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  */
  31
  32 #define DEBUG_SUBSYSTEM S_MDC
  33
  34 #include <linux/init.h>
  35 #include <linux/kthread.h>
  36 #include <linux/module.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/user_namespace.h>
  39 #include <linux/utsname.h>
  40 #include <linux/delay.h>
  41 #include <linux/uidgid.h>
  42 #include <linux/device.h>
  43 #include <linux/xarray.h>
  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_compat.h>
  52 #include <lustre_fid.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_ioctl_old.h>
  55 #include <lustre_kernelcomm.h>
  56 #include <lustre_lmv.h>
  57 #include <lustre_log.h>
  58 #include <lustre_osc.h>
  59 #include <lustre_swab.h>
  60 #include <lustre_quota.h>
  61 #include <obd_class.h>
  62
  63 #include "mdc_internal.h"
  64
  65 #define REQUEST_MINOR 244
  66
  67 static int mdc_cleanup(struct obd_device *obd);
  68
  69 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  70 {
  71         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  72         int rc;
  73
  74         /* obd_get_request_slot() ensures that this client has no more
  75          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  76          * against an MDT. */
  77         rc = obd_get_request_slot(cli);
  78         if (rc != 0)
  79                 return rc;
  80
  81         rc = ptlrpc_queue_wait(req);
  82         obd_put_request_slot(cli);
  83
  84         return rc;
  85 }
  86
  87 /*
  88  * Send MDS_GET_ROOT RPC to fetch root FID.
  89  *
  90  * If \a fileset is not NULL it should contain a subdirectory off
  91  * the ROOT/ directory to be mounted on the client. Return the FID
  92  * of the subdirectory to the client to mount onto its mountpoint.
  93  *
  94  * \param[in]   imp     MDC import
  95  * \param[in]   fileset fileset name, which could be NULL
  96  * \param[out]  rootfid root FID of this mountpoint
  97  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  98  *
  99  * \retval      0 on success, negative errno on failure
 100  */
 101 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 102                          struct lu_fid *rootfid)
 103 {
 104         struct ptlrpc_request   *req;
 105         struct mdt_body         *body;
 106         int                      rc;
 107
 108         ENTRY;
 109
 110         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 111                 RETURN(-EOPNOTSUPP);
 112
 113         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 114                                 &RQF_MDS_GET_ROOT);
 115         if (req == NULL)
 116                 RETURN(-ENOMEM);
 117
 118         if (fileset != NULL)
 119                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 120                                      strlen(fileset) + 1);
 121         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 122         if (rc) {
 123                 ptlrpc_request_free(req);
 124                 RETURN(rc);
 125         }
 126         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
 127         if (fileset != NULL) {
 128                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 129
 130                 memcpy(name, fileset, strlen(fileset));
 131         }
 132         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 133         req->rq_send_state = LUSTRE_IMP_FULL;
 134
 135         ptlrpc_request_set_replen(req);
 136
 137         rc = ptlrpc_queue_wait(req);
 138         if (rc)
 139                 GOTO(out, rc);
 140
 141         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 142         if (body == NULL)
 143                 GOTO(out, rc = -EPROTO);
 144
 145         *rootfid = body->mbo_fid1;
 146         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 147                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 148         EXIT;
 149 out:
 150         ptlrpc_req_put(req);
 151
 152         return rc;
 153 }
 154
 155 /*
 156  * This function now is known to always saying that it will receive 4 buffers
 157  * from server. Even for cases when acl_size and md_size is zero, RPC header
 158  * will contain 4 fields and RPC itself will contain zero size fields. This is
 159  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 160  * and thus zero, it shrinks it, making zero size. The same story about
 161  * md_size. And this is course of problem when client waits for smaller number
 162  * of fields. This issue will be fixed later when client gets aware of RPC
 163  * layouts.  --umka
 164  */
 165 static int mdc_getattr_common(struct obd_export *exp,
 166                               struct ptlrpc_request *req,
 167                               struct md_op_data *op_data)
 168 {
 169         struct req_capsule *pill = &req->rq_pill;
 170         struct mdt_body    *body;
 171         void               *eadata;
 172         int                 rc;
 173         ENTRY;
 174
 175         /* Request message already built. */
 176         rc = ptlrpc_queue_wait(req);
 177         if (rc != 0)
 178                 RETURN(rc);
 179
 180         /* sanity check for the reply */
 181         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 182         if (body == NULL)
 183                 RETURN(-EPROTO);
 184
 185         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 186
 187         mdc_update_max_ea_from_body(exp, body);
 188         if (body->mbo_eadatasize != 0) {
 189                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 190                                                       body->mbo_eadatasize);
 191                 if (eadata == NULL)
 192                         RETURN(-EPROTO);
 193         }
 194
 195         /* If encryption context was returned by MDT, put it in op_data
 196          * so that caller can set it on inode and save an extra getxattr.
 197          */
 198         if (op_data && op_data->op_valid & OBD_MD_ENCCTX &&
 199             body->mbo_valid & OBD_MD_ENCCTX) {
 200                 op_data->op_file_encctx =
 201                         req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 202                 op_data->op_file_encctx_size =
 203                         req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
 204                                              RCL_SERVER);
 205         }
 206
 207         RETURN(0);
 208 }
 209
 210 static void mdc_reset_acl_req(struct ptlrpc_request *req)
 211 {
 212         spin_lock(&req->rq_early_free_lock);
 213         sptlrpc_cli_free_repbuf(req);
 214         req->rq_repbuf = NULL;
 215         req->rq_repbuf_len = 0;
 216         req->rq_repdata = NULL;
 217         req->rq_reqdata_len = 0;
 218         spin_unlock(&req->rq_early_free_lock);
 219 }
 220
 221 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 222                        struct ptlrpc_request **request)
 223 {
 224         struct ptlrpc_request *req;
 225         struct obd_device *obd = class_exp2obd(exp);
 226         struct obd_import *imp = class_exp2cliimp(exp);
 227         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 228         int rc;
 229         ENTRY;
 230
 231         /* Single MDS without an LMV case */
 232         if (op_data->op_flags & MF_GET_MDT_IDX) {
 233                 op_data->op_mds = 0;
 234                 RETURN(0);
 235         }
 236
 237         *request = NULL;
 238         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 239         if (req == NULL)
 240                 RETURN(-ENOMEM);
 241
 242         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 243         if (rc) {
 244                 ptlrpc_request_free(req);
 245                 RETURN(rc);
 246         }
 247
 248         /* LU-15245: avoid deadlock with modifying RPCs on MDS_REQUEST_PORTAL */
 249         req->rq_request_portal = MDS_READPAGE_PORTAL;
 250
 251 again:
 252         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, op_data->op_valid,
 253                       op_data->op_mode, -1, 0);
 254         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 255         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 256                              op_data->op_mode);
 257         if (exp_connect_encrypt(exp) && op_data->op_valid & OBD_MD_ENCCTX)
 258                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
 259                                      RCL_SERVER,
 260                                      obd->u.cli.cl_max_mds_easize);
 261         else
 262                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
 263                                      RCL_SERVER, 0);
 264         ptlrpc_request_set_replen(req);
 265
 266         rc = mdc_getattr_common(exp, req, op_data);
 267         if (rc) {
 268                 if (rc == -ERANGE) {
 269                         acl_bufsize = min_t(__u32,
 270                                             imp->imp_connect_data.ocd_max_easize,
 271                                             XATTR_SIZE_MAX);
 272                         mdc_reset_acl_req(req);
 273                         goto again;
 274                 }
 275
 276                 ptlrpc_req_put(req);
 277         } else {
 278                 *request = req;
 279         }
 280
 281         RETURN(rc);
 282 }
 283
 284 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 285                             struct ptlrpc_request **request)
 286 {
 287         struct ptlrpc_request *req;
 288         struct obd_import *imp = class_exp2cliimp(exp);
 289         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 290         int rc;
 291         ENTRY;
 292
 293         *request = NULL;
 294         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 295         if (req == NULL)
 296                 RETURN(-ENOMEM);
 297
 298         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 299                              op_data->op_namelen + 1);
 300
 301         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 302         if (rc) {
 303                 ptlrpc_request_free(req);
 304                 RETURN(rc);
 305         }
 306
 307         if (op_data->op_name) {
 308                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 309                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 310                                 op_data->op_namelen);
 311                 memcpy(name, op_data->op_name, op_data->op_namelen);
 312         }
 313
 314 again:
 315         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, op_data->op_valid,
 316                       op_data->op_mode, op_data->op_suppgids[0], 0);
 317         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 318                              op_data->op_mode);
 319         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 320         req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX, RCL_SERVER, 0);
 321         ptlrpc_request_set_replen(req);
 322         if (op_data->op_bias & MDS_FID_OP) {
 323                 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
 324                                                             &RMF_MDT_BODY);
 325
 326                 if (b) {
 327                         b->mbo_valid |= OBD_MD_NAMEHASH;
 328                         b->mbo_fid2 = op_data->op_fid2;
 329                 }
 330         }
 331
 332         rc = mdc_getattr_common(exp, req, NULL);
 333         if (rc) {
 334                 if (rc == -ERANGE) {
 335                         acl_bufsize = min_t(__u32,
 336                                             imp->imp_connect_data.ocd_max_easize,
 337                                             XATTR_SIZE_MAX);
 338                         mdc_reset_acl_req(req);
 339                         goto again;
 340                 }
 341
 342                 ptlrpc_req_put(req);
 343         } else {
 344                 *request = req;
 345         }
 346
 347         RETURN(rc);
 348 }
 349
 350 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 351                             const struct lu_fid *fid, int opcode, u64 valid,
 352                             const char *xattr_name, const char *input,
 353                             int input_size, int output_size, int flags,
 354                             __u32 suppgid, struct ptlrpc_request **request)
 355 {
 356         struct ptlrpc_request *req;
 357         struct sptlrpc_sepol *sepol;
 358         int   xattr_namelen = 0;
 359         char *tmp;
 360         int   rc;
 361         ENTRY;
 362
 363         *request = NULL;
 364         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 365         if (req == NULL)
 366                 RETURN(-ENOMEM);
 367
 368         if (xattr_name) {
 369                 xattr_namelen = strlen(xattr_name) + 1;
 370                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 371                                      xattr_namelen);
 372         }
 373         if (input_size)
 374                 LASSERT(input);
 375         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 376                              input_size);
 377
 378         /* get SELinux policy info if any */
 379         sepol = sptlrpc_sepol_get(req);
 380         if (IS_ERR(sepol))
 381                 GOTO(err_free_rq, rc = PTR_ERR(sepol));
 382
 383         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 384                              sptlrpc_sepol_size(sepol));
 385
 386         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 387         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 388             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 389                 LIST_HEAD(cancels);
 390                 int count;
 391
 392                 /* Without that packing would fail */
 393                 if (input_size == 0)
 394                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 395                                              RCL_CLIENT, 0);
 396
 397                 count = mdc_resource_get_unused(exp, fid,
 398                                                 &cancels, LCK_EX,
 399                                                 MDS_INODELOCK_XATTR);
 400
 401                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 402                 if (rc)
 403                         GOTO(err_put_sepol, rc);
 404         } else {
 405                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 406                 if (rc)
 407                         GOTO(err_put_sepol, rc);
 408         }
 409
 410         if (opcode == MDS_REINT) {
 411                 struct mdt_rec_setxattr *rec;
 412
 413                 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
 414                              sizeof(struct mdt_rec_reint));
 415                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 416                 rec->sx_opcode = REINT_SETXATTR;
 417                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 418                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 419                 rec->sx_cap = ll_capability_u32(current_cap());
 420                 rec->sx_suppgid1 = suppgid;
 421                 rec->sx_suppgid2 = -1;
 422                 rec->sx_fid    = *fid;
 423                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 424                 rec->sx_time   = ktime_get_real_seconds();
 425                 rec->sx_size   = output_size;
 426                 rec->sx_flags  = flags;
 427         } else {
 428                 mdc_pack_body(&req->rq_pill, fid, valid, output_size,
 429                               suppgid, flags);
 430                 /* Avoid deadlock with modifying RPCs on MDS_REQUEST_PORTAL.
 431                  * See LU-15245.
