lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #include <linux/delay.h>
  42 #include <linux/uidgid.h>
  43
  44 #include <lustre_errno.h>
  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 <uapi/linux/lustre/lustre_ioctl.h>
  52 #include <lustre_kernelcomm.h>
  53 #include <lustre_lmv.h>
  54 #include <lustre_log.h>
  55 #include <lustre_swab.h>
  56 #include <obd_class.h>
  57 #include <lustre_osc.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 void mdc_reset_acl_req(struct ptlrpc_request *req)
 194 {
 195         spin_lock(&req->rq_early_free_lock);
 196         sptlrpc_cli_free_repbuf(req);
 197         req->rq_repbuf = NULL;
 198         req->rq_repbuf_len = 0;
 199         req->rq_repdata = NULL;
 200         req->rq_reqdata_len = 0;
 201         spin_unlock(&req->rq_early_free_lock);
 202 }
 203
 204 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 205                        struct ptlrpc_request **request)
 206 {
 207         struct ptlrpc_request *req;
 208         struct obd_import *imp = class_exp2cliimp(exp);
 209         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 210         int rc;
 211         ENTRY;
 212
 213         /* Single MDS without an LMV case */
 214         if (op_data->op_flags & MF_GET_MDT_IDX) {
 215                 op_data->op_mds = 0;
 216                 RETURN(0);
 217         }
 218
 219         *request = NULL;
 220         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 221         if (req == NULL)
 222                 RETURN(-ENOMEM);
 223
 224         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 225         if (rc) {
 226                 ptlrpc_request_free(req);
 227                 RETURN(rc);
 228         }
 229
 230 again:
 231         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 232                       op_data->op_mode, -1, 0);
 233         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 234         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 235                              op_data->op_mode);
 236         ptlrpc_request_set_replen(req);
 237
 238         rc = mdc_getattr_common(exp, req);
 239         if (rc) {
 240                 if (rc == -ERANGE) {
 241                         acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
 242                                           XATTR_SIZE_MAX);
 243                         mdc_reset_acl_req(req);
 244                         goto again;
 245                 }
 246
 247                 ptlrpc_req_finished(req);
 248         } else {
 249                 *request = req;
 250         }
 251
 252         RETURN(rc);
 253 }
 254
 255 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 256                             struct ptlrpc_request **request)
 257 {
 258         struct ptlrpc_request *req;
 259         struct obd_import *imp = class_exp2cliimp(exp);
 260         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 261         int rc;
 262         ENTRY;
 263
 264         *request = NULL;
 265         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 266         if (req == NULL)
 267                 RETURN(-ENOMEM);
 268
 269         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 270                              op_data->op_namelen + 1);
 271
 272         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 273         if (rc) {
 274                 ptlrpc_request_free(req);
 275                 RETURN(rc);
 276         }
 277
 278         if (op_data->op_name) {
 279                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 280                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 281                                 op_data->op_namelen);
 282                 memcpy(name, op_data->op_name, op_data->op_namelen);
 283         }
 284
 285 again:
 286         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 287                       op_data->op_mode, op_data->op_suppgids[0], 0);
 288         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 289                              op_data->op_mode);
 290         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 291         ptlrpc_request_set_replen(req);
 292
 293         rc = mdc_getattr_common(exp, req);
 294         if (rc) {
 295                 if (rc == -ERANGE) {
 296                         acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
 297                                           XATTR_SIZE_MAX);
 298                         mdc_reset_acl_req(req);
 299                         goto again;
 300                 }
 301
 302                 ptlrpc_req_finished(req);
 303         } else {
 304                 *request = req;
 305         }
 306
 307         RETURN(rc);
 308 }
 309
 310 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 311                             const struct lu_fid *fid, int opcode, u64 valid,
 312                             const char *xattr_name, const char *input,
 313                             int input_size, int output_size, int flags,
 314                             __u32 suppgid, struct ptlrpc_request **request)
 315 {
 316         struct ptlrpc_request *req;
 317         int   xattr_namelen = 0;
 318         char *tmp;
 319         int   rc;
 320         ENTRY;
 321
 322         *request = NULL;
 323         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 324         if (req == NULL)
 325                 RETURN(-ENOMEM);
 326
 327         if (xattr_name) {
 328                 xattr_namelen = strlen(xattr_name) + 1;
 329                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 330                                      xattr_namelen);
 331         }
 332         if (input_size)
 333                 LASSERT(input);
 334         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 335                              input_size);
 336
 337         /* get SELinux policy info if any */
 338         rc = sptlrpc_get_sepol(req);
 339         if (rc < 0) {
 340                 ptlrpc_request_free(req);
 341                 RETURN(rc);
 342         }
 343         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 344                              strlen(req->rq_sepol) ?
 345                              strlen(req->rq_sepol) + 1 : 0);
 346
 347         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 348         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 349             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 350                 LIST_HEAD(cancels);
 351                 int count;
 352
 353                 /* Without that packing would fail */
 354                 if (input_size == 0)
 355                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 356                                              RCL_CLIENT, 0);
 357
 358                 count = mdc_resource_get_unused(exp, fid,
 359                                                 &cancels, LCK_EX,
 360                                                 MDS_INODELOCK_XATTR);
 361
 362                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 363                 if (rc) {
 364                         ptlrpc_request_free(req);
 365                         RETURN(rc);
 366                 }
 367         } else {
 368                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 369                 if (rc) {
 370                         ptlrpc_request_free(req);
 371                         RETURN(rc);
 372                 }
 373         }
 374
 375         if (opcode == MDS_REINT) {
 376                 struct mdt_rec_setxattr *rec;
 377
 378                 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
 379                              sizeof(struct mdt_rec_reint));
 380                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 381                 rec->sx_opcode = REINT_SETXATTR;
 382                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 383                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 384                 rec->sx_cap    = cfs_curproc_cap_pack();
 385                 rec->sx_suppgid1 = suppgid;
 386                 rec->sx_suppgid2 = -1;
 387                 rec->sx_fid    = *fid;
 388                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 389                 rec->sx_time   = ktime_get_real_seconds();
 390                 rec->sx_size   = output_size;
 391                 rec->sx_flags  = flags;
 392         } else {
 393                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 394         }
 395
 396         if (xattr_name) {
 397                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 398                 memcpy(tmp, xattr_name, xattr_namelen);
 399         }
 400         if (input_size) {
 401                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 402                 memcpy(tmp, input, input_size);
 403         }
 404
 405         mdc_file_sepol_pack(req);
 406
 407         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 408                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 409                                      RCL_SERVER, output_size);
 410         ptlrpc_request_set_replen(req);
 411
 412         /* make rpc */
 413         if (opcode == MDS_REINT)
 414                 ptlrpc_get_mod_rpc_slot(req);
 415
 416         rc = ptlrpc_queue_wait(req);
 417
 418         if (opcode == MDS_REINT)
 419                 ptlrpc_put_mod_rpc_slot(req);
 420
 421         if (rc)
 422                 ptlrpc_req_finished(req);
 423         else
 424                 *request = req;
 425         RETURN(rc);
 426 }
 427
 428 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 429                         u64 obd_md_valid, const char *name,
 430                         const void *value, size_t value_size,
 431                         unsigned int xattr_flags, u32 suppgid,
 432                         struct ptlrpc_request **req)
 433 {
 434         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 435                 obd_md_valid == OBD_MD_FLXATTRRM);
 436
 437         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 438                                 fid, MDS_REINT, obd_md_valid, name,
 439                                 value, value_size, 0, xattr_flags, suppgid,
 440                                 req);
 441 }
 442
 443 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 444                         u64 obd_md_valid, const char *name, size_t buf_size,
 445                         struct ptlrpc_request **req)
 446 {
 447         struct mdt_body *body;
 448         int rc;
 449
 450         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 451                 obd_md_valid == OBD_MD_FLXATTRLS);
 452
 453         /* The below message is checked in sanity-selinux.sh test_20d */
 454         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 455                exp->exp_obd->obd_name, name, PFID(fid));
 456         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 457                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 458                               req);
 459         if (rc < 0)
 460                 GOTO(out, rc);
 461
 462         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 463         if (body == NULL)
 464                 GOTO(out, rc = -EPROTO);
 465
 466         /* only detect the xattr size */
 467         if (buf_size == 0) {
 468                 /* LU-11109: Older MDTs do not distinguish
 469                  * between nonexistent xattrs and zero length
 470                  * values in this case. Newer MDTs will return
 471                  * -ENODATA or set OBD_MD_FLXATTR. */
 472                 GOTO(out, rc = body->mbo_eadatasize);
 473         }
 474
 475         if (body->mbo_eadatasize == 0) {
 476                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 477                  * success so that we can distinguish between
 478                  * zero length value and nonexistent xattr.
