lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44 #include <linux/device.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_fid.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_kernelcomm.h>
  55 #include <lustre_lmv.h>
  56 #include <lustre_log.h>
  57 #include <uapi/linux/lustre/lustre_param.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 != imp->imp_connect_data.ocd_max_easize) {
 245                         acl_bufsize = imp->imp_connect_data.ocd_max_easize;
 246                         mdc_reset_acl_req(req);
 247                         goto again;
 248                 }
 249
 250                 ptlrpc_req_finished(req);
 251         } else {
 252                 *request = req;
 253         }
 254
 255         RETURN(rc);
 256 }
 257
 258 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 259                             struct ptlrpc_request **request)
 260 {
 261         struct ptlrpc_request *req;
 262         struct obd_import *imp = class_exp2cliimp(exp);
 263         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 264         int rc;
 265         ENTRY;
 266
 267         *request = NULL;
 268         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 269         if (req == NULL)
 270                 RETURN(-ENOMEM);
 271
 272         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 273                              op_data->op_namelen + 1);
 274
 275         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 276         if (rc) {
 277                 ptlrpc_request_free(req);
 278                 RETURN(rc);
 279         }
 280
 281         if (op_data->op_name) {
 282                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 283                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 284                                 op_data->op_namelen);
 285                 memcpy(name, op_data->op_name, op_data->op_namelen);
 286         }
 287
 288 again:
 289         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 290                       op_data->op_mode, op_data->op_suppgids[0], 0);
 291         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 292                              op_data->op_mode);
 293         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 294         ptlrpc_request_set_replen(req);
 295
 296         rc = mdc_getattr_common(exp, req);
 297         if (rc) {
 298                 if (rc == -ERANGE &&
 299                     acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
 300                         acl_bufsize = imp->imp_connect_data.ocd_max_easize;
 301                         mdc_reset_acl_req(req);
 302                         goto again;
 303                 }
 304
 305                 ptlrpc_req_finished(req);
 306         } else {
 307                 *request = req;
 308         }
 309
 310         RETURN(rc);
 311 }
 312
 313 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 314                             const struct lu_fid *fid, int opcode, u64 valid,
 315                             const char *xattr_name, const char *input,
 316                             int input_size, int output_size, int flags,
 317                             __u32 suppgid, struct ptlrpc_request **request)
 318 {
 319         struct ptlrpc_request *req;
 320         int   xattr_namelen = 0;
 321         char *tmp;
 322         int   rc;
 323         ENTRY;
 324
 325         *request = NULL;
 326         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 327         if (req == NULL)
 328                 RETURN(-ENOMEM);
 329
 330         if (xattr_name) {
 331                 xattr_namelen = strlen(xattr_name) + 1;
 332                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 333                                      xattr_namelen);
 334         }
 335         if (input_size)
 336                 LASSERT(input);
 337         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 338                              input_size);
 339
 340         /* get SELinux policy info if any */
 341         rc = sptlrpc_get_sepol(req);
 342         if (rc < 0) {
 343                 ptlrpc_request_free(req);
 344                 RETURN(rc);
 345         }
 346         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 347                              strlen(req->rq_sepol) ?
 348                              strlen(req->rq_sepol) + 1 : 0);
 349
 350         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 351         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 352             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 353                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 354                 int count;
 355
 356                 /* Without that packing would fail */
 357                 if (input_size == 0)
 358                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 359                                              RCL_CLIENT, 0);
 360
 361                 count = mdc_resource_get_unused(exp, fid,
 362                                                 &cancels, LCK_EX,
 363                                                 MDS_INODELOCK_XATTR);
 364
 365                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 366                 if (rc) {
 367                         ptlrpc_request_free(req);
 368                         RETURN(rc);
 369                 }
 370         } else {
 371                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 372                 if (rc) {
 373                         ptlrpc_request_free(req);
 374                         RETURN(rc);
 375                 }
 376         }
 377
 378         if (opcode == MDS_REINT) {
 379                 struct mdt_rec_setxattr *rec;
 380
 381                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 382                          sizeof(struct mdt_rec_reint));
 383                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 384                 rec->sx_opcode = REINT_SETXATTR;
 385                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 386                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 387                 rec->sx_cap    = cfs_curproc_cap_pack();
 388                 rec->sx_suppgid1 = suppgid;
 389                 rec->sx_suppgid2 = -1;
 390                 rec->sx_fid    = *fid;
 391                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 392                 rec->sx_time   = ktime_get_real_seconds();
 393                 rec->sx_size   = output_size;
 394                 rec->sx_flags  = flags;
 395         } else {
 396                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 397         }
 398
 399         if (xattr_name) {
 400                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 401                 memcpy(tmp, xattr_name, xattr_namelen);
 402         }
 403         if (input_size) {
 404                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 405                 memcpy(tmp, input, input_size);
 406         }
 407
 408         mdc_file_sepol_pack(req);
 409
 410         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 411                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 412                                      RCL_SERVER, output_size);
 413         ptlrpc_request_set_replen(req);
 414
 415         /* make rpc */
 416         if (opcode == MDS_REINT)
 417                 mdc_get_mod_rpc_slot(req, NULL);
 418
 419         rc = ptlrpc_queue_wait(req);
 420
 421         if (opcode == MDS_REINT)
 422                 mdc_put_mod_rpc_slot(req, NULL);
 423
 424         if (rc)
 425                 ptlrpc_req_finished(req);
 426         else
 427                 *request = req;
 428         RETURN(rc);
 429 }
 430
 431 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 432                         u64 obd_md_valid, const char *name,
 433                         const void *value, size_t value_size,
 434                         unsigned int xattr_flags, u32 suppgid,
 435                         struct ptlrpc_request **req)
 436 {
 437         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 438                 obd_md_valid == OBD_MD_FLXATTRRM);
 439
 440         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 441                                 fid, MDS_REINT, obd_md_valid, name,
 442                                 value, value_size, 0, xattr_flags, suppgid,
 443                                 req);
 444 }
 445
 446 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 447                         u64 obd_md_valid, const char *name, size_t buf_size,
 448                         struct ptlrpc_request **req)
 449 {
 450         struct mdt_body *body;
 451         int rc;
 452
 453         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 454                 obd_md_valid == OBD_MD_FLXATTRLS);
 455
 456         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 457                exp->exp_obd->obd_name, name, PFID(fid));
 458         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 459                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 460                               req);
 461         if (rc < 0)
 462                 GOTO(out, rc);
 463
 464         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 465         if (body == NULL)
 466                 GOTO(out, rc = -EPROTO);
 467
 468         /* only detect the xattr size */
 469         if (buf_size == 0) {
 470                 /* LU-11109: Older MDTs do not distinguish
 471                  * between nonexistent xattrs and zero length
 472                  * values in this case. Newer MDTs will return
 473                  * -ENODATA or set OBD_MD_FLXATTR. */
 474                 GOTO(out, rc = body->mbo_eadatasize);
 475         }
 476
 477         if (body->mbo_eadatasize == 0) {
 478                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 479                  * success so that we can distinguish between
 480                  * zero length value and nonexistent xattr.
 481                  *
 482                  * If OBD_MD_FLXATTR is not set then we keep
 483                  * the old behavior and return -ENODATA for
 484                  * getxattr() when mbo_eadatasize is 0. But
 485                  * -ENODATA only makes sense for getxattr()
 486                  * and not for listxattr(). */
 487                 if (body->mbo_valid & OBD_MD_FLXATTR)
 488                         GOTO(out, rc = 0);
 489                 else if (obd_md_valid == OBD_MD_FLXATTR)
 490                         GOTO(out, rc = -ENODATA);
 491                 else
 492                         GOTO(out, rc = 0);
 493         }
 494
 495         GOTO(out, rc = body->mbo_eadatasize);
 496 out:
 497         if (rc < 0) {
 498                 ptlrpc_req_finished(*req);
 499                 *req = NULL;
 500         }
 501
 502         return rc;
 503 }
 504
 505 #ifdef CONFIG_FS_POSIX_ACL
 506 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 507 {
 508         struct req_capsule     *pill = &req->rq_pill;
 509         struct mdt_body        *body = md->body;
 510         struct posix_acl       *acl;
 511         void                   *buf;
 512         int                     rc;
 513         ENTRY;
 514
 515         if (!body->mbo_aclsize)
 516                 RETURN(0);
 517
 518         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 519
 520         if (!buf)
 521                 RETURN(-EPROTO);
 522
 523         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 524         if (acl == NULL)
 525                 RETURN(0);
 526         if (IS_ERR(acl)) {
 527                 rc = PTR_ERR(acl);
 528                 CERROR("convert xattr to acl: %d\n", rc);
 529                 RETURN(rc);
 530         }
 531
 532         rc = posix_acl_valid(&init_user_ns, acl);
 533         if (rc) {
 534                 CERROR("validate acl: %d\n", rc);
 535                 posix_acl_release(acl);
 536                 RETURN(rc);
 537         }
 538
 539         md->posix_acl = acl;
 540         RETURN(0);
 541 }
 542 #else
 543 #define mdc_unpack_acl(req, md) 0
 544 #endif
 545
 546 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 547                       struct obd_export *dt_exp, struct obd_export *md_exp,
 548                       struct lustre_md *md)
 549 {
 550         struct req_capsule *pill = &req->rq_pill;
 551         int rc;
 552         ENTRY;
 553
 554         LASSERT(md);
 555         memset(md, 0, sizeof(*md));
 556
 557         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 558         LASSERT(md->body != NULL);
 559
 560         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 561                 if (!S_ISREG(md->body->mbo_mode)) {
 562                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 563                                "regular file, but is not\n");
 564                         GOTO(out, rc = -EPROTO);
 565                 }
 566
 567                 if (md->body->mbo_eadatasize == 0) {
 568                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 569                                "but eadatasize 0\n");
 570                         GOTO(out, rc = -EPROTO);
 571                 }
 572
 573                 md->layout.lb_len = md->body->mbo_eadatasize;
 574                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 575                                                         &RMF_MDT_MD,
 576                                                         md->layout.lb_len);
 577                 if (md->layout.lb_buf == NULL)
 578                         GOTO(out, rc = -EPROTO);
 579         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 580                 const union lmv_mds_md *lmv;
 581                 size_t lmv_size;
 582
 583                 if (!S_ISDIR(md->body->mbo_mode)) {
 584                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 585                                "directory, but is not\n");
 586                         GOTO(out, rc = -EPROTO);
 587                 }
 588
 589                 lmv_size = md->body->mbo_eadatasize;
 590                 if (lmv_size == 0) {
 591                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 592                                "but eadatasize 0\n");
 593                         RETURN(-EPROTO);
 594                 }
 595
 596                 if (md->body->mbo_valid & OBD_MD_MEA) {
 597                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 598                                                            lmv_size);
 599                         if (lmv == NULL)
 600                                 GOTO(out, rc = -EPROTO);
 601
 602                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 603                         if (rc < 0)
 604                                 GOTO(out, rc);
 605
 606                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 607                                 CDEBUG(D_INFO, "size too small:  "
 608                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 609                                         rc, (int)sizeof(*md->lmv));
 610                                 GOTO(out, rc = -EPROTO);
 611                         }
 612                 }
 613         }
 614         rc = 0;
 615
 616         if (md->body->mbo_valid & OBD_MD_FLACL) {
 617                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 618                  * only when aclsize != 0 there's an actual segment for ACL
 619                  * in reply buffer.
