lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #include <linux/delay.h>
  42 #include <linux/uidgid.h>
  43 #include <linux/device.h>
  44 #include <linux/xarray.h>
  45
  46 #include <lustre_errno.h>
  47
  48 #include <cl_object.h>
  49 #include <llog_swab.h>
  50 #include <lprocfs_status.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_compat.h>
  53 #include <lustre_fid.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_kernelcomm.h>
  56 #include <lustre_lmv.h>
  57 #include <lustre_log.h>
  58 #include <lustre_swab.h>
  59 #include <obd_class.h>
  60 #include <lustre_osc.h>
  61
  62 #include "mdc_internal.h"
  63
  64 #define REQUEST_MINOR 244
  65
  66 static int mdc_cleanup(struct obd_device *obd);
  67
  68 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  69 {
  70         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  71         int rc;
  72
  73         /* obd_get_request_slot() ensures that this client has no more
  74          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  75          * against an MDT. */
  76         rc = obd_get_request_slot(cli);
  77         if (rc != 0)
  78                 return rc;
  79
  80         rc = ptlrpc_queue_wait(req);
  81         obd_put_request_slot(cli);
  82
  83         return rc;
  84 }
  85
  86 /*
  87  * Send MDS_GET_ROOT RPC to fetch root FID.
  88  *
  89  * If \a fileset is not NULL it should contain a subdirectory off
  90  * the ROOT/ directory to be mounted on the client. Return the FID
  91  * of the subdirectory to the client to mount onto its mountpoint.
  92  *
  93  * \param[in]   imp     MDC import
  94  * \param[in]   fileset fileset name, which could be NULL
  95  * \param[out]  rootfid root FID of this mountpoint
  96  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  97  *
  98  * \retval      0 on success, negative errno on failure
  99  */
 100 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 101                          struct lu_fid *rootfid)
 102 {
 103         struct ptlrpc_request   *req;
 104         struct mdt_body         *body;
 105         int                      rc;
 106
 107         ENTRY;
 108
 109         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 110                 RETURN(-ENOTSUPP);
 111
 112         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 113                                 &RQF_MDS_GET_ROOT);
 114         if (req == NULL)
 115                 RETURN(-ENOMEM);
 116
 117         if (fileset != NULL)
 118                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 119                                      strlen(fileset) + 1);
 120         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 121         if (rc) {
 122                 ptlrpc_request_free(req);
 123                 RETURN(rc);
 124         }
 125         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 126         if (fileset != NULL) {
 127                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 128
 129                 memcpy(name, fileset, strlen(fileset));
 130         }
 131         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 132         req->rq_send_state = LUSTRE_IMP_FULL;
 133
 134         ptlrpc_request_set_replen(req);
 135
 136         rc = ptlrpc_queue_wait(req);
 137         if (rc)
 138                 GOTO(out, rc);
 139
 140         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 141         if (body == NULL)
 142                 GOTO(out, rc = -EPROTO);
 143
 144         *rootfid = body->mbo_fid1;
 145         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 146                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 147         EXIT;
 148 out:
 149         ptlrpc_req_finished(req);
 150
 151         return rc;
 152 }
 153
 154 /*
 155  * This function now is known to always saying that it will receive 4 buffers
 156  * from server. Even for cases when acl_size and md_size is zero, RPC header
 157  * will contain 4 fields and RPC itself will contain zero size fields. This is
 158  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 159  * and thus zero, it shrinks it, making zero size. The same story about
 160  * md_size. And this is course of problem when client waits for smaller number
 161  * of fields. This issue will be fixed later when client gets aware of RPC
 162  * layouts.  --umka
 163  */
 164 static int mdc_getattr_common(struct obd_export *exp,
 165                               struct ptlrpc_request *req)
 166 {
 167         struct req_capsule *pill = &req->rq_pill;
 168         struct mdt_body    *body;
 169         void               *eadata;
 170         int                 rc;
 171         ENTRY;
 172
 173         /* Request message already built. */
 174         rc = ptlrpc_queue_wait(req);
 175         if (rc != 0)
 176                 RETURN(rc);
 177
 178         /* sanity check for the reply */
 179         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 180         if (body == NULL)
 181                 RETURN(-EPROTO);
 182
 183         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 184
 185         mdc_update_max_ea_from_body(exp, body);
 186         if (body->mbo_eadatasize != 0) {
 187                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 188                                                       body->mbo_eadatasize);
 189                 if (eadata == NULL)
 190                         RETURN(-EPROTO);
 191         }
 192
 193         RETURN(0);
 194 }
 195
 196 static void mdc_reset_acl_req(struct ptlrpc_request *req)
 197 {
 198         spin_lock(&req->rq_early_free_lock);
 199         sptlrpc_cli_free_repbuf(req);
 200         req->rq_repbuf = NULL;
 201         req->rq_repbuf_len = 0;
 202         req->rq_repdata = NULL;
 203         req->rq_reqdata_len = 0;
 204         spin_unlock(&req->rq_early_free_lock);
 205 }
 206
 207 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 208                        struct ptlrpc_request **request)
 209 {
 210         struct ptlrpc_request *req;
 211         struct obd_import *imp = class_exp2cliimp(exp);
 212         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 213         int rc;
 214         ENTRY;
 215
 216         /* Single MDS without an LMV case */
 217         if (op_data->op_flags & MF_GET_MDT_IDX) {
 218                 op_data->op_mds = 0;
 219                 RETURN(0);
 220         }
 221
 222         *request = NULL;
 223         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 224         if (req == NULL)
 225                 RETURN(-ENOMEM);
 226
 227         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 228         if (rc) {
 229                 ptlrpc_request_free(req);
 230                 RETURN(rc);
 231         }
 232
 233 again:
 234         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 235                       op_data->op_mode, -1, 0);
 236         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 237         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 238                              op_data->op_mode);
 239         ptlrpc_request_set_replen(req);
 240
 241         rc = mdc_getattr_common(exp, req);
 242         if (rc) {
 243                 if (rc == -ERANGE) {
 244                         acl_bufsize = min_t(__u32,
 245                                             imp->imp_connect_data.ocd_max_easize,
 246                                             XATTR_SIZE_MAX);
 247                         mdc_reset_acl_req(req);
 248                         goto again;
 249                 }
 250
 251                 ptlrpc_req_finished(req);
 252         } else {
 253                 *request = req;
 254         }
 255
 256         RETURN(rc);
 257 }
 258
 259 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 260                             struct ptlrpc_request **request)
 261 {
 262         struct ptlrpc_request *req;
 263         struct obd_import *imp = class_exp2cliimp(exp);
 264         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 265         int rc;
 266         ENTRY;
 267
 268         *request = NULL;
 269         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 270         if (req == NULL)
 271                 RETURN(-ENOMEM);
 272
 273         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 274                              op_data->op_namelen + 1);
 275
 276         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 277         if (rc) {
 278                 ptlrpc_request_free(req);
 279                 RETURN(rc);
 280         }
 281
 282         if (op_data->op_name) {
 283                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 284                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 285                                 op_data->op_namelen);
 286                 memcpy(name, op_data->op_name, op_data->op_namelen);
 287         }
 288
 289 again:
 290         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 291                       op_data->op_mode, op_data->op_suppgids[0], 0);
 292         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 293                              op_data->op_mode);
 294         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 295         ptlrpc_request_set_replen(req);
 296
 297         rc = mdc_getattr_common(exp, req);
 298         if (rc) {
 299                 if (rc == -ERANGE) {
 300                         acl_bufsize = min_t(__u32,
 301                                             imp->imp_connect_data.ocd_max_easize,
 302                                             XATTR_SIZE_MAX);
 303                         mdc_reset_acl_req(req);
 304                         goto again;
 305                 }
 306
 307                 ptlrpc_req_finished(req);
 308         } else {
 309                 *request = req;
 310         }
 311
 312         RETURN(rc);
 313 }
 314
 315 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 316                             const struct lu_fid *fid, int opcode, u64 valid,
 317                             const char *xattr_name, const char *input,
 318                             int input_size, int output_size, int flags,
 319                             __u32 suppgid, struct ptlrpc_request **request)
 320 {
 321         struct ptlrpc_request *req;
 322         int   xattr_namelen = 0;
 323         char *tmp;
 324         int   rc;
 325         ENTRY;
 326
 327         *request = NULL;
 328         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 329         if (req == NULL)
 330                 RETURN(-ENOMEM);
 331
 332         if (xattr_name) {
 333                 xattr_namelen = strlen(xattr_name) + 1;
 334                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 335                                      xattr_namelen);
 336         }
 337         if (input_size)
 338                 LASSERT(input);
 339         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 340                              input_size);
 341
 342         /* get SELinux policy info if any */
 343         rc = sptlrpc_get_sepol(req);
 344         if (rc < 0) {
 345                 ptlrpc_request_free(req);
 346                 RETURN(rc);
 347         }
 348         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 349                              strlen(req->rq_sepol) ?
 350                              strlen(req->rq_sepol) + 1 : 0);
 351
 352         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 353         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 354             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 355                 LIST_HEAD(cancels);
 356                 int count;
 357
 358                 /* Without that packing would fail */
 359                 if (input_size == 0)
 360                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 361                                              RCL_CLIENT, 0);
 362
 363                 count = mdc_resource_get_unused(exp, fid,
 364                                                 &cancels, LCK_EX,
 365                                                 MDS_INODELOCK_XATTR);
 366
 367                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 368                 if (rc) {
 369                         ptlrpc_request_free(req);
 370                         RETURN(rc);
 371                 }
 372         } else {
 373                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 374                 if (rc) {
 375                         ptlrpc_request_free(req);
 376                         RETURN(rc);
 377                 }
 378         }
 379
 380         if (opcode == MDS_REINT) {
 381                 struct mdt_rec_setxattr *rec;
 382
 383                 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
 384                              sizeof(struct mdt_rec_reint));
 385                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 386                 rec->sx_opcode = REINT_SETXATTR;
 387                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 388                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 389                 rec->sx_cap    = cfs_curproc_cap_pack();
 390                 rec->sx_suppgid1 = suppgid;
 391                 rec->sx_suppgid2 = -1;
 392                 rec->sx_fid    = *fid;
 393                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 394                 rec->sx_time   = ktime_get_real_seconds();
 395                 rec->sx_size   = output_size;
 396                 rec->sx_flags  = flags;
 397         } else {
 398                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 399         }
 400
 401         if (xattr_name) {
 402                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 403                 memcpy(tmp, xattr_name, xattr_namelen);
 404         }
 405         if (input_size) {
 406                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 407                 memcpy(tmp, input, input_size);
 408         }
 409
 410         mdc_file_sepol_pack(req);
 411
 412         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 413                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 414                                      RCL_SERVER, output_size);
 415         ptlrpc_request_set_replen(req);
 416
 417         /* make rpc */
 418         if (opcode == MDS_REINT)
 419                 ptlrpc_get_mod_rpc_slot(req);
 420
 421         rc = ptlrpc_queue_wait(req);
 422
 423         if (opcode == MDS_REINT)
 424                 ptlrpc_put_mod_rpc_slot(req);
 425
 426         if (rc)
 427                 ptlrpc_req_finished(req);
 428         else
 429                 *request = req;
 430         RETURN(rc);
 431 }
 432
 433 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 434                         u64 obd_md_valid, const char *name,
 435                         const void *value, size_t value_size,
 436                         unsigned int xattr_flags, u32 suppgid,
 437                         struct ptlrpc_request **req)
 438 {
 439         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 440                 obd_md_valid == OBD_MD_FLXATTRRM);
 441
 442         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 443                                 fid, MDS_REINT, obd_md_valid, name,
 444                                 value, value_size, 0, xattr_flags, suppgid,
 445                                 req);
 446 }
 447
 448 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 449                         u64 obd_md_valid, const char *name, size_t buf_size,
 450                         struct ptlrpc_request **req)
 451 {
 452         struct mdt_body *body;
 453         int rc;
 454
 455         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 456                 obd_md_valid == OBD_MD_FLXATTRLS);
 457
 458         /* Message below is checked in sanity-selinux test_20d
 459          * and sanity-sec test_49
 460          */
 461         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 462                exp->exp_obd->obd_name, name, PFID(fid));
 463         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 464                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 465                               req);
 466         if (rc < 0)
 467                 GOTO(out, rc);
 468
 469         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 470         if (body == NULL)
 471                 GOTO(out, rc = -EPROTO);
 472
 473         /* only detect the xattr size */
 474         if (buf_size == 0) {
 475                 /* LU-11109: Older MDTs do not distinguish
 476                  * between nonexistent xattrs and zero length
 477                  * values in this case. Newer MDTs will return
 478                  * -ENODATA or set OBD_MD_FLXATTR. */
 479                 GOTO(out, rc = body->mbo_eadatasize);
 480         }
 481
 482         if (body->mbo_eadatasize == 0) {
 483                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 484                  * success so that we can distinguish between
 485                  * zero length value and nonexistent xattr.
