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, 2015, 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/miscdevice.h>
  38 #include <linux/module.h>
  39 #include <linux/pagemap.h>
  40 #include <linux/user_namespace.h>
  41 #include <linux/utsname.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45
  46 #include <cl_object.h>
  47 #include <llog_swab.h>
  48 #include <lprocfs_status.h>
  49 #include <lustre_acl.h>
  50 #include <lustre_fid.h>
  51 #include <lustre_ioctl.h>
  52 #include <lustre_kernelcomm.h>
  53 #include <lustre_lmv.h>
  54 #include <lustre_log.h>
  55 #include <lustre_param.h>
  56 #include <lustre_swab.h>
  57 #include <obd_class.h>
  58
  59 #include "mdc_internal.h"
  60
  61 #define REQUEST_MINOR 244
  62
  63 static int mdc_cleanup(struct obd_device *obd);
  64
  65 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  66 {
  67         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  68         int rc;
  69
  70         /* obd_get_request_slot() ensures that this client has no more
  71          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  72          * against an MDT. */
  73         rc = obd_get_request_slot(cli);
  74         if (rc != 0)
  75                 return rc;
  76
  77         rc = ptlrpc_queue_wait(req);
  78         obd_put_request_slot(cli);
  79
  80         return rc;
  81 }
  82
  83 /*
  84  * Send MDS_GET_ROOT RPC to fetch root FID.
  85  *
  86  * If \a fileset is not NULL it should contain a subdirectory off
  87  * the ROOT/ directory to be mounted on the client. Return the FID
  88  * of the subdirectory to the client to mount onto its mountpoint.
  89  *
  90  * \param[in]   imp     MDC import
  91  * \param[in]   fileset fileset name, which could be NULL
  92  * \param[out]  rootfid root FID of this mountpoint
  93  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  94  *
  95  * \retval      0 on success, negative errno on failure
  96  */
  97 static int mdc_get_root(struct obd_export *exp, const char *fileset,
  98                          struct lu_fid *rootfid)
  99 {
 100         struct ptlrpc_request   *req;
 101         struct mdt_body         *body;
 102         int                      rc;
 103
 104         ENTRY;
 105
 106         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 107                 RETURN(-ENOTSUPP);
 108
 109         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 110                                 &RQF_MDS_GET_ROOT);
 111         if (req == NULL)
 112                 RETURN(-ENOMEM);
 113
 114         if (fileset != NULL)
 115                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 116                                      strlen(fileset) + 1);
 117         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 118         if (rc) {
 119                 ptlrpc_request_free(req);
 120                 RETURN(rc);
 121         }
 122         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 123         if (fileset != NULL) {
 124                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 125
 126                 memcpy(name, fileset, strlen(fileset));
 127         }
 128         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 129         req->rq_send_state = LUSTRE_IMP_FULL;
 130
 131         ptlrpc_request_set_replen(req);
 132
 133         rc = ptlrpc_queue_wait(req);
 134         if (rc)
 135                 GOTO(out, rc);
 136
 137         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 138         if (body == NULL)
 139                 GOTO(out, rc = -EPROTO);
 140
 141         *rootfid = body->mbo_fid1;
 142         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 143                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 144         EXIT;
 145 out:
 146         ptlrpc_req_finished(req);
 147
 148         return rc;
 149 }
 150
 151 /*
 152  * This function now is known to always saying that it will receive 4 buffers
 153  * from server. Even for cases when acl_size and md_size is zero, RPC header
 154  * will contain 4 fields and RPC itself will contain zero size fields. This is
 155  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 156  * and thus zero, it shrinks it, making zero size. The same story about
 157  * md_size. And this is course of problem when client waits for smaller number
 158  * of fields. This issue will be fixed later when client gets aware of RPC
 159  * layouts.  --umka
 160  */
 161 static int mdc_getattr_common(struct obd_export *exp,
 162                               struct ptlrpc_request *req)
 163 {
 164         struct req_capsule *pill = &req->rq_pill;
 165         struct mdt_body    *body;
 166         void               *eadata;
 167         int                 rc;
 168         ENTRY;
 169
 170         /* Request message already built. */
 171         rc = ptlrpc_queue_wait(req);
 172         if (rc != 0)
 173                 RETURN(rc);
 174
 175         /* sanity check for the reply */
 176         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 177         if (body == NULL)
 178                 RETURN(-EPROTO);
 179
 180         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 181
 182         mdc_update_max_ea_from_body(exp, body);
 183         if (body->mbo_eadatasize != 0) {
 184                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 185                                                       body->mbo_eadatasize);
 186                 if (eadata == NULL)
 187                         RETURN(-EPROTO);
 188         }
 189
 190         RETURN(0);
 191 }
 192
 193 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 194                        struct ptlrpc_request **request)
 195 {
 196         struct ptlrpc_request *req;
 197         int                    rc;
 198         ENTRY;
 199
 200         /* Single MDS without an LMV case */
 201         if (op_data->op_flags & MF_GET_MDT_IDX) {
 202                 op_data->op_mds = 0;
 203                 RETURN(0);
 204         }
 205         *request = NULL;
 206         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 207         if (req == NULL)
 208                 RETURN(-ENOMEM);
 209
 210         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 211         if (rc) {
 212                 ptlrpc_request_free(req);
 213                 RETURN(rc);
 214         }
 215
 216         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 217                       op_data->op_mode, -1, 0);
 218
 219         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 220                              op_data->op_mode);
 221         ptlrpc_request_set_replen(req);
 222
 223         rc = mdc_getattr_common(exp, req);
 224         if (rc)
 225                 ptlrpc_req_finished(req);
 226         else
 227                 *request = req;
 228         RETURN(rc);
 229 }
 230
 231 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 232                             struct ptlrpc_request **request)
 233 {
 234         struct ptlrpc_request *req;
 235         int                    rc;
 236         ENTRY;
 237
 238         *request = NULL;
 239         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 240                                    &RQF_MDS_GETATTR_NAME);
 241         if (req == NULL)
 242                 RETURN(-ENOMEM);
 243
 244         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 245                              op_data->op_namelen + 1);
 246
 247         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 248         if (rc) {
 249                 ptlrpc_request_free(req);
 250                 RETURN(rc);
 251         }
 252
 253         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 254                       op_data->op_mode, op_data->op_suppgids[0], 0);
 255
 256         if (op_data->op_name) {
 257                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 258                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 259                                 op_data->op_namelen);
 260                 memcpy(name, op_data->op_name, op_data->op_namelen);
 261         }
 262
 263         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 264                              op_data->op_mode);
 265         ptlrpc_request_set_replen(req);
 266
 267         rc = mdc_getattr_common(exp, req);
 268         if (rc)
 269                 ptlrpc_req_finished(req);
 270         else
 271                 *request = req;
 272         RETURN(rc);
 273 }
 274
 275 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 276                             const struct lu_fid *fid, int opcode, u64 valid,
 277                             const char *xattr_name, const char *input,
 278                             int input_size, int output_size, int flags,
 279                             __u32 suppgid, struct ptlrpc_request **request)
 280 {
 281         struct ptlrpc_request *req;
 282         int   xattr_namelen = 0;
 283         char *tmp;
 284         int   rc;
 285         ENTRY;
 286
 287         *request = NULL;
 288         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 289         if (req == NULL)
 290                 RETURN(-ENOMEM);
 291
 292         if (xattr_name) {
 293                 xattr_namelen = strlen(xattr_name) + 1;
 294                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 295                                      xattr_namelen);
 296         }
 297         if (input_size) {
 298                 LASSERT(input);
 299                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 300                                      input_size);
 301         }
 302
 303         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 304         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 305             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 306                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 307                 int count;
 308
 309                 /* Without that packing would fail */
 310                 if (input_size == 0)
 311                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 312                                              RCL_CLIENT, 0);
 313
 314                 count = mdc_resource_get_unused(exp, fid,
 315                                                 &cancels, LCK_EX,
 316                                                 MDS_INODELOCK_XATTR);
 317
 318                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 319                 if (rc) {
 320                         ptlrpc_request_free(req);
 321                         RETURN(rc);
 322                 }
 323         } else {
 324                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 325                 if (rc) {
 326                         ptlrpc_request_free(req);
 327                         RETURN(rc);
 328                 }
 329         }
 330
 331         if (opcode == MDS_REINT) {
 332                 struct mdt_rec_setxattr *rec;
 333
 334                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 335                          sizeof(struct mdt_rec_reint));
 336                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 337                 rec->sx_opcode = REINT_SETXATTR;
 338                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 339                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 340                 rec->sx_cap    = cfs_curproc_cap_pack();
 341                 rec->sx_suppgid1 = suppgid;
 342                 rec->sx_suppgid2 = -1;
 343                 rec->sx_fid    = *fid;
 344                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 345                 rec->sx_time   = cfs_time_current_sec();
 346                 rec->sx_size   = output_size;
 347                 rec->sx_flags  = flags;
 348         } else {
 349                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 350         }
 351
 352         if (xattr_name) {
 353                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 354                 memcpy(tmp, xattr_name, xattr_namelen);
 355         }
 356         if (input_size) {
 357                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 358                 memcpy(tmp, input, input_size);
 359         }
 360
 361         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 362                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 363                                      RCL_SERVER, output_size);
 364         ptlrpc_request_set_replen(req);
 365
 366         /* make rpc */
 367         if (opcode == MDS_REINT)
 368                 mdc_get_mod_rpc_slot(req, NULL);
 369
 370         rc = ptlrpc_queue_wait(req);
 371
 372         if (opcode == MDS_REINT)
 373                 mdc_put_mod_rpc_slot(req, NULL);
 374
 375         if (rc)
 376                 ptlrpc_req_finished(req);
 377         else
 378                 *request = req;
 379         RETURN(rc);
 380 }
 381
 382 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 383                         u64 valid, const char *xattr_name,
 384                         const char *input, int input_size, int output_size,
 385                         int flags, __u32 suppgid,
 386                         struct ptlrpc_request **request)