 479                  *
 480                  * If OBD_MD_FLXATTR is not set then we keep
 481                  * the old behavior and return -ENODATA for
 482                  * getxattr() when mbo_eadatasize is 0. But
 483                  * -ENODATA only makes sense for getxattr()
 484                  * and not for listxattr(). */
 485                 if (body->mbo_valid & OBD_MD_FLXATTR)
 486                         GOTO(out, rc = 0);
 487                 else if (obd_md_valid == OBD_MD_FLXATTR)
 488                         GOTO(out, rc = -ENODATA);
 489                 else
 490                         GOTO(out, rc = 0);
 491         }
 492
 493         GOTO(out, rc = body->mbo_eadatasize);
 494 out:
 495         if (rc < 0) {
 496                 ptlrpc_req_finished(*req);
 497                 *req = NULL;
 498         }
 499
 500         return rc;
 501 }
 502
 503 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 504 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 505 {
 506         struct req_capsule     *pill = &req->rq_pill;
 507         struct mdt_body        *body = md->body;
 508         struct posix_acl       *acl;
 509         void                   *buf;
 510         int                     rc;
 511         ENTRY;
 512
 513         if (!body->mbo_aclsize)
 514                 RETURN(0);
 515
 516         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 517
 518         if (!buf)
 519                 RETURN(-EPROTO);
 520
 521         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 522         if (acl == NULL)
 523                 RETURN(0);
 524         if (IS_ERR(acl)) {
 525                 rc = PTR_ERR(acl);
 526                 CERROR("convert xattr to acl: %d\n", rc);
 527                 RETURN(rc);
 528         }
 529
 530         rc = posix_acl_valid(&init_user_ns, acl);
 531         if (rc) {
 532                 CERROR("validate acl: %d\n", rc);
 533                 posix_acl_release(acl);
 534                 RETURN(rc);
 535         }
 536
 537         md->posix_acl = acl;
 538         RETURN(0);
 539 }
 540 #else
 541 #define mdc_unpack_acl(req, md) 0
 542 #endif
 543
 544 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 545                       struct obd_export *dt_exp, struct obd_export *md_exp,
 546                       struct lustre_md *md)
 547 {
 548         struct req_capsule *pill = &req->rq_pill;
 549         int rc;
 550         ENTRY;
 551
 552         LASSERT(md);
 553         memset(md, 0, sizeof(*md));
 554
 555         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 556         LASSERT(md->body != NULL);
 557
 558         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 559                 if (!S_ISREG(md->body->mbo_mode)) {
 560                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 561                                "regular file, but is not\n");
 562                         GOTO(out, rc = -EPROTO);
 563                 }
 564
 565                 if (md->body->mbo_eadatasize == 0) {
 566                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 567                                "but eadatasize 0\n");
 568                         GOTO(out, rc = -EPROTO);
 569                 }
 570
 571                 md->layout.lb_len = md->body->mbo_eadatasize;
 572                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 573                                                         &RMF_MDT_MD,
 574                                                         md->layout.lb_len);
 575                 if (md->layout.lb_buf == NULL)
 576                         GOTO(out, rc = -EPROTO);
 577         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 578                 const union lmv_mds_md *lmv;
 579                 size_t lmv_size;
 580
 581                 if (!S_ISDIR(md->body->mbo_mode)) {
 582                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 583                                "directory, but is not\n");
 584                         GOTO(out, rc = -EPROTO);
 585                 }
 586
 587                 if (md->body->mbo_valid & OBD_MD_MEA) {
 588                         lmv_size = md->body->mbo_eadatasize;
 589                         if (lmv_size == 0) {
 590                                 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 591                                        "but eadatasize 0\n");
 592                                 RETURN(-EPROTO);
 593                         }
 594
 595                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 596                                                            lmv_size);
 597                         if (lmv == NULL)
 598                                 GOTO(out, rc = -EPROTO);
 599
 600                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 601                         if (rc < 0)
 602                                 GOTO(out, rc);
 603
 604                         if (rc < (int)sizeof(*md->lmv)) {
 605                                 struct lmv_foreign_md *lfm = md->lfm;
 606
 607                                 /* short (< sizeof(struct lmv_stripe_md))
 608                                  * foreign LMV case
 609                                  */
 610                                 if (lfm->lfm_magic != LMV_MAGIC_FOREIGN) {
 611                                         CDEBUG(D_INFO,
 612                                                "lmv size too small: %d < %d\n",
 613                                                rc, (int)sizeof(*md->lmv));
 614                                         GOTO(out, rc = -EPROTO);
 615                                 }
 616                         }
 617                 }
 618
 619                 /* since 2.12.58 intent_getattr fetches default LMV */
 620                 if (md->body->mbo_valid & OBD_MD_DEFAULT_MEA) {
 621                         lmv_size = sizeof(struct lmv_user_md);
 622                         lmv = req_capsule_server_sized_get(pill,
 623                                                            &RMF_DEFAULT_MDT_MD,
 624                                                            lmv_size);
 625                         if (!lmv)
 626                                 GOTO(out, rc = -EPROTO);
 627
 628                         rc = md_unpackmd(md_exp, &md->default_lmv, lmv,
 629                                          lmv_size);
 630                         if (rc < 0)
 631                                 GOTO(out, rc);
 632
 633                         if (rc < (int)sizeof(*md->default_lmv)) {
 634                                 CDEBUG(D_INFO,
 635                                        "default lmv size too small: %d < %d\n",
 636                                         rc, (int)sizeof(*md->default_lmv));
 637                                 GOTO(out, rc = -EPROTO);
 638                         }
 639                 }
 640         }
 641         rc = 0;
 642
 643         if (md->body->mbo_valid & OBD_MD_FLACL) {
 644                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 645                  * only when aclsize != 0 there's an actual segment for ACL
 646                  * in reply buffer.