 620                  */
 621                 if (md->body->mbo_aclsize) {
 622                         rc = mdc_unpack_acl(req, md);
 623                         if (rc)
 624                                 GOTO(out, rc);
 625 #ifdef CONFIG_FS_POSIX_ACL
 626                 } else {
 627                         md->posix_acl = NULL;
 628 #endif
 629                 }
 630         }
 631
 632         EXIT;
 633 out:
 634         if (rc) {
 635 #ifdef CONFIG_FS_POSIX_ACL
 636                 posix_acl_release(md->posix_acl);
 637 #endif
 638         }
 639         return rc;
 640 }
 641
 642 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 643 {
 644         ENTRY;
 645         RETURN(0);
 646 }
 647
 648 void mdc_replay_open(struct ptlrpc_request *req)
 649 {
 650         struct md_open_data *mod = req->rq_cb_data;
 651         struct ptlrpc_request *close_req;
 652         struct obd_client_handle *och;
 653         struct lustre_handle old_open_handle = { };
 654         struct mdt_body *body;
 655         ENTRY;
 656
 657         if (mod == NULL) {
 658                 DEBUG_REQ(D_ERROR, req,
 659                           "Can't properly replay without open data.");
 660                 EXIT;
 661                 return;
 662         }
 663
 664         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 665         LASSERT(body != NULL);
 666
 667         spin_lock(&req->rq_lock);
 668         och = mod->mod_och;
 669         if (och && och->och_open_handle.cookie)
 670                 req->rq_early_free_repbuf = 1;
 671         else
 672                 req->rq_early_free_repbuf = 0;
 673         spin_unlock(&req->rq_lock);
 674
 675         if (req->rq_early_free_repbuf) {
 676                 struct lustre_handle *file_open_handle;
 677
 678                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 679
 680                 file_open_handle = &och->och_open_handle;
 681                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 682                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 683                 old_open_handle = *file_open_handle;
 684                 *file_open_handle = body->mbo_open_handle;
 685         }
 686
 687         close_req = mod->mod_close_req;
 688         if (close_req) {
 689                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 690                 struct mdt_ioepoch *epoch;
 691
 692                 LASSERT(opc == MDS_CLOSE);
 693                 epoch = req_capsule_client_get(&close_req->rq_pill,
 694                                                &RMF_MDT_EPOCH);
 695                 LASSERT(epoch);
 696
 697                 if (req->rq_early_free_repbuf)
 698                         LASSERT(old_open_handle.cookie ==
 699                                 epoch->mio_open_handle.cookie);
 700
 701                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 702                 epoch->mio_open_handle = body->mbo_open_handle;
 703         }
 704         EXIT;
 705 }
 706
 707 void mdc_commit_open(struct ptlrpc_request *req)
 708 {
 709         struct md_open_data *mod = req->rq_cb_data;
 710         if (mod == NULL)
 711                 return;
 712
 713         /**
 714          * No need to touch md_open_data::mod_och, it holds a reference on
 715          * \var mod and will zero references to each other, \var mod will be
 716          * freed after that when md_open_data::mod_och will put the reference.
 717          */
 718
 719         /**
 720          * Do not let open request to disappear as it still may be needed
 721          * for close rpc to happen (it may happen on evict only, otherwise
 722          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 723          * called), just mark this rpc as committed to distinguish these 2
 724          * cases, see mdc_close() for details. The open request reference will
 725          * be put along with freeing \var mod.
 726          */
 727         ptlrpc_request_addref(req);
 728         spin_lock(&req->rq_lock);
 729         req->rq_committed = 1;
 730         spin_unlock(&req->rq_lock);
 731         req->rq_cb_data = NULL;
 732         obd_mod_put(mod);
 733 }
 734
 735 int mdc_set_open_replay_data(struct obd_export *exp,
 736                              struct obd_client_handle *och,
 737                              struct lookup_intent *it)
 738 {
 739         struct md_open_data     *mod;
 740         struct mdt_rec_create   *rec;
 741         struct mdt_body         *body;
 742         struct ptlrpc_request   *open_req = it->it_request;
 743         struct obd_import       *imp = open_req->rq_import;
 744         ENTRY;
 745
 746         if (!open_req->rq_replay)
 747                 RETURN(0);
 748
 749         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 750         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 751         LASSERT(rec != NULL);
 752         /* Incoming message in my byte order (it's been swabbed). */
 753         /* Outgoing messages always in my byte order. */
 754         LASSERT(body != NULL);
 755
 756         /* Only if the import is replayable, we set replay_open data */
 757         if (och && imp->imp_replayable) {
 758                 mod = obd_mod_alloc();
 759                 if (mod == NULL) {
 760                         DEBUG_REQ(D_ERROR, open_req,
 761                                   "Can't allocate md_open_data");
 762                         RETURN(0);
 763                 }
 764
 765                 /**
 766                  * Take a reference on \var mod, to be freed on mdc_close().
 767                  * It protects \var mod from being freed on eviction (commit
 768                  * callback is called despite rq_replay flag).
 769                  * Another reference for \var och.
 770                  */
 771                 obd_mod_get(mod);
 772                 obd_mod_get(mod);
 773
 774                 spin_lock(&open_req->rq_lock);
 775                 och->och_mod = mod;
 776                 mod->mod_och = och;
 777                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 778                                      it_disposition(it, DISP_OPEN_STRIPE);
 779                 mod->mod_open_req = open_req;
 780                 open_req->rq_cb_data = mod;
 781                 open_req->rq_commit_cb = mdc_commit_open;
 782                 open_req->rq_early_free_repbuf = 1;
 783                 spin_unlock(&open_req->rq_lock);
 784         }
 785
 786         rec->cr_fid2 = body->mbo_fid1;
 787         rec->cr_open_handle_old = body->mbo_open_handle;
 788         open_req->rq_replay_cb = mdc_replay_open;
 789         if (!fid_is_sane(&body->mbo_fid1)) {
 790                 DEBUG_REQ(D_ERROR, open_req,
 791                           "saving replay request with insane FID " DFID,
 792                           PFID(&body->mbo_fid1));
 793                 LBUG();
 794         }
 795
 796         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 797         RETURN(0);
 798 }
 799
 800 static void mdc_free_open(struct md_open_data *mod)
 801 {
 802         int committed = 0;
 803
 804         if (mod->mod_is_create == 0 &&
 805             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 806                 committed = 1;
 807
 808         /**
 809          * No reason to asssert here if the open request has
 810          * rq_replay == 1. It means that mdc_close failed, and
 811          * close request wasn`t sent. It is not fatal to client.
 812          * The worst thing is eviction if the client gets open lock
 813          **/
 814
 815         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 816                   "= %d\n", mod->mod_open_req->rq_replay);
 817
 818         ptlrpc_request_committed(mod->mod_open_req, committed);
 819         if (mod->mod_close_req)
 820                 ptlrpc_request_committed(mod->mod_close_req, committed);
 821 }
 822
 823 int mdc_clear_open_replay_data(struct obd_export *exp,
 824                                struct obd_client_handle *och)
 825 {
 826         struct md_open_data *mod = och->och_mod;
 827         ENTRY;
 828
 829         /**
 830          * It is possible to not have \var mod in a case of eviction between
 831          * lookup and ll_file_open().