 387 {
 388         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 389                                 fid, MDS_REINT, valid, xattr_name,
 390                                 input, input_size, output_size, flags,
 391                                 suppgid, request);
 392 }
 393
 394 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 395                         u64 valid, const char *xattr_name,
 396                         const char *input, int input_size, int output_size,
 397                         int flags, struct ptlrpc_request **request)
 398 {
 399         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 400                                 fid, MDS_GETXATTR, valid, xattr_name,
 401                                 input, input_size, output_size, flags,
 402                                 -1, request);
 403 }
 404
 405 #ifdef CONFIG_FS_POSIX_ACL
 406 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 407 {
 408         struct req_capsule     *pill = &req->rq_pill;
 409         struct mdt_body        *body = md->body;
 410         struct posix_acl       *acl;
 411         void                   *buf;
 412         int                     rc;
 413         ENTRY;
 414
 415         if (!body->mbo_aclsize)
 416                 RETURN(0);
 417
 418         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 419
 420         if (!buf)
 421                 RETURN(-EPROTO);
 422
 423         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 424         if (acl == NULL)
 425                 RETURN(0);
 426         if (IS_ERR(acl)) {
 427                 rc = PTR_ERR(acl);
 428                 CERROR("convert xattr to acl: %d\n", rc);
 429                 RETURN(rc);
 430         }
 431
 432         rc = posix_acl_valid(acl);
 433         if (rc) {
 434                 CERROR("validate acl: %d\n", rc);
 435                 posix_acl_release(acl);
 436                 RETURN(rc);
 437         }
 438
 439         md->posix_acl = acl;
 440         RETURN(0);
 441 }
 442 #else
 443 #define mdc_unpack_acl(req, md) 0
 444 #endif
 445
 446 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 447                       struct obd_export *dt_exp, struct obd_export *md_exp,
 448                       struct lustre_md *md)
 449 {
 450         struct req_capsule *pill = &req->rq_pill;
 451         int rc;
 452         ENTRY;
 453
 454         LASSERT(md);
 455         memset(md, 0, sizeof(*md));
 456
 457         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 458         LASSERT(md->body != NULL);
 459
 460         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 461                 if (!S_ISREG(md->body->mbo_mode)) {
 462                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 463                                "regular file, but is not\n");
 464                         GOTO(out, rc = -EPROTO);
 465                 }
 466
 467                 if (md->body->mbo_eadatasize == 0) {
 468                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 469                                "but eadatasize 0\n");
 470                         GOTO(out, rc = -EPROTO);
 471                 }
 472
 473                 md->layout.lb_len = md->body->mbo_eadatasize;
 474                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 475                                                         &RMF_MDT_MD,
 476                                                         md->layout.lb_len);
 477                 if (md->layout.lb_buf == NULL)
 478                         GOTO(out, rc = -EPROTO);
 479         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 480                 const union lmv_mds_md *lmv;
 481                 size_t lmv_size;
 482
 483                 if (!S_ISDIR(md->body->mbo_mode)) {
 484                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 485                                "directory, but is not\n");
 486                         GOTO(out, rc = -EPROTO);
 487                 }
 488
 489                 lmv_size = md->body->mbo_eadatasize;
 490                 if (lmv_size == 0) {
 491                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 492                                "but eadatasize 0\n");
 493                         RETURN(-EPROTO);
 494                 }
 495
 496                 if (md->body->mbo_valid & OBD_MD_MEA) {
 497                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 498                                                            lmv_size);
 499                         if (lmv == NULL)
 500                                 GOTO(out, rc = -EPROTO);
 501
 502                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 503                         if (rc < 0)
 504                                 GOTO(out, rc);
 505
 506                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 507                                 CDEBUG(D_INFO, "size too small:  "
 508                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 509                                         rc, (int)sizeof(*md->lmv));
 510                                 GOTO(out, rc = -EPROTO);
 511                         }
 512                 }
 513         }
 514         rc = 0;
 515
 516         if (md->body->mbo_valid & OBD_MD_FLACL) {
 517                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 518                  * only when aclsize != 0 there's an actual segment for ACL
 519                  * in reply buffer.
 520                  */
 521                 if (md->body->mbo_aclsize) {
 522                         rc = mdc_unpack_acl(req, md);
 523                         if (rc)
 524                                 GOTO(out, rc);
 525 #ifdef CONFIG_FS_POSIX_ACL
 526                 } else {
 527                         md->posix_acl = NULL;
 528 #endif
 529                 }
 530         }
 531
 532         EXIT;
 533 out:
 534         if (rc) {
 535 #ifdef CONFIG_FS_POSIX_ACL
 536                 posix_acl_release(md->posix_acl);
 537 #endif
 538         }
 539         return rc;
 540 }
 541
 542 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 543 {
 544         ENTRY;
 545         RETURN(0);
 546 }
 547
 548 void mdc_replay_open(struct ptlrpc_request *req)
 549 {
 550         struct md_open_data *mod = req->rq_cb_data;
 551         struct ptlrpc_request *close_req;
 552         struct obd_client_handle *och;
 553         struct lustre_handle old;
 554         struct mdt_body *body;
 555         ENTRY;
 556
 557         if (mod == NULL) {
 558                 DEBUG_REQ(D_ERROR, req,
 559                           "Can't properly replay without open data.");
 560                 EXIT;
 561                 return;
 562         }
 563
 564         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 565         LASSERT(body != NULL);
 566
 567         och = mod->mod_och;
 568         if (och != NULL) {
 569                 struct lustre_handle *file_fh;
 570
 571                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 572
 573                 file_fh = &och->och_fh;
 574                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 575                        file_fh->cookie, body->mbo_handle.cookie);
 576                 old = *file_fh;
 577                 *file_fh = body->mbo_handle;
 578         }
 579         close_req = mod->mod_close_req;
 580         if (close_req != NULL) {
 581                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 582                 struct mdt_ioepoch *epoch;
 583
 584                 LASSERT(opc == MDS_CLOSE);
 585                 epoch = req_capsule_client_get(&close_req->rq_pill,
 586                                                &RMF_MDT_EPOCH);
 587                 LASSERT(epoch);
 588
 589                 if (och != NULL)
 590                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 591
 592                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 593                 epoch->mio_handle = body->mbo_handle;
 594         }
 595         EXIT;
 596 }
 597
 598 void mdc_commit_open(struct ptlrpc_request *req)
 599 {
 600         struct md_open_data *mod = req->rq_cb_data;
 601         if (mod == NULL)
 602                 return;
 603
 604         /**
 605          * No need to touch md_open_data::mod_och, it holds a reference on
 606          * \var mod and will zero references to each other, \var mod will be
 607          * freed after that when md_open_data::mod_och will put the reference.
 608          */
 609
 610         /**
 611          * Do not let open request to disappear as it still may be needed
 612          * for close rpc to happen (it may happen on evict only, otherwise
 613          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 614          * called), just mark this rpc as committed to distinguish these 2
 615          * cases, see mdc_close() for details. The open request reference will
 616          * be put along with freeing \var mod.
 617          */
 618         ptlrpc_request_addref(req);
 619         spin_lock(&req->rq_lock);
 620         req->rq_committed = 1;
 621         spin_unlock(&req->rq_lock);
 622         req->rq_cb_data = NULL;
 623         obd_mod_put(mod);
 624 }
 625
 626 int mdc_set_open_replay_data(struct obd_export *exp,
 627                              struct obd_client_handle *och,
 628                              struct lookup_intent *it)
 629 {
 630         struct md_open_data     *mod;
 631         struct mdt_rec_create   *rec;
 632         struct mdt_body         *body;
 633         struct ptlrpc_request   *open_req = it->it_request;
 634         struct obd_import       *imp = open_req->rq_import;
 635         ENTRY;
 636
 637         if (!open_req->rq_replay)
 638                 RETURN(0);
 639
 640         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 641         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 642         LASSERT(rec != NULL);
 643         /* Incoming message in my byte order (it's been swabbed). */
 644         /* Outgoing messages always in my byte order. */
 645         LASSERT(body != NULL);
 646
 647         /* Only if the import is replayable, we set replay_open data */
 648         if (och && imp->imp_replayable) {
 649                 mod = obd_mod_alloc();
 650                 if (mod == NULL) {
 651                         DEBUG_REQ(D_ERROR, open_req,
 652                                   "Can't allocate md_open_data");
 653                         RETURN(0);
 654                 }
 655
 656                 /**
 657                  * Take a reference on \var mod, to be freed on mdc_close().
 658                  * It protects \var mod from being freed on eviction (commit
 659                  * callback is called despite rq_replay flag).
 660                  * Another reference for \var och.
 661                  */
 662                 obd_mod_get(mod);
 663                 obd_mod_get(mod);
 664
 665                 spin_lock(&open_req->rq_lock);
 666                 och->och_mod = mod;
 667                 mod->mod_och = och;
 668                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 669                                      it_disposition(it, DISP_OPEN_STRIPE);
 670                 mod->mod_open_req = open_req;
 671                 open_req->rq_cb_data = mod;
 672                 open_req->rq_commit_cb = mdc_commit_open;
 673                 spin_unlock(&open_req->rq_lock);
 674         }
 675
 676         rec->cr_fid2 = body->mbo_fid1;
 677         rec->cr_ioepoch = body->mbo_ioepoch;
 678         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 679         open_req->rq_replay_cb = mdc_replay_open;
 680         if (!fid_is_sane(&body->mbo_fid1)) {
 681                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 682                           "insane fid");
 683                 LBUG();
 684         }
 685
 686         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 687         RETURN(0);
 688 }
 689
 690 static void mdc_free_open(struct md_open_data *mod)
 691 {
 692         int committed = 0;
 693
 694         if (mod->mod_is_create == 0 &&
 695             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 696                 committed = 1;
 697
 698         /**
 699          * No reason to asssert here if the open request has
 700          * rq_replay == 1. It means that mdc_close failed, and
 701          * close request wasn`t sent. It is not fatal to client.
 702          * The worst thing is eviction if the client gets open lock
 703          **/
 704
 705         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 706                   "= %d\n", mod->mod_open_req->rq_replay);
 707
 708         ptlrpc_request_committed(mod->mod_open_req, committed);
 709         if (mod->mod_close_req)
 710                 ptlrpc_request_committed(mod->mod_close_req, committed);
 711 }
 712
 713 int mdc_clear_open_replay_data(struct obd_export *exp,
 714                                struct obd_client_handle *och)
 715 {
 716         struct md_open_data *mod = och->och_mod;
 717         ENTRY;
 718
 719         /**
 720          * It is possible to not have \var mod in a case of eviction between
 721          * lookup and ll_file_open().