 486                  *
 487                  * If OBD_MD_FLXATTR is not set then we keep
 488                  * the old behavior and return -ENODATA for
 489                  * getxattr() when mbo_eadatasize is 0. But
 490                  * -ENODATA only makes sense for getxattr()
 491                  * and not for listxattr(). */
 492                 if (body->mbo_valid & OBD_MD_FLXATTR)
 493                         GOTO(out, rc = 0);
 494                 else if (obd_md_valid == OBD_MD_FLXATTR)
 495                         GOTO(out, rc = -ENODATA);
 496                 else
 497                         GOTO(out, rc = 0);
 498         }
 499
 500         GOTO(out, rc = body->mbo_eadatasize);
 501 out:
 502         if (rc < 0) {
 503                 ptlrpc_req_finished(*req);
 504                 *req = NULL;
 505         }
 506
 507         return rc;
 508 }
 509
 510 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 511                       struct obd_export *dt_exp, 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 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         ENTRY;
 641
 642         if (mod == NULL) {
 643                 DEBUG_REQ(D_ERROR, req,
 644                           "cannot properly replay without open data");
 645                 EXIT;
 646                 return;
 647         }
 648
 649         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 650         LASSERT(body != NULL);
 651
 652         spin_lock(&req->rq_lock);
 653         och = mod->mod_och;
 654         if (och && och->och_open_handle.cookie)
 655                 req->rq_early_free_repbuf = 1;
 656         else
 657                 req->rq_early_free_repbuf = 0;
 658         spin_unlock(&req->rq_lock);
 659
 660         if (req->rq_early_free_repbuf) {
 661                 struct lustre_handle *file_open_handle;
 662
 663                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 664
 665                 file_open_handle = &och->och_open_handle;
 666                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 667                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 668                 old_open_handle = *file_open_handle;
 669                 *file_open_handle = body->mbo_open_handle;
 670         }
 671
 672         close_req = mod->mod_close_req;
 673         if (close_req) {
 674                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 675                 struct mdt_ioepoch *epoch;
 676
 677                 LASSERT(opc == MDS_CLOSE);
 678                 epoch = req_capsule_client_get(&close_req->rq_pill,
 679                                                &RMF_MDT_EPOCH);
 680                 LASSERT(epoch);
 681
 682                 if (req->rq_early_free_repbuf)
 683                         LASSERT(old_open_handle.cookie ==
 684                                 epoch->mio_open_handle.cookie);
 685
 686                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 687                 epoch->mio_open_handle = body->mbo_open_handle;
 688         }
 689         EXIT;
 690 }
 691
 692 void mdc_commit_open(struct ptlrpc_request *req)
 693 {
 694         struct md_open_data *mod = req->rq_cb_data;
 695         if (mod == NULL)
 696                 return;
 697
 698         /**
 699          * No need to touch md_open_data::mod_och, it holds a reference on
 700          * \var mod and will zero references to each other, \var mod will be
 701          * freed after that when md_open_data::mod_och will put the reference.
 702          */
 703
 704         /**
 705          * Do not let open request to disappear as it still may be needed
 706          * for close rpc to happen (it may happen on evict only, otherwise
 707          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 708          * called), just mark this rpc as committed to distinguish these 2
 709          * cases, see mdc_close() for details. The open request reference will
 710          * be put along with freeing \var mod.
 711          */
 712         ptlrpc_request_addref(req);
 713         spin_lock(&req->rq_lock);
 714         req->rq_committed = 1;
 715         spin_unlock(&req->rq_lock);
 716         req->rq_cb_data = NULL;
 717         obd_mod_put(mod);
 718 }
 719
 720 int mdc_set_open_replay_data(struct obd_export *exp,
 721                              struct obd_client_handle *och,
 722                              struct lookup_intent *it)
 723 {
 724         struct md_open_data     *mod;
 725         struct mdt_rec_create   *rec;
 726         struct mdt_body         *body;
 727         struct ptlrpc_request   *open_req = it->it_request;
 728         struct obd_import       *imp = open_req->rq_import;
 729         ENTRY;
 730
 731         if (!open_req->rq_replay)
 732                 RETURN(0);
 733
 734         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 735         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 736         LASSERT(rec != NULL);
 737         /* Incoming message in my byte order (it's been swabbed). */
 738         /* Outgoing messages always in my byte order. */
 739         LASSERT(body != NULL);
 740
 741         /* Only if the import is replayable, we set replay_open data */
 742         if (och && imp->imp_replayable) {
 743                 mod = obd_mod_alloc();
 744                 if (mod == NULL) {
 745                         DEBUG_REQ(D_ERROR, open_req,
 746                                   "cannot allocate md_open_data");
 747                         RETURN(0);
 748                 }
 749
 750                 /**
 751                  * Take a reference on \var mod, to be freed on mdc_close().
 752                  * It protects \var mod from being freed on eviction (commit
 753                  * callback is called despite rq_replay flag).
 754                  * Another reference for \var och.
 755                  */
 756                 obd_mod_get(mod);
 757                 obd_mod_get(mod);
 758
 759                 spin_lock(&open_req->rq_lock);
 760                 och->och_mod = mod;
 761                 mod->mod_och = och;
 762                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 763                                      it_disposition(it, DISP_OPEN_STRIPE);
 764                 mod->mod_open_req = open_req;
 765                 open_req->rq_cb_data = mod;
 766                 open_req->rq_commit_cb = mdc_commit_open;
 767                 open_req->rq_early_free_repbuf = 1;
 768                 spin_unlock(&open_req->rq_lock);
 769         }
 770
 771         rec->cr_fid2 = body->mbo_fid1;
 772         rec->cr_open_handle_old = body->mbo_open_handle;
 773         open_req->rq_replay_cb = mdc_replay_open;
 774         if (!fid_is_sane(&body->mbo_fid1)) {
 775                 DEBUG_REQ(D_ERROR, open_req,
 776                           "saving replay request with insane FID " DFID,
 777                           PFID(&body->mbo_fid1));
 778                 LBUG();
 779         }
 780
 781         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 782         RETURN(0);
 783 }
 784
 785 static void mdc_free_open(struct md_open_data *mod)
 786 {
 787         int committed = 0;
 788
 789         if (mod->mod_is_create == 0 &&
 790             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 791                 committed = 1;
 792
 793         /**
 794          * No reason to asssert here if the open request has
 795          * rq_replay == 1. It means that mdc_close failed, and
 796          * close request wasn`t sent. It is not fatal to client.
 797          * The worst thing is eviction if the client gets open lock
 798          **/
 799
 800         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 801                   "free open request, rq_replay=%d",
 802                   mod->mod_open_req->rq_replay);
 803
 804         ptlrpc_request_committed(mod->mod_open_req, committed);
 805         if (mod->mod_close_req)
 806                 ptlrpc_request_committed(mod->mod_close_req, committed);
 807 }
 808
 809 int mdc_clear_open_replay_data(struct obd_export *exp,
 810                                struct obd_client_handle *och)
 811 {
 812         struct md_open_data *mod = och->och_mod;
 813         ENTRY;
 814
 815         /**
 816          * It is possible to not have \var mod in a case of eviction between
 817          * lookup and ll_file_open().