 832          **/
 833         if (mod == NULL)
 834                 RETURN(0);
 835
 836         LASSERT(mod != LP_POISON);
 837         LASSERT(mod->mod_open_req != NULL);
 838
 839         spin_lock(&mod->mod_open_req->rq_lock);
 840         if (mod->mod_och)
 841                 mod->mod_och->och_open_handle.cookie = 0;
 842         mod->mod_open_req->rq_early_free_repbuf = 0;
 843         spin_unlock(&mod->mod_open_req->rq_lock);
 844         mdc_free_open(mod);
 845
 846         mod->mod_och = NULL;
 847         och->och_mod = NULL;
 848         obd_mod_put(mod);
 849
 850         RETURN(0);
 851 }
 852
 853 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 854                      struct md_open_data *mod, struct ptlrpc_request **request)
 855 {
 856         struct obd_device     *obd = class_exp2obd(exp);
 857         struct ptlrpc_request *req;
 858         struct req_format     *req_fmt;
 859         size_t                 u32_count = 0;
 860         int                    rc;
 861         int                    saved_rc = 0;
 862         ENTRY;
 863
 864         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 865                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 866                op_data->op_bias);
 867
 868         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 869                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 870                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 871                         /* allocate a FID for volatile file */
 872                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 873                                            op_data);
 874                         if (rc < 0) {
 875                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 876                                        obd->obd_name, PFID(&op_data->op_fid1),
 877                                        rc);
 878                                 /* save the errcode and proceed to close */
 879                                 saved_rc = rc;
 880                         }
 881                 }
 882                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 883                         size_t count = op_data->op_data_size / sizeof(__u32);
 884
 885                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 886                                 u32_count = count;
 887                 }
 888         } else {
 889                 req_fmt = &RQF_MDS_CLOSE;
 890         }
 891
 892         *request = NULL;
 893         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 894                 req = NULL;
 895         else
 896                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 897
 898         /* Ensure that this close's handle is fixed up during replay. */
 899         if (likely(mod != NULL)) {
 900                 LASSERTF(mod->mod_open_req != NULL &&
 901                          mod->mod_open_req->rq_type != LI_POISON,
 902                          "POISONED open %p!\n", mod->mod_open_req);
 903
 904                 mod->mod_close_req = req;
 905
 906                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 907                 /* We no longer want to preserve this open for replay even
 908                  * though the open was committed. b=3632, b=3633 */
 909                 spin_lock(&mod->mod_open_req->rq_lock);
 910                 mod->mod_open_req->rq_replay = 0;
 911                 spin_unlock(&mod->mod_open_req->rq_lock);
 912         } else {
 913                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 914         }
 915         if (req == NULL) {
 916                 /**
 917                  * TODO: repeat close after errors
 918                  */
 919                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 920                       "dropped when this client unmounts or is evicted\n",
 921                       obd->obd_name, PFID(&op_data->op_fid1));
 922                 GOTO(out, rc = -ENOMEM);
 923         }
 924
 925         if (u32_count > 0)
 926                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 927                                      u32_count * sizeof(__u32));
 928
 929         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 930         if (rc) {
 931                 ptlrpc_request_free(req);
 932                 req = NULL;
 933                 GOTO(out, rc);
 934         }
 935
 936         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 937          * portal whose threads are not taking any DLM locks and are therefore
 938          * always progressing */
 939         req->rq_request_portal = MDS_READPAGE_PORTAL;
 940         ptlrpc_at_set_req_timeout(req);
 941
 942         if (!(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 943                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 944                                         OP_XVALID_LAZYBLOCKS);
 945
 946         mdc_close_pack(req, op_data);
 947
 948         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 949                              obd->u.cli.cl_default_mds_easize);
 950
 951         ptlrpc_request_set_replen(req);
 952
 953         mdc_get_mod_rpc_slot(req, NULL);
 954         rc = ptlrpc_queue_wait(req);
 955         mdc_put_mod_rpc_slot(req, NULL);
 956
 957         if (req->rq_repmsg == NULL) {
 958                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 959                        req->rq_status);
 960                 if (rc == 0)
 961                         rc = req->rq_status ?: -EIO;
 962         } else if (rc == 0 || rc == -EAGAIN) {
 963                 struct mdt_body *body;
 964
 965                 rc = lustre_msg_get_status(req->rq_repmsg);
 966                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 967                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 968                                   "= %d", rc);
 969                         if (rc > 0)
 970                                 rc = -rc;
 971                 }
 972                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 973                 if (body == NULL)
 974                         rc = -EPROTO;
 975         } else if (rc == -ESTALE) {
 976                 /**
 977                  * it can be allowed error after 3633 if open was committed and
 978                  * server failed before close was sent. Let's check if mod
 979                  * exists and return no error in that case
 980                  */
 981                 if (mod) {
 982                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 983                         LASSERT(mod->mod_open_req != NULL);
 984                         if (mod->mod_open_req->rq_committed)
 985                                 rc = 0;
 986                 }
 987         }
 988
 989 out:
 990         if (mod) {
 991                 if (rc != 0)
 992                         mod->mod_close_req = NULL;
 993                 /* Since now, mod is accessed through open_req only,
 994                  * thus close req does not keep a reference on mod anymore. */
 995                 obd_mod_put(mod);
 996         }
 997         *request = req;
 998
 999         RETURN(rc < 0 ? rc : saved_rc);
1000 }
1001
1002 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1003                        u64 offset, struct page **pages, int npages,
1004                        struct ptlrpc_request **request)
1005 {
1006         struct ptlrpc_request   *req;
1007         struct ptlrpc_bulk_desc *desc;
1008         int                      i;
1009         wait_queue_head_t        waitq;
1010         int                      resends = 0;
1011         struct l_wait_info       lwi;
1012         int                      rc;
1013         ENTRY;
1014
1015         *request = NULL;
1016         init_waitqueue_head(&waitq);
1017
1018 restart_bulk:
1019         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1020         if (req == NULL)
1021                 RETURN(-ENOMEM);
1022
1023         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1024         if (rc) {
1025                 ptlrpc_request_free(req);
1026                 RETURN(rc);
1027         }
1028
1029         req->rq_request_portal = MDS_READPAGE_PORTAL;
1030         ptlrpc_at_set_req_timeout(req);
1031
1032         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1033                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
1034                                     MDS_BULK_PORTAL,
1035                                     &ptlrpc_bulk_kiov_pin_ops);
1036         if (desc == NULL) {
1037                 ptlrpc_req_finished(req);
1038                 RETURN(-ENOMEM);
1039         }
1040
1041         /* NB req now owns desc and will free it when it gets freed */
1042         for (i = 0; i < npages; i++)
1043                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1044                                                  PAGE_SIZE);
1045
1046         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
1047
1048         ptlrpc_request_set_replen(req);
1049         rc = ptlrpc_queue_wait(req);
1050         if (rc) {
1051                 ptlrpc_req_finished(req);
1052                 if (rc != -ETIMEDOUT)
1053                         RETURN(rc);
1054
1055                 resends++;
1056                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1057                         CERROR("%s: too many resend retries: rc = %d\n",
1058                                exp->exp_obd->obd_name, -EIO);
1059                         RETURN(-EIO);
1060                 }
1061                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1062                                        NULL);
1063                 l_wait_event(waitq, 0, &lwi);
1064
1065                 goto restart_bulk;
1066         }
1067
1068         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1069                                           req->rq_bulk->bd_nob_transferred);
1070         if (rc < 0) {
1071                 ptlrpc_req_finished(req);
1072                 RETURN(rc);
1073         }
1074
1075         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1076                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1077                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1078                        PAGE_SIZE * npages);
1079                 ptlrpc_req_finished(req);
1080                 RETURN(-EPROTO);
1081         }
1082
1083         *request = req;
1084         RETURN(0);
1085 }
1086
1087 static void mdc_release_page(struct page *page, int remove)
1088 {
1089         if (remove) {
1090                 lock_page(page);
1091                 if (likely(page->mapping != NULL))
1092                         truncate_complete_page(page->mapping, page);
1093                 unlock_page(page);
1094         }
1095         put_page(page);
1096 }
1097
1098 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1099                                     __u64 *start, __u64 *end, int hash64)
1100 {
1101         /*
1102          * Complement of hash is used as an index so that
1103          * radix_tree_gang_lookup() can be used to find a page with starting
1104          * hash _smaller_ than one we are looking for.
1105          */
1106         unsigned long offset = hash_x_index(*hash, hash64);
1107         struct page *page;
1108         int found;
1109
1110         xa_lock_irq(&mapping->i_pages);
1111         found = radix_tree_gang_lookup(&mapping->page_tree,
1112                                        (void **)&page, offset, 1);
1113         if (found > 0 && !xa_is_value(page)) {
1114                 struct lu_dirpage *dp;
1115
1116                 get_page(page);
1117                 xa_unlock_irq(&mapping->i_pages);
1118                 /*
1119                  * In contrast to find_lock_page() we are sure that directory
1120                  * page cannot be truncated (while DLM lock is held) and,
1121                  * hence, can avoid restart.
1122                  *
1123                  * In fact, page cannot be locked here at all, because
1124                  * mdc_read_page_remote does synchronous io.
1125                  */
1126                 wait_on_page_locked(page);
1127                 if (PageUptodate(page)) {
1128                         dp = kmap(page);
1129                         if (BITS_PER_LONG == 32 && hash64) {
1130                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1131                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1132                                 *hash  = *hash >> 32;
1133                         } else {
1134                                 *start = le64_to_cpu(dp->ldp_hash_start);
1135                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1136                         }
1137                         if (unlikely(*start == 1 && *hash == 0))
1138                                 *hash = *start;
1139                         else
1140                                 LASSERTF(*start <= *hash, "start = %#llx"
1141                                          ",end = %#llx,hash = %#llx\n",
1142                                          *start, *end, *hash);
1143                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1144                               " hash %#llx\n", offset, *start, *end, *hash);
1145                         if (*hash > *end) {
1146                                 kunmap(page);
1147                                 mdc_release_page(page, 0);
1148                                 page = NULL;
1149                         } else if (*end != *start && *hash == *end) {
1150                                 /*
1151                                  * upon hash collision, remove this page,
1152                                  * otherwise put page reference, and
1153                                  * mdc_read_page_remote() will issue RPC to
1154                                  * fetch the page we want.
1155                                  */
1156                                 kunmap(page);
1157                                 mdc_release_page(page,
1158                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1159                                 page = NULL;
1160                         }
1161                 } else {
1162                         put_page(page);
1163                         page = ERR_PTR(-EIO);
1164                 }
1165         } else {
1166                 xa_unlock_irq(&mapping->i_pages);
1167                 page = NULL;
1168         }
1169         return page;
1170 }
1171
1172 /*
1173  * Adjust a set of pages, each page containing an array of lu_dirpages,
1174  * so that each page can be used as a single logical lu_dirpage.
1175  *
1176  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1177  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1178  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1179  * value is used as a cookie to request the next lu_dirpage in a
1180  * directory listing that spans multiple pages (two in this example):
1181  *   ________
1182  *  |        |
1183  * .|--------v-------   -----.
1184  * |s|e|f|p|ent|ent| ... |ent|
1185  * '--|--------------   -----'   Each PAGE contains a single
1186  *    '------.                   lu_dirpage.
1187  * .---------v-------   -----.
1188  * |s|e|f|p|ent| 0 | ... | 0 |
1189  * '-----------------   -----'
1190  *
1191  * However, on hosts where the native VM page size (PAGE_SIZE) is
1192  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1193  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1194  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1195  * after it in the same PAGE (arrows simplified for brevity, but
1196  * in general e0==s1, e1==s2, etc.):
1197  *
1198  * .--------------------   -----.
1199  * |s0|e0|f0|p|ent|ent| ... |ent|
1200  * |---v----------------   -----|
1201  * |s1|e1|f1|p|ent|ent| ... |ent|
1202  * |---v----------------   -----|  Here, each PAGE contains
1203  *             ...                 multiple lu_dirpages.
1204  * |---v----------------   -----|
1205  * |s'|e'|f'|p|ent|ent| ... |ent|
1206  * '---|----------------   -----'
1207  *     v
1208  * .----------------------------.
1209  * |        next PAGE           |
1210  *
1211  * This structure is transformed into a single logical lu_dirpage as follows:
1212  *
1213  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1214  *   labeled 'next PAGE'.
1215  *
1216  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1217  *   a hash collision with the next page exists.
1218  *
1219  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1220  *   to the first entry of the next lu_dirpage.
