lustre/mdc/mdc_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) 2001, 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_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #include <lustre_errno.h>
  46
  47 #include <cl_object.h>
  48 #include <llog_swab.h>
  49 #include <lprocfs_status.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_fid.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_kernelcomm.h>
  54 #include <lustre_lmv.h>
  55 #include <lustre_log.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_swab.h>
  58 #include <obd_class.h>
  59 #include <lustre_osc.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 static int mdc_cleanup(struct obd_device *obd);
  66
  67 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  68 {
  69         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  70         int rc;
  71
  72         /* obd_get_request_slot() ensures that this client has no more
  73          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  74          * against an MDT. */
  75         rc = obd_get_request_slot(cli);
  76         if (rc != 0)
  77                 return rc;
  78
  79         rc = ptlrpc_queue_wait(req);
  80         obd_put_request_slot(cli);
  81
  82         return rc;
  83 }
  84
  85 /*
  86  * Send MDS_GET_ROOT RPC to fetch root FID.
  87  *
  88  * If \a fileset is not NULL it should contain a subdirectory off
  89  * the ROOT/ directory to be mounted on the client. Return the FID
  90  * of the subdirectory to the client to mount onto its mountpoint.
  91  *
  92  * \param[in]   imp     MDC import
  93  * \param[in]   fileset fileset name, which could be NULL
  94  * \param[out]  rootfid root FID of this mountpoint
  95  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  96  *
  97  * \retval      0 on success, negative errno on failure
  98  */
  99 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 100                          struct lu_fid *rootfid)
 101 {
 102         struct ptlrpc_request   *req;
 103         struct mdt_body         *body;
 104         int                      rc;
 105
 106         ENTRY;
 107
 108         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 109                 RETURN(-ENOTSUPP);
 110
 111         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 112                                 &RQF_MDS_GET_ROOT);
 113         if (req == NULL)
 114                 RETURN(-ENOMEM);
 115
 116         if (fileset != NULL)
 117                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 118                                      strlen(fileset) + 1);
 119         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 120         if (rc) {
 121                 ptlrpc_request_free(req);
 122                 RETURN(rc);
 123         }
 124         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 125         if (fileset != NULL) {
 126                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 127
 128                 memcpy(name, fileset, strlen(fileset));
 129         }
 130         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 131         req->rq_send_state = LUSTRE_IMP_FULL;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         *rootfid = body->mbo_fid1;
 144         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 145                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 146         EXIT;
 147 out:
 148         ptlrpc_req_finished(req);
 149
 150         return rc;
 151 }
 152
 153 /*
 154  * This function now is known to always saying that it will receive 4 buffers
 155  * from server. Even for cases when acl_size and md_size is zero, RPC header
 156  * will contain 4 fields and RPC itself will contain zero size fields. This is
 157  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 158  * and thus zero, it shrinks it, making zero size. The same story about
 159  * md_size. And this is course of problem when client waits for smaller number
 160  * of fields. This issue will be fixed later when client gets aware of RPC
 161  * layouts.  --umka
 162  */
 163 static int mdc_getattr_common(struct obd_export *exp,
 164                               struct ptlrpc_request *req)
 165 {
 166         struct req_capsule *pill = &req->rq_pill;
 167         struct mdt_body    *body;
 168         void               *eadata;
 169         int                 rc;
 170         ENTRY;
 171
 172         /* Request message already built. */
 173         rc = ptlrpc_queue_wait(req);
 174         if (rc != 0)
 175                 RETURN(rc);
 176
 177         /* sanity check for the reply */
 178         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 179         if (body == NULL)
 180                 RETURN(-EPROTO);
 181
 182         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 183
 184         mdc_update_max_ea_from_body(exp, body);
 185         if (body->mbo_eadatasize != 0) {
 186                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 187                                                       body->mbo_eadatasize);
 188                 if (eadata == NULL)
 189                         RETURN(-EPROTO);
 190         }
 191
 192         RETURN(0);
 193 }
 194
 195 static void mdc_reset_acl_req(struct ptlrpc_request *req)
 196 {
 197         spin_lock(&req->rq_early_free_lock);
 198         sptlrpc_cli_free_repbuf(req);
 199         req->rq_repbuf = NULL;
 200         req->rq_repbuf_len = 0;
 201         req->rq_repdata = NULL;
 202         req->rq_reqdata_len = 0;
 203         spin_unlock(&req->rq_early_free_lock);
 204 }
 205
 206 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 207                        struct ptlrpc_request **request)
 208 {
 209         struct ptlrpc_request *req;
 210         struct obd_import *imp = class_exp2cliimp(exp);
 211         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 212         int rc;
 213         ENTRY;
 214
 215         /* Single MDS without an LMV case */
 216         if (op_data->op_flags & MF_GET_MDT_IDX) {
 217                 op_data->op_mds = 0;
 218                 RETURN(0);
 219         }
 220
 221         *request = NULL;
 222         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 223         if (req == NULL)
 224                 RETURN(-ENOMEM);
 225
 226         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 227         if (rc) {
 228                 ptlrpc_request_free(req);
 229                 RETURN(rc);
 230         }
 231
 232 again:
 233         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 234                       op_data->op_mode, -1, 0);
 235         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 236         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 237                              op_data->op_mode);
 238         ptlrpc_request_set_replen(req);
 239
 240         rc = mdc_getattr_common(exp, req);
 241         if (rc) {
 242                 if (rc == -ERANGE &&
 243                     acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
 244                         acl_bufsize = imp->imp_connect_data.ocd_max_easize;
 245                         mdc_reset_acl_req(req);
 246                         goto again;
 247                 }
 248
 249                 ptlrpc_req_finished(req);
 250         } else {
 251                 *request = req;
 252         }
 253
 254         RETURN(rc);
 255 }
 256
 257 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 258                             struct ptlrpc_request **request)
 259 {
 260         struct ptlrpc_request *req;
 261         struct obd_import *imp = class_exp2cliimp(exp);
 262         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 263         int rc;
 264         ENTRY;
 265
 266         *request = NULL;
 267         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 268         if (req == NULL)
 269                 RETURN(-ENOMEM);
 270
 271         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 272                              op_data->op_namelen + 1);
 273
 274         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 275         if (rc) {
 276                 ptlrpc_request_free(req);
 277                 RETURN(rc);
 278         }
 279
 280         if (op_data->op_name) {
 281                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 282                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 283                                 op_data->op_namelen);
 284                 memcpy(name, op_data->op_name, op_data->op_namelen);
 285         }
 286
 287 again:
 288         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 289                       op_data->op_mode, op_data->op_suppgids[0], 0);
 290         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 291                              op_data->op_mode);
 292         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 293         ptlrpc_request_set_replen(req);
 294
 295         rc = mdc_getattr_common(exp, req);
 296         if (rc) {
 297                 if (rc == -ERANGE &&
 298                     acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
 299                         acl_bufsize = imp->imp_connect_data.ocd_max_easize;
 300                         mdc_reset_acl_req(req);
 301                         goto again;
 302                 }
 303
 304                 ptlrpc_req_finished(req);
 305         } else {
 306                 *request = req;
 307         }
 308
 309         RETURN(rc);
 310 }
 311
 312 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 313                             const struct lu_fid *fid, int opcode, u64 valid,
 314                             const char *xattr_name, const char *input,
 315                             int input_size, int output_size, int flags,
 316                             __u32 suppgid, struct ptlrpc_request **request)
 317 {
 318         struct ptlrpc_request *req;
 319         int   xattr_namelen = 0;
 320         char *tmp;
 321         int   rc;
 322         ENTRY;
 323
 324         *request = NULL;
 325         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 326         if (req == NULL)
 327                 RETURN(-ENOMEM);
 328
 329         if (xattr_name) {
 330                 xattr_namelen = strlen(xattr_name) + 1;
 331                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 332                                      xattr_namelen);
 333         }
 334         if (input_size) {
 335                 LASSERT(input);
 336                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 337                                      input_size);
 338         }
 339
 340         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 341         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 342             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 343                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 344                 int count;
 345
 346                 /* Without that packing would fail */
 347                 if (input_size == 0)
 348                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 349                                              RCL_CLIENT, 0);
 350
 351                 count = mdc_resource_get_unused(exp, fid,
 352                                                 &cancels, LCK_EX,
 353                                                 MDS_INODELOCK_XATTR);
 354
 355                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 356                 if (rc) {
 357                         ptlrpc_request_free(req);
 358                         RETURN(rc);
 359                 }
 360         } else {
 361                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 362                 if (rc) {
 363                         ptlrpc_request_free(req);
 364                         RETURN(rc);
 365                 }
 366         }
 367
 368         if (opcode == MDS_REINT) {
 369                 struct mdt_rec_setxattr *rec;
 370
 371                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 372                          sizeof(struct mdt_rec_reint));
 373                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 374                 rec->sx_opcode = REINT_SETXATTR;
 375                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 376                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 377                 rec->sx_cap    = cfs_curproc_cap_pack();
 378                 rec->sx_suppgid1 = suppgid;
 379                 rec->sx_suppgid2 = -1;
 380                 rec->sx_fid    = *fid;
 381                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 382                 rec->sx_time   = ktime_get_real_seconds();
 383                 rec->sx_size   = output_size;
 384                 rec->sx_flags  = flags;
 385         } else {
 386                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 387         }
 388
 389         if (xattr_name) {
 390                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 391                 memcpy(tmp, xattr_name, xattr_namelen);
 392         }
 393         if (input_size) {
 394                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 395                 memcpy(tmp, input, input_size);
 396         }
 397
 398         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 399                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 400                                      RCL_SERVER, output_size);
 401         ptlrpc_request_set_replen(req);