 432                  */
 433                 req->rq_request_portal = MDS_READPAGE_PORTAL;
 434         }
 435
 436         if (xattr_name) {
 437                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 438                 memcpy(tmp, xattr_name, xattr_namelen);
 439         }
 440         if (input_size) {
 441                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 442                 memcpy(tmp, input, input_size);
 443         }
 444
 445         mdc_file_sepol_pack(&req->rq_pill, sepol);
 446         sptlrpc_sepol_put(sepol);
 447
 448         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 449                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 450                                      RCL_SERVER, output_size);
 451         ptlrpc_request_set_replen(req);
 452
 453         /* make rpc */
 454         if (opcode == MDS_REINT)
 455                 ptlrpc_get_mod_rpc_slot(req);
 456
 457         rc = ptlrpc_queue_wait(req);
 458
 459         if (opcode == MDS_REINT)
 460                 ptlrpc_put_mod_rpc_slot(req);
 461
 462         if (rc)
 463                 ptlrpc_req_put(req);
 464         else
 465                 *request = req;
 466         RETURN(rc);
 467
 468 err_put_sepol:
 469         sptlrpc_sepol_put(sepol);
 470 err_free_rq:
 471         ptlrpc_request_free(req);
 472         RETURN(rc);
 473
 474 }
 475
 476 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 477                         u64 obd_md_valid, const char *name,
 478                         const void *value, size_t value_size,
 479                         unsigned int xattr_flags, u32 suppgid,
 480                         struct ptlrpc_request **req)
 481 {
 482         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 483                 obd_md_valid == OBD_MD_FLXATTRRM);
 484
 485         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 486                                 fid, MDS_REINT, obd_md_valid, name,
 487                                 value, value_size, 0, xattr_flags, suppgid,
 488                                 req);
 489 }
 490
 491 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 492                         u64 obd_md_valid, const char *name, size_t buf_size,
 493                         struct ptlrpc_request **req)
 494 {
 495         struct mdt_body *body;
 496         int rc;
 497
 498         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 499                 obd_md_valid == OBD_MD_FLXATTRLS);
 500
 501         /* Message below is checked in sanity-selinux test_20d
 502          * and sanity-sec test_49
 503          */
 504         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 505                exp->exp_obd->obd_name, name, PFID(fid));
 506         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 507                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 508                               req);
 509         if (rc < 0)
 510                 GOTO(out, rc);
 511
 512         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 513         if (body == NULL)
 514                 GOTO(out, rc = -EPROTO);
 515
 516         /* only detect the xattr size */
 517         if (buf_size == 0) {
 518                 /* LU-11109: Older MDTs do not distinguish
 519                  * between nonexistent xattrs and zero length
 520                  * values in this case. Newer MDTs will return
 521                  * -ENODATA or set OBD_MD_FLXATTR. */
 522                 GOTO(out, rc = body->mbo_eadatasize);
 523         }
 524
 525         if (body->mbo_eadatasize == 0) {
 526                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 527                  * success so that we can distinguish between
 528                  * zero length value and nonexistent xattr.
 529                  *
 530                  * If OBD_MD_FLXATTR is not set then we keep
 531                  * the old behavior and return -ENODATA for
 532                  * getxattr() when mbo_eadatasize is 0. But
 533                  * -ENODATA only makes sense for getxattr()
 534                  * and not for listxattr(). */
 535                 if (body->mbo_valid & OBD_MD_FLXATTR)
 536                         GOTO(out, rc = 0);
 537                 else if (obd_md_valid == OBD_MD_FLXATTR)
 538                         GOTO(out, rc = -ENODATA);
 539                 else
 540                         GOTO(out, rc = 0);
 541         }
 542
 543         GOTO(out, rc = body->mbo_eadatasize);
 544 out:
 545         if (rc < 0) {
 546                 ptlrpc_req_put(*req);
 547                 *req = NULL;
 548         }
 549
 550         return rc;
 551 }
 552
 553 static int mdc_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
 554                              struct obd_export *dt_exp,
 555                              struct obd_export *md_exp,
 556                              struct lustre_md *md)
 557 {
 558         int rc;
 559         ENTRY;
 560
 561         LASSERT(md);
 562         memset(md, 0, sizeof(*md));
 563
 564         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 565         LASSERT(md->body != NULL);
 566
 567         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 568                 if (!S_ISREG(md->body->mbo_mode)) {
 569                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 570                                "regular file, but is not\n");
 571                         GOTO(out, rc = -EPROTO);
 572                 }
 573
 574                 if (md->body->mbo_eadatasize == 0) {
 575                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 576                                "but eadatasize 0\n");
 577                         GOTO(out, rc = -EPROTO);
 578                 }
 579
 580                 md->layout.lb_len = md->body->mbo_eadatasize;
 581                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 582                                                         &RMF_MDT_MD,
 583                                                         md->layout.lb_len);
 584                 if (md->layout.lb_buf == NULL)
 585                         GOTO(out, rc = -EPROTO);
 586         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 587                 const union lmv_mds_md *lmv;
 588                 size_t lmv_size;
 589
 590                 if (!S_ISDIR(md->body->mbo_mode)) {
 591                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 592                                "directory, but is not\n");
 593                         GOTO(out, rc = -EPROTO);
 594                 }
 595
 596                 if (md_exp->exp_obd->obd_type->typ_lu == &mdc_device_type) {
 597                         CERROR("%s: no LMV, upgrading from old version?\n",
 598                                md_exp->exp_obd->obd_name);
 599
 600                         GOTO(out_acl, rc = 0);
 601                 }
 602
 603                 if (md->body->mbo_valid & OBD_MD_MEA) {
 604                         lmv_size = md->body->mbo_eadatasize;
 605                         if (lmv_size == 0) {
 606                                 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 607                                        "but eadatasize 0\n");
 608                                 RETURN(-EPROTO);
 609                         }
 610
 611                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 612                                                            lmv_size);
 613                         if (lmv == NULL)
 614                                 GOTO(out, rc = -EPROTO);
 615
 616                         rc = md_stripe_object_create(md_exp, &md->lsm_obj,
 617                                                      lmv, lmv_size);
 618                         if (rc < 0)
 619                                 GOTO(out, rc);
 620                 }
 621
 622                 /* since 2.12.58 intent_getattr fetches default LMV */
 623                 if (md->body->mbo_valid & OBD_MD_DEFAULT_MEA) {
 624                         lmv_size = req_capsule_get_size(pill,
 625                                                         &RMF_DEFAULT_MDT_MD,
 626                                                         RCL_SERVER);
 627                         lmv = req_capsule_server_sized_get(pill,
 628                                                            &RMF_DEFAULT_MDT_MD,
 629                                                            lmv_size);
 630                         if (!lmv)
 631                                 GOTO(out, rc = -EPROTO);
 632
 633                         rc = md_stripe_object_create(md_exp, &md->def_lsm_obj,
 634                                                      lmv, lmv_size);
 635                         if (rc < 0)
 636                                 GOTO(out, rc);
 637                 }
 638         }
 639         rc = 0;
 640
 641 out_acl:
 642         if (md->body->mbo_valid & OBD_MD_FLACL) {
 643                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 644                  * only when aclsize != 0 there's an actual segment for ACL
 645                  * in reply buffer.
 646                  */
 647                 rc = mdc_unpack_acl(pill, md);
 648                 if (rc)
 649                         GOTO(out, rc);
 650         }
 651
 652         EXIT;
 653 out:
 654         if (rc) {
 655                 lmd_clear_acl(md);
 656                 md_put_lustre_md(md_exp, md);
 657         }
 658
 659         return rc;
 660 }
 661
 662 void mdc_replay_open(struct ptlrpc_request *req)
 663 {
 664         struct md_open_data *mod = req->rq_cb_data;
 665         struct ptlrpc_request *close_req;
 666         struct obd_client_handle *och;
 667         struct lustre_handle old_open_handle = { };
 668         struct mdt_body *body;
 669         struct ldlm_reply *rep;
 670         ENTRY;
 671
 672         if (mod == NULL) {
 673                 DEBUG_REQ(D_ERROR, req,
 674                           "cannot properly replay without open data");
 675                 EXIT;
 676                 return;
 677         }
 678
 679         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 680         LASSERT(body != NULL);
 681
 682         rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
 683         if (rep != NULL && rep->lock_policy_res2 != 0)
 684                 DEBUG_REQ(D_ERROR, req, "Open request replay failed with %ld ",
 685                         (long int)rep->lock_policy_res2);
 686
 687         spin_lock(&req->rq_lock);
 688         och = mod->mod_och;
 689         if (och && och->och_open_handle.cookie)
 690                 req->rq_early_free_repbuf = 1;
 691         else
 692                 req->rq_early_free_repbuf = 0;
 693         spin_unlock(&req->rq_lock);
 694
 695         if (req->rq_early_free_repbuf) {
 696                 struct lustre_handle *file_open_handle;
 697
 698                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 699
 700                 file_open_handle = &och->och_open_handle;
 701                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 702                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 703                 old_open_handle = *file_open_handle;
 704                 *file_open_handle = body->mbo_open_handle;
 705         }
 706
 707         close_req = mod->mod_close_req;
 708         if (close_req) {
 709                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 710                 struct mdt_ioepoch *epoch;
 711
 712                 LASSERT(opc == MDS_CLOSE);
 713                 epoch = req_capsule_client_get(&close_req->rq_pill,
 714                                                &RMF_MDT_EPOCH);
 715                 LASSERT(epoch);
 716
 717                 if (req->rq_early_free_repbuf)
 718                         LASSERT(old_open_handle.cookie ==
 719                                 epoch->mio_open_handle.cookie);
 720
 721                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 722                 epoch->mio_open_handle = body->mbo_open_handle;
 723         }
 724         EXIT;
 725 }
 726
 727 void mdc_commit_open(struct ptlrpc_request *req)
 728 {
 729         struct md_open_data *mod = req->rq_cb_data;
 730         if (mod == NULL)
 731                 return;
 732
 733         /**
 734          * No need to touch md_open_data::mod_och, it holds a reference on
 735          * \var mod and will zero references to each other, \var mod will be
 736          * freed after that when md_open_data::mod_och will put the reference.
 737          */
 738
 739         /**
 740          * Do not let open request to disappear as it still may be needed
 741          * for close rpc to happen (it may happen on evict only, otherwise
 742          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 743          * called), just mark this rpc as committed to distinguish these 2
 744          * cases, see mdc_close() for details. The open request reference will
 745          * be put along with freeing \var mod.
 746          */
 747         ptlrpc_request_addref(req);
 748         spin_lock(&req->rq_lock);
 749         req->rq_committed = 1;
 750         spin_unlock(&req->rq_lock);
 751         req->rq_cb_data = NULL;
 752         obd_mod_put(mod);
 753 }
 754
 755 int mdc_set_open_replay_data(struct obd_export *exp,
 756                              struct obd_client_handle *och,
 757                              struct lookup_intent *it)
 758 {
 759         struct md_open_data     *mod;
 760         struct mdt_rec_create   *rec;
 761         struct mdt_body         *body;
 762         struct ptlrpc_request   *open_req = it->it_request;
 763         struct obd_import       *imp = open_req->rq_import;
 764         ENTRY;
 765
 766         if (!open_req->rq_replay)
 767                 RETURN(0);
 768
 769         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 770         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 771         LASSERT(rec != NULL);
 772         /* Incoming message in my byte order (it's been swabbed). */
 773         /* Outgoing messages always in my byte order. */
 774         LASSERT(body != NULL);
 775
 776         /* Only if the import is replayable, we set replay_open data */
 777         if (och && imp->imp_replayable) {
 778                 mod = obd_mod_alloc();
 779                 if (mod == NULL) {
 780                         DEBUG_REQ(D_ERROR, open_req,
 781                                   "cannot allocate md_open_data");
 782                         RETURN(0);
 783                 }
 784
 785                 /**
 786                  * Take a reference on \var mod, to be freed on mdc_close().
 787                  * It protects \var mod from being freed on eviction (commit
 788                  * callback is called despite rq_replay flag).
 789                  * Another reference for \var och.
 790                  */
 791                 obd_mod_get(mod);
 792                 obd_mod_get(mod);
 793
 794                 spin_lock(&open_req->rq_lock);
 795                 och->och_mod = mod;
 796                 mod->mod_och = och;
 797                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 798                                      it_disposition(it, DISP_OPEN_STRIPE);
 799                 mod->mod_open_req = open_req;
 800                 open_req->rq_cb_data = mod;
 801                 open_req->rq_commit_cb = mdc_commit_open;
 802                 open_req->rq_early_free_repbuf = 1;
 803                 spin_unlock(&open_req->rq_lock);
 804         }
 805
 806         rec->cr_fid2 = body->mbo_fid1;
 807         rec->cr_open_handle_old = body->mbo_open_handle;
 808         open_req->rq_replay_cb = mdc_replay_open;
 809         if (!fid_is_sane(&body->mbo_fid1)) {
 810                 DEBUG_REQ(D_ERROR, open_req,
 811                           "saving replay request with insane FID " DFID,
 812                           PFID(&body->mbo_fid1));
 813                 LBUG();
 814         }
 815
 816         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 817         RETURN(0);
 818 }
 819
 820 static void mdc_free_open(struct md_open_data *mod)
 821 {
 822         int committed = 0;
 823
 824         if (mod->mod_is_create == 0 &&
 825             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 826                 committed = 1;
 827
 828         /**
 829          * No reason to asssert here if the open request has
 830          * rq_replay == 1. It means that mdc_close failed, and
 831          * close request wasn`t sent. It is not fatal to client.