 722          **/
 723         if (mod == NULL)
 724                 RETURN(0);
 725
 726         LASSERT(mod != LP_POISON);
 727         LASSERT(mod->mod_open_req != NULL);
 728         mdc_free_open(mod);
 729
 730         mod->mod_och = NULL;
 731         och->och_mod = NULL;
 732         obd_mod_put(mod);
 733
 734         RETURN(0);
 735 }
 736
 737 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 738                      struct md_open_data *mod, struct ptlrpc_request **request)
 739 {
 740         struct obd_device     *obd = class_exp2obd(exp);
 741         struct ptlrpc_request *req;
 742         struct req_format     *req_fmt;
 743         int                    rc;
 744         int                    saved_rc = 0;
 745         ENTRY;
 746
 747         if (op_data->op_bias & MDS_HSM_RELEASE) {
 748                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 749
 750                 /* allocate a FID for volatile file */
 751                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 752                 if (rc < 0) {
 753                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 754                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 755                         /* save the errcode and proceed to close */
 756                         saved_rc = rc;
 757                 }
 758         } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
 759                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 760         } else {
 761                 req_fmt = &RQF_MDS_CLOSE;
 762         }
 763
 764         *request = NULL;
 765         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 766                 req = NULL;
 767         else
 768                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 769
 770         /* Ensure that this close's handle is fixed up during replay. */
 771         if (likely(mod != NULL)) {
 772                 LASSERTF(mod->mod_open_req != NULL &&
 773                          mod->mod_open_req->rq_type != LI_POISON,
 774                          "POISONED open %p!\n", mod->mod_open_req);
 775
 776                 mod->mod_close_req = req;
 777
 778                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 779                 /* We no longer want to preserve this open for replay even
 780                  * though the open was committed. b=3632, b=3633 */
 781                 spin_lock(&mod->mod_open_req->rq_lock);
 782                 mod->mod_open_req->rq_replay = 0;
 783                 spin_unlock(&mod->mod_open_req->rq_lock);
 784         } else {
 785                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 786         }
 787         if (req == NULL) {
 788                 /**
 789                  * TODO: repeat close after errors
 790                  */
 791                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 792                       "dropped when this client unmounts or is evicted\n",
 793                       obd->obd_name, PFID(&op_data->op_fid1));
 794                 GOTO(out, rc = -ENOMEM);
 795         }
 796
 797         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 798         if (rc) {
 799                 ptlrpc_request_free(req);
 800                 GOTO(out, rc);
 801         }
 802
 803         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 804          * portal whose threads are not taking any DLM locks and are therefore
 805          * always progressing */
 806         req->rq_request_portal = MDS_READPAGE_PORTAL;
 807         ptlrpc_at_set_req_timeout(req);
 808
 809
 810         mdc_close_pack(req, op_data);
 811
 812         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 813                              obd->u.cli.cl_default_mds_easize);
 814
 815         ptlrpc_request_set_replen(req);
 816
 817         mdc_get_mod_rpc_slot(req, NULL);
 818         rc = ptlrpc_queue_wait(req);
 819         mdc_put_mod_rpc_slot(req, NULL);
 820
 821         if (req->rq_repmsg == NULL) {
 822                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 823                        req->rq_status);
 824                 if (rc == 0)
 825                         rc = req->rq_status ?: -EIO;
 826         } else if (rc == 0 || rc == -EAGAIN) {
 827                 struct mdt_body *body;
 828
 829                 rc = lustre_msg_get_status(req->rq_repmsg);
 830                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 831                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 832                                   "= %d", rc);
 833                         if (rc > 0)
 834                                 rc = -rc;
 835                 }
 836                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 837                 if (body == NULL)
 838                         rc = -EPROTO;
 839         } else if (rc == -ESTALE) {
 840                 /**
 841                  * it can be allowed error after 3633 if open was committed and
 842                  * server failed before close was sent. Let's check if mod
 843                  * exists and return no error in that case
 844                  */
 845                 if (mod) {
 846                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 847                         LASSERT(mod->mod_open_req != NULL);
 848                         if (mod->mod_open_req->rq_committed)
 849                                 rc = 0;
 850                 }
 851         }
 852
 853 out:
 854         if (mod) {
 855                 if (rc != 0)
 856                         mod->mod_close_req = NULL;
 857                 /* Since now, mod is accessed through open_req only,
 858                  * thus close req does not keep a reference on mod anymore. */
 859                 obd_mod_put(mod);
 860         }
 861         *request = req;
 862
 863         RETURN(rc < 0 ? rc : saved_rc);
 864 }
 865
 866 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 867                        u64 offset, struct page **pages, int npages,
 868                        struct ptlrpc_request **request)
 869 {
 870         struct ptlrpc_request   *req;
 871         struct ptlrpc_bulk_desc *desc;
 872         int                      i;
 873         wait_queue_head_t        waitq;
 874         int                      resends = 0;
 875         struct l_wait_info       lwi;
 876         int                      rc;
 877         ENTRY;
 878
 879         *request = NULL;
 880         init_waitqueue_head(&waitq);
 881
 882 restart_bulk:
 883         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 884         if (req == NULL)
 885                 RETURN(-ENOMEM);
 886
 887         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 888         if (rc) {
 889                 ptlrpc_request_free(req);
 890                 RETURN(rc);
 891         }
 892
 893         req->rq_request_portal = MDS_READPAGE_PORTAL;
 894         ptlrpc_at_set_req_timeout(req);
 895
 896         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 897                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 898                                     MDS_BULK_PORTAL,
 899                                     &ptlrpc_bulk_kiov_pin_ops);
 900         if (desc == NULL) {
 901                 ptlrpc_request_free(req);
 902                 RETURN(-ENOMEM);
 903         }
 904
 905         /* NB req now owns desc and will free it when it gets freed */
 906         for (i = 0; i < npages; i++)
 907                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 908                                                  PAGE_SIZE);
 909
 910         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 911
 912         ptlrpc_request_set_replen(req);
 913         rc = ptlrpc_queue_wait(req);
 914         if (rc) {
 915                 ptlrpc_req_finished(req);
 916                 if (rc != -ETIMEDOUT)
 917                         RETURN(rc);
 918
 919                 resends++;
 920                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 921                         CERROR("%s: too many resend retries: rc = %d\n",
 922                                exp->exp_obd->obd_name, -EIO);
 923                         RETURN(-EIO);
 924                 }
 925                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 926                                        NULL);
 927                 l_wait_event(waitq, 0, &lwi);
 928
 929                 goto restart_bulk;
 930         }
 931
 932         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 933                                           req->rq_bulk->bd_nob_transferred);
 934         if (rc < 0) {
 935                 ptlrpc_req_finished(req);
 936                 RETURN(rc);
 937         }
 938
 939         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
 940                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
 941                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
 942                        PAGE_SIZE * npages);
 943                 ptlrpc_req_finished(req);
 944                 RETURN(-EPROTO);
 945         }
 946
 947         *request = req;
 948         RETURN(0);
 949 }
 950
 951 static void mdc_release_page(struct page *page, int remove)
 952 {
 953         if (remove) {
 954                 lock_page(page);
 955                 if (likely(page->mapping != NULL))
 956                         truncate_complete_page(page->mapping, page);
 957                 unlock_page(page);
 958         }
 959         put_page(page);
 960 }
 961
 962 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
 963                                     __u64 *start, __u64 *end, int hash64)
 964 {
 965         /*
 966          * Complement of hash is used as an index so that
 967          * radix_tree_gang_lookup() can be used to find a page with starting
 968          * hash _smaller_ than one we are looking for.
 969          */
 970         unsigned long offset = hash_x_index(*hash, hash64);
 971         struct page *page;
 972         int found;
 973
 974         spin_lock_irq(&mapping->tree_lock);
 975         found = radix_tree_gang_lookup(&mapping->page_tree,
 976                                        (void **)&page, offset, 1);
 977         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 978                 struct lu_dirpage *dp;
 979
 980                 get_page(page);
 981                 spin_unlock_irq(&mapping->tree_lock);
 982                 /*
 983                  * In contrast to find_lock_page() we are sure that directory
 984                  * page cannot be truncated (while DLM lock is held) and,
 985                  * hence, can avoid restart.
 986                  *
 987                  * In fact, page cannot be locked here at all, because
 988                  * mdc_read_page_remote does synchronous io.
 989                  */
 990                 wait_on_page_locked(page);
 991                 if (PageUptodate(page)) {
 992                         dp = kmap(page);
 993                         if (BITS_PER_LONG == 32 && hash64) {
 994                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 995                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 996                                 *hash  = *hash >> 32;
 997                         } else {
 998                                 *start = le64_to_cpu(dp->ldp_hash_start);
 999                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1000                         }
1001                         if (unlikely(*start == 1 && *hash == 0))
1002                                 *hash = *start;
1003                         else
1004                                 LASSERTF(*start <= *hash, "start = %#llx"
1005                                          ",end = %#llx,hash = %#llx\n",
1006                                          *start, *end, *hash);
1007                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1008                               " hash %#llx\n", offset, *start, *end, *hash);
1009                         if (*hash > *end) {
1010                                 kunmap(page);
1011                                 mdc_release_page(page, 0);
1012                                 page = NULL;
1013                         } else if (*end != *start && *hash == *end) {
1014                                 /*
1015                                  * upon hash collision, remove this page,
1016                                  * otherwise put page reference, and
1017                                  * mdc_read_page_remote() will issue RPC to
1018                                  * fetch the page we want.
1019                                  */
1020                                 kunmap(page);
1021                                 mdc_release_page(page,
1022                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1023                                 page = NULL;
1024                         }
1025                 } else {
1026                         put_page(page);
1027                         page = ERR_PTR(-EIO);
1028                 }
1029         } else {
1030                 spin_unlock_irq(&mapping->tree_lock);
1031                 page = NULL;
1032         }
1033         return page;
1034 }
1035
1036 /*
1037  * Adjust a set of pages, each page containing an array of lu_dirpages,
1038  * so that each page can be used as a single logical lu_dirpage.
1039  *
1040  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1041  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1042  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1043  * value is used as a cookie to request the next lu_dirpage in a
1044  * directory listing that spans multiple pages (two in this example):
1045  *   ________
1046  *  |        |
1047  * .|--------v-------   -----.
1048  * |s|e|f|p|ent|ent| ... |ent|
1049  * '--|--------------   -----'   Each PAGE contains a single
1050  *    '------.                   lu_dirpage.
1051  * .---------v-------   -----.
1052  * |s|e|f|p|ent| 0 | ... | 0 |
1053  * '-----------------   -----'
1054  *
1055  * However, on hosts where the native VM page size (PAGE_SIZE) is
1056  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1057  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1058  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1059  * after it in the same PAGE (arrows simplified for brevity, but
1060  * in general e0==s1, e1==s2, etc.):
1061  *
1062  * .--------------------   -----.
1063  * |s0|e0|f0|p|ent|ent| ... |ent|
1064  * |---v----------------   -----|
1065  * |s1|e1|f1|p|ent|ent| ... |ent|
1066  * |---v----------------   -----|  Here, each PAGE contains
1067  *             ...                 multiple lu_dirpages.
1068  * |---v----------------   -----|
1069  * |s'|e'|f'|p|ent|ent| ... |ent|
1070  * '---|----------------   -----'
1071  *     v
1072  * .----------------------------.