1221  */
1222 #if PAGE_SIZE > LU_PAGE_SIZE
1223 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1224 {
1225         int i;
1226
1227         for (i = 0; i < cfs_pgs; i++) {
1228                 struct lu_dirpage *dp = kmap(pages[i]);
1229                 struct lu_dirpage *first = dp;
1230                 struct lu_dirent *end_dirent = NULL;
1231                 struct lu_dirent *ent;
1232                 __u64 hash_end = dp->ldp_hash_end;
1233                 __u32 flags = dp->ldp_flags;
1234
1235                 while (--lu_pgs > 0) {
1236                         ent = lu_dirent_start(dp);
1237                         for (end_dirent = ent; ent != NULL;
1238                              end_dirent = ent, ent = lu_dirent_next(ent));
1239
1240                         /* Advance dp to next lu_dirpage. */
1241                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1242
1243                         /* Check if we've reached the end of the PAGE. */
1244                         if (!((unsigned long)dp & ~PAGE_MASK))
1245                                 break;
1246
1247                         /* Save the hash and flags of this lu_dirpage. */
1248                         hash_end = dp->ldp_hash_end;
1249                         flags = dp->ldp_flags;
1250
1251                         /* Check if lu_dirpage contains no entries. */
1252                         if (end_dirent == NULL)
1253                                 break;
1254
1255                         /* Enlarge the end entry lde_reclen from 0 to
1256                          * first entry of next lu_dirpage. */
1257                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1258                         end_dirent->lde_reclen =
1259                                 cpu_to_le16((char *)(dp->ldp_entries) -
1260                                             (char *)end_dirent);
1261                 }
1262
1263                 first->ldp_hash_end = hash_end;
1264                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1265                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1266
1267                 kunmap(pages[i]);
1268         }
1269         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1270 }
1271 #else
1272 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1273 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1274
1275 /* parameters for readdir page */
1276 struct readpage_param {
1277         struct md_op_data       *rp_mod;
1278         __u64                   rp_off;
1279         int                     rp_hash64;
1280         struct obd_export       *rp_exp;
1281         struct md_callback      *rp_cb;
1282 };
1283
1284 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1285 static inline void delete_from_page_cache(struct page *page)
1286 {
1287         remove_from_page_cache(page);
1288         put_page(page);
1289 }
1290 #endif
1291
1292 /**
1293  * Read pages from server.
1294  *
1295  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1296  * a header lu_dirpage which describes the start/end hash, and whether this
1297  * page is empty (contains no dir entry) or hash collide with next page.
1298  * After client receives reply, several pages will be integrated into dir page
1299  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1300  * lu_dirpage for this integrated page will be adjusted.
1301  **/
1302 static int mdc_read_page_remote(void *data, struct page *page0)
1303 {
1304         struct readpage_param *rp = data;
1305         struct page **page_pool;
1306         struct page *page;
1307         struct lu_dirpage *dp;
1308         struct md_op_data *op_data = rp->rp_mod;
1309         struct ptlrpc_request *req;
1310         int max_pages;
1311         struct inode *inode;
1312         struct lu_fid *fid;
1313         int rd_pgs = 0; /* number of pages actually read */
1314         int npages;
1315         int i;
1316         int rc;
1317         ENTRY;
1318
1319         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1320         inode = op_data->op_data;
1321         fid = &op_data->op_fid1;
1322         LASSERT(inode != NULL);
1323
1324         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1325         if (page_pool != NULL) {
1326                 page_pool[0] = page0;
1327         } else {
1328                 page_pool = &page0;
1329                 max_pages = 1;
1330         }
1331
1332         for (npages = 1; npages < max_pages; npages++) {
1333                 page = __page_cache_alloc(mapping_gfp_mask(inode->i_mapping)
1334                                           | __GFP_COLD);
1335                 if (page == NULL)
1336                         break;
1337                 page_pool[npages] = page;
1338         }
1339
1340         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1341         if (rc < 0) {
1342                 /* page0 is special, which was added into page cache early */
1343                 delete_from_page_cache(page0);
1344         } else {
1345                 int lu_pgs;
1346
1347                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1348                         PAGE_SHIFT;
1349                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1350                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1351
1352                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1353
1354                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1355
1356                 SetPageUptodate(page0);
1357         }
1358         unlock_page(page0);
1359
1360         ptlrpc_req_finished(req);
1361         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1362         for (i = 1; i < npages; i++) {
1363                 unsigned long   offset;
1364                 __u64           hash;
1365                 int ret;
1366
1367                 page = page_pool[i];
1368
1369                 if (rc < 0 || i >= rd_pgs) {
1370                         put_page(page);
1371                         continue;
1372                 }
1373
1374                 SetPageUptodate(page);
1375
1376                 dp = kmap(page);
1377                 hash = le64_to_cpu(dp->ldp_hash_start);
1378                 kunmap(page);
1379
1380                 offset = hash_x_index(hash, rp->rp_hash64);
1381
1382                 prefetchw(&page->flags);
1383                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1384                                             GFP_KERNEL);
1385                 if (ret == 0)
1386                         unlock_page(page);
1387                 else
1388                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1389                                " rc = %d\n", offset, ret);
1390                 put_page(page);
1391         }
1392
1393         if (page_pool != &page0)
1394                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1395
1396         RETURN(rc);
1397 }
1398
1399 /**
1400  * Read dir page from cache first, if it can not find it, read it from
1401  * server and add into the cache.
1402  *
1403  * \param[in] exp       MDC export
1404  * \param[in] op_data   client MD stack parameters, transfering parameters
1405  *                      between different layers on client MD stack.
1406  * \param[in] cb_op     callback required for ldlm lock enqueue during
1407  *                      read page
1408  * \param[in] hash_offset the hash offset of the page to be read
1409  * \param[in] ppage     the page to be read
1410  *
1411  * retval               = 0 get the page successfully
1412  *                      errno(<0) get the page failed
1413  */
1414 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1415                          struct md_callback *cb_op, __u64 hash_offset,
1416                          struct page **ppage)
1417 {
1418         struct lookup_intent    it = { .it_op = IT_READDIR };
1419         struct page             *page;
1420         struct inode            *dir = op_data->op_data;
1421         struct address_space    *mapping;
1422         struct lu_dirpage       *dp;
1423         __u64                   start = 0;
1424         __u64                   end = 0;
1425         struct lustre_handle    lockh;
1426         struct ptlrpc_request   *enq_req = NULL;
1427         struct readpage_param   rp_param;
1428         int rc;
1429
1430         ENTRY;
1431
1432         *ppage = NULL;
1433
1434         LASSERT(dir != NULL);
1435         mapping = dir->i_mapping;
1436
1437         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1438                              cb_op->md_blocking_ast, 0);
1439         if (enq_req != NULL)
1440                 ptlrpc_req_finished(enq_req);
1441
1442         if (rc < 0) {
1443                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1444                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1445                 RETURN(rc);
1446         }
1447
1448         rc = 0;
1449         lockh.cookie = it.it_lock_handle;
1450         mdc_set_lock_data(exp, &lockh, dir, NULL);
1451
1452         rp_param.rp_off = hash_offset;
1453         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1454         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1455                                rp_param.rp_hash64);
1456         if (IS_ERR(page)) {
1457                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1458                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1459                        rp_param.rp_off, PTR_ERR(page));
1460                 GOTO(out_unlock, rc = PTR_ERR(page));
1461         } else if (page != NULL) {
1462                 /*
1463                  * XXX nikita: not entirely correct handling of a corner case:
1464                  * suppose hash chain of entries with hash value HASH crosses
1465                  * border between pages P0 and P1. First both P0 and P1 are
1466                  * cached, seekdir() is called for some entry from the P0 part
1467                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1468                  * happens and finds P1, as it starts with matching hash
1469                  * value. Remaining entries from P0 part of the chain are
1470                  * skipped. (Is that really a bug?)
1471                  *
1472                  * Possible solutions: 0. don't cache P1 is such case, handle
1473                  * it as an "overflow" page. 1. invalidate all pages at
1474                  * once. 2. use HASH|1 as an index for P1.
1475                  */
1476                 GOTO(hash_collision, page);
1477         }
1478
1479         rp_param.rp_exp = exp;
1480         rp_param.rp_mod = op_data;
1481         page = read_cache_page(mapping,
1482                                hash_x_index(rp_param.rp_off,
1483                                             rp_param.rp_hash64),
1484                                mdc_read_page_remote, &rp_param);
1485         if (IS_ERR(page)) {
1486                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1487                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1488                        rp_param.rp_off, PTR_ERR(page));
1489                 GOTO(out_unlock, rc = PTR_ERR(page));
1490         }
1491
1492         wait_on_page_locked(page);
1493         (void)kmap(page);
1494         if (!PageUptodate(page)) {
1495                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1496                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1497                        rp_param.rp_off, -5);
1498                 goto fail;
1499         }
1500         if (!PageChecked(page))
1501                 SetPageChecked(page);
1502         if (PageError(page)) {
1503                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1504                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1505                        rp_param.rp_off, -5);
1506                 goto fail;
1507         }
1508
1509 hash_collision:
1510         dp = page_address(page);
1511         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1512                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1513                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1514                 rp_param.rp_off = hash_offset >> 32;
1515         } else {
1516                 start = le64_to_cpu(dp->ldp_hash_start);
1517                 end   = le64_to_cpu(dp->ldp_hash_end);
1518                 rp_param.rp_off = hash_offset;
1519         }
1520         if (end == start) {
1521                 LASSERT(start == rp_param.rp_off);
1522                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1523 #if BITS_PER_LONG == 32
1524                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1525                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1526                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1527 #endif
1528
1529                 /*
1530                  * Fetch whole overflow chain...
1531                  *
1532                  * XXX not yet.