 832          * The worst thing is eviction if the client gets open lock
 833          **/
 834
 835         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 836                   "free open request, rq_replay=%d",
 837                   mod->mod_open_req->rq_replay);
 838
 839         ptlrpc_request_committed(mod->mod_open_req, committed);
 840         if (mod->mod_close_req)
 841                 ptlrpc_request_committed(mod->mod_close_req, committed);
 842 }
 843
 844 static int mdc_clear_open_replay_data(struct obd_export *exp,
 845                                       struct obd_client_handle *och)
 846 {
 847         struct md_open_data *mod = och->och_mod;
 848         ENTRY;
 849
 850         /**
 851          * It is possible to not have \var mod in a case of eviction between
 852          * lookup and ll_file_open().
 853          **/
 854         if (mod == NULL)
 855                 RETURN(0);
 856
 857         LASSERT(mod != LP_POISON);
 858         LASSERT(mod->mod_open_req != NULL);
 859
 860         spin_lock(&mod->mod_open_req->rq_lock);
 861         if (mod->mod_och)
 862                 mod->mod_och->och_open_handle.cookie = 0;
 863         mod->mod_open_req->rq_early_free_repbuf = 0;
 864         spin_unlock(&mod->mod_open_req->rq_lock);
 865         mdc_free_open(mod);
 866
 867         mod->mod_och = NULL;
 868         och->och_mod = NULL;
 869         obd_mod_put(mod);
 870
 871         RETURN(0);
 872 }
 873
 874 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 875                      struct md_open_data *mod, struct ptlrpc_request **request)
 876 {
 877         struct obd_device     *obd = class_exp2obd(exp);
 878         struct ptlrpc_request *req;
 879         struct req_format     *req_fmt;
 880         size_t                 u32_count = 0;
 881         int                    rc;
 882         int                    saved_rc = 0;
 883         ENTRY;
 884
 885         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 886                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 887                op_data->op_bias);
 888
 889         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 890                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 891                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 892                         /* allocate a FID for volatile file */
 893                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 894                                            op_data);
 895                         if (rc < 0) {
 896                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 897                                        obd->obd_name, PFID(&op_data->op_fid1),
 898                                        rc);
 899                                 /* save the errcode and proceed to close */
 900                                 saved_rc = rc;
 901                         }
 902                 }
 903                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 904                         size_t count = op_data->op_data_size / sizeof(__u32);
 905
 906                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 907                                 u32_count = count;
 908                 }
 909         } else {
 910                 req_fmt = &RQF_MDS_CLOSE;
 911         }
 912
 913         *request = NULL;
 914         if (CFS_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 915                 req = NULL;
 916         else
 917                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 918
 919         /* Ensure that this close's handle is fixed up during replay. */
 920         if (likely(mod != NULL)) {
 921                 LASSERTF(mod->mod_open_req != NULL &&
 922                          mod->mod_open_req->rq_type != LI_POISON,
 923                          "POISONED open %px!\n", mod->mod_open_req);
 924
 925                 mod->mod_close_req = req;
 926
 927                 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
 928                 /* We no longer want to preserve this open for replay even
 929                  * though the open was committed. b=3632, b=3633 */
 930                 spin_lock(&mod->mod_open_req->rq_lock);
 931                 mod->mod_open_req->rq_replay = 0;
 932                 spin_unlock(&mod->mod_open_req->rq_lock);
 933         } else {
 934                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 935         }
 936         if (req == NULL) {
 937                 /**
 938                  * TODO: repeat close after errors
 939                  */
 940                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 941                       "dropped when this client unmounts or is evicted\n",
 942                       obd->obd_name, PFID(&op_data->op_fid1));
 943                 GOTO(out, rc = -ENOMEM);
 944         }
 945
 946         if (u32_count > 0)
 947                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 948                                      u32_count * sizeof(__u32));
 949
 950         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 951         if (rc) {
 952                 ptlrpc_request_free(req);
 953                 req = NULL;
 954                 GOTO(out, rc);
 955         }
 956
 957         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 958          * portal whose threads are not taking any DLM locks and are therefore
 959          * always progressing */
 960         req->rq_request_portal = MDS_READPAGE_PORTAL;
 961         ptlrpc_at_set_req_timeout(req);
 962
 963         if (!obd->u.cli.cl_lsom_update ||
 964             !(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 965                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 966                                         OP_XVALID_LAZYBLOCKS);
 967
 968         mdc_close_pack(&req->rq_pill, op_data);
 969
 970         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 971                              obd->u.cli.cl_default_mds_easize);
 972
 973         ptlrpc_request_set_replen(req);
 974
 975         ptlrpc_get_mod_rpc_slot(req);
 976         rc = ptlrpc_queue_wait(req);
 977         ptlrpc_put_mod_rpc_slot(req);
 978
 979         if (req->rq_repmsg == NULL) {
 980                 CDEBUG(D_RPCTRACE, "request %p failed to send: rc = %d\n", req,
 981                        req->rq_status);
 982                 if (rc == 0)
 983                         rc = req->rq_status ?: -EIO;
 984         } else if (rc == 0 || rc == -EAGAIN) {
 985                 struct mdt_body *body;
 986
 987                 rc = lustre_msg_get_status(req->rq_repmsg);
 988                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 989                         DEBUG_REQ(D_ERROR, req,
 990                                   "type = PTL_RPC_MSG_ERR: rc = %d", rc);
 991                         if (rc > 0)
 992                                 rc = -rc;
 993                 }
 994                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 995                 if (body == NULL)
 996                         rc = -EPROTO;
 997         } else if (rc == -ESTALE) {
 998                 /**
 999                  * it can be allowed error after 3633 if open was committed and
1000                  * server failed before close was sent. Let's check if mod
1001                  * exists and return no error in that case
1002                  */
1003                 if (mod) {
1004                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
1005                         LASSERT(mod->mod_open_req != NULL);
1006                         if (mod->mod_open_req->rq_committed)
1007                                 rc = 0;
1008                 }
1009         }
1010
1011 out:
1012         if (mod) {
1013                 if (rc != 0)
1014                         mod->mod_close_req = NULL;
1015                 if (mod->mod_close_req)
1016                         ptlrpc_request_addref(mod->mod_close_req);
1017                 /* Since now, mod is accessed through open_req only,
1018                  * thus close req does not keep a reference on mod anymore. */
1019                 obd_mod_put(mod);
1020         }
1021         *request = req;
1022
1023         RETURN(rc < 0 ? rc : saved_rc);
1024 }
1025
1026 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1027                        u64 offset, struct page **pages, int npages,
1028                        struct ptlrpc_request **request)
1029 {
1030         struct ptlrpc_request   *req;
1031         struct ptlrpc_bulk_desc *desc;
1032         int                      i;
1033         int                      resends = 0;
1034         int                      rc;
1035         ENTRY;
1036
1037         *request = NULL;
1038
1039 restart_bulk:
1040         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1041         if (req == NULL)
1042                 RETURN(-ENOMEM);
1043
1044         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1045         if (rc) {
1046                 ptlrpc_request_free(req);
1047                 RETURN(rc);
1048         }
1049
1050         req->rq_request_portal = MDS_READPAGE_PORTAL;
1051         ptlrpc_at_set_req_timeout(req);
1052
1053         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1054                                     PTLRPC_BULK_PUT_SINK,
1055                                     MDS_BULK_PORTAL,
1056                                     &ptlrpc_bulk_kiov_pin_ops);
1057         if (desc == NULL) {
1058                 ptlrpc_req_put(req);
1059                 RETURN(-ENOMEM);
1060         }
1061
1062         /* NB req now owns desc and will free it when it gets freed */
1063         for (i = 0; i < npages; i++)
1064                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1065                                                  PAGE_SIZE);
1066
1067         mdc_readdir_pack(&req->rq_pill, offset, PAGE_SIZE * npages, fid);
1068
1069         ptlrpc_request_set_replen(req);
1070         rc = ptlrpc_queue_wait(req);
1071         if (rc) {
1072                 ptlrpc_req_put(req);
1073                 if (rc != -ETIMEDOUT)
1074                         RETURN(rc);
1075
1076                 resends++;
1077                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1078                         CERROR("%s: too many resend retries: rc = %d\n",
1079                                exp->exp_obd->obd_name, -EIO);
1080                         RETURN(-EIO);
1081                 }
1082
1083                 /* If a signal interrupts then the timeout returned will
1084                  * not be zero. In that case return -EINTR
1085                  */
1086                 if (msleep_interruptible(resends * 1000))
1087                         RETURN(-EINTR);
1088
1089                 goto restart_bulk;
1090         }
1091
1092         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1093                                           req->rq_bulk->bd_nob_transferred);
1094         if (rc < 0) {
1095                 ptlrpc_req_put(req);
1096                 RETURN(rc);
1097         }
1098
1099         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1100                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1101                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1102                        PAGE_SIZE * npages);
1103                 ptlrpc_req_put(req);
1104                 RETURN(-EPROTO);
1105         }
1106
1107         *request = req;
1108         RETURN(0);
1109 }
1110
1111 static void mdc_release_page(struct page *page, int remove)
1112 {
1113         if (remove) {
1114                 lock_page(page);
1115                 if (likely(page->mapping != NULL))
1116                         cfs_delete_from_page_cache(page);
1117                 unlock_page(page);
1118         }
1119         put_page(page);
1120 }
1121
1122 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1123                                     __u64 *start, __u64 *end, int hash64)
1124 {
1125         /*
1126          * Complement of hash is used as an index so that
1127          * radix_tree_gang_lookup() can be used to find a page with starting
1128          * hash _smaller_ than one we are looking for.
1129          */
1130         unsigned long offset = hash_x_index(*hash, hash64);
1131         struct page *page;
1132         unsigned long flags;
1133         int found;
1134
1135         ll_xa_lock_irqsave(&mapping->i_pages, flags);
1136         found = radix_tree_gang_lookup(&mapping->page_tree,
1137                                        (void **)&page, offset, 1);
1138         if (found > 0 && !ll_xa_is_value(page)) {
1139                 struct lu_dirpage *dp;
1140
1141                 get_page(page);
1142                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1143                 /*
1144                  * In contrast to find_lock_page() we are sure that directory
1145                  * page cannot be truncated (while DLM lock is held) and,
1146                  * hence, can avoid restart.
1147                  *
1148                  * In fact, page cannot be locked here at all, because
1149                  * mdc_read_page_remote does synchronous io.
1150                  */
1151                 wait_on_page_locked(page);
1152                 if (PageUptodate(page)) {
1153                         dp = kmap(page);
1154                         if (BITS_PER_LONG == 32 && hash64) {
1155                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1156                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1157                                 *hash  = *hash >> 32;
1158                         } else {
1159                                 *start = le64_to_cpu(dp->ldp_hash_start);
1160                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1161                         }
1162                         if (unlikely(*start == 1 && *hash == 0))
1163                                 *hash = *start;
1164                         else
1165                                 LASSERTF(*start <= *hash, "start = %#llx"
1166                                          ",end = %#llx,hash = %#llx\n",
1167                                          *start, *end, *hash);
1168                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1169                               " hash %#llx\n", offset, *start, *end, *hash);
1170                         if (*hash > *end) {
1171                                 kunmap(page);
1172                                 mdc_release_page(page, 0);
1173                                 page = NULL;
1174                         } else if (*end != *start && *hash == *end) {
1175                                 /*
1176                                  * upon hash collision, remove this page,
1177                                  * otherwise put page reference, and
1178                                  * mdc_read_page_remote() will issue RPC to
1179                                  * fetch the page we want.
1180                                  */
1181                                 kunmap(page);
1182                                 mdc_release_page(page,
1183                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1184                                 page = NULL;
1185                         }
1186                 } else {
1187                         put_page(page);
1188                         page = ERR_PTR(-EIO);
1189                 }
1190         } else {
1191                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1192                 page = NULL;
1193         }
1194         return page;
1195 }
1196
1197 /*
1198  * Adjust a set of pages, each page containing an array of lu_dirpages,
1199  * so that each page can be used as a single logical lu_dirpage.