1073  * |        next PAGE           |
1074  *
1075  * This structure is transformed into a single logical lu_dirpage as follows:
1076  *
1077  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1078  *   labeled 'next PAGE'.
1079  *
1080  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1081  *   a hash collision with the next page exists.
1082  *
1083  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1084  *   to the first entry of the next lu_dirpage.
1085  */
1086 #if PAGE_SIZE > LU_PAGE_SIZE
1087 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1088 {
1089         int i;
1090
1091         for (i = 0; i < cfs_pgs; i++) {
1092                 struct lu_dirpage       *dp = kmap(pages[i]);
1093                 struct lu_dirpage       *first = dp;
1094                 struct lu_dirent        *end_dirent = NULL;
1095                 struct lu_dirent        *ent;
1096                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1097                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1098
1099                 while (--lu_pgs > 0) {
1100                         ent = lu_dirent_start(dp);
1101                         for (end_dirent = ent; ent != NULL;
1102                              end_dirent = ent, ent = lu_dirent_next(ent));
1103
1104                         /* Advance dp to next lu_dirpage. */
1105                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1106
1107                         /* Check if we've reached the end of the PAGE. */
1108                         if (!((unsigned long)dp & ~PAGE_MASK))
1109                                 break;
1110
1111                         /* Save the hash and flags of this lu_dirpage. */
1112                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1113                         flags = le32_to_cpu(dp->ldp_flags);
1114
1115                         /* Check if lu_dirpage contains no entries. */
1116                         if (end_dirent == NULL)
1117                                 break;
1118
1119                         /* Enlarge the end entry lde_reclen from 0 to
1120                          * first entry of next lu_dirpage. */
1121                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1122                         end_dirent->lde_reclen =
1123                                 cpu_to_le16((char *)(dp->ldp_entries) -
1124                                             (char *)end_dirent);
1125                 }
1126
1127                 first->ldp_hash_end = hash_end;
1128                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1129                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1130
1131                 kunmap(pages[i]);
1132         }
1133         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1134 }
1135 #else
1136 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1137 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1138
1139 /* parameters for readdir page */
1140 struct readpage_param {
1141         struct md_op_data       *rp_mod;
1142         __u64                   rp_off;
1143         int                     rp_hash64;
1144         struct obd_export       *rp_exp;
1145         struct md_callback      *rp_cb;
1146 };
1147
1148 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1149 static inline void delete_from_page_cache(struct page *page)
1150 {
1151         remove_from_page_cache(page);
1152         put_page(page);
1153 }
1154 #endif
1155
1156 /**
1157  * Read pages from server.
1158  *
1159  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1160  * a header lu_dirpage which describes the start/end hash, and whether this
1161  * page is empty (contains no dir entry) or hash collide with next page.
1162  * After client receives reply, several pages will be integrated into dir page
1163  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1164  * lu_dirpage for this integrated page will be adjusted.
1165  **/
1166 static int mdc_read_page_remote(void *data, struct page *page0)
1167 {
1168         struct readpage_param   *rp = data;
1169         struct page             **page_pool;
1170         struct page             *page;
1171         struct lu_dirpage       *dp;
1172         int                     rd_pgs = 0; /* number of pages read actually */
1173         int                     npages;
1174         struct md_op_data       *op_data = rp->rp_mod;
1175         struct ptlrpc_request   *req;
1176         int                     max_pages = op_data->op_max_pages;
1177         struct inode            *inode;
1178         struct lu_fid           *fid;
1179         int                     i;
1180         int                     rc;
1181         ENTRY;
1182
1183         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1184         inode = op_data->op_data;
1185         fid = &op_data->op_fid1;
1186         LASSERT(inode != NULL);
1187
1188         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1189         if (page_pool != NULL) {
1190                 page_pool[0] = page0;
1191         } else {
1192                 page_pool = &page0;
1193                 max_pages = 1;
1194         }
1195
1196         for (npages = 1; npages < max_pages; npages++) {
1197                 page = page_cache_alloc_cold(inode->i_mapping);
1198                 if (page == NULL)
1199                         break;
1200                 page_pool[npages] = page;
1201         }
1202
1203         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1204         if (rc < 0) {
1205                 /* page0 is special, which was added into page cache early */
1206                 delete_from_page_cache(page0);
1207         } else {
1208                 int lu_pgs;
1209
1210                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1211                             PAGE_SIZE - 1) >> PAGE_SHIFT;
1212                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1213                                                         LU_PAGE_SHIFT;
1214                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1215
1216                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1217
1218                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1219
1220                 SetPageUptodate(page0);
1221         }
1222         unlock_page(page0);
1223
1224         ptlrpc_req_finished(req);
1225         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1226         for (i = 1; i < npages; i++) {
1227                 unsigned long   offset;
1228                 __u64           hash;
1229                 int ret;
1230
1231                 page = page_pool[i];
1232
1233                 if (rc < 0 || i >= rd_pgs) {
1234                         put_page(page);
1235                         continue;
1236                 }
1237
1238                 SetPageUptodate(page);
1239
1240                 dp = kmap(page);
1241                 hash = le64_to_cpu(dp->ldp_hash_start);
1242                 kunmap(page);
1243
1244                 offset = hash_x_index(hash, rp->rp_hash64);
1245
1246                 prefetchw(&page->flags);
1247                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1248                                             GFP_KERNEL);
1249                 if (ret == 0)
1250                         unlock_page(page);
1251                 else
1252                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1253                                " rc = %d\n", offset, ret);
1254                 put_page(page);
1255         }
1256
1257         if (page_pool != &page0)
1258                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1259
1260         RETURN(rc);
1261 }
1262
1263 /**
1264  * Read dir page from cache first, if it can not find it, read it from
1265  * server and add into the cache.
1266  *
1267  * \param[in] exp       MDC export
1268  * \param[in] op_data   client MD stack parameters, transfering parameters
1269  *                      between different layers on client MD stack.
1270  * \param[in] cb_op     callback required for ldlm lock enqueue during
1271  *                      read page
1272  * \param[in] hash_offset the hash offset of the page to be read
1273  * \param[in] ppage     the page to be read
1274  *
1275  * retval               = 0 get the page successfully
1276  *                      errno(<0) get the page failed
1277  */
1278 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1279                          struct md_callback *cb_op, __u64 hash_offset,
1280                          struct page **ppage)
1281 {
1282         struct lookup_intent    it = { .it_op = IT_READDIR };
1283         struct page             *page;
1284         struct inode            *dir = op_data->op_data;
1285         struct address_space    *mapping;
1286         struct lu_dirpage       *dp;
1287         __u64                   start = 0;
1288         __u64                   end = 0;
1289         struct lustre_handle    lockh;
1290         struct ptlrpc_request   *enq_req = NULL;
1291         struct readpage_param   rp_param;
1292         int rc;
1293
1294         ENTRY;
1295
1296         *ppage = NULL;
1297
1298         LASSERT(dir != NULL);
1299         mapping = dir->i_mapping;
1300
1301         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1302                              cb_op->md_blocking_ast, 0);
1303         if (enq_req != NULL)
1304                 ptlrpc_req_finished(enq_req);
1305
1306         if (rc < 0) {
1307                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1308                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1309                 RETURN(rc);
1310         }
1311
1312         rc = 0;
1313         lockh.cookie = it.it_lock_handle;
1314         mdc_set_lock_data(exp, &lockh, dir, NULL);
1315
1316         rp_param.rp_off = hash_offset;
1317         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1318         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1319                                rp_param.rp_hash64);
1320         if (IS_ERR(page)) {
1321                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1322                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1323                        rp_param.rp_off, PTR_ERR(page));
1324                 GOTO(out_unlock, rc = PTR_ERR(page));
1325         } else if (page != NULL) {
1326                 /*
1327                  * XXX nikita: not entirely correct handling of a corner case:
1328                  * suppose hash chain of entries with hash value HASH crosses
1329                  * border between pages P0 and P1. First both P0 and P1 are
1330                  * cached, seekdir() is called for some entry from the P0 part
1331                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1332                  * happens and finds P1, as it starts with matching hash
1333                  * value. Remaining entries from P0 part of the chain are
1334                  * skipped. (Is that really a bug?)
1335                  *
1336                  * Possible solutions: 0. don't cache P1 is such case, handle
1337                  * it as an "overflow" page. 1. invalidate all pages at
1338                  * once. 2. use HASH|1 as an index for P1.
1339                  */
1340                 GOTO(hash_collision, page);
1341         }
1342
1343         rp_param.rp_exp = exp;
1344         rp_param.rp_mod = op_data;
1345         page = read_cache_page(mapping,
1346                                hash_x_index(rp_param.rp_off,
1347                                             rp_param.rp_hash64),
1348                                mdc_read_page_remote, &rp_param);
1349         if (IS_ERR(page)) {
1350                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1351                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1352                        rp_param.rp_off, PTR_ERR(page));
1353                 GOTO(out_unlock, rc = PTR_ERR(page));
1354         }
1355
1356         wait_on_page_locked(page);
1357         (void)kmap(page);
1358         if (!PageUptodate(page)) {
1359                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1360                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1361                        rp_param.rp_off, -5);
1362                 goto fail;
1363         }
1364         if (!PageChecked(page))
1365                 SetPageChecked(page);
1366         if (PageError(page)) {
1367                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1368                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1369                        rp_param.rp_off, -5);
1370                 goto fail;
1371         }
1372
1373 hash_collision:
1374         dp = page_address(page);
1375         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1376                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1377                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1378                 rp_param.rp_off = hash_offset >> 32;
1379         } else {
1380                 start = le64_to_cpu(dp->ldp_hash_start);
1381                 end   = le64_to_cpu(dp->ldp_hash_end);
1382                 rp_param.rp_off = hash_offset;
1383         }
1384         if (end == start) {
1385                 LASSERT(start == rp_param.rp_off);
1386                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1387 #if BITS_PER_LONG == 32
1388                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1389                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1390                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1391 #endif
1392
1393                 /*
1394                  * Fetch whole overflow chain...
1395                  *
1396                  * XXX not yet.