 818          **/
 819         if (mod == NULL)
 820                 RETURN(0);
 821
 822         LASSERT(mod != LP_POISON);
 823         LASSERT(mod->mod_open_req != NULL);
 824
 825         spin_lock(&mod->mod_open_req->rq_lock);
 826         if (mod->mod_och)
 827                 mod->mod_och->och_open_handle.cookie = 0;
 828         mod->mod_open_req->rq_early_free_repbuf = 0;
 829         spin_unlock(&mod->mod_open_req->rq_lock);
 830         mdc_free_open(mod);
 831
 832         mod->mod_och = NULL;
 833         och->och_mod = NULL;
 834         obd_mod_put(mod);
 835
 836         RETURN(0);
 837 }
 838
 839 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 840                      struct md_open_data *mod, struct ptlrpc_request **request)
 841 {
 842         struct obd_device     *obd = class_exp2obd(exp);
 843         struct ptlrpc_request *req;
 844         struct req_format     *req_fmt;
 845         size_t                 u32_count = 0;
 846         int                    rc;
 847         int                    saved_rc = 0;
 848         ENTRY;
 849
 850         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 851                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 852                op_data->op_bias);
 853
 854         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 855                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 856                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 857                         /* allocate a FID for volatile file */
 858                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 859                                            op_data);
 860                         if (rc < 0) {
 861                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 862                                        obd->obd_name, PFID(&op_data->op_fid1),
 863                                        rc);
 864                                 /* save the errcode and proceed to close */
 865                                 saved_rc = rc;
 866                         }
 867                 }
 868                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 869                         size_t count = op_data->op_data_size / sizeof(__u32);
 870
 871                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 872                                 u32_count = count;
 873                 }
 874         } else {
 875                 req_fmt = &RQF_MDS_CLOSE;
 876         }
 877
 878         *request = NULL;
 879         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 880                 req = NULL;
 881         else
 882                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 883
 884         /* Ensure that this close's handle is fixed up during replay. */
 885         if (likely(mod != NULL)) {
 886                 LASSERTF(mod->mod_open_req != NULL &&
 887                          mod->mod_open_req->rq_type != LI_POISON,
 888                          "POISONED open %p!\n", mod->mod_open_req);
 889
 890                 mod->mod_close_req = req;
 891
 892                 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
 893                 /* We no longer want to preserve this open for replay even
 894                  * though the open was committed. b=3632, b=3633 */
 895                 spin_lock(&mod->mod_open_req->rq_lock);
 896                 mod->mod_open_req->rq_replay = 0;
 897                 spin_unlock(&mod->mod_open_req->rq_lock);
 898         } else {
 899                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 900         }
 901         if (req == NULL) {
 902                 /**
 903                  * TODO: repeat close after errors
 904                  */
 905                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 906                       "dropped when this client unmounts or is evicted\n",
 907                       obd->obd_name, PFID(&op_data->op_fid1));
 908                 GOTO(out, rc = -ENOMEM);
 909         }
 910
 911         if (u32_count > 0)
 912                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 913                                      u32_count * sizeof(__u32));
 914
 915         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 916         if (rc) {
 917                 ptlrpc_request_free(req);
 918                 req = NULL;
 919                 GOTO(out, rc);
 920         }
 921
 922         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 923          * portal whose threads are not taking any DLM locks and are therefore
 924          * always progressing */
 925         req->rq_request_portal = MDS_READPAGE_PORTAL;
 926         ptlrpc_at_set_req_timeout(req);
 927
 928         if (!(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 929                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 930                                         OP_XVALID_LAZYBLOCKS);
 931
 932         mdc_close_pack(req, op_data);
 933
 934         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 935                              obd->u.cli.cl_default_mds_easize);
 936
 937         ptlrpc_request_set_replen(req);
 938
 939         ptlrpc_get_mod_rpc_slot(req);
 940         rc = ptlrpc_queue_wait(req);
 941         ptlrpc_put_mod_rpc_slot(req);
 942
 943         if (req->rq_repmsg == NULL) {
 944                 CDEBUG(D_RPCTRACE, "request %p failed to send: rc = %d\n", req,
 945                        req->rq_status);
 946                 if (rc == 0)
 947                         rc = req->rq_status ?: -EIO;
 948         } else if (rc == 0 || rc == -EAGAIN) {
 949                 struct mdt_body *body;
 950
 951                 rc = lustre_msg_get_status(req->rq_repmsg);
 952                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 953                         DEBUG_REQ(D_ERROR, req,
 954                                   "type = PTL_RPC_MSG_ERR: rc = %d", rc);
 955                         if (rc > 0)
 956                                 rc = -rc;
 957                 }
 958                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 959                 if (body == NULL)
 960                         rc = -EPROTO;
 961         } else if (rc == -ESTALE) {
 962                 /**
 963                  * it can be allowed error after 3633 if open was committed and
 964                  * server failed before close was sent. Let's check if mod
 965                  * exists and return no error in that case
 966                  */
 967                 if (mod) {
 968                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 969                         LASSERT(mod->mod_open_req != NULL);
 970                         if (mod->mod_open_req->rq_committed)
 971                                 rc = 0;
 972                 }
 973         }
 974
 975 out:
 976         if (mod) {
 977                 if (rc != 0)
 978                         mod->mod_close_req = NULL;
 979                 /* Since now, mod is accessed through open_req only,
 980                  * thus close req does not keep a reference on mod anymore. */
 981                 obd_mod_put(mod);
 982         }
 983         *request = req;
 984
 985         RETURN(rc < 0 ? rc : saved_rc);
 986 }
 987
 988 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 989                        u64 offset, struct page **pages, int npages,
 990                        struct ptlrpc_request **request)
 991 {
 992         struct ptlrpc_request   *req;
 993         struct ptlrpc_bulk_desc *desc;
 994         int                      i;
 995         int                      resends = 0;
 996         int                      rc;
 997         ENTRY;
 998
 999         *request = NULL;
1000
1001 restart_bulk:
1002         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1003         if (req == NULL)
1004                 RETURN(-ENOMEM);
1005
1006         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1007         if (rc) {
1008                 ptlrpc_request_free(req);
1009                 RETURN(rc);
1010         }
1011
1012         req->rq_request_portal = MDS_READPAGE_PORTAL;
1013         ptlrpc_at_set_req_timeout(req);
1014
1015         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1016                                     PTLRPC_BULK_PUT_SINK,
1017                                     MDS_BULK_PORTAL,
1018                                     &ptlrpc_bulk_kiov_pin_ops);
1019         if (desc == NULL) {
1020                 ptlrpc_req_finished(req);
1021                 RETURN(-ENOMEM);
1022         }
1023
1024         /* NB req now owns desc and will free it when it gets freed */
1025         for (i = 0; i < npages; i++)
1026                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1027                                                  PAGE_SIZE);
1028
1029         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
1030
1031         ptlrpc_request_set_replen(req);
1032         rc = ptlrpc_queue_wait(req);
1033         if (rc) {
1034                 ptlrpc_req_finished(req);
1035                 if (rc != -ETIMEDOUT)
1036                         RETURN(rc);
1037
1038                 resends++;
1039                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1040                         CERROR("%s: too many resend retries: rc = %d\n",
1041                                exp->exp_obd->obd_name, -EIO);
1042                         RETURN(-EIO);
1043                 }
1044
1045                 /* If a signal interrupts then the timeout returned will
1046                  * not be zero. In that case return -EINTR
1047                  */
1048                 if (msleep_interruptible(resends * 1000))
1049                         RETURN(-EINTR);
1050
1051                 goto restart_bulk;
1052         }
1053
1054         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1055                                           req->rq_bulk->bd_nob_transferred);
1056         if (rc < 0) {
1057                 ptlrpc_req_finished(req);
1058                 RETURN(rc);
1059         }
1060
1061         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1062                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1063                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1064                        PAGE_SIZE * npages);
1065                 ptlrpc_req_finished(req);
1066                 RETURN(-EPROTO);
1067         }
1068
1069         *request = req;
1070         RETURN(0);
1071 }
1072
1073 static void mdc_release_page(struct page *page, int remove)
1074 {
1075         if (remove) {
1076                 lock_page(page);
1077                 if (likely(page->mapping != NULL))
1078                         delete_from_page_cache(page);
1079                 unlock_page(page);
1080         }
1081         put_page(page);
1082 }
1083
1084 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1085                                     __u64 *start, __u64 *end, int hash64)
1086 {
1087         /*
1088          * Complement of hash is used as an index so that
1089          * radix_tree_gang_lookup() can be used to find a page with starting
1090          * hash _smaller_ than one we are looking for.
1091          */
1092         unsigned long offset = hash_x_index(*hash, hash64);
1093         struct page *page;
1094         unsigned long flags;
1095         int found;
1096
1097         ll_xa_lock_irqsave(&mapping->i_pages, flags);
1098         found = radix_tree_gang_lookup(&mapping->page_tree,
1099                                        (void **)&page, offset, 1);
1100         if (found > 0 && !ll_xa_is_value(page)) {
1101                 struct lu_dirpage *dp;
1102
1103                 get_page(page);
1104                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1105                 /*
1106                  * In contrast to find_lock_page() we are sure that directory
1107                  * page cannot be truncated (while DLM lock is held) and,
1108                  * hence, can avoid restart.
1109                  *
1110                  * In fact, page cannot be locked here at all, because
1111                  * mdc_read_page_remote does synchronous io.
1112                  */
1113                 wait_on_page_locked(page);
1114                 if (PageUptodate(page)) {
1115                         dp = kmap(page);
1116                         if (BITS_PER_LONG == 32 && hash64) {
1117                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1118                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1119                                 *hash  = *hash >> 32;
1120                         } else {
1121                                 *start = le64_to_cpu(dp->ldp_hash_start);
1122                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1123                         }
1124                         if (unlikely(*start == 1 && *hash == 0))
1125                                 *hash = *start;
1126                         else
1127                                 LASSERTF(*start <= *hash, "start = %#llx"
1128                                          ",end = %#llx,hash = %#llx\n",
1129                                          *start, *end, *hash);
1130                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1131                               " hash %#llx\n", offset, *start, *end, *hash);
1132                         if (*hash > *end) {
1133                                 kunmap(page);
1134                                 mdc_release_page(page, 0);
1135                                 page = NULL;
1136                         } else if (*end != *start && *hash == *end) {
1137                                 /*
1138                                  * upon hash collision, remove this page,
1139                                  * otherwise put page reference, and
1140                                  * mdc_read_page_remote() will issue RPC to
1141                                  * fetch the page we want.
1142                                  */
1143                                 kunmap(page);
1144                                 mdc_release_page(page,
1145                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1146                                 page = NULL;
1147                         }
1148                 } else {
1149                         put_page(page);
1150                         page = ERR_PTR(-EIO);
1151                 }
1152         } else {
1153                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1154                 page = NULL;
1155         }
1156         return page;
1157 }
1158
1159 /*
1160  * Adjust a set of pages, each page containing an array of lu_dirpages,
1161  * so that each page can be used as a single logical lu_dirpage.
1162  *
1163  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1164  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1165  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1166  * value is used as a cookie to request the next lu_dirpage in a
1167  * directory listing that spans multiple pages (two in this example):
1168  *   ________
1169  *  |        |
1170  * .|--------v-------   -----.
1171  * |s|e|f|p|ent|ent| ... |ent|
1172  * '--|--------------   -----'   Each PAGE contains a single
1173  *    '------.                   lu_dirpage.
1174  * .---------v-------   -----.
1175  * |s|e|f|p|ent| 0 | ... | 0 |
1176  * '-----------------   -----'
1177  *
1178  * However, on hosts where the native VM page size (PAGE_SIZE) is
1179  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1180  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1181  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1182  * after it in the same PAGE (arrows simplified for brevity, but
1183  * in general e0==s1, e1==s2, etc.):
1184  *
1185  * .--------------------   -----.
1186  * |s0|e0|f0|p|ent|ent| ... |ent|
1187  * |---v----------------   -----|
1188  * |s1|e1|f1|p|ent|ent| ... |ent|
1189  * |---v----------------   -----|  Here, each PAGE contains
1190  *             ...                 multiple lu_dirpages.
1191  * |---v----------------   -----|
1192  * |s'|e'|f'|p|ent|ent| ... |ent|
1193  * '---|----------------   -----'
1194  *     v
1195  * .----------------------------.
1196  * |        next PAGE           |
1197  *
1198  * This structure is transformed into a single logical lu_dirpage as follows:
1199  *
1200  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1201  *   labeled 'next PAGE'.
1202  *
1203  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1204  *   a hash collision with the next page exists.
1205  *
1206  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1207  *   to the first entry of the next lu_dirpage.