 402
 403         /* make rpc */
 404         if (opcode == MDS_REINT)
 405                 mdc_get_mod_rpc_slot(req, NULL);
 406
 407         rc = ptlrpc_queue_wait(req);
 408
 409         if (opcode == MDS_REINT)
 410                 mdc_put_mod_rpc_slot(req, NULL);
 411
 412         if (rc)
 413                 ptlrpc_req_finished(req);
 414         else
 415                 *request = req;
 416         RETURN(rc);
 417 }
 418
 419 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 420                         u64 valid, const char *xattr_name,
 421                         const char *input, int input_size, int output_size,
 422                         int flags, __u32 suppgid,
 423                         struct ptlrpc_request **request)
 424 {
 425         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 426                                 fid, MDS_REINT, valid, xattr_name,
 427                                 input, input_size, output_size, flags,
 428                                 suppgid, request);
 429 }
 430
 431 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 432                         u64 valid, const char *xattr_name,
 433                         const char *input, int input_size, int output_size,
 434                         int flags, struct ptlrpc_request **request)
 435 {
 436         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 437                                 fid, MDS_GETXATTR, valid, xattr_name,
 438                                 input, input_size, output_size, flags,
 439                                 -1, request);
 440 }
 441
 442 #ifdef CONFIG_FS_POSIX_ACL
 443 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 444 {
 445         struct req_capsule     *pill = &req->rq_pill;
 446         struct mdt_body        *body = md->body;
 447         struct posix_acl       *acl;
 448         void                   *buf;
 449         int                     rc;
 450         ENTRY;
 451
 452         if (!body->mbo_aclsize)
 453                 RETURN(0);
 454
 455         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 456
 457         if (!buf)
 458                 RETURN(-EPROTO);
 459
 460         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 461         if (acl == NULL)
 462                 RETURN(0);
 463         if (IS_ERR(acl)) {
 464                 rc = PTR_ERR(acl);
 465                 CERROR("convert xattr to acl: %d\n", rc);
 466                 RETURN(rc);
 467         }
 468
 469         rc = posix_acl_valid(&init_user_ns, acl);
 470         if (rc) {
 471                 CERROR("validate acl: %d\n", rc);
 472                 posix_acl_release(acl);
 473                 RETURN(rc);
 474         }
 475
 476         md->posix_acl = acl;
 477         RETURN(0);
 478 }
 479 #else
 480 #define mdc_unpack_acl(req, md) 0
 481 #endif
 482
 483 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 484                       struct obd_export *dt_exp, struct obd_export *md_exp,
 485                       struct lustre_md *md)
 486 {
 487         struct req_capsule *pill = &req->rq_pill;
 488         int rc;
 489         ENTRY;
 490
 491         LASSERT(md);
 492         memset(md, 0, sizeof(*md));
 493
 494         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 495         LASSERT(md->body != NULL);
 496
 497         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 498                 if (!S_ISREG(md->body->mbo_mode)) {
 499                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 500                                "regular file, but is not\n");
 501                         GOTO(out, rc = -EPROTO);
 502                 }
 503
 504                 if (md->body->mbo_eadatasize == 0) {
 505                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 506                                "but eadatasize 0\n");
 507                         GOTO(out, rc = -EPROTO);
 508                 }
 509
 510                 md->layout.lb_len = md->body->mbo_eadatasize;
 511                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 512                                                         &RMF_MDT_MD,
 513                                                         md->layout.lb_len);
 514                 if (md->layout.lb_buf == NULL)
 515                         GOTO(out, rc = -EPROTO);
 516         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 517                 const union lmv_mds_md *lmv;
 518                 size_t lmv_size;
 519
 520                 if (!S_ISDIR(md->body->mbo_mode)) {
 521                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 522                                "directory, but is not\n");
 523                         GOTO(out, rc = -EPROTO);
 524                 }
 525
 526                 lmv_size = md->body->mbo_eadatasize;
 527                 if (lmv_size == 0) {
 528                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 529                                "but eadatasize 0\n");
 530                         RETURN(-EPROTO);
 531                 }
 532
 533                 if (md->body->mbo_valid & OBD_MD_MEA) {
 534                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 535                                                            lmv_size);
 536                         if (lmv == NULL)
 537                                 GOTO(out, rc = -EPROTO);
 538
 539                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 540                         if (rc < 0)
 541                                 GOTO(out, rc);
 542
 543                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 544                                 CDEBUG(D_INFO, "size too small:  "
 545                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 546                                         rc, (int)sizeof(*md->lmv));
 547                                 GOTO(out, rc = -EPROTO);
 548                         }
 549                 }
 550         }
 551         rc = 0;
 552
 553         if (md->body->mbo_valid & OBD_MD_FLACL) {
 554                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 555                  * only when aclsize != 0 there's an actual segment for ACL
 556                  * in reply buffer.
 557                  */
 558                 if (md->body->mbo_aclsize) {
 559                         rc = mdc_unpack_acl(req, md);
 560                         if (rc)
 561                                 GOTO(out, rc);
 562 #ifdef CONFIG_FS_POSIX_ACL
 563                 } else {
 564                         md->posix_acl = NULL;
 565 #endif
 566                 }
 567         }
 568
 569         EXIT;
 570 out:
 571         if (rc) {
 572 #ifdef CONFIG_FS_POSIX_ACL
 573                 posix_acl_release(md->posix_acl);
 574 #endif
 575         }
 576         return rc;
 577 }
 578
 579 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 580 {
 581         ENTRY;
 582         RETURN(0);
 583 }
 584
 585 void mdc_replay_open(struct ptlrpc_request *req)
 586 {
 587         struct md_open_data *mod = req->rq_cb_data;
 588         struct ptlrpc_request *close_req;
 589         struct obd_client_handle *och;
 590         struct lustre_handle old;
 591         struct mdt_body *body;
 592         ENTRY;
 593
 594         if (mod == NULL) {
 595                 DEBUG_REQ(D_ERROR, req,
 596                           "Can't properly replay without open data.");
 597                 EXIT;
 598                 return;
 599         }
 600
 601         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 602         LASSERT(body != NULL);
 603
 604         spin_lock(&req->rq_lock);
 605         och = mod->mod_och;
 606         if (och && och->och_fh.cookie)
 607                 req->rq_early_free_repbuf = 1;
 608         else
 609                 req->rq_early_free_repbuf = 0;
 610         spin_unlock(&req->rq_lock);
 611
 612         if (req->rq_early_free_repbuf) {
 613                 struct lustre_handle *file_fh;
 614
 615                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 616
 617                 file_fh = &och->och_fh;
 618                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 619                        file_fh->cookie, body->mbo_handle.cookie);
 620                 old = *file_fh;
 621                 *file_fh = body->mbo_handle;
 622         }
 623
 624         close_req = mod->mod_close_req;
 625         if (close_req) {
 626                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 627                 struct mdt_ioepoch *epoch;
 628
 629                 LASSERT(opc == MDS_CLOSE);
 630                 epoch = req_capsule_client_get(&close_req->rq_pill,
 631                                                &RMF_MDT_EPOCH);
 632                 LASSERT(epoch);
 633
 634                 if (req->rq_early_free_repbuf)
 635                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 636
 637                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 638                 epoch->mio_handle = body->mbo_handle;
 639         }
 640         EXIT;
 641 }
 642
 643 void mdc_commit_open(struct ptlrpc_request *req)
 644 {
 645         struct md_open_data *mod = req->rq_cb_data;
 646         if (mod == NULL)
 647                 return;
 648
 649         /**
 650          * No need to touch md_open_data::mod_och, it holds a reference on
 651          * \var mod and will zero references to each other, \var mod will be
 652          * freed after that when md_open_data::mod_och will put the reference.
 653          */
 654
 655         /**
 656          * Do not let open request to disappear as it still may be needed
 657          * for close rpc to happen (it may happen on evict only, otherwise
 658          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 659          * called), just mark this rpc as committed to distinguish these 2
 660          * cases, see mdc_close() for details. The open request reference will
 661          * be put along with freeing \var mod.
 662          */
 663         ptlrpc_request_addref(req);
 664         spin_lock(&req->rq_lock);
 665         req->rq_committed = 1;
 666         spin_unlock(&req->rq_lock);
 667         req->rq_cb_data = NULL;
 668         obd_mod_put(mod);
 669 }
 670
 671 int mdc_set_open_replay_data(struct obd_export *exp,
 672                              struct obd_client_handle *och,
 673                              struct lookup_intent *it)
 674 {
 675         struct md_open_data     *mod;
 676         struct mdt_rec_create   *rec;
 677         struct mdt_body         *body;
 678         struct ptlrpc_request   *open_req = it->it_request;
 679         struct obd_import       *imp = open_req->rq_import;
 680         ENTRY;
 681
 682         if (!open_req->rq_replay)
 683                 RETURN(0);
 684
 685         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 686         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 687         LASSERT(rec != NULL);
 688         /* Incoming message in my byte order (it's been swabbed). */
 689         /* Outgoing messages always in my byte order. */
 690         LASSERT(body != NULL);
 691
 692         /* Only if the import is replayable, we set replay_open data */
 693         if (och && imp->imp_replayable) {
 694                 mod = obd_mod_alloc();
 695                 if (mod == NULL) {
 696                         DEBUG_REQ(D_ERROR, open_req,
 697                                   "Can't allocate md_open_data");
 698                         RETURN(0);
 699                 }
 700
 701                 /**
 702                  * Take a reference on \var mod, to be freed on mdc_close().
 703                  * It protects \var mod from being freed on eviction (commit
 704                  * callback is called despite rq_replay flag).
 705                  * Another reference for \var och.
 706                  */
 707                 obd_mod_get(mod);
 708                 obd_mod_get(mod);
 709
 710                 spin_lock(&open_req->rq_lock);
 711                 och->och_mod = mod;
 712                 mod->mod_och = och;
 713                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 714                                      it_disposition(it, DISP_OPEN_STRIPE);
 715                 mod->mod_open_req = open_req;
 716                 open_req->rq_cb_data = mod;
 717                 open_req->rq_commit_cb = mdc_commit_open;
 718                 open_req->rq_early_free_repbuf = 1;
 719                 spin_unlock(&open_req->rq_lock);
 720         }
 721
 722         rec->cr_fid2 = body->mbo_fid1;
 723         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 724         open_req->rq_replay_cb = mdc_replay_open;
 725         if (!fid_is_sane(&body->mbo_fid1)) {
 726                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 727                           "insane fid");
 728                 LBUG();
 729         }
 730
 731         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 732         RETURN(0);
 733 }
 734
 735 static void mdc_free_open(struct md_open_data *mod)
 736 {
 737         int committed = 0;
 738
 739         if (mod->mod_is_create == 0 &&
 740             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 741                 committed = 1;
 742
 743         /**
 744          * No reason to asssert here if the open request has
 745          * rq_replay == 1. It means that mdc_close failed, and
 746          * close request wasn`t sent. It is not fatal to client.
