288
3289 static struct obd_ops osc_obd_ops = {
3290         .o_owner                = THIS_MODULE,
3291         .o_setup                = osc_setup,
3292         .o_precleanup           = osc_precleanup,
3293         .o_cleanup              = osc_cleanup_common,
3294         .o_add_conn             = client_import_add_conn,
3295         .o_del_conn             = client_import_del_conn,
3296         .o_connect              = client_connect_import,
3297         .o_reconnect            = osc_reconnect,
3298         .o_disconnect           = osc_disconnect,
3299         .o_statfs               = osc_statfs,
3300         .o_statfs_async         = osc_statfs_async,
3301         .o_create               = osc_create,
3302         .o_destroy              = osc_destroy,
3303         .o_getattr              = osc_getattr,
3304         .o_setattr              = osc_setattr,
3305         .o_iocontrol            = osc_iocontrol,
3306         .o_set_info_async       = osc_set_info_async,
3307         .o_import_event         = osc_import_event,
3308         .o_process_config       = osc_process_config,
3309         .o_quotactl             = osc_quotactl,
3310 };
3311
3312 static struct shrinker *osc_cache_shrinker;
3313 struct list_head osc_shrink_list = LIST_HEAD_INIT(osc_shrink_list);
3314 DEFINE_SPINLOCK(osc_shrink_lock);
3315
3316 #ifndef HAVE_SHRINKER_COUNT
3317 static int osc_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
3318 {
3319         struct shrink_control scv = {
3320                 .nr_to_scan = shrink_param(sc, nr_to_scan),
3321                 .gfp_mask   = shrink_param(sc, gfp_mask)
3322         };
3323 #if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL)
3324         struct shrinker *shrinker = NULL;
3325 #endif
3326
3327         (void)osc_cache_shrink_scan(shrinker, &scv);
3328
3329         return osc_cache_shrink_count(shrinker, &scv);
3330 }
3331 #endif
3332
3333 static int __init osc_init(void)
3334 {
3335         bool enable_proc = true;
3336         struct obd_type *type;
3337         unsigned int reqpool_size;
3338         unsigned int reqsize;
3339         int rc;
3340         DEF_SHRINKER_VAR(osc_shvar, osc_cache_shrink,
3341                          osc_cache_shrink_count, osc_cache_shrink_scan);
3342         ENTRY;
3343
3344         /* print an address of _any_ initialized kernel symbol from this
3345          * module, to allow debugging with gdb that doesn't support data
3346          * symbols from modules.*/
3347         CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
3348
3349         rc = lu_kmem_init(osc_caches);
3350         if (rc)
3351                 RETURN(rc);
3352
3353         type = class_search_type(LUSTRE_OSP_NAME);
3354         if (type != NULL && type->typ_procsym != NULL)
3355                 enable_proc = false;
3356
3357         rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
3358                                  LUSTRE_OSC_NAME, &osc_device_type);
3359         if (rc)
3360                 GOTO(out_kmem, rc);
3361
3362         osc_cache_shrinker = set_shrinker(DEFAULT_SEEKS, &osc_shvar);
3363
3364         /* This is obviously too much memory, only prevent overflow here */
3365         if (osc_reqpool_mem_max >= 1 << 12 || osc_reqpool_mem_max == 0)
3366                 GOTO(out_type, rc = -EINVAL);
3367
3368         reqpool_size = osc_reqpool_mem_max << 20;
3369
3370         reqsize = 1;
3371         while (reqsize < OST_IO_MAXREQSIZE)
3372                 reqsize = reqsize << 1;
3373
3374         /*
3375          * We don't enlarge the request count in OSC pool according to
3376          * cl_max_rpcs_in_flight. The allocation from the pool will only be
3377          * tried after normal allocation failed. So a small OSC pool won't
3378          * cause much performance degression in most of cases.
3379          */
3380         osc_reqpool_maxreqcount = reqpool_size / reqsize;
3381
3382         atomic_set(&osc_pool_req_count, 0);
3383         osc_rq_pool = ptlrpc_init_rq_pool(0, OST_IO_MAXREQSIZE,
3384                                           ptlrpc_add_rqs_to_pool);
3385
3386         if (osc_rq_pool != NULL)
3387                 GOTO(out, rc);
3388         rc = -ENOMEM;
3389 out_type:
3390         class_unregister_type(LUSTRE_OSC_NAME);
3391 out_kmem:
3392         lu_kmem_fini(osc_caches);
3393 out:
3394         RETURN(rc);
3395 }
3396
3397 static void __exit osc_exit(void)
3398 {
3399         remove_shrinker(osc_cache_shrinker);
3400         class_unregister_type(LUSTRE_OSC_NAME);
3401         lu_kmem_fini(osc_caches);
3402         ptlrpc_free_rq_pool(osc_rq_pool);
3403 }
3404
3405 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3406 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3407 MODULE_VERSION(LUSTRE_VERSION_STRING);
3408 MODULE_LICENSE("GPL");
3409
3410 module_init(osc_init);
3411 module_exit(osc_exit);