1200  *
1201  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1202  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1203  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1204  * value is used as a cookie to request the next lu_dirpage in a
1205  * directory listing that spans multiple pages (two in this example):
1206  *   ________
1207  *  |        |
1208  * .|--------v-------   -----.
1209  * |s|e|f|p|ent|ent| ... |ent|
1210  * '--|--------------   -----'   Each PAGE contains a single
1211  *    '------.                   lu_dirpage.
1212  * .---------v-------   -----.
1213  * |s|e|f|p|ent| 0 | ... | 0 |
1214  * '-----------------   -----'
1215  *
1216  * However, on hosts where the native VM page size (PAGE_SIZE) is
1217  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1218  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1219  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1220  * after it in the same PAGE (arrows simplified for brevity, but
1221  * in general e0==s1, e1==s2, etc.):
1222  *
1223  * .--------------------   -----.
1224  * |s0|e0|f0|p|ent|ent| ... |ent|
1225  * |---v----------------   -----|
1226  * |s1|e1|f1|p|ent|ent| ... |ent|
1227  * |---v----------------   -----|  Here, each PAGE contains
1228  *             ...                 multiple lu_dirpages.
1229  * |---v----------------   -----|
1230  * |s'|e'|f'|p|ent|ent| ... |ent|
1231  * '---|----------------   -----'
1232  *     v
1233  * .----------------------------.
1234  * |        next PAGE           |
1235  *
1236  * This structure is transformed into a single logical lu_dirpage as follows:
1237  *
1238  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1239  *   labeled 'next PAGE'.
1240  *
1241  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1242  *   a hash collision with the next page exists.
1243  *
1244  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1245  *   to the first entry of the next lu_dirpage.
1246  */
1247 #if PAGE_SIZE > LU_PAGE_SIZE
1248 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1249 {
1250         int i;
1251
1252         for (i = 0; i < cfs_pgs; i++) {
1253                 struct lu_dirpage *dp = kmap(pages[i]);
1254                 struct lu_dirpage *first = dp;
1255                 struct lu_dirent *end_dirent = NULL;
1256                 struct lu_dirent *ent;
1257                 __u64 hash_end = dp->ldp_hash_end;
1258                 __u32 flags = dp->ldp_flags;
1259
1260                 while (--lu_pgs > 0) {
1261                         ent = lu_dirent_start(dp);
1262                         for (end_dirent = ent; ent != NULL;
1263                              end_dirent = ent, ent = lu_dirent_next(ent));
1264
1265                         /* Advance dp to next lu_dirpage. */
1266                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1267
1268                         /* Check if we've reached the end of the PAGE. */
1269                         if (!((unsigned long)dp & ~PAGE_MASK))
1270                                 break;
1271
1272                         /* Save the hash and flags of this lu_dirpage. */
1273                         hash_end = dp->ldp_hash_end;
1274                         flags = dp->ldp_flags;
1275
1276                         /* Check if lu_dirpage contains no entries. */
1277                         if (end_dirent == NULL)
1278                                 break;
1279
1280                         /* Enlarge the end entry lde_reclen from 0 to
1281                          * first entry of next lu_dirpage. */
1282                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1283                         end_dirent->lde_reclen =
1284                                 cpu_to_le16((char *)(dp->ldp_entries) -
1285                                             (char *)end_dirent);
1286                 }
1287
1288                 first->ldp_hash_end = hash_end;
1289                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1290                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1291
1292                 kunmap(pages[i]);
1293         }
1294         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1295 }
1296 #else
1297 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1298 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1299
1300 /* parameters for readdir page */
1301 struct readpage_param {
1302         struct md_op_data       *rp_mod;
1303         __u64                   rp_off;
1304         int                     rp_hash64;
1305         struct obd_export       *rp_exp;
1306 };
1307
1308 /**
1309  * Read pages from server.
1310  *
1311  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1312  * a header lu_dirpage which describes the start/end hash, and whether this
1313  * page is empty (contains no dir entry) or hash collide with next page.
1314  * After client receives reply, several pages will be integrated into dir page
1315  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1316  * lu_dirpage for this integrated page will be adjusted.
1317  **/
1318 static int ll_mdc_read_page_remote(void *data, struct page *page0)
1319 {
1320         struct readpage_param *rp = data;
1321         struct page **page_pool;
1322         struct page *page;
1323         struct lu_dirpage *dp;
1324         struct md_op_data *op_data = rp->rp_mod;
1325         struct ptlrpc_request *req;
1326         int max_pages;
1327         struct inode *inode;
1328         struct lu_fid *fid;
1329         int rd_pgs = 0; /* number of pages actually read */
1330         int npages;
1331         int i;
1332         int rc;
1333         ENTRY;
1334
1335         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1336         inode = op_data->op_data;
1337         fid = &op_data->op_fid1;
1338         LASSERT(inode != NULL);
1339
1340         OBD_ALLOC_PTR_ARRAY_LARGE(page_pool, max_pages);
1341         if (page_pool != NULL) {
1342                 page_pool[0] = page0;
1343         } else {
1344                 page_pool = &page0;
1345                 max_pages = 1;
1346         }
1347
1348         for (npages = 1; npages < max_pages; npages++) {
1349                 page = page_cache_alloc(inode->i_mapping);
1350                 if (page == NULL)
1351                         break;
1352                 page_pool[npages] = page;
1353         }
1354
1355         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1356         if (rc < 0) {
1357                 /* page0 is special, which was added into page cache early */
1358                 cfs_delete_from_page_cache(page0);
1359         } else {
1360                 int lu_pgs;
1361
1362                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1363                         PAGE_SHIFT;
1364                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1365                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1366
1367                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1368
1369                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1370
1371                 SetPageUptodate(page0);
1372         }
1373         unlock_page(page0);
1374
1375         ptlrpc_req_put(req);
1376         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1377         for (i = 1; i < npages; i++) {
1378                 unsigned long   offset;
1379                 __u64           hash;
1380                 int ret;
1381
1382                 page = page_pool[i];
1383
1384                 if (rc < 0 || i >= rd_pgs) {
1385                         put_page(page);
1386                         continue;
1387                 }
1388
1389                 SetPageUptodate(page);
1390
1391                 dp = kmap(page);
1392                 hash = le64_to_cpu(dp->ldp_hash_start);
1393                 kunmap(page);
1394
1395                 offset = hash_x_index(hash, rp->rp_hash64);
1396
1397                 prefetchw(&page->flags);
1398                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1399                                             GFP_KERNEL);
1400                 if (ret == 0)
1401                         unlock_page(page);
1402                 else
1403                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1404                                " rc = %d\n", offset, ret);
1405                 put_page(page);
1406         }
1407
1408         if (page_pool != &page0)
1409                 OBD_FREE_PTR_ARRAY_LARGE(page_pool, max_pages);
1410
1411         RETURN(rc);
1412 }
1413
1414 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
1415 static inline int mdc_read_folio_remote(struct file *file, struct folio *folio)
1416 {
1417         return ll_mdc_read_page_remote(file->private_data,
1418                                        folio_page(folio, 0));
1419 }
1420 #else
1421 #define mdc_read_folio_remote   ll_mdc_read_page_remote
1422 #endif
1423
1424 /**
1425  * Read dir page from cache first, if it can not find it, read it from
1426  * server and add into the cache.
1427  *
1428  * \param[in] exp       MDC export
1429  * \param[in] op_data   client MD stack parameters, transfering parameters
1430  *                      between different layers on client MD stack.
1431  * \param[in] mrinfo    callback required for ldlm lock enqueue during
1432  *                      read page
1433  * \param[in] hash_offset the hash offset of the page to be read
1434  * \param[in] ppage     the page to be read
1435  *
1436  * retval               = 0 get the page successfully
1437  *                      errno(<0) get the page failed
1438  */
1439 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1440                          struct md_readdir_info *mrinfo, __u64 hash_offset,
1441                          struct page **ppage)
1442 {
1443         struct lookup_intent    it = { .it_op = IT_READDIR };
1444         struct page             *page;
1445         struct inode            *dir = op_data->op_data;
1446         struct address_space    *mapping;
1447         struct lu_dirpage       *dp;
1448         __u64                   start = 0;
1449         __u64                   end = 0;
1450         struct lustre_handle    lockh;
1451         struct ptlrpc_request   *enq_req = NULL;
1452         struct readpage_param   rp_param;
1453         int rc;
1454
1455         ENTRY;
1456
1457         *ppage = NULL;
1458
1459         LASSERT(dir != NULL);
1460         mapping = dir->i_mapping;
1461
1462         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1463                              mrinfo->mr_blocking_ast, 0);
1464         if (enq_req != NULL)
1465                 ptlrpc_req_put(enq_req);
1466
1467         if (rc < 0) {
1468                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1469                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1470                 RETURN(rc);
1471         }
1472
1473         rc = 0;
1474         lockh.cookie = it.it_lock_handle;
1475         mdc_set_lock_data(exp, &lockh, dir, NULL);
1476
1477         rp_param.rp_off = hash_offset;
1478         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1479         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1480                                rp_param.rp_hash64);
1481         if (IS_ERR(page)) {
1482                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1483                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1484                        rp_param.rp_off, PTR_ERR(page));
1485                 GOTO(out_unlock, rc = PTR_ERR(page));
1486         } else if (page != NULL) {
1487                 /*
1488                  * XXX nikita: not entirely correct handling of a corner case:
1489                  * suppose hash chain of entries with hash value HASH crosses
1490                  * border between pages P0 and P1. First both P0 and P1 are
1491                  * cached, seekdir() is called for some entry from the P0 part
1492                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1493                  * happens and finds P1, as it starts with matching hash
1494                  * value. Remaining entries from P0 part of the chain are
1495                  * skipped. (Is that really a bug?)
1496                  *
1497                  * Possible solutions: 0. don't cache P1 is such case, handle
1498                  * it as an "overflow" page. 1. invalidate all pages at
1499                  * once. 2. use HASH|1 as an index for P1.
1500                  */
1501                 GOTO(hash_collision, page);
1502         }
1503
1504         rp_param.rp_exp = exp;
1505         rp_param.rp_mod = op_data;
1506         page = ll_read_cache_page(mapping,
1507                                   hash_x_index(rp_param.rp_off,
1508                                                rp_param.rp_hash64),
1509                                   mdc_read_folio_remote, &rp_param);
1510         if (IS_ERR(page)) {
1511                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1512                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1513                        rp_param.rp_off, PTR_ERR(page));
1514                 GOTO(out_unlock, rc = PTR_ERR(page));
1515         }
1516
1517         wait_on_page_locked(page);
1518         (void)kmap(page);
1519         if (!PageUptodate(page)) {
1520                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1521                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1522                        rp_param.rp_off, -5);
1523                 goto fail;
1524         }
1525         if (!PageChecked(page))
1526                 SetPageChecked(page);
1527         if (PageError(page)) {
1528                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1529                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1530                        rp_param.rp_off, -5);
1531                 goto fail;
1532         }
1533
1534 hash_collision:
1535         dp = page_address(page);
1536         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1537                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1538                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1539                 rp_param.rp_off = hash_offset >> 32;
1540         } else {
1541                 start = le64_to_cpu(dp->ldp_hash_start);
1542                 end   = le64_to_cpu(dp->ldp_hash_end);
1543                 rp_param.rp_off = hash_offset;
1544         }
1545         if (end == start) {
1546                 LASSERT(start == rp_param.rp_off);
1547                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1548 #if BITS_PER_LONG == 32
1549                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1550                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1551                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1552 #endif
1553
1554                 /*
1555                  * Fetch whole overflow chain...
1556                  *
1557                  * XXX not yet.