1208  */
1209 #if PAGE_SIZE > LU_PAGE_SIZE
1210 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1211 {
1212         int i;
1213
1214         for (i = 0; i < cfs_pgs; i++) {
1215                 struct lu_dirpage *dp = kmap(pages[i]);
1216                 struct lu_dirpage *first = dp;
1217                 struct lu_dirent *end_dirent = NULL;
1218                 struct lu_dirent *ent;
1219                 __u64 hash_end = dp->ldp_hash_end;
1220                 __u32 flags = dp->ldp_flags;
1221
1222                 while (--lu_pgs > 0) {
1223                         ent = lu_dirent_start(dp);
1224                         for (end_dirent = ent; ent != NULL;
1225                              end_dirent = ent, ent = lu_dirent_next(ent));
1226
1227                         /* Advance dp to next lu_dirpage. */
1228                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1229
1230                         /* Check if we've reached the end of the PAGE. */
1231                         if (!((unsigned long)dp & ~PAGE_MASK))
1232                                 break;
1233
1234                         /* Save the hash and flags of this lu_dirpage. */
1235                         hash_end = dp->ldp_hash_end;
1236                         flags = dp->ldp_flags;
1237
1238                         /* Check if lu_dirpage contains no entries. */
1239                         if (end_dirent == NULL)
1240                                 break;
1241
1242                         /* Enlarge the end entry lde_reclen from 0 to
1243                          * first entry of next lu_dirpage. */
1244                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1245                         end_dirent->lde_reclen =
1246                                 cpu_to_le16((char *)(dp->ldp_entries) -
1247                                             (char *)end_dirent);
1248                 }
1249
1250                 first->ldp_hash_end = hash_end;
1251                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1252                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1253
1254                 kunmap(pages[i]);
1255         }
1256         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1257 }
1258 #else
1259 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1260 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1261
1262 /* parameters for readdir page */
1263 struct readpage_param {
1264         struct md_op_data       *rp_mod;
1265         __u64                   rp_off;
1266         int                     rp_hash64;
1267         struct obd_export       *rp_exp;
1268         struct md_callback      *rp_cb;
1269 };
1270
1271 /**
1272  * Read pages from server.
1273  *
1274  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1275  * a header lu_dirpage which describes the start/end hash, and whether this
1276  * page is empty (contains no dir entry) or hash collide with next page.
1277  * After client receives reply, several pages will be integrated into dir page
1278  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1279  * lu_dirpage for this integrated page will be adjusted.
1280  **/
1281 static int mdc_read_page_remote(void *data, struct page *page0)
1282 {
1283         struct readpage_param *rp = data;
1284         struct page **page_pool;
1285         struct page *page;
1286         struct lu_dirpage *dp;
1287         struct md_op_data *op_data = rp->rp_mod;
1288         struct ptlrpc_request *req;
1289         int max_pages;
1290         struct inode *inode;
1291         struct lu_fid *fid;
1292         int rd_pgs = 0; /* number of pages actually read */
1293         int npages;
1294         int i;
1295         int rc;
1296         ENTRY;
1297
1298         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1299         inode = op_data->op_data;
1300         fid = &op_data->op_fid1;
1301         LASSERT(inode != NULL);
1302
1303         OBD_ALLOC_PTR_ARRAY(page_pool, max_pages);
1304         if (page_pool != NULL) {
1305                 page_pool[0] = page0;
1306         } else {
1307                 page_pool = &page0;
1308                 max_pages = 1;
1309         }
1310
1311         for (npages = 1; npages < max_pages; npages++) {
1312                 page = page_cache_alloc(inode->i_mapping);
1313                 if (page == NULL)
1314                         break;
1315                 page_pool[npages] = page;
1316         }
1317
1318         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1319         if (rc < 0) {
1320                 /* page0 is special, which was added into page cache early */
1321                 delete_from_page_cache(page0);
1322         } else {
1323                 int lu_pgs;
1324
1325                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1326                         PAGE_SHIFT;
1327                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1328                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1329
1330                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1331
1332                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1333
1334                 SetPageUptodate(page0);
1335         }
1336         unlock_page(page0);
1337
1338         ptlrpc_req_finished(req);
1339         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1340         for (i = 1; i < npages; i++) {
1341                 unsigned long   offset;
1342                 __u64           hash;
1343                 int ret;
1344
1345                 page = page_pool[i];
1346
1347                 if (rc < 0 || i >= rd_pgs) {
1348                         put_page(page);
1349                         continue;
1350                 }
1351
1352                 SetPageUptodate(page);
1353
1354                 dp = kmap(page);
1355                 hash = le64_to_cpu(dp->ldp_hash_start);
1356                 kunmap(page);
1357
1358                 offset = hash_x_index(hash, rp->rp_hash64);
1359
1360                 prefetchw(&page->flags);
1361                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1362                                             GFP_KERNEL);
1363                 if (ret == 0)
1364                         unlock_page(page);
1365                 else
1366                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1367                                " rc = %d\n", offset, ret);
1368                 put_page(page);
1369         }
1370
1371         if (page_pool != &page0)
1372                 OBD_FREE_PTR_ARRAY(page_pool, max_pages);
1373
1374         RETURN(rc);
1375 }
1376
1377 /**
1378  * Read dir page from cache first, if it can not find it, read it from
1379  * server and add into the cache.
1380  *
1381  * \param[in] exp       MDC export
1382  * \param[in] op_data   client MD stack parameters, transfering parameters
1383  *                      between different layers on client MD stack.
1384  * \param[in] cb_op     callback required for ldlm lock enqueue during
1385  *                      read page
1386  * \param[in] hash_offset the hash offset of the page to be read
1387  * \param[in] ppage     the page to be read
1388  *
1389  * retval               = 0 get the page successfully
1390  *                      errno(<0) get the page failed
1391  */
1392 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1393                          struct md_callback *cb_op, __u64 hash_offset,
1394                          struct page **ppage)
1395 {
1396         struct lookup_intent    it = { .it_op = IT_READDIR };
1397         struct page             *page;
1398         struct inode            *dir = op_data->op_data;
1399         struct address_space    *mapping;
1400         struct lu_dirpage       *dp;
1401         __u64                   start = 0;
1402         __u64                   end = 0;
1403         struct lustre_handle    lockh;
1404         struct ptlrpc_request   *enq_req = NULL;
1405         struct readpage_param   rp_param;
1406         int rc;
1407
1408         ENTRY;
1409
1410         *ppage = NULL;
1411
1412         LASSERT(dir != NULL);
1413         mapping = dir->i_mapping;
1414
1415         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1416                              cb_op->md_blocking_ast, 0);
1417         if (enq_req != NULL)
1418                 ptlrpc_req_finished(enq_req);
1419
1420         if (rc < 0) {
1421                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1422                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1423                 RETURN(rc);
1424         }
1425
1426         rc = 0;
1427         lockh.cookie = it.it_lock_handle;
1428         mdc_set_lock_data(exp, &lockh, dir, NULL);
1429
1430         rp_param.rp_off = hash_offset;
1431         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1432         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1433                                rp_param.rp_hash64);
1434         if (IS_ERR(page)) {
1435                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1436                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1437                        rp_param.rp_off, PTR_ERR(page));
1438                 GOTO(out_unlock, rc = PTR_ERR(page));
1439         } else if (page != NULL) {
1440                 /*
1441                  * XXX nikita: not entirely correct handling of a corner case:
1442                  * suppose hash chain of entries with hash value HASH crosses
1443                  * border between pages P0 and P1. First both P0 and P1 are
1444                  * cached, seekdir() is called for some entry from the P0 part
1445                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1446                  * happens and finds P1, as it starts with matching hash
1447                  * value. Remaining entries from P0 part of the chain are
1448                  * skipped. (Is that really a bug?)
1449                  *
1450                  * Possible solutions: 0. don't cache P1 is such case, handle
1451                  * it as an "overflow" page. 1. invalidate all pages at
1452                  * once. 2. use HASH|1 as an index for P1.
1453                  */
1454                 GOTO(hash_collision, page);
1455         }
1456
1457         rp_param.rp_exp = exp;
1458         rp_param.rp_mod = op_data;
1459         page = read_cache_page(mapping,
1460                                hash_x_index(rp_param.rp_off,
1461                                             rp_param.rp_hash64),
1462                                mdc_read_page_remote, &rp_param);
1463         if (IS_ERR(page)) {
1464                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1465                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1466                        rp_param.rp_off, PTR_ERR(page));
1467                 GOTO(out_unlock, rc = PTR_ERR(page));
1468         }
1469
1470         wait_on_page_locked(page);
1471         (void)kmap(page);
1472         if (!PageUptodate(page)) {
1473                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1474                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1475                        rp_param.rp_off, -5);
1476                 goto fail;
1477         }
1478         if (!PageChecked(page))
1479                 SetPageChecked(page);
1480         if (PageError(page)) {
1481                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1482                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1483                        rp_param.rp_off, -5);
1484                 goto fail;
1485         }
1486
1487 hash_collision:
1488         dp = page_address(page);
1489         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1490                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1491                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1492                 rp_param.rp_off = hash_offset >> 32;
1493         } else {
1494                 start = le64_to_cpu(dp->ldp_hash_start);
1495                 end   = le64_to_cpu(dp->ldp_hash_end);
1496                 rp_param.rp_off = hash_offset;
1497         }
1498         if (end == start) {
1499                 LASSERT(start == rp_param.rp_off);
1500                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1501 #if BITS_PER_LONG == 32
1502                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1503                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1504                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1505 #endif
1506
1507                 /*
1508                  * Fetch whole overflow chain...
1509                  *
1510                  * XXX not yet.