 647                  */
 648                 if (md->body->mbo_aclsize) {
 649                         rc = mdc_unpack_acl(req, md);
 650                         if (rc)
 651                                 GOTO(out, rc);
 652 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 653                 } else {
 654                         md->posix_acl = NULL;
 655 #endif
 656                 }
 657         }
 658
 659         EXIT;
 660 out:
 661         if (rc) {
 662 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 663                 posix_acl_release(md->posix_acl);
 664 #endif
 665         }
 666         return rc;
 667 }
 668
 669 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 670 {
 671         ENTRY;
 672         RETURN(0);
 673 }
 674
 675 void mdc_replay_open(struct ptlrpc_request *req)
 676 {
 677         struct md_open_data *mod = req->rq_cb_data;
 678         struct ptlrpc_request *close_req;
 679         struct obd_client_handle *och;
 680         struct lustre_handle old_open_handle = { };
 681         struct mdt_body *body;
 682         ENTRY;
 683
 684         if (mod == NULL) {
 685                 DEBUG_REQ(D_ERROR, req,
 686                           "cannot properly replay without open data");
 687                 EXIT;
 688                 return;
 689         }
 690
 691         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 692         LASSERT(body != NULL);
 693
 694         spin_lock(&req->rq_lock);
 695         och = mod->mod_och;
 696         if (och && och->och_open_handle.cookie)
 697                 req->rq_early_free_repbuf = 1;
 698         else
 699                 req->rq_early_free_repbuf = 0;
 700         spin_unlock(&req->rq_lock);
 701
 702         if (req->rq_early_free_repbuf) {
 703                 struct lustre_handle *file_open_handle;
 704
 705                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 706
 707                 file_open_handle = &och->och_open_handle;
 708                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 709                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 710                 old_open_handle = *file_open_handle;
 711                 *file_open_handle = body->mbo_open_handle;
 712         }
 713
 714         close_req = mod->mod_close_req;
 715         if (close_req) {
 716                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 717                 struct mdt_ioepoch *epoch;
 718
 719                 LASSERT(opc == MDS_CLOSE);
 720                 epoch = req_capsule_client_get(&close_req->rq_pill,
 721                                                &RMF_MDT_EPOCH);
 722                 LASSERT(epoch);
 723
 724                 if (req->rq_early_free_repbuf)
 725                         LASSERT(old_open_handle.cookie ==
 726                                 epoch->mio_open_handle.cookie);
 727
 728                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 729                 epoch->mio_open_handle = body->mbo_open_handle;
 730         }
 731         EXIT;
 732 }
 733
 734 void mdc_commit_open(struct ptlrpc_request *req)
 735 {
 736         struct md_open_data *mod = req->rq_cb_data;
 737         if (mod == NULL)
 738                 return;
 739
 740         /**
 741          * No need to touch md_open_data::mod_och, it holds a reference on
 742          * \var mod and will zero references to each other, \var mod will be
 743          * freed after that when md_open_data::mod_och will put the reference.
 744          */
 745
 746         /**
 747          * Do not let open request to disappear as it still may be needed
 748          * for close rpc to happen (it may happen on evict only, otherwise
 749          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 750          * called), just mark this rpc as committed to distinguish these 2
 751          * cases, see mdc_close() for details. The open request reference will
 752          * be put along with freeing \var mod.
 753          */
 754         ptlrpc_request_addref(req);
 755         spin_lock(&req->rq_lock);
 756         req->rq_committed = 1;
 757         spin_unlock(&req->rq_lock);
 758         req->rq_cb_data = NULL;
 759         obd_mod_put(mod);
 760 }
 761
 762 int mdc_set_open_replay_data(struct obd_export *exp,
 763                              struct obd_client_handle *och,
 764                              struct lookup_intent *it)
 765 {
 766         struct md_open_data     *mod;
 767         struct mdt_rec_create   *rec;
 768         struct mdt_body         *body;
 769         struct ptlrpc_request   *open_req = it->it_request;
 770         struct obd_import       *imp = open_req->rq_import;
 771         ENTRY;
 772
 773         if (!open_req->rq_replay)
 774                 RETURN(0);
 775
 776         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 777         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 778         LASSERT(rec != NULL);
 779         /* Incoming message in my byte order (it's been swabbed). */
 780         /* Outgoing messages always in my byte order. */
 781         LASSERT(body != NULL);
 782
 783         /* Only if the import is replayable, we set replay_open data */
 784         if (och && imp->imp_replayable) {
 785                 mod = obd_mod_alloc();
 786                 if (mod == NULL) {
 787                         DEBUG_REQ(D_ERROR, open_req,
 788                                   "cannot allocate md_open_data");
 789                         RETURN(0);
 790                 }
 791
 792                 /**
 793                  * Take a reference on \var mod, to be freed on mdc_close().
 794                  * It protects \var mod from being freed on eviction (commit
 795                  * callback is called despite rq_replay flag).
 796                  * Another reference for \var och.
 797                  */
 798                 obd_mod_get(mod);
 799                 obd_mod_get(mod);
 800
 801                 spin_lock(&open_req->rq_lock);
 802                 och->och_mod = mod;
 803                 mod->mod_och = och;
 804                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 805                                      it_disposition(it, DISP_OPEN_STRIPE);
 806                 mod->mod_open_req = open_req;
 807                 open_req->rq_cb_data = mod;
 808                 open_req->rq_commit_cb = mdc_commit_open;
 809                 open_req->rq_early_free_repbuf = 1;
 810                 spin_unlock(&open_req->rq_lock);
 811         }
 812
 813         rec->cr_fid2 = body->mbo_fid1;
 814         rec->cr_open_handle_old = body->mbo_open_handle;
 815         open_req->rq_replay_cb = mdc_replay_open;
 816         if (!fid_is_sane(&body->mbo_fid1)) {
 817                 DEBUG_REQ(D_ERROR, open_req,
 818                           "saving replay request with insane FID " DFID,
 819                           PFID(&body->mbo_fid1));
 820                 LBUG();
 821         }
 822
 823         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 824         RETURN(0);
 825 }
 826
 827 static void mdc_free_open(struct md_open_data *mod)
 828 {
 829         int committed = 0;
 830
 831         if (mod->mod_is_create == 0 &&
 832             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 833                 committed = 1;
 834
 835         /**
 836          * No reason to asssert here if the open request has
 837          * rq_replay == 1. It means that mdc_close failed, and
 838          * close request wasn`t sent. It is not fatal to client.
 839          * The worst thing is eviction if the client gets open lock
 840          **/
 841
 842         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 843                   "free open request, rq_replay=%d",
 844                   mod->mod_open_req->rq_replay);
 845
 846         ptlrpc_request_committed(mod->mod_open_req, committed);
 847         if (mod->mod_close_req)
 848                 ptlrpc_request_committed(mod->mod_close_req, committed);
 849 }
 850
 851 int mdc_clear_open_replay_data(struct obd_export *exp,
 852                                struct obd_client_handle *och)
 853 {
 854         struct md_open_data *mod = och->och_mod;
 855         ENTRY;
 856
 857         /**
 858          * It is possible to not have \var mod in a case of eviction between
 859          * lookup and ll_file_open().