1397                  */
1398                 goto fail;
1399         }
1400         *ppage = page;
1401 out_unlock:
1402         ldlm_lock_decref(&lockh, it.it_lock_mode);
1403         return rc;
1404 fail:
1405         kunmap(page);
1406         mdc_release_page(page, 1);
1407         rc = -EIO;
1408         goto out_unlock;
1409 }
1410
1411
1412 static int mdc_statfs(const struct lu_env *env,
1413                       struct obd_export *exp, struct obd_statfs *osfs,
1414                       __u64 max_age, __u32 flags)
1415 {
1416         struct obd_device     *obd = class_exp2obd(exp);
1417         struct ptlrpc_request *req;
1418         struct obd_statfs     *msfs;
1419         struct obd_import     *imp = NULL;
1420         int                    rc;
1421         ENTRY;
1422
1423         /*
1424          * Since the request might also come from lprocfs, so we need
1425          * sync this with client_disconnect_export Bug15684
1426          */
1427         down_read(&obd->u.cli.cl_sem);
1428         if (obd->u.cli.cl_import)
1429                 imp = class_import_get(obd->u.cli.cl_import);
1430         up_read(&obd->u.cli.cl_sem);
1431         if (!imp)
1432                 RETURN(-ENODEV);
1433
1434         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1435                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1436         if (req == NULL)
1437                 GOTO(output, rc = -ENOMEM);
1438
1439         ptlrpc_request_set_replen(req);
1440
1441         if (flags & OBD_STATFS_NODELAY) {
1442                 /* procfs requests not want stay in wait for avoid deadlock */
1443                 req->rq_no_resend = 1;
1444                 req->rq_no_delay = 1;
1445         }
1446
1447         rc = ptlrpc_queue_wait(req);
1448         if (rc) {
1449                 /* check connection error first */
1450                 if (imp->imp_connect_error)
1451                         rc = imp->imp_connect_error;
1452                 GOTO(out, rc);
1453         }
1454
1455         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1456         if (msfs == NULL)
1457                 GOTO(out, rc = -EPROTO);
1458
1459         *osfs = *msfs;
1460         EXIT;
1461 out:
1462         ptlrpc_req_finished(req);
1463 output:
1464         class_import_put(imp);
1465         return rc;
1466 }
1467
1468 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1469 {
1470         __u32 keylen, vallen;
1471         void *key;
1472         int rc;
1473
1474         if (gf->gf_pathlen > PATH_MAX)
1475                 RETURN(-ENAMETOOLONG);
1476         if (gf->gf_pathlen < 2)
1477                 RETURN(-EOVERFLOW);
1478
1479         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1480         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1481                                 sizeof(struct lu_fid));
1482         OBD_ALLOC(key, keylen);
1483         if (key == NULL)
1484                 RETURN(-ENOMEM);
1485         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1486         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1487         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1488                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1489         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1490                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1491
1492         if (!fid_is_sane(&gf->gf_fid))
1493                 GOTO(out, rc = -EINVAL);
1494
1495         /* Val is struct getinfo_fid2path result plus path */
1496         vallen = sizeof(*gf) + gf->gf_pathlen;
1497
1498         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1499         if (rc != 0 && rc != -EREMOTE)
1500                 GOTO(out, rc);
1501
1502         if (vallen <= sizeof(*gf))
1503                 GOTO(out, rc = -EPROTO);
1504         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1505                 GOTO(out, rc = -EOVERFLOW);
1506
1507         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1508                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1509                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1510                /* only log the last 512 characters of the path */
1511                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1512
1513 out:
1514         OBD_FREE(key, keylen);
1515         return rc;
1516 }
1517
1518 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1519                                 struct hsm_progress_kernel *hpk)
1520 {
1521         struct obd_import               *imp = class_exp2cliimp(exp);
1522         struct hsm_progress_kernel      *req_hpk;
1523         struct ptlrpc_request           *req;
1524         int                              rc;
1525         ENTRY;
1526
1527         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1528                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1529         if (req == NULL)
1530                 GOTO(out, rc = -ENOMEM);
1531
1532         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1533
1534         /* Copy hsm_progress struct */
1535         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1536         if (req_hpk == NULL)
1537                 GOTO(out, rc = -EPROTO);
1538
1539         *req_hpk = *hpk;
1540         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1541
1542         ptlrpc_request_set_replen(req);
1543
1544         mdc_get_mod_rpc_slot(req, NULL);
1545         rc = ptlrpc_queue_wait(req);
1546         mdc_put_mod_rpc_slot(req, NULL);
1547
1548         GOTO(out, rc);
1549 out:
1550         ptlrpc_req_finished(req);
1551         return rc;
1552 }
1553
1554 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1555 {
1556         __u32                   *archive_mask;
1557         struct ptlrpc_request   *req;
1558         int                      rc;
1559         ENTRY;
1560
1561         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1562                                         LUSTRE_MDS_VERSION,
1563                                         MDS_HSM_CT_REGISTER);
1564         if (req == NULL)
1565                 GOTO(out, rc = -ENOMEM);
1566
1567         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1568
1569         /* Copy hsm_progress struct */
1570         archive_mask = req_capsule_client_get(&req->rq_pill,
1571                                               &RMF_MDS_HSM_ARCHIVE);
1572         if (archive_mask == NULL)
1573                 GOTO(out, rc = -EPROTO);
1574
1575         *archive_mask = archives;
1576
1577         ptlrpc_request_set_replen(req);
1578
1579         rc = mdc_queue_wait(req);
1580         GOTO(out, rc);
1581 out:
1582         ptlrpc_req_finished(req);
1583         return rc;
1584 }
1585
1586 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1587                                       struct md_op_data *op_data)
1588 {
1589         struct hsm_current_action       *hca = op_data->op_data;
1590         struct hsm_current_action       *req_hca;
1591         struct ptlrpc_request           *req;
1592         int                              rc;
1593         ENTRY;
1594
1595         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1596                                    &RQF_MDS_HSM_ACTION);
1597         if (req == NULL)
1598                 RETURN(-ENOMEM);
1599
1600         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1601         if (rc) {
1602                 ptlrpc_request_free(req);
1603                 RETURN(rc);
1604         }
1605
1606         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1607                       op_data->op_suppgids[0], 0);
1608
1609         ptlrpc_request_set_replen(req);
1610
1611         rc = mdc_queue_wait(req);
1612         if (rc)
1613                 GOTO(out, rc);
1614
1615         req_hca = req_capsule_server_get(&req->rq_pill,
1616                                          &RMF_MDS_HSM_CURRENT_ACTION);
1617         if (req_hca == NULL)
1618                 GOTO(out, rc = -EPROTO);
1619
1620         *hca = *req_hca;
1621
1622         EXIT;
1623 out:
1624         ptlrpc_req_finished(req);
1625         return rc;
1626 }
1627
1628 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1629 {
1630         struct ptlrpc_request   *req;
1631         int                      rc;
1632         ENTRY;
1633
1634         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1635                                         LUSTRE_MDS_VERSION,
1636                                         MDS_HSM_CT_UNREGISTER);
1637         if (req == NULL)
1638                 GOTO(out, rc = -ENOMEM);
1639
1640         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1641
1642         ptlrpc_request_set_replen(req);
1643
1644         rc = mdc_queue_wait(req);
1645         GOTO(out, rc);
1646 out:
1647         ptlrpc_req_finished(req);
1648         return rc;
1649 }
1650
1651 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1652                                  struct md_op_data *op_data)
1653 {
1654         struct hsm_user_state   *hus = op_data->op_data;
1655         struct hsm_user_state   *req_hus;
1656         struct ptlrpc_request   *req;
1657         int                      rc;
1658         ENTRY;
1659
1660         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1661                                    &RQF_MDS_HSM_STATE_GET);
1662         if (req == NULL)
1663                 RETURN(-ENOMEM);
1664
1665         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1666         if (rc != 0) {
1667                 ptlrpc_request_free(req);
1668                 RETURN(rc);
1669         }
1670
1671         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1672                       op_data->op_suppgids[0], 0);
1673
1674         ptlrpc_request_set_replen(req);
1675
1676         rc = mdc_queue_wait(req);
1677         if (rc)
1678                 GOTO(out, rc);
1679
1680         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1681         if (req_hus == NULL)
1682                 GOTO(out, rc = -EPROTO);
1683
1684         *hus = *req_hus;
1685
1686         EXIT;
1687 out:
1688         ptlrpc_req_finished(req);
1689         return rc;
1690 }
1691
1692 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1693                                  struct md_op_data *op_data)
1694 {
1695         struct hsm_state_set    *hss = op_data->op_data;
1696         struct hsm_state_set    *req_hss;
1697         struct ptlrpc_request   *req;
1698         int                      rc;
1699         ENTRY;
1700
1701         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1702                                    &RQF_MDS_HSM_STATE_SET);
1703         if (req == NULL)
1704                 RETURN(-ENOMEM);
1705
1706         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1707         if (rc) {
1708                 ptlrpc_request_free(req);
1709                 RETURN(rc);
1710         }
1711
1712         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1713                       op_data->op_suppgids[0], 0);
1714
1715         /* Copy states */
1716         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1717         if (req_hss == NULL)
1718                 GOTO(out, rc = -EPROTO);
1719         *req_hss = *hss;
1720
1721         ptlrpc_request_set_replen(req);
1722
1723         mdc_get_mod_rpc_slot(req, NULL);
1724         rc = ptlrpc_queue_wait(req);
1725         mdc_put_mod_rpc_slot(req, NULL);
1726
1727         GOTO(out, rc);
1728 out:
1729         ptlrpc_req_finished(req);
1730         return rc;
1731 }
1732
1733 static int mdc_ioc_hsm_request(struct obd_export *exp,
1734                                struct hsm_user_request *hur)
1735 {
1736         struct obd_import       *imp = class_exp2cliimp(exp);
1737         struct ptlrpc_request   *req;
1738         struct hsm_request      *req_hr;
1739         struct hsm_user_item    *req_hui;
1740         char                    *req_opaque;
1741         int                      rc;
1742         ENTRY;
1743
1744         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1745         if (req == NULL)
1746                 GOTO(out, rc = -ENOMEM);
1747
1748         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1749                              hur->hur_request.hr_itemcount
1750                              * sizeof(struct hsm_user_item));
1751         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1752                              hur->hur_request.hr_data_len);
1753
1754         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1755         if (rc) {
1756                 ptlrpc_request_free(req);
1757                 RETURN(rc);
1758         }
1759
1760         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1761
1762         /* Copy hsm_request struct */
1763         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1764         if (req_hr == NULL)
1765                 GOTO(out, rc = -EPROTO);
1766         *req_hr = hur->hur_request;
1767
1768         /* Copy hsm_user_item structs */
1769         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1770         if (req_hui == NULL)
1771                 GOTO(out, rc = -EPROTO);
1772         memcpy(req_hui, hur->hur_user_item,
1773                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1774
1775         /* Copy opaque field */