 747          * The worst thing is eviction if the client gets open lock
 748          **/
 749
 750         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 751                   "= %d\n", mod->mod_open_req->rq_replay);
 752
 753         ptlrpc_request_committed(mod->mod_open_req, committed);
 754         if (mod->mod_close_req)
 755                 ptlrpc_request_committed(mod->mod_close_req, committed);
 756 }
 757
 758 int mdc_clear_open_replay_data(struct obd_export *exp,
 759                                struct obd_client_handle *och)
 760 {
 761         struct md_open_data *mod = och->och_mod;
 762         ENTRY;
 763
 764         /**
 765          * It is possible to not have \var mod in a case of eviction between
 766          * lookup and ll_file_open().
 767          **/
 768         if (mod == NULL)
 769                 RETURN(0);
 770
 771         LASSERT(mod != LP_POISON);
 772         LASSERT(mod->mod_open_req != NULL);
 773
 774         spin_lock(&mod->mod_open_req->rq_lock);
 775         if (mod->mod_och)
 776                 mod->mod_och->och_fh.cookie = 0;
 777         mod->mod_open_req->rq_early_free_repbuf = 0;
 778         spin_unlock(&mod->mod_open_req->rq_lock);
 779         mdc_free_open(mod);
 780
 781         mod->mod_och = NULL;
 782         och->och_mod = NULL;
 783         obd_mod_put(mod);
 784
 785         RETURN(0);
 786 }
 787
 788 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 789                      struct md_open_data *mod, struct ptlrpc_request **request)
 790 {
 791         struct obd_device     *obd = class_exp2obd(exp);
 792         struct ptlrpc_request *req;
 793         struct req_format     *req_fmt;
 794         size_t                 u32_count = 0;
 795         int                    rc;
 796         int                    saved_rc = 0;
 797         ENTRY;
 798
 799         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 800                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 801                op_data->op_bias);
 802
 803         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 804                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 805                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 806                         /* allocate a FID for volatile file */
 807                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 808                                            op_data);
 809                         if (rc < 0) {
 810                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 811                                        obd->obd_name, PFID(&op_data->op_fid1),
 812                                        rc);
 813                                 /* save the errcode and proceed to close */
 814                                 saved_rc = rc;
 815                         }
 816                 }
 817                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 818                         size_t count = op_data->op_data_size / sizeof(__u32);
 819
 820                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 821                                 u32_count = count;
 822                 }
 823         } else {
 824                 req_fmt = &RQF_MDS_CLOSE;
 825         }
 826
 827         *request = NULL;
 828         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 829                 req = NULL;
 830         else
 831                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 832
 833         /* Ensure that this close's handle is fixed up during replay. */
 834         if (likely(mod != NULL)) {
 835                 LASSERTF(mod->mod_open_req != NULL &&
 836                          mod->mod_open_req->rq_type != LI_POISON,
 837                          "POISONED open %p!\n", mod->mod_open_req);
 838
 839                 mod->mod_close_req = req;
 840
 841                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 842                 /* We no longer want to preserve this open for replay even
 843                  * though the open was committed. b=3632, b=3633 */
 844                 spin_lock(&mod->mod_open_req->rq_lock);
 845                 mod->mod_open_req->rq_replay = 0;
 846                 spin_unlock(&mod->mod_open_req->rq_lock);
 847         } else {
 848                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 849         }
 850         if (req == NULL) {
 851                 /**
 852                  * TODO: repeat close after errors
 853                  */
 854                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 855                       "dropped when this client unmounts or is evicted\n",
 856                       obd->obd_name, PFID(&op_data->op_fid1));
 857                 GOTO(out, rc = -ENOMEM);
 858         }
 859
 860         if (u32_count > 0)
 861                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 862                                      u32_count * sizeof(__u32));
 863
 864         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 865         if (rc) {
 866                 ptlrpc_request_free(req);
 867                 req = NULL;
 868                 GOTO(out, rc);
 869         }
 870
 871         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 872          * portal whose threads are not taking any DLM locks and are therefore
 873          * always progressing */
 874         req->rq_request_portal = MDS_READPAGE_PORTAL;
 875         ptlrpc_at_set_req_timeout(req);
 876
 877
 878         mdc_close_pack(req, op_data);
 879
 880         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 881                              obd->u.cli.cl_default_mds_easize);
 882
 883         ptlrpc_request_set_replen(req);
 884
 885         mdc_get_mod_rpc_slot(req, NULL);
 886         rc = ptlrpc_queue_wait(req);
 887         mdc_put_mod_rpc_slot(req, NULL);
 888
 889         if (req->rq_repmsg == NULL) {
 890                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 891                        req->rq_status);
 892                 if (rc == 0)
 893                         rc = req->rq_status ?: -EIO;
 894         } else if (rc == 0 || rc == -EAGAIN) {
 895                 struct mdt_body *body;
 896
 897                 rc = lustre_msg_get_status(req->rq_repmsg);
 898                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 899                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 900                                   "= %d", rc);
 901                         if (rc > 0)
 902                                 rc = -rc;
 903                 }
 904                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 905                 if (body == NULL)
 906                         rc = -EPROTO;
 907         } else if (rc == -ESTALE) {
 908                 /**
 909                  * it can be allowed error after 3633 if open was committed and
 910                  * server failed before close was sent. Let's check if mod
 911                  * exists and return no error in that case
 912                  */
 913                 if (mod) {
 914                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 915                         LASSERT(mod->mod_open_req != NULL);
 916                         if (mod->mod_open_req->rq_committed)
 917                                 rc = 0;
 918                 }
 919         }
 920
 921 out:
 922         if (mod) {
 923                 if (rc != 0)
 924                         mod->mod_close_req = NULL;
 925                 /* Since now, mod is accessed through open_req only,
 926                  * thus close req does not keep a reference on mod anymore. */
 927                 obd_mod_put(mod);
 928         }
 929         *request = req;
 930
 931         RETURN(rc < 0 ? rc : saved_rc);
 932 }
 933
 934 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 935                        u64 offset, struct page **pages, int npages,
 936                        struct ptlrpc_request **request)
 937 {
 938         struct ptlrpc_request   *req;
 939         struct ptlrpc_bulk_desc *desc;
 940         int                      i;
 941         wait_queue_head_t        waitq;
 942         int                      resends = 0;
 943         struct l_wait_info       lwi;
 944         int                      rc;
 945         ENTRY;
 946
 947         *request = NULL;
 948         init_waitqueue_head(&waitq);
 949
 950 restart_bulk:
 951         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 952         if (req == NULL)
 953                 RETURN(-ENOMEM);
 954
 955         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 956         if (rc) {
 957                 ptlrpc_request_free(req);
 958                 RETURN(rc);
 959         }
 960
 961         req->rq_request_portal = MDS_READPAGE_PORTAL;
 962         ptlrpc_at_set_req_timeout(req);
 963
 964         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 965                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 966                                     MDS_BULK_PORTAL,
 967                                     &ptlrpc_bulk_kiov_pin_ops);
 968         if (desc == NULL) {
 969                 ptlrpc_req_finished(req);
 970                 RETURN(-ENOMEM);
 971         }
 972
 973         /* NB req now owns desc and will free it when it gets freed */
 974         for (i = 0; i < npages; i++)
 975                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 976                                                  PAGE_SIZE);
 977
 978         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 979
 980         ptlrpc_request_set_replen(req);
 981         rc = ptlrpc_queue_wait(req);
 982         if (rc) {
 983                 ptlrpc_req_finished(req);
 984                 if (rc != -ETIMEDOUT)
 985                         RETURN(rc);
 986
 987                 resends++;
 988                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 989                         CERROR("%s: too many resend retries: rc = %d\n",
 990                                exp->exp_obd->obd_name, -EIO);
 991                         RETURN(-EIO);
 992                 }
 993                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 994                                        NULL);
 995                 l_wait_event(waitq, 0, &lwi);
 996
 997                 goto restart_bulk;
 998         }
 999
1000         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1001                                           req->rq_bulk->bd_nob_transferred);
1002         if (rc < 0) {
1003                 ptlrpc_req_finished(req);
1004                 RETURN(rc);
1005         }
1006
1007         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1008                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1009                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1010                        PAGE_SIZE * npages);
1011                 ptlrpc_req_finished(req);
1012                 RETURN(-EPROTO);
1013         }
1014
1015         *request = req;
1016         RETURN(0);
1017 }
1018
1019 static void mdc_release_page(struct page *page, int remove)
1020 {
1021         if (remove) {
1022                 lock_page(page);
1023                 if (likely(page->mapping != NULL))
1024                         truncate_complete_page(page->mapping, page);
1025                 unlock_page(page);
1026         }
1027         put_page(page);
1028 }
1029
1030 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1031                                     __u64 *start, __u64 *end, int hash64)
1032 {
1033         /*
1034          * Complement of hash is used as an index so that
1035          * radix_tree_gang_lookup() can be used to find a page with starting
1036          * hash _smaller_ than one we are looking for.
1037          */
1038         unsigned long offset = hash_x_index(*hash, hash64);
1039         struct page *page;
1040         int found;
1041
1042         spin_lock_irq(&mapping->tree_lock);
1043         found = radix_tree_gang_lookup(&mapping->page_tree,
1044                                        (void **)&page, offset, 1);
1045         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1046                 struct lu_dirpage *dp;
1047
1048                 get_page(page);
1049                 spin_unlock_irq(&mapping->tree_lock);
1050                 /*
1051                  * In contrast to find_lock_page() we are sure that directory
1052                  * page cannot be truncated (while DLM lock is held) and,
1053                  * hence, can avoid restart.
1054                  *
1055                  * In fact, page cannot be locked here at all, because
1056                  * mdc_read_page_remote does synchronous io.