1533                  */
1534                 goto fail;
1535         }
1536         *ppage = page;
1537 out_unlock:
1538         ldlm_lock_decref(&lockh, it.it_lock_mode);
1539         return rc;
1540 fail:
1541         kunmap(page);
1542         mdc_release_page(page, 1);
1543         rc = -EIO;
1544         goto out_unlock;
1545 }
1546
1547 static int mdc_statfs(const struct lu_env *env,
1548                       struct obd_export *exp, struct obd_statfs *osfs,
1549                       time64_t max_age, __u32 flags)
1550 {
1551         struct obd_device *obd = class_exp2obd(exp);
1552         struct req_format *fmt;
1553         struct ptlrpc_request *req;
1554         struct obd_statfs *msfs;
1555         struct obd_import *imp = NULL;
1556         int rc;
1557         ENTRY;
1558
1559         /*
1560          * Since the request might also come from lprocfs, so we need
1561          * sync this with client_disconnect_export Bug15684
1562          */
1563         down_read(&obd->u.cli.cl_sem);
1564         if (obd->u.cli.cl_import)
1565                 imp = class_import_get(obd->u.cli.cl_import);
1566         up_read(&obd->u.cli.cl_sem);
1567         if (!imp)
1568                 RETURN(-ENODEV);
1569
1570         fmt = &RQF_MDS_STATFS;
1571         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1572             (flags & OBD_STATFS_SUM))
1573                 fmt = &RQF_MDS_STATFS_NEW;
1574         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1575                                         MDS_STATFS);
1576         if (req == NULL)
1577                 GOTO(output, rc = -ENOMEM);
1578
1579         if ((flags & OBD_STATFS_SUM) &&
1580             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1581                 /* request aggregated states */
1582                 struct mdt_body *body;
1583
1584                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1585                 if (body == NULL)
1586                         GOTO(out, rc = -EPROTO);
1587                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1588         }
1589
1590         ptlrpc_request_set_replen(req);
1591
1592         if (flags & OBD_STATFS_NODELAY) {
1593                 /* procfs requests not want stay in wait for avoid deadlock */
1594                 req->rq_no_resend = 1;
1595                 req->rq_no_delay = 1;
1596         }
1597
1598         rc = ptlrpc_queue_wait(req);
1599         if (rc) {
1600                 /* check connection error first */
1601                 if (imp->imp_connect_error)
1602                         rc = imp->imp_connect_error;
1603                 GOTO(out, rc);
1604         }
1605
1606         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1607         if (msfs == NULL)
1608                 GOTO(out, rc = -EPROTO);
1609
1610         *osfs = *msfs;
1611         EXIT;
1612 out:
1613         ptlrpc_req_finished(req);
1614 output:
1615         class_import_put(imp);
1616         return rc;
1617 }
1618
1619 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1620 {
1621         __u32 keylen, vallen;
1622         void *key;
1623         int rc;
1624
1625         if (gf->gf_pathlen > PATH_MAX)
1626                 RETURN(-ENAMETOOLONG);
1627         if (gf->gf_pathlen < 2)
1628                 RETURN(-EOVERFLOW);
1629
1630         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1631         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1632                                 sizeof(struct lu_fid));
1633         OBD_ALLOC(key, keylen);
1634         if (key == NULL)
1635                 RETURN(-ENOMEM);
1636         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1637         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1638         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1639                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1640         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1641                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1642
1643         if (!fid_is_sane(&gf->gf_fid))
1644                 GOTO(out, rc = -EINVAL);
1645
1646         /* Val is struct getinfo_fid2path result plus path */
1647         vallen = sizeof(*gf) + gf->gf_pathlen;
1648
1649         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1650         if (rc != 0 && rc != -EREMOTE)
1651                 GOTO(out, rc);
1652
1653         if (vallen <= sizeof(*gf))
1654                 GOTO(out, rc = -EPROTO);
1655         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1656                 GOTO(out, rc = -EOVERFLOW);
1657
1658         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1659                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1660                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1661                /* only log the last 512 characters of the path */
1662                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1663
1664 out:
1665         OBD_FREE(key, keylen);
1666         return rc;
1667 }
1668
1669 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1670                                 struct hsm_progress_kernel *hpk)
1671 {
1672         struct obd_import               *imp = class_exp2cliimp(exp);
1673         struct hsm_progress_kernel      *req_hpk;
1674         struct ptlrpc_request           *req;
1675         int                              rc;
1676         ENTRY;
1677
1678         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1679                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1680         if (req == NULL)
1681                 GOTO(out, rc = -ENOMEM);
1682
1683         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1684
1685         /* Copy hsm_progress struct */
1686         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1687         if (req_hpk == NULL)
1688                 GOTO(out, rc = -EPROTO);
1689
1690         *req_hpk = *hpk;
1691         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1692
1693         ptlrpc_request_set_replen(req);
1694
1695         mdc_get_mod_rpc_slot(req, NULL);
1696         rc = ptlrpc_queue_wait(req);
1697         mdc_put_mod_rpc_slot(req, NULL);
1698
1699         GOTO(out, rc);
1700 out:
1701         ptlrpc_req_finished(req);
1702         return rc;
1703 }
1704 /**
1705  * Send hsm_ct_register to MDS
1706  *
1707  * \param[in]   imp             import
1708  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1709  *                              else it is the count of archive_ids
1710  * \param[in]   archives        if in bitmap format, it is NULL,
1711  *                              else it is archive_id lists
1712  */
1713 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1714                                    __u32 *archives)
1715 {
1716         struct ptlrpc_request *req;
1717         __u32 *archive_array;
1718         size_t archives_size;
1719         int rc;
1720         ENTRY;
1721
1722         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1723         if (req == NULL)
1724                 RETURN(-ENOMEM);
1725
1726         if (archives != NULL)
1727                 archives_size = sizeof(*archive_array) * archive_count;
1728         else
1729                 archives_size = sizeof(archive_count);
1730
1731         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1732                              RCL_CLIENT, archives_size);
1733
1734         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1735         if (rc) {
1736                 ptlrpc_request_free(req);
1737                 RETURN(-ENOMEM);
1738         }
1739
1740         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1741
1742         archive_array = req_capsule_client_get(&req->rq_pill,
1743                                                &RMF_MDS_HSM_ARCHIVE);
1744         if (archive_array == NULL)
1745                 GOTO(out, rc = -EPROTO);
1746
1747         if (archives != NULL)
1748                 memcpy(archive_array, archives, archives_size);
1749         else
1750                 *archive_array = archive_count;
1751
1752         ptlrpc_request_set_replen(req);
1753
1754         rc = mdc_queue_wait(req);
1755         GOTO(out, rc);
1756 out:
1757         ptlrpc_req_finished(req);
1758         return rc;
1759 }
1760
1761 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1762                                       struct md_op_data *op_data)
1763 {
1764         struct hsm_current_action       *hca = op_data->op_data;
1765         struct hsm_current_action       *req_hca;
1766         struct ptlrpc_request           *req;
1767         int                              rc;
1768         ENTRY;
1769
1770         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1771                                    &RQF_MDS_HSM_ACTION);
1772         if (req == NULL)
1773                 RETURN(-ENOMEM);
1774
1775         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1776         if (rc) {
1777                 ptlrpc_request_free(req);
1778                 RETURN(rc);
1779         }
1780
1781         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1782                       op_data->op_suppgids[0], 0);
1783
1784         ptlrpc_request_set_replen(req);
1785
1786         rc = mdc_queue_wait(req);
1787         if (rc)
1788                 GOTO(out, rc);
1789
1790         req_hca = req_capsule_server_get(&req->rq_pill,
1791                                          &RMF_MDS_HSM_CURRENT_ACTION);
1792         if (req_hca == NULL)
1793                 GOTO(out, rc = -EPROTO);
1794
1795         *hca = *req_hca;
1796
1797         EXIT;
1798 out:
1799         ptlrpc_req_finished(req);
1800         return rc;
1801 }
1802
1803 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1804 {
1805         struct ptlrpc_request   *req;
1806         int                      rc;
1807         ENTRY;
1808
1809         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1810                                         LUSTRE_MDS_VERSION,
1811                                         MDS_HSM_CT_UNREGISTER);
1812         if (req == NULL)
1813                 GOTO(out, rc = -ENOMEM);
1814
1815         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1816
1817         ptlrpc_request_set_replen(req);
1818
1819         rc = mdc_queue_wait(req);
1820         GOTO(out, rc);
1821 out:
1822         ptlrpc_req_finished(req);
1823         return rc;
1824 }
1825
1826 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1827                                  struct md_op_data *op_data)
1828 {
1829         struct hsm_user_state   *hus = op_data->op_data;
1830         struct hsm_user_state   *req_hus;
1831         struct ptlrpc_request   *req;
1832         int                      rc;
1833         ENTRY;
1834
1835         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1836                                    &RQF_MDS_HSM_STATE_GET);
1837         if (req == NULL)
1838                 RETURN(-ENOMEM);
1839
1840         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1841         if (rc != 0) {
1842                 ptlrpc_request_free(req);
1843                 RETURN(rc);
1844         }
1845
1846         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1847                       op_data->op_suppgids[0], 0);
1848
1849         ptlrpc_request_set_replen(req);
1850
1851         rc = mdc_queue_wait(req);
1852         if (rc)
1853                 GOTO(out, rc);
1854
1855         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1856         if (req_hus == NULL)
1857                 GOTO(out, rc = -EPROTO);
1858
1859         *hus = *req_hus;
1860
1861         EXIT;
1862 out:
1863         ptlrpc_req_finished(req);
1864         return rc;
1865 }
1866
1867 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1868                                  struct md_op_data *op_data)
1869 {
1870         struct hsm_state_set    *hss = op_data->op_data;
1871         struct hsm_state_set    *req_hss;
1872         struct ptlrpc_request   *req;
1873         int                      rc;
1874         ENTRY;
1875
1876         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1877                                    &RQF_MDS_HSM_STATE_SET);
1878         if (req == NULL)
1879                 RETURN(-ENOMEM);
1880
1881         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1882         if (rc) {
1883                 ptlrpc_request_free(req);
1884                 RETURN(rc);
1885         }
1886
1887         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1888                       op_data->op_suppgids[0], 0);
1889
1890         /* Copy states */
1891         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1892         if (req_hss == NULL)
1893                 GOTO(out, rc = -EPROTO);
1894         *req_hss = *hss;
1895
1896         ptlrpc_request_set_replen(req);
1897
1898         mdc_get_mod_rpc_slot(req, NULL);
1899         rc = ptlrpc_queue_wait(req);
1900         mdc_put_mod_rpc_slot(req, NULL);
1901
1902         GOTO(out, rc);
1903 out:
1904         ptlrpc_req_finished(req);
1905         return rc;
1906 }
1907
1908 static int mdc_ioc_hsm_request(struct obd_export *exp,
1909                                struct hsm_user_request *hur)
1910 {
1911         struct obd_import       *imp = class_exp2cliimp(exp);
1912         struct ptlrpc_request   *req;
1913         struct hsm_request      *req_hr;
1914         struct hsm_user_item    *req_hui;
1915         char                    *req_opaque;
1916         int                      rc;
1917         ENTRY;
1918
1919         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1920         if (req == NULL)
1921                 GOTO(out, rc = -ENOMEM);
1922
1923         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1924                              hur->hur_request.hr_itemcount
1925                              * sizeof(struct hsm_user_item));
1926         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1927                              hur->hur_request.hr_data_len);
1928
1929         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1930         if (rc) {
1931                 ptlrpc_request_free(req);
1932                 RETURN(rc);
1933         }
1934
1935         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1936
1937         /* Copy hsm_request struct */
1938         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1939         if (req_hr == NULL)
1940                 GOTO(out, rc = -EPROTO);
1941         *req_hr = hur->hur_request;
1942
1943         /* Copy hsm_user_item structs */
1944         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1945         if (req_hui == NULL)
1946                 GOTO(out, rc = -EPROTO);
1947         memcpy(req_hui, hur->hur_user_item,
1948                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1949
1950         /* Copy opaque field */
1951         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1952         if (req_opaque == NULL)
1953                 GOTO(out, rc = -EPROTO);
1954         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1955
1956         ptlrpc_request_set_replen(req);
1957
1958         mdc_get_mod_rpc_slot(req, NULL);
1959         rc = ptlrpc_queue_wait(req);
1960         mdc_put_mod_rpc_slot(req, NULL);
1961
1962         GOTO(out, rc);
1963
1964 out:
1965         ptlrpc_req_finished(req);
1966         return rc;
1967 }
1968
1969 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1970                                 struct lustre_kernelcomm *lk);
1971
1972 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1973                         struct obd_quotactl *oqctl)
1974 {
1975         struct ptlrpc_request   *req;
1976         struct obd_quotactl     *oqc;
1977         int                      rc;
1978         ENTRY;
1979
1980         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1981                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1982                                         MDS_QUOTACTL);
1983         if (req == NULL)
1984                 RETURN(-ENOMEM);
1985
1986         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1987         *oqc = *oqctl;
1988
1989         ptlrpc_request_set_replen(req);
1990         ptlrpc_at_set_req_timeout(req);
1991
1992         rc = ptlrpc_queue_wait(req);
1993         if (rc)
1994                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1995
1996         if (req->rq_repmsg &&
1997             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1998                 *oqctl = *oqc;
1999         } else if (!rc) {
2000                 CERROR ("Can't unpack obd_quotactl\n");
2001                 rc = -EPROTO;
2002         }
2003         ptlrpc_req_finished(req);
2004
2005         RETURN(rc);
2006 }
2007
2008 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2009                                 struct md_op_data *op_data)
2010 {
2011         struct list_head cancels = LIST_HEAD_INIT(cancels);
2012         struct ptlrpc_request   *req;
2013         int                      rc, count;
2014         struct mdc_swap_layouts *msl, *payload;
2015         ENTRY;
2016
2017         msl = op_data->op_data;
2018
2019         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2020          * first thing it will do is to cancel the 2 layout
2021          * locks held by this client.