1511                  */
1512                 goto fail;
1513         }
1514         *ppage = page;
1515 out_unlock:
1516         ldlm_lock_decref(&lockh, it.it_lock_mode);
1517         return rc;
1518 fail:
1519         kunmap(page);
1520         mdc_release_page(page, 1);
1521         rc = -EIO;
1522         goto out_unlock;
1523 }
1524
1525 static int mdc_statfs_interpret(const struct lu_env *env,
1526                                 struct ptlrpc_request *req, void *args, int rc)
1527 {
1528         struct obd_info *oinfo = args;
1529         struct obd_statfs *osfs;
1530
1531         if (!rc) {
1532                 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1533                 if (!osfs)
1534                         return -EPROTO;
1535
1536                 oinfo->oi_osfs = osfs;
1537
1538                 CDEBUG(D_CACHE, "blocks=%llu free=%llu avail=%llu "
1539                        "objects=%llu free=%llu state=%x\n",
1540                         osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
1541                         osfs->os_files, osfs->os_ffree, osfs->os_state);
1542         }
1543
1544         oinfo->oi_cb_up(oinfo, rc);
1545
1546         return rc;
1547 }
1548
1549 static int mdc_statfs_async(struct obd_export *exp,
1550                             struct obd_info *oinfo, time64_t max_age,
1551                             struct ptlrpc_request_set *unused)
1552 {
1553         struct ptlrpc_request *req;
1554         struct obd_info *aa;
1555
1556         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_STATFS,
1557                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1558         if (req == NULL)
1559                 return -ENOMEM;
1560
1561         ptlrpc_request_set_replen(req);
1562         req->rq_interpret_reply = mdc_statfs_interpret;
1563
1564         aa = ptlrpc_req_async_args(aa, req);
1565         *aa = *oinfo;
1566
1567         ptlrpcd_add_req(req);
1568
1569         return 0;
1570 }
1571
1572 static int mdc_statfs(const struct lu_env *env,
1573                       struct obd_export *exp, struct obd_statfs *osfs,
1574                       time64_t max_age, __u32 flags)
1575 {
1576         struct obd_device *obd = class_exp2obd(exp);
1577         struct req_format *fmt;
1578         struct ptlrpc_request *req;
1579         struct obd_statfs *msfs;
1580         struct obd_import *imp = NULL;
1581         int rc;
1582         ENTRY;
1583
1584         /*
1585          * Since the request might also come from lprocfs, so we need
1586          * sync this with client_disconnect_export Bug15684
1587          */
1588         down_read(&obd->u.cli.cl_sem);
1589         if (obd->u.cli.cl_import)
1590                 imp = class_import_get(obd->u.cli.cl_import);
1591         up_read(&obd->u.cli.cl_sem);
1592         if (!imp)
1593                 RETURN(-ENODEV);
1594
1595         fmt = &RQF_MDS_STATFS;
1596         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1597             (flags & OBD_STATFS_SUM))
1598                 fmt = &RQF_MDS_STATFS_NEW;
1599         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1600                                         MDS_STATFS);
1601         if (req == NULL)
1602                 GOTO(output, rc = -ENOMEM);
1603
1604         if ((flags & OBD_STATFS_SUM) &&
1605             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1606                 /* request aggregated states */
1607                 struct mdt_body *body;
1608
1609                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1610                 if (body == NULL)
1611                         GOTO(out, rc = -EPROTO);
1612                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1613         }
1614
1615         ptlrpc_request_set_replen(req);
1616
1617         if (flags & OBD_STATFS_NODELAY) {
1618                 /* procfs requests not want stay in wait for avoid deadlock */
1619                 req->rq_no_resend = 1;
1620                 req->rq_no_delay = 1;
1621         }
1622
1623         rc = ptlrpc_queue_wait(req);
1624         if (rc) {
1625                 /* check connection error first */
1626                 if (imp->imp_connect_error)
1627                         rc = imp->imp_connect_error;
1628                 GOTO(out, rc);
1629         }
1630
1631         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1632         if (msfs == NULL)
1633                 GOTO(out, rc = -EPROTO);
1634
1635         *osfs = *msfs;
1636         EXIT;
1637 out:
1638         ptlrpc_req_finished(req);
1639 output:
1640         class_import_put(imp);
1641         return rc;
1642 }
1643
1644 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1645 {
1646         __u32 keylen, vallen;
1647         void *key;
1648         int rc;
1649
1650         if (gf->gf_pathlen > PATH_MAX)
1651                 RETURN(-ENAMETOOLONG);
1652         if (gf->gf_pathlen < 2)
1653                 RETURN(-EOVERFLOW);
1654
1655         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1656         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1657                                 sizeof(struct lu_fid));
1658         OBD_ALLOC(key, keylen);
1659         if (key == NULL)
1660                 RETURN(-ENOMEM);
1661         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1662         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1663         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1664                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1665         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1666                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1667
1668         if (!fid_is_sane(&gf->gf_fid))
1669                 GOTO(out, rc = -EINVAL);
1670
1671         /* Val is struct getinfo_fid2path result plus path */
1672         vallen = sizeof(*gf) + gf->gf_pathlen;
1673
1674         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1675         if (rc != 0 && rc != -EREMOTE)
1676                 GOTO(out, rc);
1677
1678         if (vallen <= sizeof(*gf))
1679                 GOTO(out, rc = -EPROTO);
1680         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1681                 GOTO(out, rc = -EOVERFLOW);
1682
1683         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1684                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1685                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1686                /* only log the last 512 characters of the path */
1687                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1688
1689 out:
1690         OBD_FREE(key, keylen);
1691         return rc;
1692 }
1693
1694 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1695                                 struct hsm_progress_kernel *hpk)
1696 {
1697         struct obd_import               *imp = class_exp2cliimp(exp);
1698         struct hsm_progress_kernel      *req_hpk;
1699         struct ptlrpc_request           *req;
1700         int                              rc;
1701         ENTRY;
1702
1703         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1704                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1705         if (req == NULL)
1706                 GOTO(out, rc = -ENOMEM);
1707
1708         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1709
1710         /* Copy hsm_progress struct */
1711         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1712         if (req_hpk == NULL)
1713                 GOTO(out, rc = -EPROTO);
1714
1715         *req_hpk = *hpk;
1716         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1717
1718         ptlrpc_request_set_replen(req);
1719
1720         ptlrpc_get_mod_rpc_slot(req);
1721         rc = ptlrpc_queue_wait(req);
1722         ptlrpc_put_mod_rpc_slot(req);
1723
1724         GOTO(out, rc);
1725 out:
1726         ptlrpc_req_finished(req);
1727         return rc;
1728 }
1729 /**
1730  * Send hsm_ct_register to MDS
1731  *
1732  * \param[in]   imp             import
1733  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1734  *                              else it is the count of archive_ids
1735  * \param[in]   archives        if in bitmap format, it is NULL,
1736  *                              else it is archive_id lists
1737  */
1738 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1739                                    __u32 *archives)
1740 {
1741         struct ptlrpc_request *req;
1742         __u32 *archive_array;
1743         size_t archives_size;
1744         int rc;
1745         ENTRY;
1746
1747         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1748         if (req == NULL)
1749                 RETURN(-ENOMEM);
1750
1751         if (archives != NULL)
1752                 archives_size = sizeof(*archive_array) * archive_count;
1753         else
1754                 archives_size = sizeof(archive_count);
1755
1756         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1757                              RCL_CLIENT, archives_size);
1758
1759         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1760         if (rc) {
1761                 ptlrpc_request_free(req);
1762                 RETURN(-ENOMEM);
1763         }
1764
1765         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1766
1767         archive_array = req_capsule_client_get(&req->rq_pill,
1768                                                &RMF_MDS_HSM_ARCHIVE);
1769         if (archive_array == NULL)
1770                 GOTO(out, rc = -EPROTO);
1771
1772         if (archives != NULL)
1773                 memcpy(archive_array, archives, archives_size);
1774         else
1775                 *archive_array = archive_count;
1776
1777         ptlrpc_request_set_replen(req);
1778         req->rq_no_resend = 1;
1779
1780         rc = mdc_queue_wait(req);
1781         GOTO(out, rc);
1782 out:
1783         ptlrpc_req_finished(req);
1784         return rc;
1785 }
1786
1787 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1788                                       struct md_op_data *op_data)
1789 {
1790         struct hsm_current_action       *hca = op_data->op_data;
1791         struct hsm_current_action       *req_hca;
1792         struct ptlrpc_request           *req;
1793         int                              rc;
1794         ENTRY;
1795
1796         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1797                                    &RQF_MDS_HSM_ACTION);
1798         if (req == NULL)
1799                 RETURN(-ENOMEM);
1800
1801         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1802         if (rc) {
1803                 ptlrpc_request_free(req);
1804                 RETURN(rc);
1805         }
1806
1807         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1808                       op_data->op_suppgids[0], 0);
1809
1810         ptlrpc_request_set_replen(req);
1811
1812         rc = mdc_queue_wait(req);
1813         if (rc)
1814                 GOTO(out, rc);
1815
1816         req_hca = req_capsule_server_get(&req->rq_pill,
1817                                          &RMF_MDS_HSM_CURRENT_ACTION);
1818         if (req_hca == NULL)
1819                 GOTO(out, rc = -EPROTO);
1820
1821         *hca = *req_hca;
1822
1823         EXIT;
1824 out:
1825         ptlrpc_req_finished(req);
1826         return rc;
1827 }
1828
1829 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1830 {
1831         struct ptlrpc_request   *req;
1832         int                      rc;
1833         ENTRY;
1834
1835         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1836                                         LUSTRE_MDS_VERSION,
1837                                         MDS_HSM_CT_UNREGISTER);
1838         if (req == NULL)
1839                 GOTO(out, rc = -ENOMEM);
1840
1841         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1842
1843         ptlrpc_request_set_replen(req);
1844
1845         rc = mdc_queue_wait(req);
1846         GOTO(out, rc);
1847 out:
1848         ptlrpc_req_finished(req);
1849         return rc;
1850 }
1851
1852 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1853                                  struct md_op_data *op_data)
1854 {
1855         struct hsm_user_state   *hus = op_data->op_data;
1856         struct hsm_user_state   *req_hus;
1857         struct ptlrpc_request   *req;
1858         int                      rc;
1859         ENTRY;
1860
1861         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1862                                    &RQF_MDS_HSM_STATE_GET);
1863         if (req == NULL)
1864                 RETURN(-ENOMEM);
1865
1866         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1867         if (rc != 0) {
1868                 ptlrpc_request_free(req);
1869                 RETURN(rc);
1870         }
1871
1872         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1873                       op_data->op_suppgids[0], 0);
1874
1875         ptlrpc_request_set_replen(req);
1876
1877         rc = mdc_queue_wait(req);
1878         if (rc)
1879                 GOTO(out, rc);
1880
1881         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1882         if (req_hus == NULL)
1883                 GOTO(out, rc = -EPROTO);
1884
1885         *hus = *req_hus;
1886
1887         EXIT;
1888 out:
1889         ptlrpc_req_finished(req);
1890         return rc;
1891 }
1892
1893 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1894                                  struct md_op_data *op_data)
1895 {
1896         struct hsm_state_set    *hss = op_data->op_data;
1897         struct hsm_state_set    *req_hss;
1898         struct ptlrpc_request   *req;
1899         int                      rc;
1900         ENTRY;
1901
1902         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1903                                    &RQF_MDS_HSM_STATE_SET);
1904         if (req == NULL)
1905                 RETURN(-ENOMEM);
1906
1907         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1908         if (rc) {
1909                 ptlrpc_request_free(req);
1910                 RETURN(rc);
1911         }
1912
1913         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1914                       op_data->op_suppgids[0], 0);
1915
1916         /* Copy states */
1917         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1918         if (req_hss == NULL)
1919                 GOTO(out, rc = -EPROTO);
1920         *req_hss = *hss;
1921
1922         ptlrpc_request_set_replen(req);
1923
1924         ptlrpc_get_mod_rpc_slot(req);
1925         rc = ptlrpc_queue_wait(req);
1926         ptlrpc_put_mod_rpc_slot(req);
1927
1928         GOTO(out, rc);
1929 out:
1930         ptlrpc_req_finished(req);
1931         return rc;
1932 }
1933
1934 static int mdc_ioc_hsm_request(struct obd_export *exp,
1935                                struct hsm_user_request *hur)
1936 {
1937         struct obd_import       *imp = class_exp2cliimp(exp);
1938         struct ptlrpc_request   *req;
1939         struct hsm_request      *req_hr;
1940         struct hsm_user_item    *req_hui;
1941         char                    *req_opaque;
1942         int                      rc;
1943         ENTRY;
1944
1945         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1946         if (req == NULL)
1947                 GOTO(out, rc = -ENOMEM);
1948
1949         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1950                              hur->hur_request.hr_itemcount
1951                              * sizeof(struct hsm_user_item));
1952         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1953                              hur->hur_request.hr_data_len);
1954
1955         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1956         if (rc) {
1957                 ptlrpc_request_free(req);
1958                 RETURN(rc);
1959         }
1960