1776         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1777         if (req_opaque == NULL)
1778                 GOTO(out, rc = -EPROTO);
1779         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1780
1781         ptlrpc_request_set_replen(req);
1782
1783         mdc_get_mod_rpc_slot(req, NULL);
1784         rc = ptlrpc_queue_wait(req);
1785         mdc_put_mod_rpc_slot(req, NULL);
1786
1787         GOTO(out, rc);
1788
1789 out:
1790         ptlrpc_req_finished(req);
1791         return rc;
1792 }
1793
1794 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1795 {
1796         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1797
1798         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1799
1800         lh->kuc_magic = KUC_MAGIC;
1801         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1802         lh->kuc_flags = flags;
1803         lh->kuc_msgtype = CL_RECORD;
1804         lh->kuc_msglen = len;
1805         return lh;
1806 }
1807
1808 struct changelog_show {
1809         __u64                            cs_startrec;
1810         enum changelog_send_flag         cs_flags;
1811         struct file                     *cs_fp;
1812         char                            *cs_buf;
1813         struct obd_device               *cs_obd;
1814 };
1815
1816 static inline char *cs_obd_name(struct changelog_show *cs)
1817 {
1818         return cs->cs_obd->obd_name;
1819 }
1820
1821 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1822                              struct llog_rec_hdr *hdr, void *data)
1823 {
1824         struct changelog_show           *cs = data;
1825         struct llog_changelog_rec       *rec = (struct llog_changelog_rec *)hdr;
1826         struct kuc_hdr                  *lh;
1827         size_t                           len;
1828         int                              rc;
1829         ENTRY;
1830
1831         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1832                 rc = -EINVAL;
1833                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1834                        cs_obd_name(cs), rec->cr_hdr.lrh_type,
1835                        rec->cr.cr_type, rc);
1836                 RETURN(rc);
1837         }
1838
1839         if (rec->cr.cr_index < cs->cs_startrec) {
1840                 /* Skip entries earlier than what we are interested in */
1841                 CDEBUG(D_HSM, "rec=%llu start=%llu\n",
1842                        rec->cr.cr_index, cs->cs_startrec);
1843                 RETURN(0);
1844         }
1845
1846         CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID" %.*s\n",
1847                rec->cr.cr_index, rec->cr.cr_type,
1848                changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1849                rec->cr.cr_flags & CLF_FLAGMASK,
1850                PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1851                rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1852
1853         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1854
1855         /* Set up the message */
1856         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1857         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1858
1859         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1860         CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1861
1862         RETURN(rc);
1863 }
1864
1865 static int mdc_changelog_send_thread(void *csdata)
1866 {
1867         struct changelog_show   *cs = csdata;
1868         struct llog_ctxt        *ctxt = NULL;
1869         struct llog_handle      *llh = NULL;
1870         struct kuc_hdr          *kuch;
1871         enum llog_flag           flags = LLOG_F_IS_CAT;
1872         int                      rc;
1873
1874         CDEBUG(D_HSM, "changelog to fp=%p start %llu\n",
1875                cs->cs_fp, cs->cs_startrec);
1876
1877         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1878         if (cs->cs_buf == NULL)
1879                 GOTO(out, rc = -ENOMEM);
1880
1881         /* Set up the remote catalog handle */
1882         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1883         if (ctxt == NULL)
1884                 GOTO(out, rc = -ENOENT);
1885         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1886                        LLOG_OPEN_EXISTS);
1887         if (rc) {
1888                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1889                        cs_obd_name(cs), rc);
1890                 GOTO(out, rc);
1891         }
1892
1893         if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1894                 flags |= LLOG_F_EXT_JOBID;
1895
1896         rc = llog_init_handle(NULL, llh, flags, NULL);
1897         if (rc) {
1898                 CERROR("llog_init_handle failed %d\n", rc);
1899                 GOTO(out, rc);
1900         }
1901
1902         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1903
1904         /* Send EOF no matter what our result */
1905         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1906         kuch->kuc_msgtype = CL_EOF;
1907         libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1908
1909 out:
1910         fput(cs->cs_fp);
1911         if (llh)
1912                 llog_cat_close(NULL, llh);
1913         if (ctxt)
1914                 llog_ctxt_put(ctxt);
1915         if (cs->cs_buf)
1916                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1917         OBD_FREE_PTR(cs);
1918         return rc;
1919 }
1920
1921 static int mdc_ioc_changelog_send(struct obd_device *obd,
1922                                   struct ioc_changelog *icc)
1923 {
1924         struct changelog_show *cs;
1925         struct task_struct *task;
1926         int rc;
1927
1928         /* Freed in mdc_changelog_send_thread */
1929         OBD_ALLOC_PTR(cs);
1930         if (!cs)
1931                 return -ENOMEM;
1932
1933         cs->cs_obd = obd;
1934         cs->cs_startrec = icc->icc_recno;
1935         /* matching fput in mdc_changelog_send_thread */
1936         cs->cs_fp = fget(icc->icc_id);
1937         cs->cs_flags = icc->icc_flags;
1938
1939         /*
1940          * New thread because we should return to user app before
1941          * writing into our pipe
1942          */
1943         task = kthread_run(mdc_changelog_send_thread, cs,
1944                            "mdc_clg_send_thread");
1945         if (IS_ERR(task)) {
1946                 rc = PTR_ERR(task);
1947                 CERROR("%s: cannot start changelog thread: rc = %d\n",
1948                        cs_obd_name(cs), rc);
1949                 OBD_FREE_PTR(cs);
1950         } else {
1951                 rc = 0;
1952                 CDEBUG(D_HSM, "%s: started changelog thread\n",
1953                        cs_obd_name(cs));
1954         }
1955
1956         return rc;
1957 }
1958
1959 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1960                                 struct lustre_kernelcomm *lk);
1961
1962 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1963                         struct obd_quotactl *oqctl)
1964 {
1965         struct ptlrpc_request   *req;
1966         struct obd_quotactl     *oqc;
1967         int                      rc;
1968         ENTRY;
1969
1970         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1971                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1972                                         MDS_QUOTACTL);
1973         if (req == NULL)
1974                 RETURN(-ENOMEM);
1975
1976         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1977         *oqc = *oqctl;
1978
1979         ptlrpc_request_set_replen(req);
1980         ptlrpc_at_set_req_timeout(req);
1981         req->rq_no_resend = 1;
1982
1983         rc = ptlrpc_queue_wait(req);
1984         if (rc)
1985                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1986
1987         if (req->rq_repmsg &&
1988             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1989                 *oqctl = *oqc;
1990         } else if (!rc) {
1991                 CERROR ("Can't unpack obd_quotactl\n");
1992                 rc = -EPROTO;
1993         }
1994         ptlrpc_req_finished(req);
1995
1996         RETURN(rc);
1997 }
1998
1999 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2000                                 struct md_op_data *op_data)
2001 {
2002         struct list_head cancels = LIST_HEAD_INIT(cancels);
2003         struct ptlrpc_request   *req;
2004         int                      rc, count;
2005         struct mdc_swap_layouts *msl, *payload;
2006         ENTRY;
2007
2008         msl = op_data->op_data;
2009
2010         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2011          * first thing it will do is to cancel the 2 layout
2012          * locks held by this client.
2013          * So the client must cancel its layout locks on the 2 fids
2014          * with the request RPC to avoid extra RPC round trips.
2015          */
2016         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2017                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2018                                         MDS_INODELOCK_XATTR);
2019         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2020                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2021                                          MDS_INODELOCK_XATTR);
2022
2023         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2024                                    &RQF_MDS_SWAP_LAYOUTS);
2025         if (req == NULL) {
2026                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2027                 RETURN(-ENOMEM);
2028         }
2029
2030         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2031         if (rc) {
2032                 ptlrpc_request_free(req);
2033                 RETURN(rc);
2034         }
2035
2036         mdc_swap_layouts_pack(req, op_data);
2037
2038         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2039         LASSERT(payload);
2040
2041         *payload = *msl;
2042
2043         ptlrpc_request_set_replen(req);
2044
2045         rc = ptlrpc_queue_wait(req);
2046         if (rc)
2047                 GOTO(out, rc);
2048         EXIT;
2049
2050 out:
2051         ptlrpc_req_finished(req);
2052         return rc;
2053 }
2054
2055 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2056                          void *karg, void __user *uarg)
2057 {
2058         struct obd_device *obd = exp->exp_obd;
2059         struct obd_ioctl_data *data = karg;
2060         struct obd_import *imp = obd->u.cli.cl_import;
2061         int rc;
2062         ENTRY;
2063
2064         if (!try_module_get(THIS_MODULE)) {
2065                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2066                        module_name(THIS_MODULE));
2067                 return -EINVAL;
2068         }
2069         switch (cmd) {
2070         case OBD_IOC_CHANGELOG_SEND:
2071                 rc = mdc_ioc_changelog_send(obd, karg);
2072                 GOTO(out, rc);
2073         case OBD_IOC_CHANGELOG_CLEAR: {
2074                 struct ioc_changelog *icc = karg;
2075                 struct changelog_setinfo cs =
2076                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2077                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2078                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2079                                         NULL);
2080                 GOTO(out, rc);
2081         }
2082         case OBD_IOC_FID2PATH:
2083                 rc = mdc_ioc_fid2path(exp, karg);
2084                 GOTO(out, rc);
2085         case LL_IOC_HSM_CT_START:
2086                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2087                 /* ignore if it was already registered on this MDS. */
2088                 if (rc == -EEXIST)
2089                         rc = 0;
2090                 GOTO(out, rc);
2091         case LL_IOC_HSM_PROGRESS:
2092                 rc = mdc_ioc_hsm_progress(exp, karg);
2093                 GOTO(out, rc);
2094         case LL_IOC_HSM_STATE_GET:
2095                 rc = mdc_ioc_hsm_state_get(exp, karg);
2096                 GOTO(out, rc);
2097         case LL_IOC_HSM_STATE_SET:
2098                 rc = mdc_ioc_hsm_state_set(exp, karg);
2099                 GOTO(out, rc);
2100         case LL_IOC_HSM_ACTION:
2101                 rc = mdc_ioc_hsm_current_action(exp, karg);
2102                 GOTO(out, rc);
2103         case LL_IOC_HSM_REQUEST:
2104                 rc = mdc_ioc_hsm_request(exp, karg);
2105                 GOTO(out, rc);
2106         case OBD_IOC_CLIENT_RECOVER:
2107                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2108                 if (rc < 0)
2109                         GOTO(out, rc);
2110                 GOTO(out, rc = 0);
2111         case IOC_OSC_SET_ACTIVE:
2112                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2113                 GOTO(out, rc);
2114         case OBD_IOC_PING_TARGET:
2115                 rc = ptlrpc_obd_ping(obd);
2116                 GOTO(out, rc);
2117         /*
2118          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2119          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2120          * there'd be no LMV layer thus we might be called here. Eventually
2121          * this code should be removed.
2122          * bz20731, LU-592.