1558                  */
1559                 goto fail;
1560         }
1561         *ppage = page;
1562 out_unlock:
1563         ldlm_lock_decref(&lockh, it.it_lock_mode);
1564         return rc;
1565 fail:
1566         kunmap(page);
1567         mdc_release_page(page, 1);
1568         rc = -EIO;
1569         goto out_unlock;
1570 }
1571
1572 static int mdc_statfs_interpret(const struct lu_env *env,
1573                                 struct ptlrpc_request *req, void *args, int rc)
1574 {
1575         struct obd_info *oinfo = args;
1576         struct obd_statfs *osfs;
1577
1578         if (!rc) {
1579                 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1580                 if (!osfs)
1581                         return -EPROTO;
1582
1583                 oinfo->oi_osfs = osfs;
1584
1585                 CDEBUG(D_CACHE, "blocks=%llu free=%llu avail=%llu "
1586                        "objects=%llu free=%llu state=%x\n",
1587                         osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
1588                         osfs->os_files, osfs->os_ffree, osfs->os_state);
1589         }
1590
1591         oinfo->oi_cb_up(oinfo, rc);
1592
1593         return rc;
1594 }
1595
1596 static int mdc_statfs_async(struct obd_export *exp,
1597                             struct obd_info *oinfo, time64_t max_age,
1598                             struct ptlrpc_request_set *unused)
1599 {
1600         struct ptlrpc_request *req;
1601         struct obd_info *aa;
1602
1603         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_STATFS,
1604                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1605         if (IS_ERR(req))
1606                 return PTR_ERR(req);
1607
1608         ptlrpc_request_set_replen(req);
1609         req->rq_interpret_reply = mdc_statfs_interpret;
1610
1611         aa = ptlrpc_req_async_args(aa, req);
1612         *aa = *oinfo;
1613
1614         ptlrpcd_add_req(req);
1615
1616         return 0;
1617 }
1618
1619 static int mdc_statfs(const struct lu_env *env,
1620                       struct obd_export *exp, struct obd_statfs *osfs,
1621                       time64_t max_age, __u32 flags)
1622 {
1623         struct obd_device *obd = class_exp2obd(exp);
1624         struct req_format *fmt;
1625         struct ptlrpc_request *req;
1626         struct obd_statfs *msfs;
1627         struct obd_import *imp, *imp0;
1628         int rc;
1629         ENTRY;
1630
1631         /*
1632          * Since the request might also come from lprocfs, so we need
1633          * sync this with client_disconnect_export Bug15684
1634          */
1635         with_imp_locked(obd, imp0, rc)
1636                 imp = class_import_get(imp0);
1637         if (rc)
1638                 RETURN(rc);
1639
1640         fmt = &RQF_MDS_STATFS;
1641         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1642             (flags & OBD_STATFS_SUM))
1643                 fmt = &RQF_MDS_STATFS_NEW;
1644         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1645                                         MDS_STATFS);
1646         if (IS_ERR(req))
1647                 GOTO(output, rc = PTR_ERR(req));
1648         req->rq_allow_intr = 1;
1649
1650         if ((flags & OBD_STATFS_SUM) &&
1651             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1652                 /* request aggregated states */
1653                 struct mdt_body *body;
1654
1655                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1656                 if (body == NULL)
1657                         GOTO(out, rc = -EPROTO);
1658                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1659         }
1660
1661         ptlrpc_request_set_replen(req);
1662
1663         if (flags & OBD_STATFS_NODELAY) {
1664                 /* procfs requests not want stay in wait for avoid deadlock */
1665                 req->rq_no_resend = 1;
1666                 req->rq_no_delay = 1;
1667         }
1668
1669         rc = ptlrpc_queue_wait(req);
1670         if (rc) {
1671                 /* check connection error first */
1672                 if (imp->imp_connect_error)
1673                         rc = imp->imp_connect_error;
1674                 GOTO(out, rc);
1675         }
1676
1677         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1678         if (msfs == NULL)
1679                 GOTO(out, rc = -EPROTO);
1680
1681         *osfs = *msfs;
1682         EXIT;
1683 out:
1684         ptlrpc_req_put(req);
1685 output:
1686         class_import_put(imp);
1687         return rc;
1688 }
1689
1690 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1691 {
1692         __u32 keylen, vallen;
1693         void *key;
1694         int rc;
1695
1696         if (gf->gf_pathlen < 2)
1697                 RETURN(-EOVERFLOW);
1698
1699         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1700         keylen = round_up(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1701                                 sizeof(struct lu_fid), 8);
1702         OBD_ALLOC(key, keylen);
1703         if (key == NULL)
1704                 RETURN(-ENOMEM);
1705         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1706         memcpy(key + round_up(sizeof(KEY_FID2PATH), 8), gf, sizeof(*gf));
1707         memcpy(key + round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*gf),
1708                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1709         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1710                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1711
1712         if (!fid_is_sane(&gf->gf_fid))
1713                 GOTO(out, rc = -EINVAL);
1714
1715         /* Val is struct getinfo_fid2path result plus path */
1716         vallen = sizeof(*gf) + gf->gf_pathlen;
1717
1718         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1719         if (rc != 0 && rc != -EREMOTE)
1720                 GOTO(out, rc);
1721
1722         if (vallen <= sizeof(*gf))
1723                 GOTO(out, rc = -EPROTO);
1724         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1725                 GOTO(out, rc = -EOVERFLOW);
1726
1727         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %.*s\n",
1728                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1729                /* only log the first 512 characters of the path */
1730                512, gf->gf_u.gf_path);
1731
1732 out:
1733         OBD_FREE(key, keylen);
1734         return rc;
1735 }
1736
1737 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1738                                 struct hsm_progress_kernel *hpk)
1739 {
1740         struct obd_import               *imp = class_exp2cliimp(exp);
1741         struct hsm_progress_kernel      *req_hpk;
1742         struct ptlrpc_request           *req;
1743         int                              rc;
1744         ENTRY;
1745
1746         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1747                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1748         if (IS_ERR(req))
1749                 RETURN(PTR_ERR(req));
1750
1751         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1752
1753         /* Copy hsm_progress struct */
1754         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1755         if (req_hpk == NULL)
1756                 GOTO(out, rc = -EPROTO);
1757
1758         *req_hpk = *hpk;
1759         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1760
1761         ptlrpc_request_set_replen(req);
1762
1763         ptlrpc_get_mod_rpc_slot(req);
1764         rc = ptlrpc_queue_wait(req);
1765         ptlrpc_put_mod_rpc_slot(req);
1766
1767         GOTO(out, rc);
1768 out:
1769         ptlrpc_req_put(req);
1770         return rc;
1771 }
1772
1773 /**
1774  * Send hsm_ct_register to MDS
1775  *
1776  * \param[in]   imp             import
1777  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1778  *                              else it is the count of archive_ids
1779  * \param[in]   archives        if in bitmap format, it is NULL,
1780  *                              else it is archive_id lists
1781  */
1782 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1783                                    __u32 *archives)
1784 {
1785         struct ptlrpc_request *req;
1786         __u32 *archive_array;
1787         size_t archives_size;
1788         int rc;
1789         ENTRY;
1790
1791         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1792         if (req == NULL)
1793                 RETURN(-ENOMEM);
1794
1795         if (archives != NULL)
1796                 archives_size = sizeof(*archive_array) * archive_count;
1797         else
1798                 archives_size = sizeof(archive_count);
1799
1800         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1801                              RCL_CLIENT, archives_size);
1802
1803         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1804         if (rc) {
1805                 ptlrpc_request_free(req);
1806                 RETURN(-ENOMEM);
1807         }
1808
1809         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1810
1811         archive_array = req_capsule_client_get(&req->rq_pill,
1812                                                &RMF_MDS_HSM_ARCHIVE);
1813         if (archive_array == NULL)
1814                 GOTO(out, rc = -EPROTO);
1815
1816         if (archives != NULL)
1817                 memcpy(archive_array, archives, archives_size);
1818         else
1819                 *archive_array = archive_count;
1820
1821         ptlrpc_request_set_replen(req);
1822
1823         rc = mdc_queue_wait(req);
1824         GOTO(out, rc);
1825 out:
1826         ptlrpc_req_put(req);
1827         return rc;
1828 }
1829
1830 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1831                                       struct md_op_data *op_data)
1832 {
1833         struct hsm_current_action       *hca = op_data->op_data;
1834         struct hsm_current_action       *req_hca;
1835         struct ptlrpc_request           *req;
1836         int                              rc;
1837         ENTRY;
1838
1839         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1840                                    &RQF_MDS_HSM_ACTION);
1841         if (req == NULL)
1842                 RETURN(-ENOMEM);
1843
1844         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1845         if (rc) {
1846                 ptlrpc_request_free(req);
1847                 RETURN(rc);
1848         }
1849
1850         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1851                       op_data->op_suppgids[0], 0);
1852
1853         ptlrpc_request_set_replen(req);
1854
1855         rc = mdc_queue_wait(req);
1856         if (rc)
1857                 GOTO(out, rc);
1858
1859         req_hca = req_capsule_server_get(&req->rq_pill,
1860                                          &RMF_MDS_HSM_CURRENT_ACTION);
1861         if (req_hca == NULL)
1862                 GOTO(out, rc = -EPROTO);
1863
1864         *hca = *req_hca;
1865
1866         EXIT;
1867 out:
1868         ptlrpc_req_put(req);
1869         return rc;
1870 }
1871
1872 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1873 {
1874         struct ptlrpc_request   *req;
1875         int                      rc;
1876         ENTRY;
1877
1878         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1879                                         LUSTRE_MDS_VERSION,
1880                                         MDS_HSM_CT_UNREGISTER);
1881         if (IS_ERR(req))
1882                 RETURN(PTR_ERR(req));
1883
1884         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1885
1886         ptlrpc_request_set_replen(req);
1887
1888         rc = mdc_queue_wait(req);
1889
1890         ptlrpc_req_put(req);
1891
1892         RETURN(rc);
1893 }
1894
1895 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1896                                  struct md_op_data *op_data)
1897 {
1898         struct hsm_user_state   *hus = op_data->op_data;
1899         struct hsm_user_state   *req_hus;
1900         struct ptlrpc_request   *req;
1901         int                      rc;
1902         ENTRY;
1903
1904         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1905                                    &RQF_MDS_HSM_STATE_GET);
1906         if (req == NULL)
1907                 RETURN(-ENOMEM);
1908
1909         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1910         if (rc != 0) {
1911                 ptlrpc_request_free(req);
1912                 RETURN(rc);
1913         }
1914
1915         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1916                       op_data->op_suppgids[0], 0);
1917
1918         ptlrpc_request_set_replen(req);
1919
1920         rc = mdc_queue_wait(req);
1921         if (rc)
1922                 GOTO(out, rc);
1923
1924         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1925         if (req_hus == NULL)
1926                 GOTO(out, rc = -EPROTO);
1927
1928         *hus = *req_hus;
1929
1930         EXIT;
1931 out:
1932         ptlrpc_req_put(req);
1933         return rc;
1934 }
1935
1936 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1937                                  struct md_op_data *op_data)
1938 {
1939         struct hsm_state_set    *hss = op_data->op_data;
1940         struct hsm_state_set    *req_hss;
1941         struct ptlrpc_request   *req;
1942         int                      rc;
1943         ENTRY;
1944
1945         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1946                                    &RQF_MDS_HSM_STATE_SET);
1947         if (req == NULL)
1948                 RETURN(-ENOMEM);
1949
1950         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1951         if (rc) {
1952                 ptlrpc_request_free(req);
1953                 RETURN(rc);
1954         }
1955
1956         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1957                       op_data->op_suppgids[0], 0);
1958
1959         /* Copy states */
1960         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1961         if (req_hss == NULL)
1962                 GOTO(out, rc = -EPROTO);
1963         *req_hss = *hss;
1964
1965         ptlrpc_request_set_replen(req);
1966
1967         ptlrpc_get_mod_rpc_slot(req);
1968         rc = ptlrpc_queue_wait(req);
1969         ptlrpc_put_mod_rpc_slot(req);
1970
1971         GOTO(out, rc);
1972 out:
1973         ptlrpc_req_put(req);
1974         return rc;
1975 }
1976
1977 /* For RESTORE and RELEASE the mdt will take EX lock on the file layout.
1978  * So we can use early cancel on client side locks for that resource.