 860          **/
 861         if (mod == NULL)
 862                 RETURN(0);
 863
 864         LASSERT(mod != LP_POISON);
 865         LASSERT(mod->mod_open_req != NULL);
 866
 867         spin_lock(&mod->mod_open_req->rq_lock);
 868         if (mod->mod_och)
 869                 mod->mod_och->och_open_handle.cookie = 0;
 870         mod->mod_open_req->rq_early_free_repbuf = 0;
 871         spin_unlock(&mod->mod_open_req->rq_lock);
 872         mdc_free_open(mod);
 873
 874         mod->mod_och = NULL;
 875         och->och_mod = NULL;
 876         obd_mod_put(mod);
 877
 878         RETURN(0);
 879 }
 880
 881 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 882                      struct md_open_data *mod, struct ptlrpc_request **request)
 883 {
 884         struct obd_device     *obd = class_exp2obd(exp);
 885         struct ptlrpc_request *req;
 886         struct req_format     *req_fmt;
 887         size_t                 u32_count = 0;
 888         int                    rc;
 889         int                    saved_rc = 0;
 890         ENTRY;
 891
 892         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 893                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 894                op_data->op_bias);
 895
 896         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 897                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 898                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 899                         /* allocate a FID for volatile file */
 900                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 901                                            op_data);
 902                         if (rc < 0) {
 903                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 904                                        obd->obd_name, PFID(&op_data->op_fid1),
 905                                        rc);
 906                                 /* save the errcode and proceed to close */
 907                                 saved_rc = rc;
 908                         }
 909                 }
 910                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 911                         size_t count = op_data->op_data_size / sizeof(__u32);
 912
 913                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 914                                 u32_count = count;
 915                 }
 916         } else {
 917                 req_fmt = &RQF_MDS_CLOSE;
 918         }
 919
 920         *request = NULL;
 921         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 922                 req = NULL;
 923         else
 924                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 925
 926         /* Ensure that this close's handle is fixed up during replay. */
 927         if (likely(mod != NULL)) {
 928                 LASSERTF(mod->mod_open_req != NULL &&
 929                          mod->mod_open_req->rq_type != LI_POISON,
 930                          "POISONED open %p!\n", mod->mod_open_req);
 931
 932                 mod->mod_close_req = req;
 933
 934                 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
 935                 /* We no longer want to preserve this open for replay even
 936                  * though the open was committed. b=3632, b=3633 */
 937                 spin_lock(&mod->mod_open_req->rq_lock);
 938                 mod->mod_open_req->rq_replay = 0;
 939                 spin_unlock(&mod->mod_open_req->rq_lock);
 940         } else {
 941                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 942         }
 943         if (req == NULL) {
 944                 /**
 945                  * TODO: repeat close after errors
 946                  */
 947                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 948                       "dropped when this client unmounts or is evicted\n",
 949                       obd->obd_name, PFID(&op_data->op_fid1));
 950                 GOTO(out, rc = -ENOMEM);
 951         }
 952
 953         if (u32_count > 0)
 954                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 955                                      u32_count * sizeof(__u32));
 956
 957         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 958         if (rc) {
 959                 ptlrpc_request_free(req);
 960                 req = NULL;
 961                 GOTO(out, rc);
 962         }
 963
 964         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 965          * portal whose threads are not taking any DLM locks and are therefore
 966          * always progressing */
 967         req->rq_request_portal = MDS_READPAGE_PORTAL;
 968         ptlrpc_at_set_req_timeout(req);
 969
 970         if (!(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 971                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 972                                         OP_XVALID_LAZYBLOCKS);
 973
 974         mdc_close_pack(req, op_data);
 975
 976         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 977                              obd->u.cli.cl_default_mds_easize);
 978
 979         ptlrpc_request_set_replen(req);
 980
 981         ptlrpc_get_mod_rpc_slot(req);
 982         rc = ptlrpc_queue_wait(req);
 983         ptlrpc_put_mod_rpc_slot(req);
 984
 985         if (req->rq_repmsg == NULL) {
 986                 CDEBUG(D_RPCTRACE, "request %p failed to send: rc = %d\n", req,
 987                        req->rq_status);
 988                 if (rc == 0)
 989                         rc = req->rq_status ?: -EIO;
 990         } else if (rc == 0 || rc == -EAGAIN) {
 991                 struct mdt_body *body;
 992
 993                 rc = lustre_msg_get_status(req->rq_repmsg);
 994                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 995                         DEBUG_REQ(D_ERROR, req,
 996                                   "type = PTL_RPC_MSG_ERR: rc = %d", rc);
 997                         if (rc > 0)
 998                                 rc = -rc;
 999                 }
1000                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1001                 if (body == NULL)
1002                         rc = -EPROTO;
1003         } else if (rc == -ESTALE) {
1004                 /**
1005                  * it can be allowed error after 3633 if open was committed and
1006                  * server failed before close was sent. Let's check if mod
1007                  * exists and return no error in that case
1008                  */
1009                 if (mod) {
1010                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
1011                         LASSERT(mod->mod_open_req != NULL);
1012                         if (mod->mod_open_req->rq_committed)
1013                                 rc = 0;
1014                 }
1015         }
1016
1017 out:
1018         if (mod) {
1019                 if (rc != 0)
1020                         mod->mod_close_req = NULL;
1021                 /* Since now, mod is accessed through open_req only,
1022                  * thus close req does not keep a reference on mod anymore. */
1023                 obd_mod_put(mod);
1024         }
1025         *request = req;
1026
1027         RETURN(rc < 0 ? rc : saved_rc);
1028 }
1029
1030 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1031                        u64 offset, struct page **pages, int npages,
1032                        struct ptlrpc_request **request)
1033 {
1034         struct ptlrpc_request   *req;
1035         struct ptlrpc_bulk_desc *desc;
1036         int                      i;
1037         int                      resends = 0;
1038         int                      rc;
1039         ENTRY;
1040
1041         *request = NULL;
1042
1043 restart_bulk:
1044         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1045         if (req == NULL)
1046                 RETURN(-ENOMEM);
1047
1048         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1049         if (rc) {
1050                 ptlrpc_request_free(req);
1051                 RETURN(rc);
1052         }
1053
1054         req->rq_request_portal = MDS_READPAGE_PORTAL;
1055         ptlrpc_at_set_req_timeout(req);
1056
1057         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1058                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
1059                                     MDS_BULK_PORTAL,
1060                                     &ptlrpc_bulk_kiov_pin_ops);
1061         if (desc == NULL) {
1062                 ptlrpc_req_finished(req);
1063                 RETURN(-ENOMEM);
1064         }
1065
1066         /* NB req now owns desc and will free it when it gets freed */
1067         for (i = 0; i < npages; i++)
1068                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1069                                                  PAGE_SIZE);
1070
1071         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
1072
1073         ptlrpc_request_set_replen(req);
1074         rc = ptlrpc_queue_wait(req);
1075         if (rc) {
1076                 ptlrpc_req_finished(req);
1077                 if (rc != -ETIMEDOUT)
1078                         RETURN(rc);
1079
1080                 resends++;
1081                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1082                         CERROR("%s: too many resend retries: rc = %d\n",
1083                                exp->exp_obd->obd_name, -EIO);
1084                         RETURN(-EIO);
1085                 }
1086                 ssleep(resends);
1087
1088                 goto restart_bulk;
1089         }
1090
1091         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1092                                           req->rq_bulk->bd_nob_transferred);
1093         if (rc < 0) {
1094                 ptlrpc_req_finished(req);
1095                 RETURN(rc);
1096         }
1097
1098         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1099                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1100                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1101                        PAGE_SIZE * npages);
1102                 ptlrpc_req_finished(req);
1103                 RETURN(-EPROTO);
1104         }
1105
1106         *request = req;
1107         RETURN(0);
1108 }
1109
1110 static void mdc_release_page(struct page *page, int remove)
1111 {
1112         if (remove) {
1113                 lock_page(page);
1114                 if (likely(page->mapping != NULL))
1115                         delete_from_page_cache(page);
1116                 unlock_page(page);
1117         }
1118         put_page(page);
1119 }
1120
1121 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1122                                     __u64 *start, __u64 *end, int hash64)
1123 {
1124         /*
1125          * Complement of hash is used as an index so that
1126          * radix_tree_gang_lookup() can be used to find a page with starting
1127          * hash _smaller_ than one we are looking for.
1128          */
1129         unsigned long offset = hash_x_index(*hash, hash64);
1130         struct page *page;
1131         unsigned long flags;
1132         int found;
1133
1134         xa_lock_irqsave(&mapping->i_pages, flags);
1135         found = radix_tree_gang_lookup(&mapping->page_tree,
1136                                        (void **)&page, offset, 1);
1137         if (found > 0 && !xa_is_value(page)) {
1138                 struct lu_dirpage *dp;
1139
1140                 get_page(page);
1141                 xa_unlock_irqrestore(&mapping->i_pages, flags);
1142                 /*
1143                  * In contrast to find_lock_page() we are sure that directory
1144                  * page cannot be truncated (while DLM lock is held) and,
1145                  * hence, can avoid restart.
1146                  *
1147                  * In fact, page cannot be locked here at all, because
1148                  * mdc_read_page_remote does synchronous io.