1057                  */
1058                 wait_on_page_locked(page);
1059                 if (PageUptodate(page)) {
1060                         dp = kmap(page);
1061                         if (BITS_PER_LONG == 32 && hash64) {
1062                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1063                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1064                                 *hash  = *hash >> 32;
1065                         } else {
1066                                 *start = le64_to_cpu(dp->ldp_hash_start);
1067                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1068                         }
1069                         if (unlikely(*start == 1 && *hash == 0))
1070                                 *hash = *start;
1071                         else
1072                                 LASSERTF(*start <= *hash, "start = %#llx"
1073                                          ",end = %#llx,hash = %#llx\n",
1074                                          *start, *end, *hash);
1075                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1076                               " hash %#llx\n", offset, *start, *end, *hash);
1077                         if (*hash > *end) {
1078                                 kunmap(page);
1079                                 mdc_release_page(page, 0);
1080                                 page = NULL;
1081                         } else if (*end != *start && *hash == *end) {
1082                                 /*
1083                                  * upon hash collision, remove this page,
1084                                  * otherwise put page reference, and
1085                                  * mdc_read_page_remote() will issue RPC to
1086                                  * fetch the page we want.
1087                                  */
1088                                 kunmap(page);
1089                                 mdc_release_page(page,
1090                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1091                                 page = NULL;
1092                         }
1093                 } else {
1094                         put_page(page);
1095                         page = ERR_PTR(-EIO);
1096                 }
1097         } else {
1098                 spin_unlock_irq(&mapping->tree_lock);
1099                 page = NULL;
1100         }
1101         return page;
1102 }
1103
1104 /*
1105  * Adjust a set of pages, each page containing an array of lu_dirpages,
1106  * so that each page can be used as a single logical lu_dirpage.
1107  *
1108  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1109  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1110  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1111  * value is used as a cookie to request the next lu_dirpage in a
1112  * directory listing that spans multiple pages (two in this example):
1113  *   ________
1114  *  |        |
1115  * .|--------v-------   -----.
1116  * |s|e|f|p|ent|ent| ... |ent|
1117  * '--|--------------   -----'   Each PAGE contains a single
1118  *    '------.                   lu_dirpage.
1119  * .---------v-------   -----.
1120  * |s|e|f|p|ent| 0 | ... | 0 |
1121  * '-----------------   -----'
1122  *
1123  * However, on hosts where the native VM page size (PAGE_SIZE) is
1124  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1125  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1126  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1127  * after it in the same PAGE (arrows simplified for brevity, but
1128  * in general e0==s1, e1==s2, etc.):
1129  *
1130  * .--------------------   -----.
1131  * |s0|e0|f0|p|ent|ent| ... |ent|
1132  * |---v----------------   -----|
1133  * |s1|e1|f1|p|ent|ent| ... |ent|
1134  * |---v----------------   -----|  Here, each PAGE contains
1135  *             ...                 multiple lu_dirpages.
1136  * |---v----------------   -----|
1137  * |s'|e'|f'|p|ent|ent| ... |ent|
1138  * '---|----------------   -----'
1139  *     v
1140  * .----------------------------.
1141  * |        next PAGE           |
1142  *
1143  * This structure is transformed into a single logical lu_dirpage as follows:
1144  *
1145  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1146  *   labeled 'next PAGE'.
1147  *
1148  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1149  *   a hash collision with the next page exists.
1150  *
1151  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1152  *   to the first entry of the next lu_dirpage.
1153  */
1154 #if PAGE_SIZE > LU_PAGE_SIZE
1155 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1156 {
1157         int i;
1158
1159         for (i = 0; i < cfs_pgs; i++) {
1160                 struct lu_dirpage       *dp = kmap(pages[i]);
1161                 struct lu_dirpage       *first = dp;
1162                 struct lu_dirent        *end_dirent = NULL;
1163                 struct lu_dirent        *ent;
1164                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1165                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1166
1167                 while (--lu_pgs > 0) {
1168                         ent = lu_dirent_start(dp);
1169                         for (end_dirent = ent; ent != NULL;
1170                              end_dirent = ent, ent = lu_dirent_next(ent));
1171
1172                         /* Advance dp to next lu_dirpage. */
1173                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1174
1175                         /* Check if we've reached the end of the PAGE. */
1176                         if (!((unsigned long)dp & ~PAGE_MASK))
1177                                 break;
1178
1179                         /* Save the hash and flags of this lu_dirpage. */
1180                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1181                         flags = le32_to_cpu(dp->ldp_flags);
1182
1183                         /* Check if lu_dirpage contains no entries. */
1184                         if (end_dirent == NULL)
1185                                 break;
1186
1187                         /* Enlarge the end entry lde_reclen from 0 to
1188                          * first entry of next lu_dirpage. */
1189                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1190                         end_dirent->lde_reclen =
1191                                 cpu_to_le16((char *)(dp->ldp_entries) -
1192                                             (char *)end_dirent);
1193                 }
1194
1195                 first->ldp_hash_end = hash_end;
1196                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1197                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1198
1199                 kunmap(pages[i]);
1200         }
1201         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1202 }
1203 #else
1204 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1205 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1206
1207 /* parameters for readdir page */
1208 struct readpage_param {
1209         struct md_op_data       *rp_mod;
1210         __u64                   rp_off;
1211         int                     rp_hash64;
1212         struct obd_export       *rp_exp;
1213         struct md_callback      *rp_cb;
1214 };
1215
1216 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1217 static inline void delete_from_page_cache(struct page *page)
1218 {
1219         remove_from_page_cache(page);
1220         put_page(page);
1221 }
1222 #endif
1223
1224 /**
1225  * Read pages from server.
1226  *
1227  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1228  * a header lu_dirpage which describes the start/end hash, and whether this
1229  * page is empty (contains no dir entry) or hash collide with next page.
1230  * After client receives reply, several pages will be integrated into dir page
1231  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1232  * lu_dirpage for this integrated page will be adjusted.
1233  **/
1234 static int mdc_read_page_remote(void *data, struct page *page0)
1235 {
1236         struct readpage_param *rp = data;
1237         struct page **page_pool;
1238         struct page *page;
1239         struct lu_dirpage *dp;
1240         struct md_op_data *op_data = rp->rp_mod;
1241         struct ptlrpc_request *req;
1242         int max_pages;
1243         struct inode *inode;
1244         struct lu_fid *fid;
1245         int rd_pgs = 0; /* number of pages actually read */
1246         int npages;
1247         int i;
1248         int rc;
1249         ENTRY;
1250
1251         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1252         inode = op_data->op_data;
1253         fid = &op_data->op_fid1;
1254         LASSERT(inode != NULL);
1255
1256         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1257         if (page_pool != NULL) {
1258                 page_pool[0] = page0;
1259         } else {
1260                 page_pool = &page0;
1261                 max_pages = 1;
1262         }
1263
1264         for (npages = 1; npages < max_pages; npages++) {
1265                 page = __page_cache_alloc(mapping_gfp_mask(inode->i_mapping)
1266                                           | __GFP_COLD);
1267                 if (page == NULL)
1268                         break;
1269                 page_pool[npages] = page;
1270         }
1271
1272         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1273         if (rc < 0) {
1274                 /* page0 is special, which was added into page cache early */
1275                 delete_from_page_cache(page0);
1276         } else {
1277                 int lu_pgs;
1278
1279                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1280                         PAGE_SHIFT;
1281                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1282                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1283
1284                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1285
1286                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1287
1288                 SetPageUptodate(page0);
1289         }
1290         unlock_page(page0);
1291
1292         ptlrpc_req_finished(req);
1293         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1294         for (i = 1; i < npages; i++) {
1295                 unsigned long   offset;
1296                 __u64           hash;
1297                 int ret;
1298
1299                 page = page_pool[i];
1300
1301                 if (rc < 0 || i >= rd_pgs) {
1302                         put_page(page);
1303                         continue;
1304                 }
1305
1306                 SetPageUptodate(page);
1307
1308                 dp = kmap(page);
1309                 hash = le64_to_cpu(dp->ldp_hash_start);
1310                 kunmap(page);
1311
1312                 offset = hash_x_index(hash, rp->rp_hash64);
1313
1314                 prefetchw(&page->flags);
1315                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1316                                             GFP_KERNEL);
1317                 if (ret == 0)
1318                         unlock_page(page);
1319                 else
1320                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1321                                " rc = %d\n", offset, ret);
1322                 put_page(page);
1323         }
1324
1325         if (page_pool != &page0)
1326                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1327
1328         RETURN(rc);
1329 }
1330
1331 /**
1332  * Read dir page from cache first, if it can not find it, read it from
1333  * server and add into the cache.
1334  *
1335  * \param[in] exp       MDC export
1336  * \param[in] op_data   client MD stack parameters, transfering parameters
1337  *                      between different layers on client MD stack.
1338  * \param[in] cb_op     callback required for ldlm lock enqueue during
1339  *                      read page
1340  * \param[in] hash_offset the hash offset of the page to be read
1341  * \param[in] ppage     the page to be read
1342  *
1343  * retval               = 0 get the page successfully
1344  *                      errno(<0) get the page failed
1345  */
1346 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1347                          struct md_callback *cb_op, __u64 hash_offset,
1348                          struct page **ppage)
1349 {
1350         struct lookup_intent    it = { .it_op = IT_READDIR };
1351         struct page             *page;
1352         struct inode            *dir = op_data->op_data;
1353         struct address_space    *mapping;
1354         struct lu_dirpage       *dp;
1355         __u64                   start = 0;
1356         __u64                   end = 0;
1357         struct lustre_handle    lockh;
1358         struct ptlrpc_request   *enq_req = NULL;
1359         struct readpage_param   rp_param;
1360         int rc;
1361
1362         ENTRY;
1363
1364         *ppage = NULL;
1365
1366         LASSERT(dir != NULL);
1367         mapping = dir->i_mapping;
1368
1369         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1370                              cb_op->md_blocking_ast, 0);
1371         if (enq_req != NULL)
1372                 ptlrpc_req_finished(enq_req);
1373
1374         if (rc < 0) {
1375                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1376                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1377                 RETURN(rc);
1378         }
1379
1380         rc = 0;
1381         lockh.cookie = it.it_lock_handle;
1382         mdc_set_lock_data(exp, &lockh, dir, NULL);
1383
1384         rp_param.rp_off = hash_offset;
1385         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1386         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1387                                rp_param.rp_hash64);
1388         if (IS_ERR(page)) {
1389                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1390                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1391                        rp_param.rp_off, PTR_ERR(page));
1392                 GOTO(out_unlock, rc = PTR_ERR(page));
1393         } else if (page != NULL) {
1394                 /*
1395                  * XXX nikita: not entirely correct handling of a corner case:
1396                  * suppose hash chain of entries with hash value HASH crosses
1397                  * border between pages P0 and P1. First both P0 and P1 are
1398                  * cached, seekdir() is called for some entry from the P0 part
1399                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1400                  * happens and finds P1, as it starts with matching hash
1401                  * value. Remaining entries from P0 part of the chain are
1402                  * skipped. (Is that really a bug?)