1979  */
1980 static inline int mdc_hsm_request_lock_to_cancel(struct obd_export *exp,
1981                                                  struct hsm_user_request *hur,
1982                                                  struct list_head *cancels)
1983 {
1984         struct hsm_user_item *hui = &hur->hur_user_item[0];
1985         struct hsm_request *req_hr = &hur->hur_request;
1986         int count = 0;
1987         int i;
1988
1989         if (req_hr->hr_action != HUA_RESTORE &&
1990             req_hr->hr_action != HUA_RELEASE)
1991                 return 0;
1992
1993         for (i = 0; i < req_hr->hr_itemcount; i++, hui++) {
1994                 if (!fid_is_sane(&hui->hui_fid))
1995                         continue;
1996                 count += mdc_resource_get_unused(exp, &hui->hui_fid, cancels,
1997                                                  LCK_EX, MDS_INODELOCK_LAYOUT);
1998         }
1999
2000         return count;
2001 }
2002
2003 static int mdc_ioc_hsm_request(struct obd_export *exp,
2004                                struct hsm_user_request *hur)
2005 {
2006         struct obd_import *imp = class_exp2cliimp(exp);
2007         struct ptlrpc_request *req;
2008         struct hsm_request *req_hr;
2009         struct hsm_user_item *req_hui;
2010         char *req_opaque;
2011         LIST_HEAD(cancels);
2012         int count;
2013         int rc;
2014         ENTRY;
2015
2016         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2017         if (req == NULL)
2018                 RETURN(-ENOMEM);
2019
2020         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2021                              hur->hur_request.hr_itemcount
2022                              * sizeof(struct hsm_user_item));
2023         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2024                              hur->hur_request.hr_data_len);
2025
2026         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2027         if (rc) {
2028                 ptlrpc_request_free(req);
2029                 RETURN(rc);
2030         }
2031
2032         /* Cancel existing locks */
2033         count = mdc_hsm_request_lock_to_cancel(exp, hur, &cancels);
2034         ldlm_cli_cancel_list(&cancels, count, NULL, 0);
2035         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
2036
2037         /* Copy hsm_request struct */
2038         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2039         if (req_hr == NULL)
2040                 GOTO(out, rc = -EPROTO);
2041         *req_hr = hur->hur_request;
2042
2043         /* Copy hsm_user_item structs */
2044         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2045         if (req_hui == NULL)
2046                 GOTO(out, rc = -EPROTO);
2047         memcpy(req_hui, hur->hur_user_item,
2048                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2049
2050         /* Copy opaque field */
2051         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2052         if (req_opaque == NULL)
2053                 GOTO(out, rc = -EPROTO);
2054         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2055
2056         ptlrpc_request_set_replen(req);
2057
2058         ptlrpc_get_mod_rpc_slot(req);
2059         rc = ptlrpc_queue_wait(req);
2060         ptlrpc_put_mod_rpc_slot(req);
2061
2062         GOTO(out, rc);
2063
2064 out:
2065         ptlrpc_req_put(req);
2066         return rc;
2067 }
2068
2069 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2070                                 struct lustre_kernelcomm *lk);
2071 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2072                         struct obd_quotactl *oqctl)
2073 {
2074         struct ptlrpc_request *req;
2075         struct obd_quotactl *oqc;
2076         int rc;
2077         ENTRY;
2078
2079         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_QUOTACTL);
2080         if (req == NULL)
2081                 RETURN(-ENOMEM);
2082
2083
2084         if (LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd))
2085                 req_capsule_set_size(&req->rq_pill,
2086                                      &RMF_OBD_QUOTACTL,
2087                                      RCL_CLIENT,
2088                                      sizeof(*oqc) + LOV_MAXPOOLNAME + 1);
2089
2090         if (oqctl->qc_cmd == LUSTRE_Q_ITERQUOTA ||
2091             oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA)
2092                 req_capsule_set_size(&req->rq_pill, &RMF_OBD_QUOTA_ITER,
2093                                      RCL_SERVER, LQUOTA_ITER_BUFLEN);
2094         else
2095                 req_capsule_set_size(&req->rq_pill, &RMF_OBD_QUOTA_ITER,
2096                                      RCL_SERVER, 0);
2097
2098         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION,
2099                                  MDS_QUOTACTL);
2100         if (rc) {
2101                 ptlrpc_request_free(req);
2102                 RETURN(rc);
2103         }
2104
2105         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2106         QCTL_COPY(oqc, oqctl);
2107
2108         ptlrpc_request_set_replen(req);
2109         ptlrpc_at_set_req_timeout(req);
2110
2111         rc = ptlrpc_queue_wait(req);
2112         if (rc) {
2113                 CERROR("%s: ptlrpc_queue_wait failed: rc = %d\n",
2114                        exp->exp_obd->obd_name, rc);
2115                 GOTO(out, rc);
2116         }
2117
2118         if (req->rq_repmsg &&
2119             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2120                 struct list_head *lst = (struct list_head *)oqctl->qc_iter_list;
2121
2122                 QCTL_COPY(oqctl, oqc);
2123
2124                 if (oqctl->qc_cmd == LUSTRE_Q_ITERQUOTA ||
2125                     oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA) {
2126                         void *buffer;
2127                         struct lquota_iter *iter;
2128
2129                         buffer = req_capsule_server_get(&req->rq_pill,
2130                                                         &RMF_OBD_QUOTA_ITER);
2131
2132                         if (buffer == NULL) {
2133                                 CDEBUG(D_QUOTA, "%s: no buffer in iter req\n",
2134                                        exp->exp_obd->obd_name);
2135
2136                                 rc = -EPROTO;
2137                                 GOTO(out, rc);
2138                         }
2139
2140                         OBD_ALLOC_LARGE(iter,
2141                                sizeof(struct lquota_iter) + LQUOTA_ITER_BUFLEN);
2142                         if (iter == NULL)
2143                                 GOTO(out, rc = -ENOMEM);
2144
2145                         INIT_LIST_HEAD(&iter->li_link);
2146                         list_add(&iter->li_link, lst);
2147
2148                         memcpy(iter->li_buffer, buffer, LQUOTA_ITER_BUFLEN);
2149                         iter->li_md_size = oqctl->qc_iter_md_buflen;
2150                         if (oqctl->qc_cmd == LUSTRE_Q_ITERQUOTA)
2151                                 iter->li_dt_size = oqctl->qc_iter_dt_buflen;
2152
2153                         oqctl->qc_iter_md_buflen = 0;
2154                         oqctl->qc_iter_dt_buflen = 0;
2155                 }
2156         } else if (!rc) {
2157                 rc = -EPROTO;
2158                 CERROR("%s: cannot unpack obd_quotactl: rc = %d\n",
2159                         exp->exp_obd->obd_name, rc);
2160         }
2161 out:
2162         ptlrpc_req_put(req);
2163
2164         RETURN(rc);
2165 }
2166
2167 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2168                                 struct md_op_data *op_data)
2169 {
2170         LIST_HEAD(cancels);
2171         struct ptlrpc_request   *req;
2172         int                      rc, count;
2173         struct mdc_swap_layouts *msl, *payload;
2174         ENTRY;
2175
2176         msl = op_data->op_data;
2177
2178         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2179          * first thing it will do is to cancel the 2 layout
2180          * locks held by this client.
2181          * So the client must cancel its layout locks on the 2 fids
2182          * with the request RPC to avoid extra RPC round trips.
2183          */
2184         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2185                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2186                                         MDS_INODELOCK_XATTR);
2187         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2188                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2189                                          MDS_INODELOCK_XATTR);
2190
2191         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2192                                    &RQF_MDS_SWAP_LAYOUTS);
2193         if (req == NULL) {
2194                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2195                 RETURN(-ENOMEM);
2196         }
2197
2198         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2199         if (rc) {
2200                 ptlrpc_request_free(req);
2201                 RETURN(rc);
2202         }
2203
2204         mdc_swap_layouts_pack(&req->rq_pill, op_data);
2205
2206         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2207         LASSERT(payload);
2208
2209         *payload = *msl;
2210
2211         ptlrpc_request_set_replen(req);
2212
2213         rc = ptlrpc_queue_wait(req);
2214         if (rc)
2215                 GOTO(out, rc);
2216         EXIT;
2217
2218 out:
2219         ptlrpc_req_put(req);
2220         return rc;
2221 }
2222
2223 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2224                          void *karg, void __user *uarg)
2225 {
2226         struct obd_device *obd = exp->exp_obd;
2227         struct obd_ioctl_data *data;
2228         struct obd_import *imp = obd->u.cli.cl_import;
2229         int rc;
2230
2231         ENTRY;
2232         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
2233                obd->obd_name, cmd, len, karg, uarg);
2234
2235         /* handle commands that do not need @karg first */
2236         switch (cmd) {
2237         case LL_IOC_GET_CONNECT_FLAGS:
2238                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2239                                  sizeof(*exp_connect_flags_ptr(exp))))
2240                         RETURN(-EFAULT);
2241                 RETURN(0);
2242         }
2243
2244         if (unlikely(karg == NULL))
2245                 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", -EINVAL));
2246         data = karg;
2247
2248         if (!try_module_get(THIS_MODULE)) {
2249                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2250                        module_name(THIS_MODULE));
2251                 return -EINVAL;
2252         }
2253         switch (cmd) {
2254         case OBD_IOC_FID2PATH:
2255                 rc = mdc_ioc_fid2path(exp, karg);
2256                 GOTO(out, rc);
2257         case LL_IOC_HSM_CT_START:
2258                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2259                 /* ignore if it was already registered on this MDS. */
2260                 if (rc == -EEXIST)
2261                         rc = 0;
2262                 GOTO(out, rc);
2263         case LL_IOC_HSM_PROGRESS:
2264                 rc = mdc_ioc_hsm_progress(exp, karg);
2265                 GOTO(out, rc);
2266         case LL_IOC_HSM_STATE_GET:
2267                 rc = mdc_ioc_hsm_state_get(exp, karg);
2268                 GOTO(out, rc);
2269         case LL_IOC_HSM_STATE_SET:
2270                 rc = mdc_ioc_hsm_state_set(exp, karg);
2271                 GOTO(out, rc);
2272         case LL_IOC_HSM_ACTION:
2273                 rc = mdc_ioc_hsm_current_action(exp, karg);
2274                 GOTO(out, rc);
2275         case LL_IOC_HSM_REQUEST:
2276                 rc = mdc_ioc_hsm_request(exp, karg);
2277                 GOTO(out, rc);
2278         case OBD_IOC_CLIENT_RECOVER:
2279                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2280                 if (rc < 0)
2281                         GOTO(out, rc);
2282                 GOTO(out, rc = 0);
2283 #ifdef IOC_OSC_SET_ACTIVE
2284         case_OBD_IOC_DEPRECATED_FT(IOC_OSC_SET_ACTIVE, obd->obd_name, 2, 17);
2285 #endif
2286         case OBD_IOC_SET_ACTIVE:
2287                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2288                 GOTO(out, rc);
2289         /*
2290          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2291          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2292          * there'd be no LMV layer thus we might be called here. Eventually
2293          * this code should be removed.
2294          * bz20731, LU-592.
2295          */
2296         case IOC_OBD_STATFS: {
2297                 struct obd_statfs stat_buf = {0};
2298
2299                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2300                         GOTO(out, rc = -ENODEV);
2301
2302                 /* copy UUID */
2303                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2304                                  min((int)data->ioc_plen2,
2305                                      (int)sizeof(struct obd_uuid))))
2306                         GOTO(out, rc = -EFAULT);
2307
2308                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2309                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2310                                 0);
2311                 if (rc != 0)
2312                         GOTO(out, rc);
2313
2314                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2315                                      min((int) data->ioc_plen1,
2316                                          (int) sizeof(stat_buf))))
2317                         GOTO(out, rc = -EFAULT);
2318
2319                 GOTO(out, rc = 0);
2320         }
2321         case OBD_IOC_QUOTACTL: {
2322                 struct if_quotactl *qctl = karg;
2323                 struct obd_quotactl *oqctl;
2324
2325                 OBD_ALLOC_PTR(oqctl);
2326                 if (oqctl == NULL)
2327                         GOTO(out, rc = -ENOMEM);
2328
2329                 QCTL_COPY(oqctl, qctl);
2330                 rc = obd_quotactl(exp, oqctl);
2331                 if (rc == 0) {
2332                         QCTL_COPY_NO_PNAME(qctl, oqctl);
2333                         qctl->qc_valid = QC_MDTIDX;
2334                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2335                 }
2336
2337                 OBD_FREE_PTR(oqctl);
2338                 GOTO(out, rc);
2339         }
2340         case LL_IOC_LOV_SWAP_LAYOUTS:
2341                 rc = mdc_ioc_swap_layouts(exp, karg);
2342                 GOTO(out, rc);
2343         default:
2344                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
2345                 break;
2346         }
2347 out:
2348         module_put(THIS_MODULE);
2349
2350         return rc;
2351 }
2352
2353 static int mdc_get_info_rpc(struct obd_export *exp,
2354                             u32 keylen, void *key,
2355                             u32 vallen, void *val)
2356 {
2357         struct obd_import      *imp = class_exp2cliimp(exp);
2358         struct ptlrpc_request  *req;
2359         char                   *tmp;
2360         int                     rc = -EINVAL;
2361         ENTRY;
2362
2363         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2364         if (req == NULL)
2365                 RETURN(-ENOMEM);
2366
2367         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2368                              RCL_CLIENT, keylen);
2369         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2370                              RCL_CLIENT, sizeof(vallen));
2371
2372         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2373         if (rc) {
2374                 ptlrpc_request_free(req);
2375                 RETURN(rc);
2376         }
2377
2378         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2379         memcpy(tmp, key, keylen);
2380         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2381         memcpy(tmp, &vallen, sizeof(vallen));
2382
2383         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2384                              RCL_SERVER, vallen);
2385         ptlrpc_request_set_replen(req);
2386
2387         /* if server failed to resolve FID, and OI scrub not able to fix it, it
2388          * will return -EINPROGRESS, ptlrpc_queue_wait() will keep retrying,
2389          * set request interruptible to avoid deadlock.