1149                  */
1150                 wait_on_page_locked(page);
1151                 if (PageUptodate(page)) {
1152                         dp = kmap(page);
1153                         if (BITS_PER_LONG == 32 && hash64) {
1154                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1155                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1156                                 *hash  = *hash >> 32;
1157                         } else {
1158                                 *start = le64_to_cpu(dp->ldp_hash_start);
1159                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1160                         }
1161                         if (unlikely(*start == 1 && *hash == 0))
1162                                 *hash = *start;
1163                         else
1164                                 LASSERTF(*start <= *hash, "start = %#llx"
1165                                          ",end = %#llx,hash = %#llx\n",
1166                                          *start, *end, *hash);
1167                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1168                               " hash %#llx\n", offset, *start, *end, *hash);
1169                         if (*hash > *end) {
1170                                 kunmap(page);
1171                                 mdc_release_page(page, 0);
1172                                 page = NULL;
1173                         } else if (*end != *start && *hash == *end) {
1174                                 /*
1175                                  * upon hash collision, remove this page,
1176                                  * otherwise put page reference, and
1177                                  * mdc_read_page_remote() will issue RPC to
1178                                  * fetch the page we want.
1179                                  */
1180                                 kunmap(page);
1181                                 mdc_release_page(page,
1182                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1183                                 page = NULL;
1184                         }
1185                 } else {
1186                         put_page(page);
1187                         page = ERR_PTR(-EIO);
1188                 }
1189         } else {
1190                 xa_unlock_irqrestore(&mapping->i_pages, flags);
1191                 page = NULL;
1192         }
1193         return page;
1194 }
1195
1196 /*
1197  * Adjust a set of pages, each page containing an array of lu_dirpages,
1198  * so that each page can be used as a single logical lu_dirpage.
1199  *
1200  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1201  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1202  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1203  * value is used as a cookie to request the next lu_dirpage in a
1204  * directory listing that spans multiple pages (two in this example):
1205  *   ________
1206  *  |        |
1207  * .|--------v-------   -----.
1208  * |s|e|f|p|ent|ent| ... |ent|
1209  * '--|--------------   -----'   Each PAGE contains a single
1210  *    '------.                   lu_dirpage.
1211  * .---------v-------   -----.
1212  * |s|e|f|p|ent| 0 | ... | 0 |
1213  * '-----------------   -----'
1214  *
1215  * However, on hosts where the native VM page size (PAGE_SIZE) is
1216  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1217  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1218  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1219  * after it in the same PAGE (arrows simplified for brevity, but
1220  * in general e0==s1, e1==s2, etc.):
1221  *
1222  * .--------------------   -----.
1223  * |s0|e0|f0|p|ent|ent| ... |ent|
1224  * |---v----------------   -----|
1225  * |s1|e1|f1|p|ent|ent| ... |ent|
1226  * |---v----------------   -----|  Here, each PAGE contains
1227  *             ...                 multiple lu_dirpages.
1228  * |---v----------------   -----|
1229  * |s'|e'|f'|p|ent|ent| ... |ent|
1230  * '---|----------------   -----'
1231  *     v
1232  * .----------------------------.
1233  * |        next PAGE           |
1234  *
1235  * This structure is transformed into a single logical lu_dirpage as follows:
1236  *
1237  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1238  *   labeled 'next PAGE'.
1239  *
1240  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1241  *   a hash collision with the next page exists.
1242  *
1243  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1244  *   to the first entry of the next lu_dirpage.
1245  */
1246 #if PAGE_SIZE > LU_PAGE_SIZE
1247 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1248 {
1249         int i;
1250
1251         for (i = 0; i < cfs_pgs; i++) {
1252                 struct lu_dirpage *dp = kmap(pages[i]);
1253                 struct lu_dirpage *first = dp;
1254                 struct lu_dirent *end_dirent = NULL;
1255                 struct lu_dirent *ent;
1256                 __u64 hash_end = dp->ldp_hash_end;
1257                 __u32 flags = dp->ldp_flags;
1258
1259                 while (--lu_pgs > 0) {
1260                         ent = lu_dirent_start(dp);
1261                         for (end_dirent = ent; ent != NULL;
1262                              end_dirent = ent, ent = lu_dirent_next(ent));
1263
1264                         /* Advance dp to next lu_dirpage. */
1265                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1266
1267                         /* Check if we've reached the end of the PAGE. */
1268                         if (!((unsigned long)dp & ~PAGE_MASK))
1269                                 break;
1270
1271                         /* Save the hash and flags of this lu_dirpage. */
1272                         hash_end = dp->ldp_hash_end;
1273                         flags = dp->ldp_flags;
1274
1275                         /* Check if lu_dirpage contains no entries. */
1276                         if (end_dirent == NULL)
1277                                 break;
1278
1279                         /* Enlarge the end entry lde_reclen from 0 to
1280                          * first entry of next lu_dirpage. */
1281                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1282                         end_dirent->lde_reclen =
1283                                 cpu_to_le16((char *)(dp->ldp_entries) -
1284                                             (char *)end_dirent);
1285                 }
1286
1287                 first->ldp_hash_end = hash_end;
1288                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1289                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1290
1291                 kunmap(pages[i]);
1292         }
1293         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1294 }
1295 #else
1296 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1297 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1298
1299 /* parameters for readdir page */
1300 struct readpage_param {
1301         struct md_op_data       *rp_mod;
1302         __u64                   rp_off;
1303         int                     rp_hash64;
1304         struct obd_export       *rp_exp;
1305         struct md_callback      *rp_cb;
1306 };
1307
1308 /**
1309  * Read pages from server.
1310  *
1311  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1312  * a header lu_dirpage which describes the start/end hash, and whether this
1313  * page is empty (contains no dir entry) or hash collide with next page.
1314  * After client receives reply, several pages will be integrated into dir page
1315  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1316  * lu_dirpage for this integrated page will be adjusted.
1317  **/
1318 static int mdc_read_page_remote(void *data, struct page *page0)
1319 {
1320         struct readpage_param *rp = data;
1321         struct page **page_pool;
1322         struct page *page;
1323         struct lu_dirpage *dp;
1324         struct md_op_data *op_data = rp->rp_mod;
1325         struct ptlrpc_request *req;
1326         int max_pages;
1327         struct inode *inode;
1328         struct lu_fid *fid;
1329         int rd_pgs = 0; /* number of pages actually read */
1330         int npages;
1331         int i;
1332         int rc;
1333         ENTRY;
1334
1335         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1336         inode = op_data->op_data;
1337         fid = &op_data->op_fid1;
1338         LASSERT(inode != NULL);
1339
1340         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1341         if (page_pool != NULL) {
1342                 page_pool[0] = page0;
1343         } else {
1344                 page_pool = &page0;
1345                 max_pages = 1;
1346         }
1347
1348         for (npages = 1; npages < max_pages; npages++) {
1349                 page = __page_cache_alloc(mapping_gfp_mask(inode->i_mapping)
1350                                           | __GFP_COLD);
1351                 if (page == NULL)
1352                         break;
1353                 page_pool[npages] = page;
1354         }
1355
1356         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1357         if (rc < 0) {
1358                 /* page0 is special, which was added into page cache early */
1359                 delete_from_page_cache(page0);
1360         } else {
1361                 int lu_pgs;
1362
1363                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1364                         PAGE_SHIFT;
1365                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1366                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1367
1368                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1369
1370                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1371
1372                 SetPageUptodate(page0);
1373         }
1374         unlock_page(page0);
1375
1376         ptlrpc_req_finished(req);
1377         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1378         for (i = 1; i < npages; i++) {
1379                 unsigned long   offset;
1380                 __u64           hash;
1381                 int ret;
1382
1383                 page = page_pool[i];
1384
1385                 if (rc < 0 || i >= rd_pgs) {
1386                         put_page(page);
1387                         continue;
1388                 }
1389
1390                 SetPageUptodate(page);
1391
1392                 dp = kmap(page);
1393                 hash = le64_to_cpu(dp->ldp_hash_start);
1394                 kunmap(page);
1395
1396                 offset = hash_x_index(hash, rp->rp_hash64);
1397
1398                 prefetchw(&page->flags);
1399                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1400                                             GFP_KERNEL);
1401                 if (ret == 0)
1402                         unlock_page(page);
1403                 else
1404                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1405                                " rc = %d\n", offset, ret);
1406                 put_page(page);
1407         }
1408
1409         if (page_pool != &page0)
1410                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1411
1412         RETURN(rc);
1413 }
1414
1415 /**
1416  * Read dir page from cache first, if it can not find it, read it from
1417  * server and add into the cache.