2022          * So the client must cancel its layout locks on the 2 fids
2023          * with the request RPC to avoid extra RPC round trips.
2024          */
2025         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2026                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2027                                         MDS_INODELOCK_XATTR);
2028         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2029                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2030                                          MDS_INODELOCK_XATTR);
2031
2032         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2033                                    &RQF_MDS_SWAP_LAYOUTS);
2034         if (req == NULL) {
2035                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2036                 RETURN(-ENOMEM);
2037         }
2038
2039         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2040         if (rc) {
2041                 ptlrpc_request_free(req);
2042                 RETURN(rc);
2043         }
2044
2045         mdc_swap_layouts_pack(req, op_data);
2046
2047         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2048         LASSERT(payload);
2049
2050         *payload = *msl;
2051
2052         ptlrpc_request_set_replen(req);
2053
2054         rc = ptlrpc_queue_wait(req);
2055         if (rc)
2056                 GOTO(out, rc);
2057         EXIT;
2058
2059 out:
2060         ptlrpc_req_finished(req);
2061         return rc;
2062 }
2063
2064 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2065                          void *karg, void __user *uarg)
2066 {
2067         struct obd_device *obd = exp->exp_obd;
2068         struct obd_ioctl_data *data = karg;
2069         struct obd_import *imp = obd->u.cli.cl_import;
2070         int rc;
2071         ENTRY;
2072
2073         if (!try_module_get(THIS_MODULE)) {
2074                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2075                        module_name(THIS_MODULE));
2076                 return -EINVAL;
2077         }
2078         switch (cmd) {
2079         case OBD_IOC_FID2PATH:
2080                 rc = mdc_ioc_fid2path(exp, karg);
2081                 GOTO(out, rc);
2082         case LL_IOC_HSM_CT_START:
2083                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2084                 /* ignore if it was already registered on this MDS. */
2085                 if (rc == -EEXIST)
2086                         rc = 0;
2087                 GOTO(out, rc);
2088         case LL_IOC_HSM_PROGRESS:
2089                 rc = mdc_ioc_hsm_progress(exp, karg);
2090                 GOTO(out, rc);
2091         case LL_IOC_HSM_STATE_GET:
2092                 rc = mdc_ioc_hsm_state_get(exp, karg);
2093                 GOTO(out, rc);
2094         case LL_IOC_HSM_STATE_SET:
2095                 rc = mdc_ioc_hsm_state_set(exp, karg);
2096                 GOTO(out, rc);
2097         case LL_IOC_HSM_ACTION:
2098                 rc = mdc_ioc_hsm_current_action(exp, karg);
2099                 GOTO(out, rc);
2100         case LL_IOC_HSM_REQUEST:
2101                 rc = mdc_ioc_hsm_request(exp, karg);
2102                 GOTO(out, rc);
2103         case OBD_IOC_CLIENT_RECOVER:
2104                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2105                 if (rc < 0)
2106                         GOTO(out, rc);
2107                 GOTO(out, rc = 0);
2108         case IOC_OSC_SET_ACTIVE:
2109                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2110                 GOTO(out, rc);
2111         case OBD_IOC_PING_TARGET:
2112                 rc = ptlrpc_obd_ping(obd);
2113                 GOTO(out, rc);
2114         /*
2115          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2116          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2117          * there'd be no LMV layer thus we might be called here. Eventually
2118          * this code should be removed.
2119          * bz20731, LU-592.
2120          */
2121         case IOC_OBD_STATFS: {
2122                 struct obd_statfs stat_buf = {0};
2123
2124                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2125                         GOTO(out, rc = -ENODEV);
2126
2127                 /* copy UUID */
2128                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2129                                  min((int)data->ioc_plen2,
2130                                      (int)sizeof(struct obd_uuid))))
2131                         GOTO(out, rc = -EFAULT);
2132
2133                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2134                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2135                                 0);
2136                 if (rc != 0)
2137                         GOTO(out, rc);
2138
2139                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2140                                      min((int) data->ioc_plen1,
2141                                          (int) sizeof(stat_buf))))
2142                         GOTO(out, rc = -EFAULT);
2143
2144                 GOTO(out, rc = 0);
2145         }
2146         case OBD_IOC_QUOTACTL: {
2147                 struct if_quotactl *qctl = karg;
2148                 struct obd_quotactl *oqctl;
2149
2150                 OBD_ALLOC_PTR(oqctl);
2151                 if (oqctl == NULL)
2152                         GOTO(out, rc = -ENOMEM);
2153
2154                 QCTL_COPY(oqctl, qctl);
2155                 rc = obd_quotactl(exp, oqctl);
2156                 if (rc == 0) {
2157                         QCTL_COPY(qctl, oqctl);
2158                         qctl->qc_valid = QC_MDTIDX;
2159                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2160                 }
2161
2162                 OBD_FREE_PTR(oqctl);
2163                 GOTO(out, rc);
2164         }
2165         case LL_IOC_GET_CONNECT_FLAGS:
2166                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2167                                  sizeof(*exp_connect_flags_ptr(exp))))
2168                         GOTO(out, rc = -EFAULT);
2169
2170                 GOTO(out, rc = 0);
2171         case LL_IOC_LOV_SWAP_LAYOUTS:
2172                 rc = mdc_ioc_swap_layouts(exp, karg);
2173                 GOTO(out, rc);
2174         default:
2175                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2176                 GOTO(out, rc = -ENOTTY);
2177         }
2178 out:
2179         module_put(THIS_MODULE);
2180
2181         return rc;
2182 }
2183
2184 static int mdc_get_info_rpc(struct obd_export *exp,
2185                             u32 keylen, void *key,
2186                             u32 vallen, void *val)
2187 {
2188         struct obd_import      *imp = class_exp2cliimp(exp);
2189         struct ptlrpc_request  *req;
2190         char                   *tmp;
2191         int                     rc = -EINVAL;
2192         ENTRY;
2193
2194         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2195         if (req == NULL)
2196                 RETURN(-ENOMEM);
2197
2198         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2199                              RCL_CLIENT, keylen);
2200         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2201                              RCL_CLIENT, sizeof(vallen));
2202
2203         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2204         if (rc) {
2205                 ptlrpc_request_free(req);
2206                 RETURN(rc);
2207         }
2208
2209         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2210         memcpy(tmp, key, keylen);
2211         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2212         memcpy(tmp, &vallen, sizeof(vallen));
2213
2214         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2215                              RCL_SERVER, vallen);
2216         ptlrpc_request_set_replen(req);
2217
2218         rc = ptlrpc_queue_wait(req);
2219         /* -EREMOTE means the get_info result is partial, and it needs to
2220          * continue on another MDT, see fid2path part in lmv_iocontrol */
2221         if (rc == 0 || rc == -EREMOTE) {
2222                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2223                 memcpy(val, tmp, vallen);
2224                 if (ptlrpc_rep_need_swab(req)) {
2225                         if (KEY_IS(KEY_FID2PATH))
2226                                 lustre_swab_fid2path(val);
2227                 }
2228         }
2229         ptlrpc_req_finished(req);
2230
2231         RETURN(rc);
2232 }
2233
2234 static void lustre_swab_hai(struct hsm_action_item *h)
2235 {
2236         __swab32s(&h->hai_len);
2237         __swab32s(&h->hai_action);
2238         lustre_swab_lu_fid(&h->hai_fid);
2239         lustre_swab_lu_fid(&h->hai_dfid);
2240         __swab64s(&h->hai_cookie);
2241         __swab64s(&h->hai_extent.offset);
2242         __swab64s(&h->hai_extent.length);
2243         __swab64s(&h->hai_gid);
2244 }
2245
2246 static void lustre_swab_hal(struct hsm_action_list *h)
2247 {
2248         struct hsm_action_item  *hai;
2249         __u32                    i;
2250
2251         __swab32s(&h->hal_version);
2252         __swab32s(&h->hal_count);
2253         __swab32s(&h->hal_archive_id);
2254         __swab64s(&h->hal_flags);
2255         hai = hai_first(h);
2256         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2257                 lustre_swab_hai(hai);
2258 }
2259
2260 static void lustre_swab_kuch(struct kuc_hdr *l)
2261 {
2262         __swab16s(&l->kuc_magic);
2263         /* __u8 l->kuc_transport */
2264         __swab16s(&l->kuc_msgtype);
2265         __swab16s(&l->kuc_msglen);
2266 }
2267
2268 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2269                                 struct lustre_kernelcomm *lk)
2270 {
2271         struct obd_import *imp = class_exp2cliimp(exp);
2272         int rc = 0;
2273
2274         if (lk->lk_group != KUC_GRP_HSM) {
2275                 CERROR("Bad copytool group %d\n", lk->lk_group);
2276                 return -EINVAL;
2277         }
2278
2279         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2280                lk->lk_uid, lk->lk_group, lk->lk_flags);
2281
2282         if (lk->lk_flags & LK_FLG_STOP) {
2283                 /* Unregister with the coordinator */
2284                 rc = mdc_ioc_hsm_ct_unregister(imp);
2285         } else {
2286                 __u32 *archives = NULL;
2287
2288                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2289                         archives = lk->lk_data;
2290
2291                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2292         }
2293
2294         return rc;
2295 }
2296
2297 /**
2298  * Send a message to any listening copytools
2299  * @param val KUC message (kuc_hdr + hsm_action_list)
2300  * @param len total length of message
2301  */
2302 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2303                                  size_t len, void *val)
2304 {
2305         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2306         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2307         int                      rc;
2308         ENTRY;
2309
2310         if (len < sizeof(*lh) + sizeof(*hal)) {
2311                 CERROR("Short HSM message %zu < %zu\n", len,
2312                        sizeof(*lh) + sizeof(*hal));
2313                 RETURN(-EPROTO);
2314         }
2315         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2316                 lustre_swab_kuch(lh);
2317                 lustre_swab_hal(hal);
2318         } else if (lh->kuc_magic != KUC_MAGIC) {
2319                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2320                 RETURN(-EPROTO);
2321         }
2322
2323         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2324                "on %s\n",
2325                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2326                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2327
2328         /* Broadcast to HSM listeners */
2329         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2330
2331         RETURN(rc);
2332 }
2333
2334 /**
2335  * callback function passed to kuc for re-registering each HSM copytool
2336  * running on MDC, after MDT shutdown/recovery.