2390          */
2391         if (KEY_IS(KEY_FID2PATH))
2392                 req->rq_allow_intr = 1;
2393
2394         rc = ptlrpc_queue_wait(req);
2395         /* -EREMOTE means the get_info result is partial, and it needs to
2396          * continue on another MDT, see fid2path part in lmv_iocontrol */
2397         if (rc == 0 || rc == -EREMOTE) {
2398                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2399                 memcpy(val, tmp, vallen);
2400                 if (req_capsule_rep_need_swab(&req->rq_pill)) {
2401                         if (KEY_IS(KEY_FID2PATH))
2402                                 lustre_swab_fid2path(val);
2403                 }
2404         }
2405         ptlrpc_req_put(req);
2406
2407         RETURN(rc);
2408 }
2409
2410 static void lustre_swab_hai(struct hsm_action_item *h)
2411 {
2412         __swab32s(&h->hai_len);
2413         __swab32s(&h->hai_action);
2414         lustre_swab_lu_fid(&h->hai_fid);
2415         lustre_swab_lu_fid(&h->hai_dfid);
2416         __swab64s(&h->hai_cookie);
2417         __swab64s(&h->hai_extent.offset);
2418         __swab64s(&h->hai_extent.length);
2419         __swab64s(&h->hai_gid);
2420 }
2421
2422 static void lustre_swab_hal(struct hsm_action_list *h)
2423 {
2424         struct hsm_action_item  *hai;
2425         __u32                    i;
2426
2427         __swab32s(&h->hal_version);
2428         __swab32s(&h->hal_count);
2429         __swab32s(&h->hal_archive_id);
2430         __swab64s(&h->hal_flags);
2431         hai = hai_first(h);
2432         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2433                 lustre_swab_hai(hai);
2434 }
2435
2436 static void lustre_swab_kuch(struct kuc_hdr *l)
2437 {
2438         __swab16s(&l->kuc_magic);
2439         /* __u8 l->kuc_transport */
2440         __swab16s(&l->kuc_msgtype);
2441         __swab16s(&l->kuc_msglen);
2442 }
2443
2444 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2445                                 struct lustre_kernelcomm *lk)
2446 {
2447         struct obd_import *imp = class_exp2cliimp(exp);
2448         int rc = 0;
2449
2450         if (lk->lk_group != KUC_GRP_HSM) {
2451                 CERROR("Bad copytool group %d\n", lk->lk_group);
2452                 return -EINVAL;
2453         }
2454
2455         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2456                lk->lk_uid, lk->lk_group, lk->lk_flags);
2457
2458         if (lk->lk_flags & LK_FLG_STOP) {
2459                 /* Unregister with the coordinator */
2460                 rc = mdc_ioc_hsm_ct_unregister(imp);
2461         } else {
2462                 __u32 *archives = NULL;
2463
2464                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2465                         archives = lk->lk_data;
2466
2467                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2468         }
2469
2470         return rc;
2471 }
2472
2473 /**
2474  * Send a message to any listening copytools
2475  * @param val KUC message (kuc_hdr + hsm_action_list)
2476  * @param len total length of message
2477  */
2478 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2479                                  size_t len, void *val)
2480 {
2481         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2482         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2483         int                      rc;
2484         ENTRY;
2485
2486         if (len < sizeof(*lh) + sizeof(*hal)) {
2487                 CERROR("Short HSM message %zu < %zu\n", len,
2488                        sizeof(*lh) + sizeof(*hal));
2489                 RETURN(-EPROTO);
2490         }
2491         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2492                 lustre_swab_kuch(lh);
2493                 lustre_swab_hal(hal);
2494         } else if (lh->kuc_magic != KUC_MAGIC) {
2495                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2496                 RETURN(-EPROTO);
2497         }
2498
2499         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2500                "on %s\n",
2501                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2502                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2503
2504         /* Broadcast to HSM listeners */
2505         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2506
2507         RETURN(rc);
2508 }
2509
2510 /**
2511  * callback function passed to kuc for re-registering each HSM copytool
2512  * running on MDC, after MDT shutdown/recovery.
2513  * @param data copytool registration data
2514  * @param cb_arg callback argument (obd_import)
2515  */
2516 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2517 {
2518         struct obd_import *imp = (struct obd_import *)cb_arg;
2519         struct kkuc_ct_data *kcd = data;
2520         __u32 *archives = NULL;
2521         int rc;
2522
2523         if (kcd == NULL ||
2524             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2525              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2526                 return -EPROTO;
2527
2528         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2529                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2530                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2531                        kcd->kcd_nr_archives);
2532         } else {
2533                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2534                        "(archive nr = %u)\n",
2535                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2536                 if (kcd->kcd_nr_archives != 0)
2537                         archives = kcd->kcd_archives;
2538         }
2539
2540         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2541         /* ignore error if the copytool is already registered */
2542         return (rc == -EEXIST) ? 0 : rc;
2543 }
2544
2545 static int mdc_kuc_reregister_thread(void *data)
2546 {
2547         struct obd_import *imp = data;
2548         int rc;
2549         ENTRY;
2550
2551         /* re-register HSM agents */
2552         rc = libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2553                                        mdc_hsm_ct_reregister, imp);
2554         if (rc < 0 && rc != -EEXIST)
2555                 CWARN("%s: Failed to re-register HSM agents (uuid: %s): rc = %d\n",
2556                       imp->imp_obd->obd_name,
2557                       obd_uuid2str(&imp->imp_obd->obd_uuid), rc);
2558
2559         class_import_put(imp);
2560         RETURN(rc);
2561 }
2562
2563 /**
2564  * Re-establish all kuc contexts with MDT
2565  * after MDT shutdown/recovery.
2566  * This is done in background.
2567  */
2568 static int mdc_kuc_reregister(struct obd_import *imp)
2569 {
2570         struct task_struct *task;
2571         int rc = 0;
2572
2573         class_import_get(imp);
2574         task = kthread_run(mdc_kuc_reregister_thread, imp, "kuc_reregister");
2575
2576         if (IS_ERR(task)) {
2577                 class_import_put(imp);
2578                 rc = PTR_ERR(task);
2579         }
2580
2581         return rc;
2582 }
2583
2584 static int mdc_set_info_async(const struct lu_env *env,
2585                               struct obd_export *exp,
2586                               u32 keylen, void *key,
2587                               u32 vallen, void *val,
2588                               struct ptlrpc_request_set *set)
2589 {
2590         struct obd_import       *imp = class_exp2cliimp(exp);
2591         int                      rc;
2592         ENTRY;
2593
2594         if (KEY_IS(KEY_READ_ONLY)) {
2595                 if (vallen != sizeof(int))
2596                         RETURN(-EINVAL);
2597
2598                 spin_lock(&imp->imp_lock);
2599                 if (*((int *)val)) {
2600                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2601                         imp->imp_connect_data.ocd_connect_flags |=
2602                                                         OBD_CONNECT_RDONLY;
2603                 } else {
2604                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2605                         imp->imp_connect_data.ocd_connect_flags &=
2606                                                         ~OBD_CONNECT_RDONLY;
2607                 }
2608                 spin_unlock(&imp->imp_lock);
2609
2610                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2611                                        keylen, key, vallen, val, set);
2612                 RETURN(rc);
2613         }
2614         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2615                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2616                                        keylen, key, vallen, val, set);
2617                 RETURN(rc);
2618         }
2619         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2620                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2621                                            val);
2622                 RETURN(rc);
2623         }
2624
2625         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2626                 __u32 *default_easize = val;
2627
2628                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2629                 RETURN(0);
2630         }
2631
2632         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2633         RETURN(rc);
2634 }
2635
2636 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2637                         __u32 keylen, void *key, __u32 *vallen, void *val)
2638 {
2639         int rc = -EINVAL;
2640
2641         if (KEY_IS(KEY_MAX_EASIZE)) {
2642                 __u32 mdsize, *max_easize;
2643
2644                 if (*vallen != sizeof(int))
2645                         RETURN(-EINVAL);
2646                 mdsize = *(__u32 *)val;
2647                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2648                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2649                 max_easize = val;
2650                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2651                 RETURN(0);
2652         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2653                 __u32 *default_easize;
2654
2655                 if (*vallen != sizeof(int))
2656                         RETURN(-EINVAL);
2657                 default_easize = val;
2658                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2659                 RETURN(0);
2660         } else if (KEY_IS(KEY_CONN_DATA)) {
2661                 struct obd_import *imp = class_exp2cliimp(exp);
2662                 struct obd_connect_data *data = val;
2663
2664                 if (*vallen != sizeof(*data))
2665                         RETURN(-EINVAL);
2666
2667                 *data = imp->imp_connect_data;
2668                 RETURN(0);
2669         } else if (KEY_IS(KEY_TGT_COUNT)) {
2670                 *((__u32 *)val) = 1;
2671                 RETURN(0);
2672         }
2673
2674         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2675
2676         RETURN(rc);
2677 }
2678
2679 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2680                      struct ptlrpc_request **request)
2681 {
2682         struct ptlrpc_request *req;
2683         int                    rc;
2684         ENTRY;
2685
2686         *request = NULL;
2687         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2688         if (req == NULL)
2689                 RETURN(-ENOMEM);
2690
2691         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2692         if (rc) {
2693                 ptlrpc_request_free(req);
2694                 RETURN(rc);
2695         }
2696
2697         mdc_pack_body(&req->rq_pill, fid, 0, 0, -1, 0);
2698
2699         ptlrpc_request_set_replen(req);
2700
2701         rc = ptlrpc_queue_wait(req);
2702         if (rc)
2703                 ptlrpc_req_put(req);
2704         else
2705                 *request = req;
2706         RETURN(rc);
2707 }
2708
2709 struct mdc_rmfid_args {
2710         int *mra_rcs;
2711         int mra_nr;
2712 };
2713
2714 static int mdc_rmfid_interpret(const struct lu_env *env,
2715                                struct ptlrpc_request *req,
2716                                void *args, int rc)
2717 {
2718         struct mdc_rmfid_args *aa;
2719         int *rcs, size;
2720         ENTRY;
2721
2722         if (!rc) {
2723                 aa = ptlrpc_req_async_args(aa, req);
2724
2725                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2726                                             RCL_SERVER);
2727                 LASSERT(size == sizeof(int) * aa->mra_nr);
2728                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2729                 LASSERT(rcs);
2730                 LASSERT(aa->mra_rcs);
2731                 LASSERT(aa->mra_nr);
2732                 memcpy(aa->mra_rcs, rcs, size);
2733         }
2734
2735         RETURN(rc);
2736 }
2737
2738 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2739                      int *rcs, struct ptlrpc_request_set *set)
2740 {
2741         struct ptlrpc_request *req;
2742         struct mdc_rmfid_args *aa;
2743         struct mdt_body *b;
2744         struct lu_fid *tmp;
2745         int rc, flen;
2746         ENTRY;
2747
2748         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2749         if (req == NULL)
2750                 RETURN(-ENOMEM);
2751
2752         flen = fa->fa_nr * sizeof(struct lu_fid);
2753         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2754                              RCL_CLIENT, flen);
2755         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2756                              RCL_SERVER, flen);
2757         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2758                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2759         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2760         if (rc) {
2761                 ptlrpc_request_free(req);
2762                 RETURN(rc);
2763         }
2764         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2765         memcpy(tmp, fa->fa_fids, flen);
2766
2767         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
2768         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2769         b->mbo_ctime = ktime_get_real_seconds();
2770
2771         ptlrpc_request_set_replen(req);
2772
2773         LASSERT(rcs);
2774         aa = ptlrpc_req_async_args(aa, req);
2775         aa->mra_rcs = rcs;
2776         aa->mra_nr = fa->fa_nr;
2777         req->rq_interpret_reply = mdc_rmfid_interpret;
2778
2779         ptlrpc_set_add_req(set, req);
2780         ptlrpc_check_set(NULL, set);
2781
2782         RETURN(rc);
2783 }
2784
2785 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2786                             enum obd_import_event event)
2787 {
2788         struct client_obd *cli;
2789         int rc = 0;
2790
2791         ENTRY;
2792         if (WARN_ON_ONCE(!obd || !imp || imp->imp_obd != obd))
2793                 RETURN(-ENODEV);
2794
2795         cli = &obd->u.cli;
2796         if (!cli)
2797                 RETURN(-ENODEV);
2798
2799         switch (event) {
2800         case IMP_EVENT_DISCON:
2801                 spin_lock(&cli->cl_loi_list_lock);
2802                 cli->cl_avail_grant = 0;
2803                 cli->cl_lost_grant = 0;
2804                 spin_unlock(&cli->cl_loi_list_lock);
2805                 break;
2806         case IMP_EVENT_INACTIVE:
2807                 /*
2808                  * Flush current sequence to make client obtain new one
2809                  * from server in case of disconnect/reconnect.