1418  *
1419  * \param[in] exp       MDC export
1420  * \param[in] op_data   client MD stack parameters, transfering parameters
1421  *                      between different layers on client MD stack.
1422  * \param[in] cb_op     callback required for ldlm lock enqueue during
1423  *                      read page
1424  * \param[in] hash_offset the hash offset of the page to be read
1425  * \param[in] ppage     the page to be read
1426  *
1427  * retval               = 0 get the page successfully
1428  *                      errno(<0) get the page failed
1429  */
1430 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1431                          struct md_callback *cb_op, __u64 hash_offset,
1432                          struct page **ppage)
1433 {
1434         struct lookup_intent    it = { .it_op = IT_READDIR };
1435         struct page             *page;
1436         struct inode            *dir = op_data->op_data;
1437         struct address_space    *mapping;
1438         struct lu_dirpage       *dp;
1439         __u64                   start = 0;
1440         __u64                   end = 0;
1441         struct lustre_handle    lockh;
1442         struct ptlrpc_request   *enq_req = NULL;
1443         struct readpage_param   rp_param;
1444         int rc;
1445
1446         ENTRY;
1447
1448         *ppage = NULL;
1449
1450         LASSERT(dir != NULL);
1451         mapping = dir->i_mapping;
1452
1453         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1454                              cb_op->md_blocking_ast, 0);
1455         if (enq_req != NULL)
1456                 ptlrpc_req_finished(enq_req);
1457
1458         if (rc < 0) {
1459                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1460                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1461                 RETURN(rc);
1462         }
1463
1464         rc = 0;
1465         lockh.cookie = it.it_lock_handle;
1466         mdc_set_lock_data(exp, &lockh, dir, NULL);
1467
1468         rp_param.rp_off = hash_offset;
1469         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1470         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1471                                rp_param.rp_hash64);
1472         if (IS_ERR(page)) {
1473                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1474                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1475                        rp_param.rp_off, PTR_ERR(page));
1476                 GOTO(out_unlock, rc = PTR_ERR(page));
1477         } else if (page != NULL) {
1478                 /*
1479                  * XXX nikita: not entirely correct handling of a corner case:
1480                  * suppose hash chain of entries with hash value HASH crosses
1481                  * border between pages P0 and P1. First both P0 and P1 are
1482                  * cached, seekdir() is called for some entry from the P0 part
1483                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1484                  * happens and finds P1, as it starts with matching hash
1485                  * value. Remaining entries from P0 part of the chain are
1486                  * skipped. (Is that really a bug?)
1487                  *
1488                  * Possible solutions: 0. don't cache P1 is such case, handle
1489                  * it as an "overflow" page. 1. invalidate all pages at
1490                  * once. 2. use HASH|1 as an index for P1.
1491                  */
1492                 GOTO(hash_collision, page);
1493         }
1494
1495         rp_param.rp_exp = exp;
1496         rp_param.rp_mod = op_data;
1497         page = read_cache_page(mapping,
1498                                hash_x_index(rp_param.rp_off,
1499                                             rp_param.rp_hash64),
1500                                mdc_read_page_remote, &rp_param);
1501         if (IS_ERR(page)) {
1502                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1503                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1504                        rp_param.rp_off, PTR_ERR(page));
1505                 GOTO(out_unlock, rc = PTR_ERR(page));
1506         }
1507
1508         wait_on_page_locked(page);
1509         (void)kmap(page);
1510         if (!PageUptodate(page)) {
1511                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1512                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1513                        rp_param.rp_off, -5);
1514                 goto fail;
1515         }
1516         if (!PageChecked(page))
1517                 SetPageChecked(page);
1518         if (PageError(page)) {
1519                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1520                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1521                        rp_param.rp_off, -5);
1522                 goto fail;
1523         }
1524
1525 hash_collision:
1526         dp = page_address(page);
1527         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1528                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1529                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1530                 rp_param.rp_off = hash_offset >> 32;
1531         } else {
1532                 start = le64_to_cpu(dp->ldp_hash_start);
1533                 end   = le64_to_cpu(dp->ldp_hash_end);
1534                 rp_param.rp_off = hash_offset;
1535         }
1536         if (end == start) {
1537                 LASSERT(start == rp_param.rp_off);
1538                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1539 #if BITS_PER_LONG == 32
1540                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1541                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1542                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1543 #endif
1544
1545                 /*
1546                  * Fetch whole overflow chain...
1547                  *
1548                  * XXX not yet.
1549                  */
1550                 goto fail;
1551         }
1552         *ppage = page;
1553 out_unlock:
1554         ldlm_lock_decref(&lockh, it.it_lock_mode);
1555         return rc;
1556 fail:
1557         kunmap(page);
1558         mdc_release_page(page, 1);
1559         rc = -EIO;
1560         goto out_unlock;
1561 }
1562
1563 static int mdc_statfs_interpret(const struct lu_env *env,
1564                                 struct ptlrpc_request *req, void *args, int rc)
1565 {
1566         struct obd_info *oinfo = args;
1567         struct obd_statfs *osfs;
1568
1569         if (!rc) {
1570                 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1571                 if (!osfs)
1572                         return -EPROTO;
1573
1574                 oinfo->oi_osfs = osfs;
1575
1576                 CDEBUG(D_CACHE, "blocks=%llu free=%llu avail=%llu "
1577                        "objects=%llu free=%llu state=%x\n",
1578                         osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
1579                         osfs->os_files, osfs->os_ffree, osfs->os_state);
1580         }
1581
1582         oinfo->oi_cb_up(oinfo, rc);
1583
1584         return rc;
1585 }
1586
1587 static int mdc_statfs_async(struct obd_export *exp,
1588                             struct obd_info *oinfo, time64_t max_age,
1589                             struct ptlrpc_request_set *unused)
1590 {
1591         struct ptlrpc_request *req;
1592         struct obd_info *aa;
1593
1594         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_STATFS,
1595                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1596         if (req == NULL)
1597                 return -ENOMEM;
1598
1599         ptlrpc_request_set_replen(req);
1600         req->rq_interpret_reply = mdc_statfs_interpret;
1601
1602         aa = ptlrpc_req_async_args(aa, req);
1603         *aa = *oinfo;
1604
1605         ptlrpcd_add_req(req);
1606
1607         return 0;
1608 }
1609
1610 static int mdc_statfs(const struct lu_env *env,
1611                       struct obd_export *exp, struct obd_statfs *osfs,
1612                       time64_t max_age, __u32 flags)
1613 {
1614         struct obd_device *obd = class_exp2obd(exp);
1615         struct req_format *fmt;
1616         struct ptlrpc_request *req;
1617         struct obd_statfs *msfs;
1618         struct obd_import *imp = NULL;
1619         int rc;
1620         ENTRY;
1621
1622         /*
1623          * Since the request might also come from lprocfs, so we need
1624          * sync this with client_disconnect_export Bug15684
1625          */
1626         down_read(&obd->u.cli.cl_sem);
1627         if (obd->u.cli.cl_import)
1628                 imp = class_import_get(obd->u.cli.cl_import);
1629         up_read(&obd->u.cli.cl_sem);
1630         if (!imp)
1631                 RETURN(-ENODEV);
1632
1633         fmt = &RQF_MDS_STATFS;
1634         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1635             (flags & OBD_STATFS_SUM))
1636                 fmt = &RQF_MDS_STATFS_NEW;
1637         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1638                                         MDS_STATFS);
1639         if (req == NULL)
1640                 GOTO(output, rc = -ENOMEM);
1641
1642         if ((flags & OBD_STATFS_SUM) &&