1403                  *
1404                  * Possible solutions: 0. don't cache P1 is such case, handle
1405                  * it as an "overflow" page. 1. invalidate all pages at
1406                  * once. 2. use HASH|1 as an index for P1.
1407                  */
1408                 GOTO(hash_collision, page);
1409         }
1410
1411         rp_param.rp_exp = exp;
1412         rp_param.rp_mod = op_data;
1413         page = read_cache_page(mapping,
1414                                hash_x_index(rp_param.rp_off,
1415                                             rp_param.rp_hash64),
1416                                mdc_read_page_remote, &rp_param);
1417         if (IS_ERR(page)) {
1418                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1419                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1420                        rp_param.rp_off, PTR_ERR(page));
1421                 GOTO(out_unlock, rc = PTR_ERR(page));
1422         }
1423
1424         wait_on_page_locked(page);
1425         (void)kmap(page);
1426         if (!PageUptodate(page)) {
1427                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1428                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1429                        rp_param.rp_off, -5);
1430                 goto fail;
1431         }
1432         if (!PageChecked(page))
1433                 SetPageChecked(page);
1434         if (PageError(page)) {
1435                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1436                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1437                        rp_param.rp_off, -5);
1438                 goto fail;
1439         }
1440
1441 hash_collision:
1442         dp = page_address(page);
1443         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1444                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1445                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1446                 rp_param.rp_off = hash_offset >> 32;
1447         } else {
1448                 start = le64_to_cpu(dp->ldp_hash_start);
1449                 end   = le64_to_cpu(dp->ldp_hash_end);
1450                 rp_param.rp_off = hash_offset;
1451         }
1452         if (end == start) {
1453                 LASSERT(start == rp_param.rp_off);
1454                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1455 #if BITS_PER_LONG == 32
1456                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1457                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1458                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1459 #endif
1460
1461                 /*
1462                  * Fetch whole overflow chain...
1463                  *
1464                  * XXX not yet.
1465                  */
1466                 goto fail;
1467         }
1468         *ppage = page;
1469 out_unlock:
1470         ldlm_lock_decref(&lockh, it.it_lock_mode);
1471         return rc;
1472 fail:
1473         kunmap(page);
1474         mdc_release_page(page, 1);
1475         rc = -EIO;
1476         goto out_unlock;
1477 }
1478
1479 static int mdc_statfs(const struct lu_env *env,
1480                       struct obd_export *exp, struct obd_statfs *osfs,
1481                       time64_t max_age, __u32 flags)
1482 {
1483         struct obd_device     *obd = class_exp2obd(exp);
1484         struct ptlrpc_request *req;
1485         struct obd_statfs     *msfs;
1486         struct obd_import     *imp = NULL;
1487         int                    rc;
1488         ENTRY;
1489
1490         /*
1491          * Since the request might also come from lprocfs, so we need
1492          * sync this with client_disconnect_export Bug15684
1493          */
1494         down_read(&obd->u.cli.cl_sem);
1495         if (obd->u.cli.cl_import)
1496                 imp = class_import_get(obd->u.cli.cl_import);
1497         up_read(&obd->u.cli.cl_sem);
1498         if (!imp)
1499                 RETURN(-ENODEV);
1500
1501         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1502                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1503         if (req == NULL)
1504                 GOTO(output, rc = -ENOMEM);
1505
1506         ptlrpc_request_set_replen(req);
1507
1508         if (flags & OBD_STATFS_NODELAY) {
1509                 /* procfs requests not want stay in wait for avoid deadlock */
1510                 req->rq_no_resend = 1;
1511                 req->rq_no_delay = 1;
1512         }
1513
1514         rc = ptlrpc_queue_wait(req);
1515         if (rc) {
1516                 /* check connection error first */
1517                 if (imp->imp_connect_error)
1518                         rc = imp->imp_connect_error;
1519                 GOTO(out, rc);
1520         }
1521
1522         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1523         if (msfs == NULL)
1524                 GOTO(out, rc = -EPROTO);
1525
1526         *osfs = *msfs;
1527         EXIT;
1528 out:
1529         ptlrpc_req_finished(req);
1530 output:
1531         class_import_put(imp);
1532         return rc;
1533 }
1534
1535 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1536 {
1537         __u32 keylen, vallen;
1538         void *key;
1539         int rc;
1540
1541         if (gf->gf_pathlen > PATH_MAX)
1542                 RETURN(-ENAMETOOLONG);
1543         if (gf->gf_pathlen < 2)
1544                 RETURN(-EOVERFLOW);
1545
1546         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1547         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1548                                 sizeof(struct lu_fid));
1549         OBD_ALLOC(key, keylen);
1550         if (key == NULL)
1551                 RETURN(-ENOMEM);
1552         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1553         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1554         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1555                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1556         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1557                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1558
1559         if (!fid_is_sane(&gf->gf_fid))
1560                 GOTO(out, rc = -EINVAL);
1561
1562         /* Val is struct getinfo_fid2path result plus path */
1563         vallen = sizeof(*gf) + gf->gf_pathlen;
1564
1565         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1566         if (rc != 0 && rc != -EREMOTE)
1567                 GOTO(out, rc);
1568
1569         if (vallen <= sizeof(*gf))
1570                 GOTO(out, rc = -EPROTO);
1571         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1572                 GOTO(out, rc = -EOVERFLOW);
1573
1574         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1575                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1576                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1577                /* only log the last 512 characters of the path */
1578                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1579
1580 out:
1581         OBD_FREE(key, keylen);
1582         return rc;
1583 }
1584
1585 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1586                                 struct hsm_progress_kernel *hpk)
1587 {
1588         struct obd_import               *imp = class_exp2cliimp(exp);
1589         struct hsm_progress_kernel      *req_hpk;
1590         struct ptlrpc_request           *req;
1591         int                              rc;
1592         ENTRY;
1593
1594         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1595                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1596         if (req == NULL)
1597                 GOTO(out, rc = -ENOMEM);
1598
1599         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1600
1601         /* Copy hsm_progress struct */
1602         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1603         if (req_hpk == NULL)
1604                 GOTO(out, rc = -EPROTO);
1605
1606         *req_hpk = *hpk;
1607         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1608
1609         ptlrpc_request_set_replen(req);
1610
1611         mdc_get_mod_rpc_slot(req, NULL);
1612         rc = ptlrpc_queue_wait(req);
1613         mdc_put_mod_rpc_slot(req, NULL);
1614
1615         GOTO(out, rc);
1616 out:
1617         ptlrpc_req_finished(req);
1618         return rc;
1619 }
1620
1621 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1622 {
1623         __u32                   *archive_mask;
1624         struct ptlrpc_request   *req;
1625         int                      rc;
1626         ENTRY;
1627
1628         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1629                                         LUSTRE_MDS_VERSION,
1630                                         MDS_HSM_CT_REGISTER);
1631         if (req == NULL)
1632                 GOTO(out, rc = -ENOMEM);
1633
1634         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1635
1636         /* Copy hsm_progress struct */
1637         archive_mask = req_capsule_client_get(&req->rq_pill,
1638                                               &RMF_MDS_HSM_ARCHIVE);
1639         if (archive_mask == NULL)
1640                 GOTO(out, rc = -EPROTO);
1641
1642         *archive_mask = archives;
1643
1644         ptlrpc_request_set_replen(req);
1645
1646         rc = mdc_queue_wait(req);
1647         GOTO(out, rc);
1648 out:
1649         ptlrpc_req_finished(req);
1650         return rc;
1651 }
1652
1653 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1654                                       struct md_op_data *op_data)
1655 {
1656         struct hsm_current_action       *hca = op_data->op_data;
1657         struct hsm_current_action       *req_hca;
1658         struct ptlrpc_request           *req;
1659         int                              rc;
1660         ENTRY;
1661
1662         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1663                                    &RQF_MDS_HSM_ACTION);
1664         if (req == NULL)
1665                 RETURN(-ENOMEM);
1666
1667         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1668         if (rc) {
1669                 ptlrpc_request_free(req);
1670                 RETURN(rc);
1671         }
1672
1673         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1674                       op_data->op_suppgids[0], 0);
1675
1676         ptlrpc_request_set_replen(req);
1677
1678         rc = mdc_queue_wait(req);
1679         if (rc)
1680                 GOTO(out, rc);
1681
1682         req_hca = req_capsule_server_get(&req->rq_pill,
1683                                          &RMF_MDS_HSM_CURRENT_ACTION);
1684         if (req_hca == NULL)
1685                 GOTO(out, rc = -EPROTO);
1686
1687         *hca = *req_hca;
1688
1689         EXIT;
1690 out:
1691         ptlrpc_req_finished(req);
1692         return rc;
1693 }
1694
1695 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1696 {
1697         struct ptlrpc_request   *req;
1698         int                      rc;
1699         ENTRY;
1700
1701         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1702                                         LUSTRE_MDS_VERSION,
1703                                         MDS_HSM_CT_UNREGISTER);
