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