2810                  */
2811                 down_read(&cli->cl_seq_rwsem);
2812                 if (cli->cl_seq)
2813                         seq_client_flush(cli->cl_seq);
2814                 up_read(&cli->cl_seq_rwsem);
2815
2816                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2817                 break;
2818         case IMP_EVENT_INVALIDATE: {
2819                 struct ldlm_namespace *ns = obd->obd_namespace;
2820                 struct lu_env *env;
2821                 __u16 refcheck;
2822
2823                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2824
2825                 env = cl_env_get(&refcheck);
2826                 if (!IS_ERR(env)) {
2827                         /* Reset grants. All pages go to failing rpcs due to
2828                          * the invalid import.
2829                          */
2830                         osc_io_unplug(env, cli, NULL);
2831
2832                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2833                                                  osc_ldlm_resource_invalidate,
2834                                                  env, 0);
2835                         cl_env_put(env, &refcheck);
2836                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2837                 } else {
2838                         rc = PTR_ERR(env);
2839                 }
2840                 break;
2841         }
2842         case IMP_EVENT_ACTIVE:
2843                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2844                 /* redo the kuc registration after reconnecting */
2845                 if (rc == 0)
2846                         rc = mdc_kuc_reregister(imp);
2847                 break;
2848         case IMP_EVENT_OCD: {
2849                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2850
2851                 if (OCD_HAS_FLAG(ocd, GRANT))
2852                         osc_init_grant(cli, ocd);
2853
2854                 md_init_ea_size(obd->obd_self_export, ocd->ocd_max_easize, 0);
2855                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2856                 break;
2857         }
2858         case IMP_EVENT_DEACTIVATE:
2859         case IMP_EVENT_ACTIVATE:
2860                 break;
2861         default:
2862                 CERROR("Unknown import event %x\n", event);
2863                 LBUG();
2864         }
2865         RETURN(rc);
2866 }
2867
2868 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2869                   struct lu_fid *fid, struct md_op_data *op_data)
2870 {
2871         struct client_obd *cli = &exp->exp_obd->u.cli;
2872         int rc = -EIO;
2873
2874         ENTRY;
2875
2876         down_read(&cli->cl_seq_rwsem);
2877         if (cli->cl_seq)
2878                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2879         up_read(&cli->cl_seq_rwsem);
2880
2881         RETURN(rc);
2882 }
2883
2884 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2885 {
2886         struct client_obd *cli = &exp->exp_obd->u.cli;
2887         return &cli->cl_target_uuid;
2888 }
2889
2890 /**
2891  * Determine whether the lock can be canceled before replaying it during
2892  * recovery, non zero value will be return if the lock can be canceled,
2893  * or zero returned for not
2894  */
2895 static int mdc_cancel_weight(struct ldlm_lock *lock)
2896 {
2897         if (lock->l_resource->lr_type != LDLM_IBITS)
2898                 RETURN(0);
2899
2900         /* FIXME: if we ever get into a situation where there are too many
2901          * opened files with open locks on a single node, then we really
2902          * should replay these open locks to reget it */
2903         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2904                 RETURN(0);
2905
2906         /* Special case for DoM locks, cancel only unused and granted locks */
2907         if (ldlm_has_dom(lock) &&
2908             (lock->l_granted_mode != lock->l_req_mode ||
2909              osc_ldlm_weigh_ast(lock) != 0))
2910                 RETURN(0);
2911
2912         RETURN(1);
2913 }
2914
2915 static int mdc_resource_inode_free(struct ldlm_resource *res)
2916 {
2917         if (res->lr_lvb_inode)
2918                 res->lr_lvb_inode = NULL;
2919
2920         return 0;
2921 }
2922
2923 static struct ldlm_valblock_ops inode_lvbo = {
2924         .lvbo_free = mdc_resource_inode_free
2925 };
2926
2927 static int mdc_llog_init(struct obd_device *obd)
2928 {
2929         struct obd_llog_group   *olg = &obd->obd_olg;
2930         struct llog_ctxt        *ctxt;
2931         int                      rc;
2932
2933         ENTRY;
2934
2935         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2936                         &llog_client_ops);
2937         if (rc < 0)
2938                 RETURN(rc);
2939
2940         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2941         llog_initiator_connect(ctxt);
2942         llog_ctxt_put(ctxt);
2943
2944         RETURN(0);
2945 }
2946
2947 static void mdc_llog_finish(struct obd_device *obd)
2948 {
2949         struct llog_ctxt *ctxt;
2950
2951         ENTRY;
2952
2953         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2954         if (ctxt != NULL)
2955                 llog_cleanup(NULL, ctxt);
2956
2957         EXIT;
2958 }
2959
2960 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2961 {
2962         int rc;
2963
2964         ENTRY;
2965
2966         rc = osc_setup_common(obd, cfg);
2967         if (rc < 0)
2968                 RETURN(rc);
2969
2970         rc = mdc_tunables_init(obd);
2971         if (rc)
2972                 GOTO(err_osc_cleanup, rc);
2973
2974         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2975         obd->u.cli.cl_lsom_update = true;
2976
2977         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2978
2979         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2980
2981         rc = mdc_llog_init(obd);
2982         if (rc) {
2983                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2984                        obd->obd_name, rc);
2985                 GOTO(err_llog_cleanup, rc);
2986         }
2987
2988         rc = mdc_changelog_cdev_init(obd);
2989         if (rc) {
2990                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2991                        obd->obd_name, rc);
2992                 GOTO(err_changelog_cleanup, rc);
2993         }
2994
2995         RETURN(rc);
2996
2997 err_changelog_cleanup:
2998         mdc_llog_finish(obd);
2999 err_llog_cleanup:
3000         lprocfs_free_md_stats(obd);
3001         ptlrpc_lprocfs_unregister_obd(obd);
3002 err_osc_cleanup:
3003         osc_cleanup_common(obd);
3004         return rc;
3005 }
3006
3007 /* Initialize the default and maximum LOV EA sizes.  This allows
3008  * us to make MDS RPCs with large enough reply buffers to hold a default
3009  * sized EA without having to calculate this (via a call into the
3010  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
3011  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
3012  * a large number of stripes is possible.  If a larger reply buffer is
3013  * required it will be reallocated in the ptlrpc layer due to overflow.
3014  */
3015 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
3016                             __u32 def_easize)
3017 {
3018         struct obd_device *obd = exp->exp_obd;
3019         struct client_obd *cli = &obd->u.cli;
3020         ENTRY;
3021
3022         if (cli->cl_max_mds_easize < easize)
3023                 cli->cl_max_mds_easize = easize;
3024
3025         if (cli->cl_default_mds_easize < def_easize)
3026                 cli->cl_default_mds_easize = def_easize;
3027
3028         RETURN(0);
3029 }
3030
3031 static int mdc_precleanup(struct obd_device *obd)
3032 {
3033         ENTRY;
3034
3035         osc_precleanup_common(obd);
3036
3037         mdc_changelog_cdev_finish(obd);
3038         mdc_llog_finish(obd);
3039         lprocfs_free_md_stats(obd);
3040         ptlrpc_lprocfs_unregister_obd(obd);
3041
3042         RETURN(0);
3043 }
3044
3045 static int mdc_cleanup(struct obd_device *obd)
3046 {
3047         struct client_obd *cli = &obd->u.cli;
3048         LASSERT(cli->cl_mod_rpcs_in_flight == 0);
3049         return osc_cleanup_common(obd);
3050 }
3051
3052 static const struct obd_ops mdc_obd_ops = {
3053         .o_owner            = THIS_MODULE,
3054         .o_setup            = mdc_setup,
3055         .o_precleanup       = mdc_precleanup,
3056         .o_cleanup          = mdc_cleanup,
3057         .o_add_conn         = client_import_add_conn,
3058         .o_del_conn         = client_import_del_conn,
3059         .o_connect          = client_connect_import,
3060         .o_reconnect        = osc_reconnect,
3061         .o_disconnect       = osc_disconnect,
3062         .o_iocontrol        = mdc_iocontrol,
3063         .o_set_info_async   = mdc_set_info_async,
3064         .o_statfs           = mdc_statfs,
3065         .o_statfs_async     = mdc_statfs_async,
3066         .o_fid_init         = client_fid_init,
3067         .o_fid_fini         = client_fid_fini,
3068         .o_fid_alloc        = mdc_fid_alloc,
3069         .o_import_event     = mdc_import_event,
3070         .o_get_info         = mdc_get_info,
3071         .o_get_uuid         = mdc_get_uuid,
3072         .o_quotactl         = mdc_quotactl,
3073 };
3074
3075 static const struct md_ops mdc_md_ops = {
3076         .m_get_root         = mdc_get_root,
3077         .m_null_inode       = mdc_null_inode,
3078         .m_close            = mdc_close,
3079         .m_create           = mdc_create,
3080         .m_enqueue          = mdc_enqueue,
3081         .m_getattr          = mdc_getattr,
3082         .m_getattr_name     = mdc_getattr_name,
3083         .m_intent_lock      = mdc_intent_lock,
3084         .m_link             = mdc_link,
3085         .m_rename           = mdc_rename,
3086         .m_setattr          = mdc_setattr,
3087         .m_setxattr         = mdc_setxattr,
3088         .m_getxattr         = mdc_getxattr,
3089         .m_fsync                = mdc_fsync,
3090         .m_file_resync          = mdc_file_resync,
3091         .m_read_page            = mdc_read_page,
3092         .m_unlink           = mdc_unlink,
3093         .m_cancel_unused    = mdc_cancel_unused,
3094         .m_init_ea_size     = mdc_init_ea_size,
3095         .m_set_lock_data    = mdc_set_lock_data,
3096         .m_lock_match       = mdc_lock_match,
3097         .m_get_lustre_md    = mdc_get_lustre_md,
3098         .m_set_open_replay_data = mdc_set_open_replay_data,
3099         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3100         .m_intent_getattr_async = mdc_intent_getattr_async,
3101         .m_revalidate_lock      = mdc_revalidate_lock,
3102         .m_rmfid                = mdc_rmfid,
3103         .m_batch_create         = cli_batch_create,
3104         .m_batch_stop           = cli_batch_stop,
3105         .m_batch_flush          = cli_batch_flush,
3106         .m_batch_add            = mdc_batch_add,
3107 };
3108
3109 dev_t mdc_changelog_dev;
3110 struct class *mdc_changelog_class;
3111 static int __init mdc_init(void)
3112 {
3113         int rc = 0;
3114
3115         rc = libcfs_setup();
3116         if (rc)
3117                 return rc;
3118
3119         rc = alloc_chrdev_region(&mdc_changelog_dev, 0,
3120                                  MDC_CHANGELOG_DEV_COUNT,
3121                                  MDC_CHANGELOG_DEV_NAME);
3122         if (rc)
3123                 return rc;
3124
3125         mdc_changelog_class = ll_class_create(MDC_CHANGELOG_DEV_NAME);
3126         if (IS_ERR(mdc_changelog_class)) {
3127                 rc = PTR_ERR(mdc_changelog_class);
3128                 goto out_dev;
3129         }
3130
3131         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, true,
3132                                  LUSTRE_MDC_NAME, &mdc_device_type);
3133         if (rc)
3134                 goto out_class;
3135
3136         return 0;
3137
3138 out_class:
3139         class_destroy(mdc_changelog_class);
3140 out_dev:
3141         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3142         return rc;
3143 }
3144
3145 static void __exit mdc_exit(void)
3146 {
3147         class_unregister_type(LUSTRE_MDC_NAME);
3148         class_destroy(mdc_changelog_class);
3149         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3150         idr_destroy(&mdc_changelog_minor_idr);
3151 }
3152
3153 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3154 MODULE_DESCRIPTION("Lustre Metadata Client");
3155 MODULE_VERSION(LUSTRE_VERSION_STRING);
3156 MODULE_LICENSE("GPL");
3157
3158 module_init(mdc_init);
3159 module_exit(mdc_exit);