1704         if (req == NULL)
1705                 GOTO(out, rc = -ENOMEM);
1706
1707         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1708
1709         ptlrpc_request_set_replen(req);
1710
1711         rc = mdc_queue_wait(req);
1712         GOTO(out, rc);
1713 out:
1714         ptlrpc_req_finished(req);
1715         return rc;
1716 }
1717
1718 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1719                                  struct md_op_data *op_data)
1720 {
1721         struct hsm_user_state   *hus = op_data->op_data;
1722         struct hsm_user_state   *req_hus;
1723         struct ptlrpc_request   *req;
1724         int                      rc;
1725         ENTRY;
1726
1727         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1728                                    &RQF_MDS_HSM_STATE_GET);
1729         if (req == NULL)
1730                 RETURN(-ENOMEM);
1731
1732         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1733         if (rc != 0) {
1734                 ptlrpc_request_free(req);
1735                 RETURN(rc);
1736         }
1737
1738         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1739                       op_data->op_suppgids[0], 0);
1740
1741         ptlrpc_request_set_replen(req);
1742
1743         rc = mdc_queue_wait(req);
1744         if (rc)
1745                 GOTO(out, rc);
1746
1747         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1748         if (req_hus == NULL)
1749                 GOTO(out, rc = -EPROTO);
1750
1751         *hus = *req_hus;
1752
1753         EXIT;
1754 out:
1755         ptlrpc_req_finished(req);
1756         return rc;
1757 }
1758
1759 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1760                                  struct md_op_data *op_data)
1761 {
1762         struct hsm_state_set    *hss = op_data->op_data;
1763         struct hsm_state_set    *req_hss;
1764         struct ptlrpc_request   *req;
1765         int                      rc;
1766         ENTRY;
1767
1768         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1769                                    &RQF_MDS_HSM_STATE_SET);
1770         if (req == NULL)
1771                 RETURN(-ENOMEM);
1772
1773         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1774         if (rc) {
1775                 ptlrpc_request_free(req);
1776                 RETURN(rc);
1777         }
1778
1779         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1780                       op_data->op_suppgids[0], 0);
1781
1782         /* Copy states */
1783         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1784         if (req_hss == NULL)
1785                 GOTO(out, rc = -EPROTO);
1786         *req_hss = *hss;
1787
1788         ptlrpc_request_set_replen(req);
1789
1790         mdc_get_mod_rpc_slot(req, NULL);
1791         rc = ptlrpc_queue_wait(req);
1792         mdc_put_mod_rpc_slot(req, NULL);
1793
1794         GOTO(out, rc);
1795 out:
1796         ptlrpc_req_finished(req);
1797         return rc;
1798 }
1799
1800 static int mdc_ioc_hsm_request(struct obd_export *exp,
1801                                struct hsm_user_request *hur)
1802 {
1803         struct obd_import       *imp = class_exp2cliimp(exp);
1804         struct ptlrpc_request   *req;
1805         struct hsm_request      *req_hr;
1806         struct hsm_user_item    *req_hui;
1807         char                    *req_opaque;
1808         int                      rc;
1809         ENTRY;
1810
1811         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1812         if (req == NULL)
1813                 GOTO(out, rc = -ENOMEM);
1814
1815         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1816                              hur->hur_request.hr_itemcount
1817                              * sizeof(struct hsm_user_item));
1818         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1819                              hur->hur_request.hr_data_len);
1820
1821         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1822         if (rc) {
1823                 ptlrpc_request_free(req);
1824                 RETURN(rc);
1825         }
1826
1827         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1828
1829         /* Copy hsm_request struct */
1830         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1831         if (req_hr == NULL)
1832                 GOTO(out, rc = -EPROTO);
1833         *req_hr = hur->hur_request;
1834
1835         /* Copy hsm_user_item structs */
1836         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1837         if (req_hui == NULL)
1838                 GOTO(out, rc = -EPROTO);
1839         memcpy(req_hui, hur->hur_user_item,
1840                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1841
1842         /* Copy opaque field */
1843         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1844         if (req_opaque == NULL)
1845                 GOTO(out, rc = -EPROTO);
1846         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1847
1848         ptlrpc_request_set_replen(req);
1849
1850         mdc_get_mod_rpc_slot(req, NULL);
1851         rc = ptlrpc_queue_wait(req);
1852         mdc_put_mod_rpc_slot(req, NULL);
1853
1854         GOTO(out, rc);
1855
1856 out:
1857         ptlrpc_req_finished(req);
1858         return rc;
1859 }
1860
1861 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1862                                 struct lustre_kernelcomm *lk);
1863
1864 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1865                         struct obd_quotactl *oqctl)
1866 {
1867         struct ptlrpc_request   *req;
1868         struct obd_quotactl     *oqc;
1869         int                      rc;
1870         ENTRY;
1871
1872         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1873                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1874                                         MDS_QUOTACTL);
1875         if (req == NULL)
1876                 RETURN(-ENOMEM);
1877
1878         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1879         *oqc = *oqctl;
1880
1881         ptlrpc_request_set_replen(req);
1882         ptlrpc_at_set_req_timeout(req);
1883         req->rq_no_resend = 1;
1884
1885         rc = ptlrpc_queue_wait(req);
1886         if (rc)
1887                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1888
1889         if (req->rq_repmsg &&
1890             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1891                 *oqctl = *oqc;
1892         } else if (!rc) {
1893                 CERROR ("Can't unpack obd_quotactl\n");
1894                 rc = -EPROTO;
1895         }
1896         ptlrpc_req_finished(req);
1897
1898         RETURN(rc);
1899 }
1900
1901 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1902                                 struct md_op_data *op_data)
1903 {
1904         struct list_head cancels = LIST_HEAD_INIT(cancels);
1905         struct ptlrpc_request   *req;
1906         int                      rc, count;
1907         struct mdc_swap_layouts *msl, *payload;
1908         ENTRY;
1909
1910         msl = op_data->op_data;
1911
1912         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1913          * first thing it will do is to cancel the 2 layout
1914          * locks held by this client.
1915          * So the client must cancel its layout locks on the 2 fids
1916          * with the request RPC to avoid extra RPC round trips.
1917          */
1918         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1919                                         LCK_EX, MDS_INODELOCK_LAYOUT |
1920                                         MDS_INODELOCK_XATTR);
1921         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1922                                          LCK_EX, MDS_INODELOCK_LAYOUT |
1923                                          MDS_INODELOCK_XATTR);
1924
1925         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1926                                    &RQF_MDS_SWAP_LAYOUTS);
1927         if (req == NULL) {
1928                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1929                 RETURN(-ENOMEM);
1930         }
1931
1932         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1933         if (rc) {
1934                 ptlrpc_request_free(req);
1935                 RETURN(rc);
1936         }
1937
1938         mdc_swap_layouts_pack(req, op_data);
1939
1940         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1941         LASSERT(payload);
1942
1943         *payload = *msl;
1944
1945         ptlrpc_request_set_replen(req);
1946
1947         rc = ptlrpc_queue_wait(req);
1948         if (rc)
1949                 GOTO(out, rc);
1950         EXIT;
1951
1952 out:
1953         ptlrpc_req_finished(req);
1954         return rc;
1955 }
1956
1957 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1958                          void *karg, void __user *uarg)
1959 {
1960         struct obd_device *obd = exp->exp_obd;
1961         struct obd_ioctl_data *data = karg;
1962         struct obd_import *imp = obd->u.cli.cl_import;
1963         int rc;
1964         ENTRY;
1965
1966         if (!try_module_get(THIS_MODULE)) {
1967                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1968                        module_name(THIS_MODULE));
1969                 return -EINVAL;
1970         }
1971         switch (cmd) {
1972         case OBD_IOC_FID2PATH:
1973                 rc = mdc_ioc_fid2path(exp, karg);
1974                 GOTO(out, rc);
1975         case LL_IOC_HSM_CT_START:
1976                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1977                 /* ignore if it was already registered on this MDS. */
1978                 if (rc == -EEXIST)
1979                         rc = 0;
1980                 GOTO(out, rc);
1981         case LL_IOC_HSM_PROGRESS:
1982                 rc = mdc_ioc_hsm_progress(exp, karg);
1983                 GOTO(out, rc);
1984         case LL_IOC_HSM_STATE_GET:
1985                 rc = mdc_ioc_hsm_state_get(exp, karg);
1986                 GOTO(out, rc);
1987         case LL_IOC_HSM_STATE_SET:
1988                 rc = mdc_ioc_hsm_state_set(exp, karg);
1989                 GOTO(out, rc);
1990         case LL_IOC_HSM_ACTION:
1991                 rc = mdc_ioc_hsm_current_action(exp, karg);
1992                 GOTO(out, rc);
1993         case LL_IOC_HSM_REQUEST:
1994                 rc = mdc_ioc_hsm_request(exp, karg);
1995                 GOTO(out, rc);
1996         case OBD_IOC_CLIENT_RECOVER:
1997                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1998                 if (rc < 0)
1999                         GOTO(out, rc);
2000                 GOTO(out, rc = 0);
2001         case IOC_OSC_SET_ACTIVE:
2002                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2003                 GOTO(out, rc);
2004         case OBD_IOC_PING_TARGET:
2005                 rc = ptlrpc_obd_ping(obd);
2006                 GOTO(out, rc);
2007         /*
2008          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2009          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2010          * there'd be no LMV layer thus we might be called here. Eventually
2011          * this code should be removed.
2012          * bz20731, LU-592.