1961         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1962
1963         /* Copy hsm_request struct */
1964         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1965         if (req_hr == NULL)
1966                 GOTO(out, rc = -EPROTO);
1967         *req_hr = hur->hur_request;
1968
1969         /* Copy hsm_user_item structs */
1970         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1971         if (req_hui == NULL)
1972                 GOTO(out, rc = -EPROTO);
1973         memcpy(req_hui, hur->hur_user_item,
1974                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1975
1976         /* Copy opaque field */
1977         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1978         if (req_opaque == NULL)
1979                 GOTO(out, rc = -EPROTO);
1980         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1981
1982         ptlrpc_request_set_replen(req);
1983
1984         ptlrpc_get_mod_rpc_slot(req);
1985         rc = ptlrpc_queue_wait(req);
1986         ptlrpc_put_mod_rpc_slot(req);
1987
1988         GOTO(out, rc);
1989
1990 out:
1991         ptlrpc_req_finished(req);
1992         return rc;
1993 }
1994
1995 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1996                                 struct lustre_kernelcomm *lk);
1997
1998 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1999                         struct obd_quotactl *oqctl)
2000 {
2001         struct ptlrpc_request *req;
2002         struct obd_quotactl *oqc;
2003         int rc;
2004         ENTRY;
2005
2006         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_QUOTACTL);
2007         if (req == NULL)
2008                 RETURN(-ENOMEM);
2009
2010
2011         if (LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd))
2012                 req_capsule_set_size(&req->rq_pill,
2013                                      &RMF_OBD_QUOTACTL,
2014                                      RCL_CLIENT,
2015                                      sizeof(*oqc) + LOV_MAXPOOLNAME + 1);
2016
2017         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION,
2018                                  MDS_QUOTACTL);
2019         if (rc) {
2020                 ptlrpc_request_free(req);
2021                 RETURN(rc);
2022         }
2023
2024         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2025         QCTL_COPY(oqc, oqctl);
2026
2027         ptlrpc_request_set_replen(req);
2028         ptlrpc_at_set_req_timeout(req);
2029
2030         rc = ptlrpc_queue_wait(req);
2031         if (rc) {
2032                 CERROR("%s: ptlrpc_queue_wait failed: rc = %d\n",
2033                        exp->exp_obd->obd_name, rc);
2034                 GOTO(out, rc);
2035         }
2036
2037         if (req->rq_repmsg &&
2038             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2039                 QCTL_COPY(oqctl, oqc);
2040         } else if (!rc) {
2041                 rc = -EPROTO;
2042                 CERROR("%s: cannot unpack obd_quotactl: rc = %d\n",
2043                         exp->exp_obd->obd_name, rc);
2044         }
2045 out:
2046         ptlrpc_req_finished(req);
2047
2048         RETURN(rc);
2049 }
2050
2051 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2052                                 struct md_op_data *op_data)
2053 {
2054         LIST_HEAD(cancels);
2055         struct ptlrpc_request   *req;
2056         int                      rc, count;
2057         struct mdc_swap_layouts *msl, *payload;
2058         ENTRY;
2059
2060         msl = op_data->op_data;
2061
2062         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2063          * first thing it will do is to cancel the 2 layout
2064          * locks held by this client.
2065          * So the client must cancel its layout locks on the 2 fids
2066          * with the request RPC to avoid extra RPC round trips.
2067          */
2068         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2069                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2070                                         MDS_INODELOCK_XATTR);
2071         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2072                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2073                                          MDS_INODELOCK_XATTR);
2074
2075         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2076                                    &RQF_MDS_SWAP_LAYOUTS);
2077         if (req == NULL) {
2078                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2079                 RETURN(-ENOMEM);
2080         }
2081
2082         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2083         if (rc) {
2084                 ptlrpc_request_free(req);
2085                 RETURN(rc);
2086         }
2087
2088         mdc_swap_layouts_pack(req, op_data);
2089
2090         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2091         LASSERT(payload);
2092
2093         *payload = *msl;
2094
2095         ptlrpc_request_set_replen(req);
2096
2097         rc = ptlrpc_queue_wait(req);
2098         if (rc)
2099                 GOTO(out, rc);
2100         EXIT;
2101
2102 out:
2103         ptlrpc_req_finished(req);
2104         return rc;
2105 }
2106
2107 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2108                          void *karg, void __user *uarg)
2109 {
2110         struct obd_device *obd = exp->exp_obd;
2111         struct obd_ioctl_data *data = karg;
2112         struct obd_import *imp = obd->u.cli.cl_import;
2113         int rc;
2114         ENTRY;
2115
2116         if (!try_module_get(THIS_MODULE)) {
2117                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2118                        module_name(THIS_MODULE));
2119                 return -EINVAL;
2120         }
2121         switch (cmd) {
2122         case OBD_IOC_FID2PATH:
2123                 rc = mdc_ioc_fid2path(exp, karg);
2124                 GOTO(out, rc);
2125         case LL_IOC_HSM_CT_START:
2126                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2127                 /* ignore if it was already registered on this MDS. */
2128                 if (rc == -EEXIST)
2129                         rc = 0;
2130                 GOTO(out, rc);
2131         case LL_IOC_HSM_PROGRESS:
2132                 rc = mdc_ioc_hsm_progress(exp, karg);
2133                 GOTO(out, rc);
2134         case LL_IOC_HSM_STATE_GET:
2135                 rc = mdc_ioc_hsm_state_get(exp, karg);
2136                 GOTO(out, rc);
2137         case LL_IOC_HSM_STATE_SET:
2138                 rc = mdc_ioc_hsm_state_set(exp, karg);
2139                 GOTO(out, rc);
2140         case LL_IOC_HSM_ACTION:
2141                 rc = mdc_ioc_hsm_current_action(exp, karg);
2142                 GOTO(out, rc);
2143         case LL_IOC_HSM_REQUEST:
2144                 rc = mdc_ioc_hsm_request(exp, karg);
2145                 GOTO(out, rc);
2146         case OBD_IOC_CLIENT_RECOVER:
2147                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2148                 if (rc < 0)
2149                         GOTO(out, rc);
2150                 GOTO(out, rc = 0);
2151         case IOC_OSC_SET_ACTIVE:
2152                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2153                 GOTO(out, rc);
2154         /*
2155          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2156          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2157          * there'd be no LMV layer thus we might be called here. Eventually
2158          * this code should be removed.
2159          * bz20731, LU-592.
2160          */
2161         case IOC_OBD_STATFS: {
2162                 struct obd_statfs stat_buf = {0};
2163
2164                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2165                         GOTO(out, rc = -ENODEV);
2166
2167                 /* copy UUID */
2168                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2169                                  min((int)data->ioc_plen2,
2170                                      (int)sizeof(struct obd_uuid))))
2171                         GOTO(out, rc = -EFAULT);
2172
2173                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2174                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2175                                 0);
2176                 if (rc != 0)
2177                         GOTO(out, rc);
2178
2179                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2180                                      min((int) data->ioc_plen1,
2181                                          (int) sizeof(stat_buf))))
2182                         GOTO(out, rc = -EFAULT);
2183
2184                 GOTO(out, rc = 0);
2185         }
2186         case OBD_IOC_QUOTACTL: {
2187                 struct if_quotactl *qctl = karg;
2188                 struct obd_quotactl *oqctl;
2189
2190                 OBD_ALLOC_PTR(oqctl);
2191                 if (oqctl == NULL)
2192                         GOTO(out, rc = -ENOMEM);
2193
2194                 QCTL_COPY(oqctl, qctl);
2195                 rc = obd_quotactl(exp, oqctl);
2196                 if (rc == 0) {
2197                         QCTL_COPY(qctl, oqctl);
2198                         qctl->qc_valid = QC_MDTIDX;
2199                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2200                 }
2201
2202                 OBD_FREE_PTR(oqctl);
2203                 GOTO(out, rc);
2204         }
2205         case LL_IOC_GET_CONNECT_FLAGS:
2206                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2207                                  sizeof(*exp_connect_flags_ptr(exp))))
2208                         GOTO(out, rc = -EFAULT);
2209
2210                 GOTO(out, rc = 0);
2211         case LL_IOC_LOV_SWAP_LAYOUTS:
2212                 rc = mdc_ioc_swap_layouts(exp, karg);
2213                 GOTO(out, rc);
2214         default:
2215                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2216                 GOTO(out, rc = -ENOTTY);
2217         }
2218 out:
2219         module_put(THIS_MODULE);
2220
2221         return rc;
2222 }
2223
2224 static int mdc_get_info_rpc(struct obd_export *exp,
2225                             u32 keylen, void *key,
2226                             u32 vallen, void *val)
2227 {
2228         struct obd_import      *imp = class_exp2cliimp(exp);
2229         struct ptlrpc_request  *req;
2230         char                   *tmp;
2231         int                     rc = -EINVAL;
2232         ENTRY;
2233
2234         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2235         if (req == NULL)
2236                 RETURN(-ENOMEM);
2237
2238         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2239                              RCL_CLIENT, keylen);
2240         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2241                              RCL_CLIENT, sizeof(vallen));
2242
2243         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2244         if (rc) {
2245                 ptlrpc_request_free(req);
2246                 RETURN(rc);
2247         }
2248
2249         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2250         memcpy(tmp, key, keylen);
2251         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2252         memcpy(tmp, &vallen, sizeof(vallen));
2253
2254         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2255                              RCL_SERVER, vallen);
2256         ptlrpc_request_set_replen(req);
2257
2258         rc = ptlrpc_queue_wait(req);
2259         /* -EREMOTE means the get_info result is partial, and it needs to
2260          * continue on another MDT, see fid2path part in lmv_iocontrol */
2261         if (rc == 0 || rc == -EREMOTE) {
2262                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2263                 memcpy(val, tmp, vallen);
2264                 if (ptlrpc_rep_need_swab(req)) {
2265                         if (KEY_IS(KEY_FID2PATH))
2266                                 lustre_swab_fid2path(val);
2267                 }
2268         }
2269         ptlrpc_req_finished(req);
2270
2271         RETURN(rc);
2272 }
2273
2274 static void lustre_swab_hai(struct hsm_action_item *h)
2275 {
2276         __swab32s(&h->hai_len);
2277         __swab32s(&h->hai_action);
2278         lustre_swab_lu_fid(&h->hai_fid);
2279         lustre_swab_lu_fid(&h->hai_dfid);
2280         __swab64s(&h->hai_cookie);
2281         __swab64s(&h->hai_extent.offset);
2282         __swab64s(&h->hai_extent.length);
2283         __swab64s(&h->hai_gid);
2284 }
2285
2286 static void lustre_swab_hal(struct hsm_action_list *h)
2287 {
2288         struct hsm_action_item  *hai;
2289         __u32                    i;
2290
2291         __swab32s(&h->hal_version);
2292         __swab32s(&h->hal_count);
2293         __swab32s(&h->hal_archive_id);
2294         __swab64s(&h->hal_flags);
2295         hai = hai_first(h);
2296         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2297                 lustre_swab_hai(hai);
2298 }
2299
2300 static void lustre_swab_kuch(struct kuc_hdr *l)
2301 {
2302         __swab16s(&l->kuc_magic);
2303         /* __u8 l->kuc_transport */
2304         __swab16s(&l->kuc_msgtype);
2305         __swab16s(&l->kuc_msglen);
2306 }
2307
2308 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2309                                 struct lustre_kernelcomm *lk)
2310 {
2311         struct obd_import *imp = class_exp2cliimp(exp);
2312         int rc = 0;
2313
2314         if (lk->lk_group != KUC_GRP_HSM) {
2315                 CERROR("Bad copytool group %d\n", lk->lk_group);
2316                 return -EINVAL;
2317         }
2318
2319         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2320                lk->lk_uid, lk->lk_group, lk->lk_flags);
2321
2322         if (lk->lk_flags & LK_FLG_STOP) {
2323                 /* Unregister with the coordinator */
2324                 rc = mdc_ioc_hsm_ct_unregister(imp);
2325         } else {
2326                 __u32 *archives = NULL;
2327
2328                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2329                         archives = lk->lk_data;
2330
2331                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2332         }
2333
2334         return rc;
2335 }
2336
2337 /**
2338  * Send a message to any listening copytools
2339  * @param val KUC message (kuc_hdr + hsm_action_list)
2340  * @param len total length of message
2341  */
2342 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2343                                  size_t len, void *val)
2344 {
2345         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2346         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2347         int                      rc;
2348         ENTRY;
2349
2350         if (len < sizeof(*lh) + sizeof(*hal)) {
2351                 CERROR("Short HSM message %zu < %zu\n", len,
2352                        sizeof(*lh) + sizeof(*hal));
2353                 RETURN(-EPROTO);
2354         }
2355         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2356                 lustre_swab_kuch(lh);
2357                 lustre_swab_hal(hal);
2358         } else if (lh->kuc_magic != KUC_MAGIC) {
2359                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2360                 RETURN(-EPROTO);
2361         }
2362
2363         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2364                "on %s\n",
2365                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2366                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2367
2368         /* Broadcast to HSM listeners */
2369         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2370
2371         RETURN(rc);
2372 }
2373
2374 /**
2375  * callback function passed to kuc for re-registering each HSM copytool
2376  * running on MDC, after MDT shutdown/recovery.