2123          */
2124         case IOC_OBD_STATFS: {
2125                 struct obd_statfs stat_buf = {0};
2126
2127                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2128                         GOTO(out, rc = -ENODEV);
2129
2130                 /* copy UUID */
2131                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2132                                  min((int)data->ioc_plen2,
2133                                      (int)sizeof(struct obd_uuid))))
2134                         GOTO(out, rc = -EFAULT);
2135
2136                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2137                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2138                                 0);
2139                 if (rc != 0)
2140                         GOTO(out, rc);
2141
2142                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2143                                      min((int) data->ioc_plen1,
2144                                          (int) sizeof(stat_buf))))
2145                         GOTO(out, rc = -EFAULT);
2146
2147                 GOTO(out, rc = 0);
2148         }
2149         case OBD_IOC_QUOTACTL: {
2150                 struct if_quotactl *qctl = karg;
2151                 struct obd_quotactl *oqctl;
2152
2153                 OBD_ALLOC_PTR(oqctl);
2154                 if (oqctl == NULL)
2155                         GOTO(out, rc = -ENOMEM);
2156
2157                 QCTL_COPY(oqctl, qctl);
2158                 rc = obd_quotactl(exp, oqctl);
2159                 if (rc == 0) {
2160                         QCTL_COPY(qctl, oqctl);
2161                         qctl->qc_valid = QC_MDTIDX;
2162                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2163                 }
2164
2165                 OBD_FREE_PTR(oqctl);
2166                 GOTO(out, rc);
2167         }
2168         case LL_IOC_GET_CONNECT_FLAGS:
2169                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2170                                  sizeof(*exp_connect_flags_ptr(exp))))
2171                         GOTO(out, rc = -EFAULT);
2172
2173                 GOTO(out, rc = 0);
2174         case LL_IOC_LOV_SWAP_LAYOUTS:
2175                 rc = mdc_ioc_swap_layouts(exp, karg);
2176                 GOTO(out, rc);
2177         default:
2178                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2179                 GOTO(out, rc = -ENOTTY);
2180         }
2181 out:
2182         module_put(THIS_MODULE);
2183
2184         return rc;
2185 }
2186
2187 static int mdc_get_info_rpc(struct obd_export *exp,
2188                             u32 keylen, void *key,
2189                             u32 vallen, void *val)
2190 {
2191         struct obd_import      *imp = class_exp2cliimp(exp);
2192         struct ptlrpc_request  *req;
2193         char                   *tmp;
2194         int                     rc = -EINVAL;
2195         ENTRY;
2196
2197         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2198         if (req == NULL)
2199                 RETURN(-ENOMEM);
2200
2201         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2202                              RCL_CLIENT, keylen);
2203         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2204                              RCL_CLIENT, sizeof(vallen));
2205
2206         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2207         if (rc) {
2208                 ptlrpc_request_free(req);
2209                 RETURN(rc);
2210         }
2211
2212         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2213         memcpy(tmp, key, keylen);
2214         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2215         memcpy(tmp, &vallen, sizeof(vallen));
2216
2217         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2218                              RCL_SERVER, vallen);
2219         ptlrpc_request_set_replen(req);
2220
2221         rc = ptlrpc_queue_wait(req);
2222         /* -EREMOTE means the get_info result is partial, and it needs to
2223          * continue on another MDT, see fid2path part in lmv_iocontrol */
2224         if (rc == 0 || rc == -EREMOTE) {
2225                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2226                 memcpy(val, tmp, vallen);
2227                 if (ptlrpc_rep_need_swab(req)) {
2228                         if (KEY_IS(KEY_FID2PATH))
2229                                 lustre_swab_fid2path(val);
2230                 }
2231         }
2232         ptlrpc_req_finished(req);
2233
2234         RETURN(rc);
2235 }
2236
2237 static void lustre_swab_hai(struct hsm_action_item *h)
2238 {
2239         __swab32s(&h->hai_len);
2240         __swab32s(&h->hai_action);
2241         lustre_swab_lu_fid(&h->hai_fid);
2242         lustre_swab_lu_fid(&h->hai_dfid);
2243         __swab64s(&h->hai_cookie);
2244         __swab64s(&h->hai_extent.offset);
2245         __swab64s(&h->hai_extent.length);
2246         __swab64s(&h->hai_gid);
2247 }
2248
2249 static void lustre_swab_hal(struct hsm_action_list *h)
2250 {
2251         struct hsm_action_item  *hai;
2252         __u32                    i;
2253
2254         __swab32s(&h->hal_version);
2255         __swab32s(&h->hal_count);
2256         __swab32s(&h->hal_archive_id);
2257         __swab64s(&h->hal_flags);
2258         hai = hai_first(h);
2259         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2260                 lustre_swab_hai(hai);
2261 }
2262
2263 static void lustre_swab_kuch(struct kuc_hdr *l)
2264 {
2265         __swab16s(&l->kuc_magic);
2266         /* __u8 l->kuc_transport */
2267         __swab16s(&l->kuc_msgtype);
2268         __swab16s(&l->kuc_msglen);
2269 }
2270
2271 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2272                                 struct lustre_kernelcomm *lk)
2273 {
2274         struct obd_import  *imp = class_exp2cliimp(exp);
2275         __u32               archive = lk->lk_data;
2276         int                 rc = 0;
2277
2278         if (lk->lk_group != KUC_GRP_HSM) {
2279                 CERROR("Bad copytool group %d\n", lk->lk_group);
2280                 return -EINVAL;
2281         }
2282
2283         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2284                lk->lk_uid, lk->lk_group, lk->lk_flags);
2285
2286         if (lk->lk_flags & LK_FLG_STOP) {
2287                 /* Unregister with the coordinator */
2288                 rc = mdc_ioc_hsm_ct_unregister(imp);
2289         } else {
2290                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2291         }
2292
2293         return rc;
2294 }
2295
2296 /**
2297  * Send a message to any listening copytools
2298  * @param val KUC message (kuc_hdr + hsm_action_list)
2299  * @param len total length of message
2300  */
2301 static int mdc_hsm_copytool_send(size_t len, void *val)
2302 {
2303         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2304         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2305         int                      rc;
2306         ENTRY;
2307
2308         if (len < sizeof(*lh) + sizeof(*hal)) {
2309                 CERROR("Short HSM message %zu < %zu\n", len,
2310                        sizeof(*lh) + sizeof(*hal));
2311                 RETURN(-EPROTO);
2312         }
2313         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2314                 lustre_swab_kuch(lh);
2315                 lustre_swab_hal(hal);
2316         } else if (lh->kuc_magic != KUC_MAGIC) {
2317                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2318                 RETURN(-EPROTO);
2319         }
2320
2321         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2322                "on %s\n",
2323                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2324                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2325
2326         /* Broadcast to HSM listeners */
2327         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2328
2329         RETURN(rc);
2330 }
2331
2332 /**
2333  * callback function passed to kuc for re-registering each HSM copytool
2334  * running on MDC, after MDT shutdown/recovery.
2335  * @param data copytool registration data
2336  * @param cb_arg callback argument (obd_import)
2337  */
2338 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2339 {
2340         struct kkuc_ct_data     *kcd = data;
2341         struct obd_import       *imp = (struct obd_import *)cb_arg;
2342         int                      rc;
2343
2344         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2345                 return -EPROTO;
2346
2347         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2348                 return 0;
2349
2350         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2351                imp->imp_obd->obd_name, kcd->kcd_archive);
2352         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2353
2354         /* ignore error if the copytool is already registered */
2355         return (rc == -EEXIST) ? 0 : rc;
2356 }
2357
2358 /**
2359  * Re-establish all kuc contexts with MDT
2360  * after MDT shutdown/recovery.
2361  */
2362 static int mdc_kuc_reregister(struct obd_import *imp)
2363 {
2364         /* re-register HSM agents */
2365         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2366                                          (void *)imp);
2367 }
2368
2369 static int mdc_set_info_async(const struct lu_env *env,
2370                               struct obd_export *exp,
2371                               u32 keylen, void *key,
2372                               u32 vallen, void *val,
2373                               struct ptlrpc_request_set *set)
2374 {
2375         struct obd_import       *imp = class_exp2cliimp(exp);
2376         int                      rc;
2377         ENTRY;
2378
2379         if (KEY_IS(KEY_READ_ONLY)) {
2380                 if (vallen != sizeof(int))
2381                         RETURN(-EINVAL);
2382
2383                 spin_lock(&imp->imp_lock);
2384                 if (*((int *)val)) {
2385                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2386                         imp->imp_connect_data.ocd_connect_flags |=
2387                                                         OBD_CONNECT_RDONLY;
2388                 } else {
2389                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2390                         imp->imp_connect_data.ocd_connect_flags &=
2391                                                         ~OBD_CONNECT_RDONLY;
2392                 }
2393                 spin_unlock(&imp->imp_lock);
2394
2395                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2396                                        keylen, key, vallen, val, set);
2397                 RETURN(rc);
2398         }
2399         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2400                 sptlrpc_conf_client_adapt(exp->exp_obd);
2401                 RETURN(0);
2402         }
2403         if (KEY_IS(KEY_FLUSH_CTX)) {
2404                 sptlrpc_import_flush_my_ctx(imp);
2405                 RETURN(0);
2406         }
2407         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2408                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2409                                        keylen, key, vallen, val, set);
2410                 RETURN(rc);
2411         }
2412         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2413                 rc = mdc_hsm_copytool_send(vallen, val);
2414                 RETURN(rc);
2415         }
2416
2417         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2418                 __u32 *default_easize = val;
2419
2420                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2421                 RETURN(0);
2422         }
2423
2424         CERROR("Unknown key %s\n", (char *)key);
2425         RETURN(-EINVAL);
2426 }
2427
2428 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2429                         __u32 keylen, void *key, __u32 *vallen, void *val)
2430 {
2431         int rc = -EINVAL;
2432
2433         if (KEY_IS(KEY_MAX_EASIZE)) {
2434                 __u32 mdsize, *max_easize;
2435
2436                 if (*vallen != sizeof(int))
2437                         RETURN(-EINVAL);
2438                 mdsize = *(__u32 *)val;
2439                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2440                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2441                 max_easize = val;
2442                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2443                 RETURN(0);
2444         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2445                 __u32 *default_easize;
2446
2447                 if (*vallen != sizeof(int))
2448                         RETURN(-EINVAL);
2449                 default_easize = val;
2450                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2451                 RETURN(0);
2452         } else if (KEY_IS(KEY_CONN_DATA)) {
2453                 struct obd_import *imp = class_exp2cliimp(exp);
2454                 struct obd_connect_data *data = val;
2455
2456                 if (*vallen != sizeof(*data))
2457                         RETURN(-EINVAL);
2458
2459                 *data = imp->imp_connect_data;
2460                 RETURN(0);
2461         } else if (KEY_IS(KEY_TGT_COUNT)) {
2462                 *((__u32 *)val) = 1;
2463                 RETURN(0);
2464         }
2465
2466         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2467
2468         RETURN(rc);
2469 }
2470
2471 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2472                      struct ptlrpc_request **request)
2473 {
2474         struct ptlrpc_request *req;
2475         int                    rc;
2476         ENTRY;
2477
2478         *request = NULL;
2479         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2480         if (req == NULL)
2481                 RETURN(-ENOMEM);
2482
2483         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2484         if (rc) {
2485                 ptlrpc_request_free(req);
2486                 RETURN(rc);
2487         }
2488
2489         mdc_pack_body(req, fid, 0, 0, -1, 0);
2490
2491         ptlrpc_request_set_replen(req);
2492
2493         rc = ptlrpc_queue_wait(req);
2494         if (rc)
2495                 ptlrpc_req_finished(req);
2496         else
2497                 *request = req;
2498         RETURN(rc);
2499 }
2500
2501 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2502                             enum obd_import_event event)
2503 {
2504         int rc = 0;
2505
2506         LASSERT(imp->imp_obd == obd);
2507
2508         switch (event) {
2509
2510         case IMP_EVENT_INACTIVE: {
2511                 struct client_obd *cli = &obd->u.cli;
2512                 /*
2513                  * Flush current sequence to make client obtain new one
2514                  * from server in case of disconnect/reconnect.