2013          */
2014         case IOC_OBD_STATFS: {
2015                 struct obd_statfs stat_buf = {0};
2016
2017                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2018                         GOTO(out, rc = -ENODEV);
2019
2020                 /* copy UUID */
2021                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2022                                  min((int)data->ioc_plen2,
2023                                      (int)sizeof(struct obd_uuid))))
2024                         GOTO(out, rc = -EFAULT);
2025
2026                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2027                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2028                                 0);
2029                 if (rc != 0)
2030                         GOTO(out, rc);
2031
2032                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2033                                      min((int) data->ioc_plen1,
2034                                          (int) sizeof(stat_buf))))
2035                         GOTO(out, rc = -EFAULT);
2036
2037                 GOTO(out, rc = 0);
2038         }
2039         case OBD_IOC_QUOTACTL: {
2040                 struct if_quotactl *qctl = karg;
2041                 struct obd_quotactl *oqctl;
2042
2043                 OBD_ALLOC_PTR(oqctl);
2044                 if (oqctl == NULL)
2045                         GOTO(out, rc = -ENOMEM);
2046
2047                 QCTL_COPY(oqctl, qctl);
2048                 rc = obd_quotactl(exp, oqctl);
2049                 if (rc == 0) {
2050                         QCTL_COPY(qctl, oqctl);
2051                         qctl->qc_valid = QC_MDTIDX;
2052                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2053                 }
2054
2055                 OBD_FREE_PTR(oqctl);
2056                 GOTO(out, rc);
2057         }
2058         case LL_IOC_GET_CONNECT_FLAGS:
2059                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2060                                  sizeof(*exp_connect_flags_ptr(exp))))
2061                         GOTO(out, rc = -EFAULT);
2062
2063                 GOTO(out, rc = 0);
2064         case LL_IOC_LOV_SWAP_LAYOUTS:
2065                 rc = mdc_ioc_swap_layouts(exp, karg);
2066                 GOTO(out, rc);
2067         default:
2068                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2069                 GOTO(out, rc = -ENOTTY);
2070         }
2071 out:
2072         module_put(THIS_MODULE);
2073
2074         return rc;
2075 }
2076
2077 static int mdc_get_info_rpc(struct obd_export *exp,
2078                             u32 keylen, void *key,
2079                             u32 vallen, void *val)
2080 {
2081         struct obd_import      *imp = class_exp2cliimp(exp);
2082         struct ptlrpc_request  *req;
2083         char                   *tmp;
2084         int                     rc = -EINVAL;
2085         ENTRY;
2086
2087         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2088         if (req == NULL)
2089                 RETURN(-ENOMEM);
2090
2091         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2092                              RCL_CLIENT, keylen);
2093         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2094                              RCL_CLIENT, sizeof(vallen));
2095
2096         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2097         if (rc) {
2098                 ptlrpc_request_free(req);
2099                 RETURN(rc);
2100         }
2101
2102         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2103         memcpy(tmp, key, keylen);
2104         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2105         memcpy(tmp, &vallen, sizeof(vallen));
2106
2107         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2108                              RCL_SERVER, vallen);
2109         ptlrpc_request_set_replen(req);
2110
2111         rc = ptlrpc_queue_wait(req);
2112         /* -EREMOTE means the get_info result is partial, and it needs to
2113          * continue on another MDT, see fid2path part in lmv_iocontrol */
2114         if (rc == 0 || rc == -EREMOTE) {
2115                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2116                 memcpy(val, tmp, vallen);
2117                 if (ptlrpc_rep_need_swab(req)) {
2118                         if (KEY_IS(KEY_FID2PATH))
2119                                 lustre_swab_fid2path(val);
2120                 }
2121         }
2122         ptlrpc_req_finished(req);
2123
2124         RETURN(rc);
2125 }
2126
2127 static void lustre_swab_hai(struct hsm_action_item *h)
2128 {
2129         __swab32s(&h->hai_len);
2130         __swab32s(&h->hai_action);
2131         lustre_swab_lu_fid(&h->hai_fid);
2132         lustre_swab_lu_fid(&h->hai_dfid);
2133         __swab64s(&h->hai_cookie);
2134         __swab64s(&h->hai_extent.offset);
2135         __swab64s(&h->hai_extent.length);
2136         __swab64s(&h->hai_gid);
2137 }
2138
2139 static void lustre_swab_hal(struct hsm_action_list *h)
2140 {
2141         struct hsm_action_item  *hai;
2142         __u32                    i;
2143
2144         __swab32s(&h->hal_version);
2145         __swab32s(&h->hal_count);
2146         __swab32s(&h->hal_archive_id);
2147         __swab64s(&h->hal_flags);
2148         hai = hai_first(h);
2149         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2150                 lustre_swab_hai(hai);
2151 }
2152
2153 static void lustre_swab_kuch(struct kuc_hdr *l)
2154 {
2155         __swab16s(&l->kuc_magic);
2156         /* __u8 l->kuc_transport */
2157         __swab16s(&l->kuc_msgtype);
2158         __swab16s(&l->kuc_msglen);
2159 }
2160
2161 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2162                                 struct lustre_kernelcomm *lk)
2163 {
2164         struct obd_import  *imp = class_exp2cliimp(exp);
2165         __u32               archive = lk->lk_data;
2166         int                 rc = 0;
2167
2168         if (lk->lk_group != KUC_GRP_HSM) {
2169                 CERROR("Bad copytool group %d\n", lk->lk_group);
2170                 return -EINVAL;
2171         }
2172
2173         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2174                lk->lk_uid, lk->lk_group, lk->lk_flags);
2175
2176         if (lk->lk_flags & LK_FLG_STOP) {
2177                 /* Unregister with the coordinator */
2178                 rc = mdc_ioc_hsm_ct_unregister(imp);
2179         } else {
2180                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2181         }
2182
2183         return rc;
2184 }
2185
2186 /**
2187  * Send a message to any listening copytools
2188  * @param val KUC message (kuc_hdr + hsm_action_list)
2189  * @param len total length of message
2190  */
2191 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2192                                  size_t len, void *val)
2193 {
2194         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2195         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2196         int                      rc;
2197         ENTRY;
2198
2199         if (len < sizeof(*lh) + sizeof(*hal)) {
2200                 CERROR("Short HSM message %zu < %zu\n", len,
2201                        sizeof(*lh) + sizeof(*hal));
2202                 RETURN(-EPROTO);
2203         }
2204         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2205                 lustre_swab_kuch(lh);
2206                 lustre_swab_hal(hal);
2207         } else if (lh->kuc_magic != KUC_MAGIC) {
2208                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2209                 RETURN(-EPROTO);
2210         }
2211
2212         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2213                "on %s\n",
2214                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2215                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2216
2217         /* Broadcast to HSM listeners */
2218         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2219
2220         RETURN(rc);
2221 }
2222
2223 /**
2224  * callback function passed to kuc for re-registering each HSM copytool
2225  * running on MDC, after MDT shutdown/recovery.
2226  * @param data copytool registration data
2227  * @param cb_arg callback argument (obd_import)
2228  */
2229 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2230 {
2231         struct kkuc_ct_data     *kcd = data;
2232         struct obd_import       *imp = (struct obd_import *)cb_arg;
2233         int                      rc;
2234
2235         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2236                 return -EPROTO;
2237
2238         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2239                imp->imp_obd->obd_name, kcd->kcd_archive);
2240         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2241
2242         /* ignore error if the copytool is already registered */
2243         return (rc == -EEXIST) ? 0 : rc;
2244 }
2245
2246 /**
2247  * Re-establish all kuc contexts with MDT
2248  * after MDT shutdown/recovery.
2249  */
2250 static int mdc_kuc_reregister(struct obd_import *imp)
2251 {
2252         /* re-register HSM agents */
2253         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2254                                          mdc_hsm_ct_reregister, imp);
2255 }
2256
2257 static int mdc_set_info_async(const struct lu_env *env,
2258                               struct obd_export *exp,
2259                               u32 keylen, void *key,
2260                               u32 vallen, void *val,
2261                               struct ptlrpc_request_set *set)
2262 {
2263         struct obd_import       *imp = class_exp2cliimp(exp);
2264         int                      rc;
2265         ENTRY;
2266
2267         if (KEY_IS(KEY_READ_ONLY)) {
2268                 if (vallen != sizeof(int))
2269                         RETURN(-EINVAL);
2270
2271                 spin_lock(&imp->imp_lock);
2272                 if (*((int *)val)) {
2273                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2274                         imp->imp_connect_data.ocd_connect_flags |=
2275                                                         OBD_CONNECT_RDONLY;
2276                 } else {
2277                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2278                         imp->imp_connect_data.ocd_connect_flags &=
2279                                                         ~OBD_CONNECT_RDONLY;
2280                 }
2281                 spin_unlock(&imp->imp_lock);
2282
2283                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2284                                        keylen, key, vallen, val, set);
2285                 RETURN(rc);
2286         }
2287         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2288                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2289                                        keylen, key, vallen, val, set);
2290                 RETURN(rc);
2291         }
2292         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2293                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2294                                            val);
2295                 RETURN(rc);
2296         }
2297
2298         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2299                 __u32 *default_easize = val;
2300
2301                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2302                 RETURN(0);
2303         }
2304
2305         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2306         RETURN(rc);
2307 }
2308
2309 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2310                         __u32 keylen, void *key, __u32 *vallen, void *val)
2311 {
2312         int rc = -EINVAL;
2313
2314         if (KEY_IS(KEY_MAX_EASIZE)) {
2315                 __u32 mdsize, *max_easize;
2316
2317                 if (*vallen != sizeof(int))
2318                         RETURN(-EINVAL);
2319                 mdsize = *(__u32 *)val;
2320                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2321                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2322                 max_easize = val;
2323                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2324                 RETURN(0);
2325         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2326                 __u32 *default_easize;
2327
2328                 if (*vallen != sizeof(int))
2329                         RETURN(-EINVAL);
2330                 default_easize = val;
2331                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2332                 RETURN(0);
2333         } else if (KEY_IS(KEY_CONN_DATA)) {
2334                 struct obd_import *imp = class_exp2cliimp(exp);
2335                 struct obd_connect_data *data = val;
2336
2337                 if (*vallen != sizeof(*data))
2338                         RETURN(-EINVAL);
2339
2340                 *data = imp->imp_connect_data;
2341                 RETURN(0);
2342         } else if (KEY_IS(KEY_TGT_COUNT)) {
2343                 *((__u32 *)val) = 1;
2344                 RETURN(0);
2345         }
2346
2347         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2348
2349         RETURN(rc);
2350 }
2351
2352 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2353                      struct ptlrpc_request **request)
2354 {
2355         struct ptlrpc_request *req;
2356         int                    rc;
2357         ENTRY;
2358
2359         *request = NULL;
2360         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2361         if (req == NULL)
2362                 RETURN(-ENOMEM);
2363
2364         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2365         if (rc) {
2366                 ptlrpc_request_free(req);
2367                 RETURN(rc);
2368         }
2369
2370         mdc_pack_body(req, fid, 0, 0, -1, 0);
2371
2372         ptlrpc_request_set_replen(req);
2373
2374         rc = ptlrpc_queue_wait(req);
2375         if (rc)
2376                 ptlrpc_req_finished(req);
2377         else
2378                 *request = req;
2379         RETURN(rc);
2380 }
2381
2382 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2383                             enum obd_import_event event)
2384 {
2385         struct client_obd *cli = &obd->u.cli;
2386         int rc = 0;
2387
2388         LASSERT(imp->imp_obd == obd);
2389
2390         switch (event) {
2391         case IMP_EVENT_DISCON:
2392                 spin_lock(&cli->cl_loi_list_lock);
2393                 cli->cl_avail_grant = 0;
2394                 cli->cl_lost_grant = 0;
2395                 spin_unlock(&cli->cl_loi_list_lock);
2396                 break;
2397         case IMP_EVENT_INACTIVE:
2398                 /*
2399                  * Flush current sequence to make client obtain new one
2400                  * from server in case of disconnect/reconnect.