2337  * @param data copytool registration data
2338  * @param cb_arg callback argument (obd_import)
2339  */
2340 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2341 {
2342         struct obd_import *imp = (struct obd_import *)cb_arg;
2343         struct kkuc_ct_data *kcd = data;
2344         __u32 *archives = NULL;
2345         int rc;
2346
2347         if (kcd == NULL ||
2348             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2349              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2350                 return -EPROTO;
2351
2352         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2353                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2354                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2355                        kcd->kcd_nr_archives);
2356         } else {
2357                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2358                        "(archive nr = %u)\n",
2359                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2360                 if (kcd->kcd_nr_archives != 0)
2361                         archives = kcd->kcd_archives;
2362         }
2363
2364         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2365         /* ignore error if the copytool is already registered */
2366         return (rc == -EEXIST) ? 0 : rc;
2367 }
2368
2369 /**
2370  * Re-establish all kuc contexts with MDT
2371  * after MDT shutdown/recovery.
2372  */
2373 static int mdc_kuc_reregister(struct obd_import *imp)
2374 {
2375         /* re-register HSM agents */
2376         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2377                                          mdc_hsm_ct_reregister, imp);
2378 }
2379
2380 static int mdc_set_info_async(const struct lu_env *env,
2381                               struct obd_export *exp,
2382                               u32 keylen, void *key,
2383                               u32 vallen, void *val,
2384                               struct ptlrpc_request_set *set)
2385 {
2386         struct obd_import       *imp = class_exp2cliimp(exp);
2387         int                      rc;
2388         ENTRY;
2389
2390         if (KEY_IS(KEY_READ_ONLY)) {
2391                 if (vallen != sizeof(int))
2392                         RETURN(-EINVAL);
2393
2394                 spin_lock(&imp->imp_lock);
2395                 if (*((int *)val)) {
2396                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2397                         imp->imp_connect_data.ocd_connect_flags |=
2398                                                         OBD_CONNECT_RDONLY;
2399                 } else {
2400                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2401                         imp->imp_connect_data.ocd_connect_flags &=
2402                                                         ~OBD_CONNECT_RDONLY;
2403                 }
2404                 spin_unlock(&imp->imp_lock);
2405
2406                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2407                                        keylen, key, vallen, val, set);
2408                 RETURN(rc);
2409         }
2410         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2411                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2412                                        keylen, key, vallen, val, set);
2413                 RETURN(rc);
2414         }
2415         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2416                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2417                                            val);
2418                 RETURN(rc);
2419         }
2420
2421         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2422                 __u32 *default_easize = val;
2423
2424                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2425                 RETURN(0);
2426         }
2427
2428         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2429         RETURN(rc);
2430 }
2431
2432 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2433                         __u32 keylen, void *key, __u32 *vallen, void *val)
2434 {
2435         int rc = -EINVAL;
2436
2437         if (KEY_IS(KEY_MAX_EASIZE)) {
2438                 __u32 mdsize, *max_easize;
2439
2440                 if (*vallen != sizeof(int))
2441                         RETURN(-EINVAL);
2442                 mdsize = *(__u32 *)val;
2443                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2444                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2445                 max_easize = val;
2446                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2447                 RETURN(0);
2448         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2449                 __u32 *default_easize;
2450
2451                 if (*vallen != sizeof(int))
2452                         RETURN(-EINVAL);
2453                 default_easize = val;
2454                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2455                 RETURN(0);
2456         } else if (KEY_IS(KEY_CONN_DATA)) {
2457                 struct obd_import *imp = class_exp2cliimp(exp);
2458                 struct obd_connect_data *data = val;
2459
2460                 if (*vallen != sizeof(*data))
2461                         RETURN(-EINVAL);
2462
2463                 *data = imp->imp_connect_data;
2464                 RETURN(0);
2465         } else if (KEY_IS(KEY_TGT_COUNT)) {
2466                 *((__u32 *)val) = 1;
2467                 RETURN(0);
2468         }
2469
2470         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2471
2472         RETURN(rc);
2473 }
2474
2475 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2476                      struct ptlrpc_request **request)
2477 {
2478         struct ptlrpc_request *req;
2479         int                    rc;
2480         ENTRY;
2481
2482         *request = NULL;
2483         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2484         if (req == NULL)
2485                 RETURN(-ENOMEM);
2486
2487         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2488         if (rc) {
2489                 ptlrpc_request_free(req);
2490                 RETURN(rc);
2491         }
2492
2493         mdc_pack_body(req, fid, 0, 0, -1, 0);
2494
2495         ptlrpc_request_set_replen(req);
2496
2497         rc = ptlrpc_queue_wait(req);
2498         if (rc)
2499                 ptlrpc_req_finished(req);
2500         else
2501                 *request = req;
2502         RETURN(rc);
2503 }
2504
2505 struct mdc_rmfid_args {
2506         int *mra_rcs;
2507         int mra_nr;
2508 };
2509
2510 int mdc_rmfid_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2511                           void *args, int rc)
2512 {
2513         struct mdc_rmfid_args *aa;
2514         int *rcs, size;
2515         ENTRY;
2516
2517         if (!rc) {
2518                 aa = ptlrpc_req_async_args(req);
2519
2520                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2521                                             RCL_SERVER);
2522                 LASSERT(size == sizeof(int) * aa->mra_nr);
2523                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2524                 LASSERT(rcs);
2525                 LASSERT(aa->mra_rcs);
2526                 LASSERT(aa->mra_nr);
2527                 memcpy(aa->mra_rcs, rcs, size);
2528         }
2529
2530         RETURN(rc);
2531 }
2532
2533 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2534                      int *rcs, struct ptlrpc_request_set *set)
2535 {
2536         struct ptlrpc_request *req;
2537         struct mdc_rmfid_args *aa;
2538         struct mdt_body *b;
2539         struct lu_fid *tmp;
2540         int rc, flen;
2541         ENTRY;
2542
2543         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2544         if (req == NULL)
2545                 RETURN(-ENOMEM);
2546
2547         flen = fa->fa_nr * sizeof(struct lu_fid);
2548         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2549                              RCL_CLIENT, flen);
2550         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2551                              RCL_SERVER, flen);
2552         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2553                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2554         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2555         if (rc) {
2556                 ptlrpc_request_free(req);
2557                 RETURN(rc);
2558         }
2559         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2560         memcpy(tmp, fa->fa_fids, flen);
2561
2562         mdc_pack_body(req, NULL, 0, 0, -1, 0);
2563         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2564         b->mbo_ctime = ktime_get_real_seconds();
2565
2566         ptlrpc_request_set_replen(req);
2567
2568         LASSERT(rcs);
2569         aa = ptlrpc_req_async_args(req);
2570         aa->mra_rcs = rcs;
2571         aa->mra_nr = fa->fa_nr;
2572         req->rq_interpret_reply = mdc_rmfid_interpret;
2573
2574         ptlrpc_set_add_req(set, req);
2575         ptlrpc_check_set(NULL, set);
2576
2577         RETURN(rc);
2578 }
2579
2580 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2581                             enum obd_import_event event)
2582 {
2583         struct client_obd *cli = &obd->u.cli;
2584         int rc = 0;
2585
2586         LASSERT(imp->imp_obd == obd);
2587
2588         switch (event) {
2589         case IMP_EVENT_DISCON:
2590                 spin_lock(&cli->cl_loi_list_lock);
2591                 cli->cl_avail_grant = 0;
2592                 cli->cl_lost_grant = 0;
2593                 spin_unlock(&cli->cl_loi_list_lock);
2594                 break;
2595         case IMP_EVENT_INACTIVE:
2596                 /*
2597                  * Flush current sequence to make client obtain new one
2598                  * from server in case of disconnect/reconnect.
2599                  */
2600                 down_read(&cli->cl_seq_rwsem);
2601                 if (cli->cl_seq)
2602                         seq_client_flush(cli->cl_seq);
2603                 up_read(&cli->cl_seq_rwsem);
2604
2605                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2606                 break;
2607         case IMP_EVENT_INVALIDATE: {
2608                 struct ldlm_namespace *ns = obd->obd_namespace;
2609                 struct lu_env *env;
2610                 __u16 refcheck;
2611
2612                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2613
2614                 env = cl_env_get(&refcheck);
2615                 if (!IS_ERR(env)) {
2616                         /* Reset grants. All pages go to failing rpcs due to
2617                          * the invalid import.