1643             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1644                 /* request aggregated states */
1645                 struct mdt_body *body;
1646
1647                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1648                 if (body == NULL)
1649                         GOTO(out, rc = -EPROTO);
1650                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1651         }
1652
1653         ptlrpc_request_set_replen(req);
1654
1655         if (flags & OBD_STATFS_NODELAY) {
1656                 /* procfs requests not want stay in wait for avoid deadlock */
1657                 req->rq_no_resend = 1;
1658                 req->rq_no_delay = 1;
1659         }
1660
1661         rc = ptlrpc_queue_wait(req);
1662         if (rc) {
1663                 /* check connection error first */
1664                 if (imp->imp_connect_error)
1665                         rc = imp->imp_connect_error;
1666                 GOTO(out, rc);
1667         }
1668
1669         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1670         if (msfs == NULL)
1671                 GOTO(out, rc = -EPROTO);
1672
1673         *osfs = *msfs;
1674         EXIT;
1675 out:
1676         ptlrpc_req_finished(req);
1677 output:
1678         class_import_put(imp);
1679         return rc;
1680 }
1681
1682 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1683 {
1684         __u32 keylen, vallen;
1685         void *key;
1686         int rc;
1687
1688         if (gf->gf_pathlen > PATH_MAX)
1689                 RETURN(-ENAMETOOLONG);
1690         if (gf->gf_pathlen < 2)
1691                 RETURN(-EOVERFLOW);
1692
1693         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1694         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1695                                 sizeof(struct lu_fid));
1696         OBD_ALLOC(key, keylen);
1697         if (key == NULL)
1698                 RETURN(-ENOMEM);
1699         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1700         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1701         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1702                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1703         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1704                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1705
1706         if (!fid_is_sane(&gf->gf_fid))
1707                 GOTO(out, rc = -EINVAL);
1708
1709         /* Val is struct getinfo_fid2path result plus path */
1710         vallen = sizeof(*gf) + gf->gf_pathlen;
1711
1712         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1713         if (rc != 0 && rc != -EREMOTE)
1714                 GOTO(out, rc);
1715
1716         if (vallen <= sizeof(*gf))
1717                 GOTO(out, rc = -EPROTO);
1718         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1719                 GOTO(out, rc = -EOVERFLOW);
1720
1721         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1722                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1723                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1724                /* only log the last 512 characters of the path */
1725                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1726
1727 out:
1728         OBD_FREE(key, keylen);
1729         return rc;
1730 }
1731
1732 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1733                                 struct hsm_progress_kernel *hpk)
1734 {
1735         struct obd_import               *imp = class_exp2cliimp(exp);
1736         struct hsm_progress_kernel      *req_hpk;
1737         struct ptlrpc_request           *req;
1738         int                              rc;
1739         ENTRY;
1740
1741         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1742                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1743         if (req == NULL)
1744                 GOTO(out, rc = -ENOMEM);
1745
1746         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1747
1748         /* Copy hsm_progress struct */
1749         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1750         if (req_hpk == NULL)
1751                 GOTO(out, rc = -EPROTO);
1752
1753         *req_hpk = *hpk;
1754         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1755
1756         ptlrpc_request_set_replen(req);
1757
1758         ptlrpc_get_mod_rpc_slot(req);
1759         rc = ptlrpc_queue_wait(req);
1760         ptlrpc_put_mod_rpc_slot(req);
1761
1762         GOTO(out, rc);
1763 out:
1764         ptlrpc_req_finished(req);
1765         return rc;
1766 }
1767 /**
1768  * Send hsm_ct_register to MDS
1769  *
1770  * \param[in]   imp             import
1771  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1772  *                              else it is the count of archive_ids
1773  * \param[in]   archives        if in bitmap format, it is NULL,
1774  *                              else it is archive_id lists
1775  */
1776 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1777                                    __u32 *archives)
1778 {
1779         struct ptlrpc_request *req;
1780         __u32 *archive_array;
1781         size_t archives_size;
1782         int rc;
1783         ENTRY;
1784
1785         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1786         if (req == NULL)
1787                 RETURN(-ENOMEM);
1788
1789         if (archives != NULL)
1790                 archives_size = sizeof(*archive_array) * archive_count;
1791         else
1792                 archives_size = sizeof(archive_count);
1793
1794         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1795                              RCL_CLIENT, archives_size);
1796
1797         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1798         if (rc) {
1799                 ptlrpc_request_free(req);
1800                 RETURN(-ENOMEM);
1801         }
1802
1803         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1804
1805         archive_array = req_capsule_client_get(&req->rq_pill,
1806                                                &RMF_MDS_HSM_ARCHIVE);
1807         if (archive_array == NULL)
1808                 GOTO(out, rc = -EPROTO);
1809
1810         if (archives != NULL)
1811                 memcpy(archive_array, archives, archives_size);
1812         else
1813                 *archive_array = archive_count;
1814
1815         ptlrpc_request_set_replen(req);
1816
1817         rc = mdc_queue_wait(req);
1818         GOTO(out, rc);
1819 out:
1820         ptlrpc_req_finished(req);
1821         return rc;
1822 }
1823
1824 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1825                                       struct md_op_data *op_data)
1826 {
1827         struct hsm_current_action       *hca = op_data->op_data;
1828         struct hsm_current_action       *req_hca;
1829         struct ptlrpc_request           *req;
1830         int                              rc;
1831         ENTRY;
1832
1833         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1834                                    &RQF_MDS_HSM_ACTION);
1835         if (req == NULL)
1836                 RETURN(-ENOMEM);
1837
1838         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1839         if (rc) {
1840                 ptlrpc_request_free(req);
1841                 RETURN(rc);
1842         }
1843
1844         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1845                       op_data->op_suppgids[0], 0);
1846
1847         ptlrpc_request_set_replen(req);
1848
1849         rc = mdc_queue_wait(req);
1850         if (rc)
1851                 GOTO(out, rc);
1852
1853         req_hca = req_capsule_server_get(&req->rq_pill,
1854                                          &RMF_MDS_HSM_CURRENT_ACTION);
1855         if (req_hca == NULL)
1856                 GOTO(out, rc = -EPROTO);
1857
1858         *hca = *req_hca;
1859
1860         EXIT;
1861 out:
1862         ptlrpc_req_finished(req);
1863         return rc;
1864 }
1865
1866 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1867 {
1868         struct ptlrpc_request   *req;
1869         int                      rc;
1870         ENTRY;
1871
1872         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1873                                         LUSTRE_MDS_VERSION,
1874                                         MDS_HSM_CT_UNREGISTER);
1875         if (req == NULL)
1876                 GOTO(out, rc = -ENOMEM);
1877
1878         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1879
1880         ptlrpc_request_set_replen(req);
1881
1882         rc = mdc_queue_wait(req);
1883         GOTO(out, rc);
1884 out:
1885         ptlrpc_req_finished(req);
1886         return rc;
1887 }
1888
1889 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1890                                  struct md_op_data *op_data)
1891 {
1892         struct hsm_user_state   *hus = op_data->op_data;
1893         struct hsm_user_state   *req_hus;
1894         struct ptlrpc_request   *req;
1895         int                      rc;
1896         ENTRY;
1897
1898         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1899                                    &RQF_MDS_HSM_STATE_GET);
1900         if (req == NULL)
1901                 RETURN(-ENOMEM);
1902