2401                  */
2402                 down_read(&cli->cl_seq_rwsem);
2403                 if (cli->cl_seq)
2404                         seq_client_flush(cli->cl_seq);
2405                 up_read(&cli->cl_seq_rwsem);
2406
2407                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2408                 break;
2409         case IMP_EVENT_INVALIDATE: {
2410                 struct ldlm_namespace *ns = obd->obd_namespace;
2411                 struct lu_env *env;
2412                 __u16 refcheck;
2413
2414                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2415
2416                 env = cl_env_get(&refcheck);
2417                 if (!IS_ERR(env)) {
2418                         /* Reset grants. All pages go to failing rpcs due to
2419                          * the invalid import.
2420                          */
2421                         osc_io_unplug(env, cli, NULL);
2422
2423                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2424                                                  osc_ldlm_resource_invalidate,
2425                                                  env, 0);
2426                         cl_env_put(env, &refcheck);
2427                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2428                 } else {
2429                         rc = PTR_ERR(env);
2430                 }
2431                 break;
2432         }
2433         case IMP_EVENT_ACTIVE:
2434                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2435                 /* redo the kuc registration after reconnecting */
2436                 if (rc == 0)
2437                         rc = mdc_kuc_reregister(imp);
2438                 break;
2439         case IMP_EVENT_OCD: {
2440                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2441
2442                 if (OCD_HAS_FLAG(ocd, GRANT))
2443                         osc_init_grant(cli, ocd);
2444
2445                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2446                 break;
2447         }
2448         case IMP_EVENT_DEACTIVATE:
2449         case IMP_EVENT_ACTIVATE:
2450                 break;
2451         default:
2452                 CERROR("Unknown import event %x\n", event);
2453                 LBUG();
2454         }
2455         RETURN(rc);
2456 }
2457
2458 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2459                   struct lu_fid *fid, struct md_op_data *op_data)
2460 {
2461         struct client_obd *cli = &exp->exp_obd->u.cli;
2462         int rc = -EIO;
2463
2464         ENTRY;
2465
2466         down_read(&cli->cl_seq_rwsem);
2467         if (cli->cl_seq)
2468                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2469         up_read(&cli->cl_seq_rwsem);
2470
2471         RETURN(rc);
2472 }
2473
2474 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2475 {
2476         struct client_obd *cli = &exp->exp_obd->u.cli;
2477         return &cli->cl_target_uuid;
2478 }
2479
2480 /**
2481  * Determine whether the lock can be canceled before replaying it during
2482  * recovery, non zero value will be return if the lock can be canceled,
2483  * or zero returned for not
2484  */
2485 static int mdc_cancel_weight(struct ldlm_lock *lock)
2486 {
2487         if (lock->l_resource->lr_type != LDLM_IBITS)
2488                 RETURN(0);
2489
2490         /* FIXME: if we ever get into a situation where there are too many
2491          * opened files with open locks on a single node, then we really
2492          * should replay these open locks to reget it */
2493         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2494                 RETURN(0);
2495
2496         RETURN(1);
2497 }
2498
2499 static int mdc_resource_inode_free(struct ldlm_resource *res)
2500 {
2501         if (res->lr_lvb_inode)
2502                 res->lr_lvb_inode = NULL;
2503
2504         return 0;
2505 }
2506
2507 static struct ldlm_valblock_ops inode_lvbo = {
2508         .lvbo_free = mdc_resource_inode_free
2509 };
2510
2511 static int mdc_llog_init(struct obd_device *obd)
2512 {
2513         struct obd_llog_group   *olg = &obd->obd_olg;
2514         struct llog_ctxt        *ctxt;
2515         int                      rc;
2516
2517         ENTRY;
2518
2519         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2520                         &llog_client_ops);
2521         if (rc < 0)
2522                 RETURN(rc);
2523
2524         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2525         llog_initiator_connect(ctxt);
2526         llog_ctxt_put(ctxt);
2527
2528         RETURN(0);
2529 }
2530
2531 static void mdc_llog_finish(struct obd_device *obd)
2532 {
2533         struct llog_ctxt *ctxt;
2534
2535         ENTRY;
2536
2537         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2538         if (ctxt != NULL)
2539                 llog_cleanup(NULL, ctxt);
2540
2541         EXIT;
2542 }
2543
2544 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2545 {
2546         int rc;
2547
2548         ENTRY;
2549
2550         rc = osc_setup_common(obd, cfg);
2551         if (rc < 0)
2552                 RETURN(rc);
2553
2554 #ifdef CONFIG_PROC_FS
2555         obd->obd_vars = lprocfs_mdc_obd_vars;
2556         lprocfs_obd_setup(obd, false);
2557         lprocfs_alloc_md_stats(obd, 0);
2558 #endif
2559
2560         sptlrpc_lprocfs_cliobd_attach(obd);
2561         ptlrpc_lprocfs_register_obd(obd);
2562
2563         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2564
2565         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2566
2567         rc = mdc_llog_init(obd);
2568         if (rc) {
2569                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2570                        obd->obd_name, rc);
2571                 GOTO(err_llog_cleanup, rc);
2572         }
2573
2574         rc = mdc_changelog_cdev_init(obd);
2575         if (rc) {
2576                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2577                        obd->obd_name, rc);
2578                 GOTO(err_changelog_cleanup, rc);
2579         }
2580
2581         RETURN(rc);
2582
2583 err_changelog_cleanup:
2584         mdc_llog_finish(obd);
2585 err_llog_cleanup:
2586         ptlrpc_lprocfs_unregister_obd(obd);
2587         lprocfs_free_md_stats(obd);
2588
2589         osc_cleanup_common(obd);
2590         return rc;
2591 }
2592
2593 /* Initialize the default and maximum LOV EA sizes.  This allows
2594  * us to make MDS RPCs with large enough reply buffers to hold a default
2595  * sized EA without having to calculate this (via a call into the
2596  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2597  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2598  * a large number of stripes is possible.  If a larger reply buffer is
2599  * required it will be reallocated in the ptlrpc layer due to overflow.
2600  */
2601 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2602                             __u32 def_easize)
2603 {
2604         struct obd_device *obd = exp->exp_obd;
2605         struct client_obd *cli = &obd->u.cli;
2606         ENTRY;
2607
2608         if (cli->cl_max_mds_easize < easize)
2609                 cli->cl_max_mds_easize = easize;
2610
2611         if (cli->cl_default_mds_easize < def_easize)
2612                 cli->cl_default_mds_easize = def_easize;
2613
2614         RETURN(0);
2615 }
2616
2617 static int mdc_precleanup(struct obd_device *obd)
2618 {
2619         ENTRY;
2620
2621         osc_precleanup_common(obd);
2622         mdc_changelog_cdev_finish(obd);
2623
2624         obd_cleanup_client_import(obd);
2625         ptlrpc_lprocfs_unregister_obd(obd);
2626         lprocfs_free_md_stats(obd);
2627         mdc_llog_finish(obd);
2628         RETURN(0);
2629 }
2630
2631 static int mdc_cleanup(struct obd_device *obd)
2632 {
2633         return osc_cleanup_common(obd);
2634 }
2635
2636 int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2637 {
2638         struct lustre_cfg *lcfg = buf;
2639         int rc;
2640
2641         rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2642         return (rc > 0 ? 0: rc);
2643 }
2644
2645 static struct obd_ops mdc_obd_ops = {
2646         .o_owner            = THIS_MODULE,
2647         .o_setup            = mdc_setup,
2648         .o_precleanup       = mdc_precleanup,
2649         .o_cleanup          = mdc_cleanup,
2650         .o_add_conn         = client_import_add_conn,
2651         .o_del_conn         = client_import_del_conn,
2652         .o_connect          = client_connect_import,
2653         .o_reconnect        = osc_reconnect,
2654         .o_disconnect       = osc_disconnect,
2655         .o_iocontrol        = mdc_iocontrol,
2656         .o_set_info_async   = mdc_set_info_async,
2657         .o_statfs           = mdc_statfs,
2658         .o_fid_init         = client_fid_init,
2659         .o_fid_fini         = client_fid_fini,
2660         .o_fid_alloc        = mdc_fid_alloc,
2661         .o_import_event     = mdc_import_event,
2662         .o_get_info         = mdc_get_info,
2663         .o_process_config   = mdc_process_config,
2664         .o_get_uuid         = mdc_get_uuid,
2665         .o_quotactl         = mdc_quotactl,
2666 };
2667
2668 static struct md_ops mdc_md_ops = {
2669         .m_get_root         = mdc_get_root,
2670         .m_null_inode       = mdc_null_inode,
2671         .m_close            = mdc_close,
2672         .m_create           = mdc_create,
2673         .m_enqueue          = mdc_enqueue,
2674         .m_getattr          = mdc_getattr,
2675         .m_getattr_name     = mdc_getattr_name,
2676         .m_intent_lock      = mdc_intent_lock,
2677         .m_link             = mdc_link,
2678         .m_rename           = mdc_rename,
2679         .m_setattr          = mdc_setattr,
2680         .m_setxattr         = mdc_setxattr,
2681         .m_getxattr         = mdc_getxattr,
2682         .m_fsync                = mdc_fsync,
2683         .m_file_resync          = mdc_file_resync,
2684         .m_read_page            = mdc_read_page,
2685         .m_unlink           = mdc_unlink,
2686         .m_cancel_unused    = mdc_cancel_unused,
2687         .m_init_ea_size     = mdc_init_ea_size,
2688         .m_set_lock_data    = mdc_set_lock_data,
2689         .m_lock_match       = mdc_lock_match,
2690         .m_get_lustre_md    = mdc_get_lustre_md,
2691         .m_free_lustre_md   = mdc_free_lustre_md,
2692         .m_set_open_replay_data = mdc_set_open_replay_data,
2693         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2694         .m_intent_getattr_async = mdc_intent_getattr_async,
2695         .m_revalidate_lock      = mdc_revalidate_lock
2696 };
2697
2698 static int __init mdc_init(void)
2699 {
2700         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2701                                    LUSTRE_MDC_NAME, &mdc_device_type);
2702 }
2703
2704 static void __exit mdc_exit(void)
2705 {
2706         class_unregister_type(LUSTRE_MDC_NAME);
2707 }
2708
2709 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2710 MODULE_DESCRIPTION("Lustre Metadata Client");
2711 MODULE_VERSION(LUSTRE_VERSION_STRING);
2712 MODULE_LICENSE("GPL");
2713
2714 module_init(mdc_init);
2715 module_exit(mdc_exit);