2377  * @param data copytool registration data
2378  * @param cb_arg callback argument (obd_import)
2379  */
2380 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2381 {
2382         struct obd_import *imp = (struct obd_import *)cb_arg;
2383         struct kkuc_ct_data *kcd = data;
2384         __u32 *archives = NULL;
2385         int rc;
2386
2387         if (kcd == NULL ||
2388             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2389              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2390                 return -EPROTO;
2391
2392         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2393                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2394                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2395                        kcd->kcd_nr_archives);
2396         } else {
2397                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2398                        "(archive nr = %u)\n",
2399                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2400                 if (kcd->kcd_nr_archives != 0)
2401                         archives = kcd->kcd_archives;
2402         }
2403
2404         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2405         /* ignore error if the copytool is already registered */
2406         return (rc == -EEXIST) ? 0 : rc;
2407 }
2408
2409 /**
2410  * Re-establish all kuc contexts with MDT
2411  * after MDT shutdown/recovery.
2412  */
2413 static int mdc_kuc_reregister(struct obd_import *imp)
2414 {
2415         /* re-register HSM agents */
2416         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2417                                          mdc_hsm_ct_reregister, imp);
2418 }
2419
2420 static int mdc_set_info_async(const struct lu_env *env,
2421                               struct obd_export *exp,
2422                               u32 keylen, void *key,
2423                               u32 vallen, void *val,
2424                               struct ptlrpc_request_set *set)
2425 {
2426         struct obd_import       *imp = class_exp2cliimp(exp);
2427         int                      rc;
2428         ENTRY;
2429
2430         if (KEY_IS(KEY_READ_ONLY)) {
2431                 if (vallen != sizeof(int))
2432                         RETURN(-EINVAL);
2433
2434                 spin_lock(&imp->imp_lock);
2435                 if (*((int *)val)) {
2436                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2437                         imp->imp_connect_data.ocd_connect_flags |=
2438                                                         OBD_CONNECT_RDONLY;
2439                 } else {
2440                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2441                         imp->imp_connect_data.ocd_connect_flags &=
2442                                                         ~OBD_CONNECT_RDONLY;
2443                 }
2444                 spin_unlock(&imp->imp_lock);
2445
2446                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2447                                        keylen, key, vallen, val, set);
2448                 RETURN(rc);
2449         }
2450         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2451                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2452                                        keylen, key, vallen, val, set);
2453                 RETURN(rc);
2454         }
2455         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2456                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2457                                            val);
2458                 RETURN(rc);
2459         }
2460
2461         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2462                 __u32 *default_easize = val;
2463
2464                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2465                 RETURN(0);
2466         }
2467
2468         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2469         RETURN(rc);
2470 }
2471
2472 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2473                         __u32 keylen, void *key, __u32 *vallen, void *val)
2474 {
2475         int rc = -EINVAL;
2476
2477         if (KEY_IS(KEY_MAX_EASIZE)) {
2478                 __u32 mdsize, *max_easize;
2479
2480                 if (*vallen != sizeof(int))
2481                         RETURN(-EINVAL);
2482                 mdsize = *(__u32 *)val;
2483                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2484                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2485                 max_easize = val;
2486                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2487                 RETURN(0);
2488         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2489                 __u32 *default_easize;
2490
2491                 if (*vallen != sizeof(int))
2492                         RETURN(-EINVAL);
2493                 default_easize = val;
2494                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2495                 RETURN(0);
2496         } else if (KEY_IS(KEY_CONN_DATA)) {
2497                 struct obd_import *imp = class_exp2cliimp(exp);
2498                 struct obd_connect_data *data = val;
2499
2500                 if (*vallen != sizeof(*data))
2501                         RETURN(-EINVAL);
2502
2503                 *data = imp->imp_connect_data;
2504                 RETURN(0);
2505         } else if (KEY_IS(KEY_TGT_COUNT)) {
2506                 *((__u32 *)val) = 1;
2507                 RETURN(0);
2508         }
2509
2510         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2511
2512         RETURN(rc);
2513 }
2514
2515 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2516                      struct ptlrpc_request **request)
2517 {
2518         struct ptlrpc_request *req;
2519         int                    rc;
2520         ENTRY;
2521
2522         *request = NULL;
2523         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2524         if (req == NULL)
2525                 RETURN(-ENOMEM);
2526
2527         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2528         if (rc) {
2529                 ptlrpc_request_free(req);
2530                 RETURN(rc);
2531         }
2532
2533         mdc_pack_body(req, fid, 0, 0, -1, 0);
2534
2535         ptlrpc_request_set_replen(req);
2536
2537         rc = ptlrpc_queue_wait(req);
2538         if (rc)
2539                 ptlrpc_req_finished(req);
2540         else
2541                 *request = req;
2542         RETURN(rc);
2543 }
2544
2545 struct mdc_rmfid_args {
2546         int *mra_rcs;
2547         int mra_nr;
2548 };
2549
2550 int mdc_rmfid_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2551                           void *args, int rc)
2552 {
2553         struct mdc_rmfid_args *aa;
2554         int *rcs, size;
2555         ENTRY;
2556
2557         if (!rc) {
2558                 aa = ptlrpc_req_async_args(aa, req);
2559
2560                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2561                                             RCL_SERVER);
2562                 LASSERT(size == sizeof(int) * aa->mra_nr);
2563                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2564                 LASSERT(rcs);
2565                 LASSERT(aa->mra_rcs);
2566                 LASSERT(aa->mra_nr);
2567                 memcpy(aa->mra_rcs, rcs, size);
2568         }
2569
2570         RETURN(rc);
2571 }
2572
2573 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2574                      int *rcs, struct ptlrpc_request_set *set)
2575 {
2576         struct ptlrpc_request *req;
2577         struct mdc_rmfid_args *aa;
2578         struct mdt_body *b;
2579         struct lu_fid *tmp;
2580         int rc, flen;
2581         ENTRY;
2582
2583         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2584         if (req == NULL)
2585                 RETURN(-ENOMEM);
2586
2587         flen = fa->fa_nr * sizeof(struct lu_fid);
2588         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2589                              RCL_CLIENT, flen);
2590         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2591                              RCL_SERVER, flen);
2592         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2593                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2594         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2595         if (rc) {
2596                 ptlrpc_request_free(req);
2597                 RETURN(rc);
2598         }
2599         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2600         memcpy(tmp, fa->fa_fids, flen);
2601
2602         mdc_pack_body(req, NULL, 0, 0, -1, 0);
2603         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2604         b->mbo_ctime = ktime_get_real_seconds();
2605
2606         ptlrpc_request_set_replen(req);
2607
2608         LASSERT(rcs);
2609         aa = ptlrpc_req_async_args(aa, req);
2610         aa->mra_rcs = rcs;
2611         aa->mra_nr = fa->fa_nr;
2612         req->rq_interpret_reply = mdc_rmfid_interpret;
2613
2614         ptlrpc_set_add_req(set, req);
2615         ptlrpc_check_set(NULL, set);
2616
2617         RETURN(rc);
2618 }
2619
2620 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2621                             enum obd_import_event event)
2622 {
2623         struct client_obd *cli = &obd->u.cli;
2624         int rc = 0;
2625
2626         LASSERT(imp->imp_obd == obd);
2627
2628         switch (event) {
2629         case IMP_EVENT_DISCON:
2630                 spin_lock(&cli->cl_loi_list_lock);
2631                 cli->cl_avail_grant = 0;
2632                 cli->cl_lost_grant = 0;
2633                 spin_unlock(&cli->cl_loi_list_lock);
2634                 break;
2635         case IMP_EVENT_INACTIVE:
2636                 /*
2637                  * Flush current sequence to make client obtain new one
2638                  * from server in case of disconnect/reconnect.
2639                  */
2640                 down_read(&cli->cl_seq_rwsem);
2641                 if (cli->cl_seq)
2642                         seq_client_flush(cli->cl_seq);
2643                 up_read(&cli->cl_seq_rwsem);
2644
2645                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2646                 break;
2647         case IMP_EVENT_INVALIDATE: {
2648                 struct ldlm_namespace *ns = obd->obd_namespace;
2649                 struct lu_env *env;
2650                 __u16 refcheck;
2651
2652                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2653
2654                 env = cl_env_get(&refcheck);
2655                 if (!IS_ERR(env)) {
2656                         /* Reset grants. All pages go to failing rpcs due to
2657                          * the invalid import.