2515                  */
2516                 if (cli->cl_seq != NULL)
2517                         seq_client_flush(cli->cl_seq);
2518
2519                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2520                 break;
2521         }
2522         case IMP_EVENT_INVALIDATE: {
2523                 struct ldlm_namespace *ns = obd->obd_namespace;
2524
2525                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2526
2527                 break;
2528         }
2529         case IMP_EVENT_ACTIVE:
2530                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2531                 /* redo the kuc registration after reconnecting */
2532                 if (rc == 0)
2533                         rc = mdc_kuc_reregister(imp);
2534                 break;
2535         case IMP_EVENT_OCD:
2536                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2537                 break;
2538         case IMP_EVENT_DISCON:
2539         case IMP_EVENT_DEACTIVATE:
2540         case IMP_EVENT_ACTIVATE:
2541                 break;
2542         default:
2543                 CERROR("Unknown import event %x\n", event);
2544                 LBUG();
2545         }
2546         RETURN(rc);
2547 }
2548
2549 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2550                   struct lu_fid *fid, struct md_op_data *op_data)
2551 {
2552         struct client_obd *cli = &exp->exp_obd->u.cli;
2553         struct lu_client_seq *seq = cli->cl_seq;
2554         ENTRY;
2555         RETURN(seq_client_alloc_fid(env, seq, fid));
2556 }
2557
2558 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2559 {
2560         struct client_obd *cli = &exp->exp_obd->u.cli;
2561         return &cli->cl_target_uuid;
2562 }
2563
2564 /**
2565  * Determine whether the lock can be canceled before replaying it during
2566  * recovery, non zero value will be return if the lock can be canceled,
2567  * or zero returned for not
2568  */
2569 static int mdc_cancel_weight(struct ldlm_lock *lock)
2570 {
2571         if (lock->l_resource->lr_type != LDLM_IBITS)
2572                 RETURN(0);
2573
2574         /* FIXME: if we ever get into a situation where there are too many
2575          * opened files with open locks on a single node, then we really
2576          * should replay these open locks to reget it */
2577         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2578                 RETURN(0);
2579
2580         RETURN(1);
2581 }
2582
2583 static int mdc_resource_inode_free(struct ldlm_resource *res)
2584 {
2585         if (res->lr_lvb_inode)
2586                 res->lr_lvb_inode = NULL;
2587
2588         return 0;
2589 }
2590
2591 static struct ldlm_valblock_ops inode_lvbo = {
2592         .lvbo_free = mdc_resource_inode_free
2593 };
2594
2595 static int mdc_llog_init(struct obd_device *obd)
2596 {
2597         struct obd_llog_group   *olg = &obd->obd_olg;
2598         struct llog_ctxt        *ctxt;
2599         int                      rc;
2600
2601         ENTRY;
2602
2603         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2604                         &llog_client_ops);
2605         if (rc < 0)
2606                 RETURN(rc);
2607
2608         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2609         llog_initiator_connect(ctxt);
2610         llog_ctxt_put(ctxt);
2611
2612         RETURN(0);
2613 }
2614
2615 static void mdc_llog_finish(struct obd_device *obd)
2616 {
2617         struct llog_ctxt *ctxt;
2618
2619         ENTRY;
2620
2621         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2622         if (ctxt != NULL)
2623                 llog_cleanup(NULL, ctxt);
2624
2625         EXIT;
2626 }
2627
2628 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2629 {
2630         int                             rc;
2631         ENTRY;
2632
2633         rc = ptlrpcd_addref();
2634         if (rc < 0)
2635                 RETURN(rc);
2636
2637         rc = client_obd_setup(obd, cfg);
2638         if (rc)
2639                 GOTO(err_ptlrpcd_decref, rc);
2640 #ifdef CONFIG_PROC_FS
2641         obd->obd_vars = lprocfs_mdc_obd_vars;
2642         lprocfs_obd_setup(obd);
2643         lprocfs_alloc_md_stats(obd, 0);
2644 #endif
2645         sptlrpc_lprocfs_cliobd_attach(obd);
2646         ptlrpc_lprocfs_register_obd(obd);
2647
2648         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2649
2650         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2651
2652         rc = mdc_llog_init(obd);
2653         if (rc) {
2654                 mdc_cleanup(obd);
2655                 CERROR("failed to setup llogging subsystems\n");
2656                 RETURN(rc);
2657         }
2658
2659         RETURN(rc);
2660
2661 err_ptlrpcd_decref:
2662         ptlrpcd_decref();
2663         RETURN(rc);
2664 }
2665
2666 /* Initialize the default and maximum LOV EA sizes.  This allows
2667  * us to make MDS RPCs with large enough reply buffers to hold a default
2668  * sized EA without having to calculate this (via a call into the
2669  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2670  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2671  * a large number of stripes is possible.  If a larger reply buffer is
2672  * required it will be reallocated in the ptlrpc layer due to overflow.
2673  */
2674 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2675                             __u32 def_easize)
2676 {
2677         struct obd_device *obd = exp->exp_obd;
2678         struct client_obd *cli = &obd->u.cli;
2679         ENTRY;
2680
2681         if (cli->cl_max_mds_easize < easize)
2682                 cli->cl_max_mds_easize = easize;
2683
2684         if (cli->cl_default_mds_easize < def_easize)
2685                 cli->cl_default_mds_easize = def_easize;
2686
2687         RETURN(0);
2688 }
2689
2690 static int mdc_precleanup(struct obd_device *obd)
2691 {
2692         ENTRY;
2693
2694         /* Failsafe, ok if racy */
2695         if (obd->obd_type->typ_refcnt <= 1)
2696                 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2697
2698         obd_cleanup_client_import(obd);
2699         ptlrpc_lprocfs_unregister_obd(obd);
2700         lprocfs_obd_cleanup(obd);
2701         lprocfs_free_md_stats(obd);
2702         mdc_llog_finish(obd);
2703         RETURN(0);
2704 }
2705
2706 static int mdc_cleanup(struct obd_device *obd)
2707 {
2708         ptlrpcd_decref();
2709
2710         return client_obd_cleanup(obd);
2711 }
2712
2713 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2714 {
2715         struct lustre_cfg *lcfg = buf;
2716         int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2717         return (rc > 0 ? 0: rc);
2718 }
2719
2720 static struct obd_ops mdc_obd_ops = {
2721         .o_owner            = THIS_MODULE,
2722         .o_setup            = mdc_setup,
2723         .o_precleanup       = mdc_precleanup,
2724         .o_cleanup          = mdc_cleanup,
2725         .o_add_conn         = client_import_add_conn,
2726         .o_del_conn         = client_import_del_conn,
2727         .o_connect          = client_connect_import,
2728         .o_disconnect       = client_disconnect_export,
2729         .o_iocontrol        = mdc_iocontrol,
2730         .o_set_info_async   = mdc_set_info_async,
2731         .o_statfs           = mdc_statfs,
2732         .o_fid_init         = client_fid_init,
2733         .o_fid_fini         = client_fid_fini,
2734         .o_fid_alloc        = mdc_fid_alloc,
2735         .o_import_event     = mdc_import_event,
2736         .o_get_info         = mdc_get_info,
2737         .o_process_config   = mdc_process_config,
2738         .o_get_uuid         = mdc_get_uuid,
2739         .o_quotactl         = mdc_quotactl,
2740 };
2741
2742 static struct md_ops mdc_md_ops = {
2743         .m_get_root         = mdc_get_root,
2744         .m_null_inode       = mdc_null_inode,
2745         .m_close            = mdc_close,
2746         .m_create           = mdc_create,
2747         .m_enqueue          = mdc_enqueue,
2748         .m_getattr          = mdc_getattr,
2749         .m_getattr_name     = mdc_getattr_name,
2750         .m_intent_lock      = mdc_intent_lock,
2751         .m_link             = mdc_link,
2752         .m_rename           = mdc_rename,
2753         .m_setattr          = mdc_setattr,
2754         .m_setxattr         = mdc_setxattr,
2755         .m_getxattr         = mdc_getxattr,
2756         .m_fsync                = mdc_fsync,
2757         .m_read_page            = mdc_read_page,
2758         .m_unlink           = mdc_unlink,
2759         .m_cancel_unused    = mdc_cancel_unused,
2760         .m_init_ea_size     = mdc_init_ea_size,
2761         .m_set_lock_data    = mdc_set_lock_data,
2762         .m_lock_match       = mdc_lock_match,
2763         .m_get_lustre_md    = mdc_get_lustre_md,
2764         .m_free_lustre_md   = mdc_free_lustre_md,
2765         .m_set_open_replay_data = mdc_set_open_replay_data,
2766         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2767         .m_intent_getattr_async = mdc_intent_getattr_async,
2768         .m_revalidate_lock      = mdc_revalidate_lock
2769 };
2770
2771 static int __init mdc_init(void)
2772 {
2773         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2774                                    LUSTRE_MDC_NAME, NULL);
2775 }
2776
2777 static void __exit mdc_exit(void)
2778 {
2779         class_unregister_type(LUSTRE_MDC_NAME);
2780 }
2781
2782 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2783 MODULE_DESCRIPTION("Lustre Metadata Client");
2784 MODULE_VERSION(LUSTRE_VERSION_STRING);
2785 MODULE_LICENSE("GPL");
2786
2787 module_init(mdc_init);
2788 module_exit(mdc_exit);