2618                          */
2619                         osc_io_unplug(env, cli, NULL);
2620
2621                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2622                                                  osc_ldlm_resource_invalidate,
2623                                                  env, 0);
2624                         cl_env_put(env, &refcheck);
2625                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2626                 } else {
2627                         rc = PTR_ERR(env);
2628                 }
2629                 break;
2630         }
2631         case IMP_EVENT_ACTIVE:
2632                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2633                 /* redo the kuc registration after reconnecting */
2634                 if (rc == 0)
2635                         rc = mdc_kuc_reregister(imp);
2636                 break;
2637         case IMP_EVENT_OCD: {
2638                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2639
2640                 if (OCD_HAS_FLAG(ocd, GRANT))
2641                         osc_init_grant(cli, ocd);
2642
2643                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2644                 break;
2645         }
2646         case IMP_EVENT_DEACTIVATE:
2647         case IMP_EVENT_ACTIVATE:
2648                 break;
2649         default:
2650                 CERROR("Unknown import event %x\n", event);
2651                 LBUG();
2652         }
2653         RETURN(rc);
2654 }
2655
2656 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2657                   struct lu_fid *fid, struct md_op_data *op_data)
2658 {
2659         struct client_obd *cli = &exp->exp_obd->u.cli;
2660         int rc = -EIO;
2661
2662         ENTRY;
2663
2664         down_read(&cli->cl_seq_rwsem);
2665         if (cli->cl_seq)
2666                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2667         up_read(&cli->cl_seq_rwsem);
2668
2669         RETURN(rc);
2670 }
2671
2672 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2673 {
2674         struct client_obd *cli = &exp->exp_obd->u.cli;
2675         return &cli->cl_target_uuid;
2676 }
2677
2678 /**
2679  * Determine whether the lock can be canceled before replaying it during
2680  * recovery, non zero value will be return if the lock can be canceled,
2681  * or zero returned for not
2682  */
2683 static int mdc_cancel_weight(struct ldlm_lock *lock)
2684 {
2685         if (lock->l_resource->lr_type != LDLM_IBITS)
2686                 RETURN(0);
2687
2688         /* FIXME: if we ever get into a situation where there are too many
2689          * opened files with open locks on a single node, then we really
2690          * should replay these open locks to reget it */
2691         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2692                 RETURN(0);
2693
2694         /* Special case for DoM locks, cancel only unused and granted locks */
2695         if (ldlm_has_dom(lock) &&
2696             (lock->l_granted_mode != lock->l_req_mode ||
2697              osc_ldlm_weigh_ast(lock) != 0))
2698                 RETURN(0);
2699
2700         RETURN(1);
2701 }
2702
2703 static int mdc_resource_inode_free(struct ldlm_resource *res)
2704 {
2705         if (res->lr_lvb_inode)
2706                 res->lr_lvb_inode = NULL;
2707
2708         return 0;
2709 }
2710
2711 static struct ldlm_valblock_ops inode_lvbo = {
2712         .lvbo_free = mdc_resource_inode_free
2713 };
2714
2715 static int mdc_llog_init(struct obd_device *obd)
2716 {
2717         struct obd_llog_group   *olg = &obd->obd_olg;
2718         struct llog_ctxt        *ctxt;
2719         int                      rc;
2720
2721         ENTRY;
2722
2723         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2724                         &llog_client_ops);
2725         if (rc < 0)
2726                 RETURN(rc);
2727
2728         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2729         llog_initiator_connect(ctxt);
2730         llog_ctxt_put(ctxt);
2731
2732         RETURN(0);
2733 }
2734
2735 static void mdc_llog_finish(struct obd_device *obd)
2736 {
2737         struct llog_ctxt *ctxt;
2738
2739         ENTRY;
2740
2741         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2742         if (ctxt != NULL)
2743                 llog_cleanup(NULL, ctxt);
2744
2745         EXIT;
2746 }
2747
2748 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2749 {
2750         int rc;
2751
2752         ENTRY;
2753
2754         rc = osc_setup_common(obd, cfg);
2755         if (rc < 0)
2756                 RETURN(rc);
2757
2758         rc = mdc_tunables_init(obd);
2759         if (rc)
2760                 GOTO(err_osc_cleanup, rc);
2761
2762         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2763
2764         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2765
2766         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2767
2768         rc = mdc_llog_init(obd);
2769         if (rc) {
2770                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2771                        obd->obd_name, rc);
2772                 GOTO(err_llog_cleanup, rc);
2773         }
2774
2775         rc = mdc_changelog_cdev_init(obd);
2776         if (rc) {
2777                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2778                        obd->obd_name, rc);
2779                 GOTO(err_changelog_cleanup, rc);
2780         }
2781
2782         RETURN(rc);
2783
2784 err_changelog_cleanup:
2785         mdc_llog_finish(obd);
2786 err_llog_cleanup:
2787         lprocfs_free_md_stats(obd);
2788         ptlrpc_lprocfs_unregister_obd(obd);
2789 err_osc_cleanup:
2790         osc_cleanup_common(obd);
2791         return rc;
2792 }
2793
2794 /* Initialize the default and maximum LOV EA sizes.  This allows
2795  * us to make MDS RPCs with large enough reply buffers to hold a default
2796  * sized EA without having to calculate this (via a call into the
2797  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2798  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2799  * a large number of stripes is possible.  If a larger reply buffer is
2800  * required it will be reallocated in the ptlrpc layer due to overflow.
2801  */
2802 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2803                             __u32 def_easize)
2804 {
2805         struct obd_device *obd = exp->exp_obd;
2806         struct client_obd *cli = &obd->u.cli;
2807         ENTRY;
2808
2809         if (cli->cl_max_mds_easize < easize)
2810                 cli->cl_max_mds_easize = easize;
2811
2812         if (cli->cl_default_mds_easize < def_easize)
2813                 cli->cl_default_mds_easize = def_easize;
2814
2815         RETURN(0);
2816 }
2817
2818 static int mdc_precleanup(struct obd_device *obd)
2819 {
2820         ENTRY;
2821
2822         osc_precleanup_common(obd);
2823         mdc_changelog_cdev_finish(obd);
2824
2825         obd_cleanup_client_import(obd);
2826         ptlrpc_lprocfs_unregister_obd(obd);
2827         lprocfs_free_md_stats(obd);
2828         mdc_llog_finish(obd);
2829         RETURN(0);
2830 }
2831
2832 static int mdc_cleanup(struct obd_device *obd)
2833 {
2834         return osc_cleanup_common(obd);
2835 }
2836
2837 int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2838 {
2839         struct lustre_cfg *lcfg = buf;
2840         size_t count  = class_modify_config(lcfg, PARAM_MDC,
2841                                             &obd->obd_kset.kobj);
2842
2843         return count > 0 ? 0 : count;
2844 }
2845
2846 static struct obd_ops mdc_obd_ops = {
2847         .o_owner            = THIS_MODULE,
2848         .o_setup            = mdc_setup,
2849         .o_precleanup       = mdc_precleanup,
2850         .o_cleanup          = mdc_cleanup,
2851         .o_add_conn         = client_import_add_conn,
2852         .o_del_conn         = client_import_del_conn,
2853         .o_connect          = client_connect_import,
2854         .o_reconnect        = osc_reconnect,
2855         .o_disconnect       = osc_disconnect,
2856         .o_iocontrol        = mdc_iocontrol,
2857         .o_set_info_async   = mdc_set_info_async,
2858         .o_statfs           = mdc_statfs,
2859         .o_fid_init         = client_fid_init,
2860         .o_fid_fini         = client_fid_fini,
2861         .o_fid_alloc        = mdc_fid_alloc,
2862         .o_import_event     = mdc_import_event,
2863         .o_get_info         = mdc_get_info,
2864         .o_process_config   = mdc_process_config,
2865         .o_get_uuid         = mdc_get_uuid,
2866         .o_quotactl         = mdc_quotactl,
2867 };
2868
2869 static struct md_ops mdc_md_ops = {
2870         .m_get_root         = mdc_get_root,
2871         .m_null_inode       = mdc_null_inode,
2872         .m_close            = mdc_close,
2873         .m_create           = mdc_create,
2874         .m_enqueue          = mdc_enqueue,
2875         .m_getattr          = mdc_getattr,
2876         .m_getattr_name     = mdc_getattr_name,
2877         .m_intent_lock      = mdc_intent_lock,
2878         .m_link             = mdc_link,
2879         .m_rename           = mdc_rename,
2880         .m_setattr          = mdc_setattr,
2881         .m_setxattr         = mdc_setxattr,
2882         .m_getxattr         = mdc_getxattr,
2883         .m_fsync                = mdc_fsync,
2884         .m_file_resync          = mdc_file_resync,
2885         .m_read_page            = mdc_read_page,
2886         .m_unlink           = mdc_unlink,
2887         .m_cancel_unused    = mdc_cancel_unused,
2888         .m_init_ea_size     = mdc_init_ea_size,
2889         .m_set_lock_data    = mdc_set_lock_data,
2890         .m_lock_match       = mdc_lock_match,
2891         .m_get_lustre_md    = mdc_get_lustre_md,
2892         .m_free_lustre_md   = mdc_free_lustre_md,
2893         .m_set_open_replay_data = mdc_set_open_replay_data,
2894         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2895         .m_intent_getattr_async = mdc_intent_getattr_async,
2896         .m_revalidate_lock      = mdc_revalidate_lock,
2897         .m_rmfid                = mdc_rmfid,
2898 };
2899
2900 dev_t mdc_changelog_dev;
2901 struct class *mdc_changelog_class;
2902 static int __init mdc_init(void)
2903 {
2904         int rc = 0;
2905         rc = alloc_chrdev_region(&mdc_changelog_dev, 0,
2906                                  MDC_CHANGELOG_DEV_COUNT,
2907                                  MDC_CHANGELOG_DEV_NAME);
2908         if (rc)
2909                 return rc;
2910
2911         mdc_changelog_class = class_create(THIS_MODULE, MDC_CHANGELOG_DEV_NAME);
2912         if (IS_ERR(mdc_changelog_class)) {
2913                 rc = PTR_ERR(mdc_changelog_class);
2914                 goto out_dev;
2915         }
2916
2917         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2918                                  LUSTRE_MDC_NAME, &mdc_device_type);
2919         if (rc)
2920                 goto out_dev;
2921
2922         return 0;
2923
2924 out_dev:
2925         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
2926         return rc;
2927 }
2928
2929 static void __exit mdc_exit(void)
2930 {
2931         class_destroy(mdc_changelog_class);
2932         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
2933         class_unregister_type(LUSTRE_MDC_NAME);
2934 }
2935
2936 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2937 MODULE_DESCRIPTION("Lustre Metadata Client");
2938 MODULE_VERSION(LUSTRE_VERSION_STRING);
2939 MODULE_LICENSE("GPL");
2940
2941 module_init(mdc_init);
2942 module_exit(mdc_exit);