2658                          */
2659                         osc_io_unplug(env, cli, NULL);
2660
2661                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2662                                                  osc_ldlm_resource_invalidate,
2663                                                  env, 0);
2664                         cl_env_put(env, &refcheck);
2665                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2666                 } else {
2667                         rc = PTR_ERR(env);
2668                 }
2669                 break;
2670         }
2671         case IMP_EVENT_ACTIVE:
2672                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2673                 /* redo the kuc registration after reconnecting */
2674                 if (rc == 0)
2675                         rc = mdc_kuc_reregister(imp);
2676                 break;
2677         case IMP_EVENT_OCD: {
2678                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2679
2680                 if (OCD_HAS_FLAG(ocd, GRANT))
2681                         osc_init_grant(cli, ocd);
2682
2683                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2684                 break;
2685         }
2686         case IMP_EVENT_DEACTIVATE:
2687         case IMP_EVENT_ACTIVATE:
2688                 break;
2689         default:
2690                 CERROR("Unknown import event %x\n", event);
2691                 LBUG();
2692         }
2693         RETURN(rc);
2694 }
2695
2696 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2697                   struct lu_fid *fid, struct md_op_data *op_data)
2698 {
2699         struct client_obd *cli = &exp->exp_obd->u.cli;
2700         int rc = -EIO;
2701
2702         ENTRY;
2703
2704         down_read(&cli->cl_seq_rwsem);
2705         if (cli->cl_seq)
2706                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2707         up_read(&cli->cl_seq_rwsem);
2708
2709         RETURN(rc);
2710 }
2711
2712 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2713 {
2714         struct client_obd *cli = &exp->exp_obd->u.cli;
2715         return &cli->cl_target_uuid;
2716 }
2717
2718 /**
2719  * Determine whether the lock can be canceled before replaying it during
2720  * recovery, non zero value will be return if the lock can be canceled,
2721  * or zero returned for not
2722  */
2723 static int mdc_cancel_weight(struct ldlm_lock *lock)
2724 {
2725         if (lock->l_resource->lr_type != LDLM_IBITS)
2726                 RETURN(0);
2727
2728         /* FIXME: if we ever get into a situation where there are too many
2729          * opened files with open locks on a single node, then we really
2730          * should replay these open locks to reget it */
2731         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2732                 RETURN(0);
2733
2734         /* Special case for DoM locks, cancel only unused and granted locks */
2735         if (ldlm_has_dom(lock) &&
2736             (lock->l_granted_mode != lock->l_req_mode ||
2737              osc_ldlm_weigh_ast(lock) != 0))
2738                 RETURN(0);
2739
2740         RETURN(1);
2741 }
2742
2743 static int mdc_resource_inode_free(struct ldlm_resource *res)
2744 {
2745         if (res->lr_lvb_inode)
2746                 res->lr_lvb_inode = NULL;
2747
2748         return 0;
2749 }
2750
2751 static struct ldlm_valblock_ops inode_lvbo = {
2752         .lvbo_free = mdc_resource_inode_free
2753 };
2754
2755 static int mdc_llog_init(struct obd_device *obd)
2756 {
2757         struct obd_llog_group   *olg = &obd->obd_olg;
2758         struct llog_ctxt        *ctxt;
2759         int                      rc;
2760
2761         ENTRY;
2762
2763         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2764                         &llog_client_ops);
2765         if (rc < 0)
2766                 RETURN(rc);
2767
2768         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2769         llog_initiator_connect(ctxt);
2770         llog_ctxt_put(ctxt);
2771
2772         RETURN(0);
2773 }
2774
2775 static void mdc_llog_finish(struct obd_device *obd)
2776 {
2777         struct llog_ctxt *ctxt;
2778
2779         ENTRY;
2780
2781         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2782         if (ctxt != NULL)
2783                 llog_cleanup(NULL, ctxt);
2784
2785         EXIT;
2786 }
2787
2788 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2789 {
2790         int rc;
2791
2792         ENTRY;
2793
2794         rc = osc_setup_common(obd, cfg);
2795         if (rc < 0)
2796                 RETURN(rc);
2797
2798         rc = mdc_tunables_init(obd);
2799         if (rc)
2800                 GOTO(err_osc_cleanup, rc);
2801
2802         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2803
2804         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2805
2806         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2807
2808         rc = mdc_llog_init(obd);
2809         if (rc) {
2810                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2811                        obd->obd_name, rc);
2812                 GOTO(err_llog_cleanup, rc);
2813         }
2814
2815         rc = mdc_changelog_cdev_init(obd);
2816         if (rc) {
2817                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2818                        obd->obd_name, rc);
2819                 GOTO(err_changelog_cleanup, rc);
2820         }
2821
2822         RETURN(rc);
2823
2824 err_changelog_cleanup:
2825         mdc_llog_finish(obd);
2826 err_llog_cleanup:
2827         lprocfs_free_md_stats(obd);
2828         ptlrpc_lprocfs_unregister_obd(obd);
2829 err_osc_cleanup:
2830         osc_cleanup_common(obd);
2831         return rc;
2832 }
2833
2834 /* Initialize the default and maximum LOV EA sizes.  This allows
2835  * us to make MDS RPCs with large enough reply buffers to hold a default
2836  * sized EA without having to calculate this (via a call into the
2837  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2838  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2839  * a large number of stripes is possible.  If a larger reply buffer is
2840  * required it will be reallocated in the ptlrpc layer due to overflow.
2841  */
2842 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2843                             __u32 def_easize)
2844 {
2845         struct obd_device *obd = exp->exp_obd;
2846         struct client_obd *cli = &obd->u.cli;
2847         ENTRY;
2848
2849         if (cli->cl_max_mds_easize < easize)
2850                 cli->cl_max_mds_easize = easize;
2851
2852         if (cli->cl_default_mds_easize < def_easize)
2853                 cli->cl_default_mds_easize = def_easize;
2854
2855         RETURN(0);
2856 }
2857
2858 static int mdc_precleanup(struct obd_device *obd)
2859 {
2860         ENTRY;
2861
2862         osc_precleanup_common(obd);
2863         mdc_changelog_cdev_finish(obd);
2864
2865         obd_cleanup_client_import(obd);
2866         ptlrpc_lprocfs_unregister_obd(obd);
2867         lprocfs_free_md_stats(obd);
2868         mdc_llog_finish(obd);
2869         RETURN(0);
2870 }
2871
2872 static int mdc_cleanup(struct obd_device *obd)
2873 {
2874         return osc_cleanup_common(obd);
2875 }
2876
2877 static const struct obd_ops mdc_obd_ops = {
2878         .o_owner            = THIS_MODULE,
2879         .o_setup            = mdc_setup,
2880         .o_precleanup       = mdc_precleanup,
2881         .o_cleanup          = mdc_cleanup,
2882         .o_add_conn         = client_import_add_conn,
2883         .o_del_conn         = client_import_del_conn,
2884         .o_connect          = client_connect_import,
2885         .o_reconnect        = osc_reconnect,
2886         .o_disconnect       = osc_disconnect,
2887         .o_iocontrol        = mdc_iocontrol,
2888         .o_set_info_async   = mdc_set_info_async,
2889         .o_statfs           = mdc_statfs,
2890         .o_statfs_async     = mdc_statfs_async,
2891         .o_fid_init         = client_fid_init,
2892         .o_fid_fini         = client_fid_fini,
2893         .o_fid_alloc        = mdc_fid_alloc,
2894         .o_import_event     = mdc_import_event,
2895         .o_get_info         = mdc_get_info,
2896         .o_get_uuid         = mdc_get_uuid,
2897         .o_quotactl         = mdc_quotactl,
2898 };
2899
2900 static const struct md_ops mdc_md_ops = {
2901         .m_get_root         = mdc_get_root,
2902         .m_null_inode       = mdc_null_inode,
2903         .m_close            = mdc_close,
2904         .m_create           = mdc_create,
2905         .m_enqueue          = mdc_enqueue,
2906         .m_getattr          = mdc_getattr,
2907         .m_getattr_name     = mdc_getattr_name,
2908         .m_intent_lock      = mdc_intent_lock,
2909         .m_link             = mdc_link,
2910         .m_rename           = mdc_rename,
2911         .m_setattr          = mdc_setattr,
2912         .m_setxattr         = mdc_setxattr,
2913         .m_getxattr         = mdc_getxattr,
2914         .m_fsync                = mdc_fsync,
2915         .m_file_resync          = mdc_file_resync,
2916         .m_read_page            = mdc_read_page,
2917         .m_unlink           = mdc_unlink,
2918         .m_cancel_unused    = mdc_cancel_unused,
2919         .m_init_ea_size     = mdc_init_ea_size,
2920         .m_set_lock_data    = mdc_set_lock_data,
2921         .m_lock_match       = mdc_lock_match,
2922         .m_get_lustre_md    = mdc_get_lustre_md,
2923         .m_free_lustre_md   = mdc_free_lustre_md,
2924         .m_set_open_replay_data = mdc_set_open_replay_data,
2925         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2926         .m_intent_getattr_async = mdc_intent_getattr_async,
2927         .m_revalidate_lock      = mdc_revalidate_lock,
2928         .m_rmfid                = mdc_rmfid,
2929 };
2930
2931 dev_t mdc_changelog_dev;
2932 struct class *mdc_changelog_class;
2933 static int __init mdc_init(void)
2934 {
2935         int rc = 0;
2936         rc = alloc_chrdev_region(&mdc_changelog_dev, 0,
2937                                  MDC_CHANGELOG_DEV_COUNT,
2938                                  MDC_CHANGELOG_DEV_NAME);
2939         if (rc)
2940                 return rc;
2941
2942         mdc_changelog_class = class_create(THIS_MODULE, MDC_CHANGELOG_DEV_NAME);
2943         if (IS_ERR(mdc_changelog_class)) {
2944                 rc = PTR_ERR(mdc_changelog_class);
2945                 goto out_dev;
2946         }
2947
2948         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2949                                  LUSTRE_MDC_NAME, &mdc_device_type);
2950         if (rc)
2951                 goto out_class;
2952
2953         return 0;
2954
2955 out_class:
2956         class_destroy(mdc_changelog_class);
2957 out_dev:
2958         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
2959         return rc;
2960 }
2961
2962 static void __exit mdc_exit(void)
2963 {
2964         class_unregister_type(LUSTRE_MDC_NAME);
2965         class_destroy(mdc_changelog_class);
2966         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
2967         idr_destroy(&mdc_changelog_minor_idr);
2968 }
2969
2970 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2971 MODULE_DESCRIPTION("Lustre Metadata Client");
2972 MODULE_VERSION(LUSTRE_VERSION_STRING);
2973 MODULE_LICENSE("GPL");
2974
2975 module_init(mdc_init);
2976 module_exit(mdc_exit);