1903         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1904         if (rc != 0) {
1905                 ptlrpc_request_free(req);
1906                 RETURN(rc);
1907         }
1908
1909         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1910                       op_data->op_suppgids[0], 0);
1911
1912         ptlrpc_request_set_replen(req);
1913
1914         rc = mdc_queue_wait(req);
1915         if (rc)
1916                 GOTO(out, rc);
1917
1918         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1919         if (req_hus == NULL)
1920                 GOTO(out, rc = -EPROTO);
1921
1922         *hus = *req_hus;
1923
1924         EXIT;
1925 out:
1926         ptlrpc_req_finished(req);
1927         return rc;
1928 }
1929
1930 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1931                                  struct md_op_data *op_data)
1932 {
1933         struct hsm_state_set    *hss = op_data->op_data;
1934         struct hsm_state_set    *req_hss;
1935         struct ptlrpc_request   *req;
1936         int                      rc;
1937         ENTRY;
1938
1939         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1940                                    &RQF_MDS_HSM_STATE_SET);
1941         if (req == NULL)
1942                 RETURN(-ENOMEM);
1943
1944         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1945         if (rc) {
1946                 ptlrpc_request_free(req);
1947                 RETURN(rc);
1948         }
1949
1950         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1951                       op_data->op_suppgids[0], 0);
1952
1953         /* Copy states */
1954         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1955         if (req_hss == NULL)
1956                 GOTO(out, rc = -EPROTO);
1957         *req_hss = *hss;
1958
1959         ptlrpc_request_set_replen(req);
1960
1961         ptlrpc_get_mod_rpc_slot(req);
1962         rc = ptlrpc_queue_wait(req);
1963         ptlrpc_put_mod_rpc_slot(req);
1964
1965         GOTO(out, rc);
1966 out:
1967         ptlrpc_req_finished(req);
1968         return rc;
1969 }
1970
1971 static int mdc_ioc_hsm_request(struct obd_export *exp,
1972                                struct hsm_user_request *hur)
1973 {
1974         struct obd_import       *imp = class_exp2cliimp(exp);
1975         struct ptlrpc_request   *req;
1976         struct hsm_request      *req_hr;
1977         struct hsm_user_item    *req_hui;
1978         char                    *req_opaque;
1979         int                      rc;
1980         ENTRY;
1981
1982         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1983         if (req == NULL)
1984                 GOTO(out, rc = -ENOMEM);
1985
1986         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1987                              hur->hur_request.hr_itemcount
1988                              * sizeof(struct hsm_user_item));
1989         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1990                              hur->hur_request.hr_data_len);
1991
1992         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1993         if (rc) {
1994                 ptlrpc_request_free(req);
1995                 RETURN(rc);
1996         }
1997
1998         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1999
2000         /* Copy hsm_request struct */
2001         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2002         if (req_hr == NULL)
2003                 GOTO(out, rc = -EPROTO);
2004         *req_hr = hur->hur_request;
2005
2006         /* Copy hsm_user_item structs */
2007         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2008         if (req_hui == NULL)
2009                 GOTO(out, rc = -EPROTO);
2010         memcpy(req_hui, hur->hur_user_item,
2011                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2012
2013         /* Copy opaque field */
2014         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2015         if (req_opaque == NULL)
2016                 GOTO(out, rc = -EPROTO);
2017         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2018
2019         ptlrpc_request_set_replen(req);
2020
2021         ptlrpc_get_mod_rpc_slot(req);
2022         rc = ptlrpc_queue_wait(req);
2023         ptlrpc_put_mod_rpc_slot(req);
2024
2025         GOTO(out, rc);
2026
2027 out:
2028         ptlrpc_req_finished(req);
2029         return rc;
2030 }
2031
2032 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2033                                 struct lustre_kernelcomm *lk);
2034
2035 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2036                         struct obd_quotactl *oqctl)
2037 {
2038         struct ptlrpc_request   *req;
2039         struct obd_quotactl     *oqc;
2040         int                      rc;
2041         ENTRY;
2042
2043         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2044                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2045                                         MDS_QUOTACTL);
2046         if (req == NULL)
2047                 RETURN(-ENOMEM);
2048
2049         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2050         *oqc = *oqctl;
2051
2052         ptlrpc_request_set_replen(req);
2053         ptlrpc_at_set_req_timeout(req);
2054
2055         rc = ptlrpc_queue_wait(req);
2056         if (rc)
2057                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2058
2059         if (req->rq_repmsg &&
2060             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2061                 *oqctl = *oqc;
2062         } else if (!rc) {
2063                 CERROR ("Can't unpack obd_quotactl\n");
2064                 rc = -EPROTO;
2065         }
2066         ptlrpc_req_finished(req);
2067
2068         RETURN(rc);
2069 }
2070
2071 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2072                                 struct md_op_data *op_data)
2073 {
2074         LIST_HEAD(cancels);
2075         struct ptlrpc_request   *req;
2076         int                      rc, count;
2077         struct mdc_swap_layouts *msl, *payload;
2078         ENTRY;
2079
2080         msl = op_data->op_data;
2081
2082         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2083          * first thing it will do is to cancel the 2 layout
2084          * locks held by this client.
2085          * So the client must cancel its layout locks on the 2 fids
2086          * with the request RPC to avoid extra RPC round trips.
2087          */
2088         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2089                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2090                                         MDS_INODELOCK_XATTR);
2091         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2092                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2093                                          MDS_INODELOCK_XATTR);
2094
2095         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2096                                    &RQF_MDS_SWAP_LAYOUTS);
2097         if (req == NULL) {
2098                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2099                 RETURN(-ENOMEM);
2100         }
2101
2102         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2103         if (rc) {
2104                 ptlrpc_request_free(req);
2105                 RETURN(rc);
2106         }
2107
2108         mdc_swap_layouts_pack(req, op_data);
2109
2110         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2111         LASSERT(payload);
2112
2113         *payload = *msl;
2114
2115         ptlrpc_request_set_replen(req);
2116
2117         rc = ptlrpc_queue_wait(req);
2118         if (rc)
2119                 GOTO(out, rc);
2120         EXIT;
2121
2122 out:
2123         ptlrpc_req_finished(req);
2124         return rc;
2125 }
2126
2127 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2128                          void *karg, void __user *uarg)
2129 {
2130         struct obd_device *obd = exp->exp_obd;
2131         struct obd_ioctl_data *data = karg;
2132         struct obd_import *imp = obd->u.cli.cl_import;
2133         int rc;
2134         ENTRY;
2135
2136         if (!try_module_get(THIS_MODULE)) {
2137                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2138                        module_name(THIS_MODULE));
2139                 return -EINVAL;
2140         }
2141         switch (cmd) {
2142         case OBD_IOC_FID2PATH:
2143                 rc = mdc_ioc_fid2path(exp, karg);
2144                 GOTO(out, rc);
2145         case LL_IOC_HSM_CT_START:
2146                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2147                 /* ignore if it was already registered on this MDS. */
2148                 if (rc == -EEXIST)
2149                         rc = 0;
2150                 GOTO(out, rc);
2151         case LL_IOC_HSM_PROGRESS:
2152                 rc = mdc_ioc_hsm_progress(exp, karg);
2153                 GOTO(out, rc);
2154         case LL_IOC_HSM_STATE_GET:
2155                 rc = mdc_ioc_hsm_state_get(exp, karg);
2156                 GOTO(out, rc);
2157         case LL_IOC_HSM_STATE_SET:
2158                 rc = mdc_ioc_hsm_state_set(exp, karg);
2159                 GOTO(out, rc);
2160         case LL_IOC_HSM_ACTION:
2161                 rc = mdc_ioc_hsm_current_action(exp, karg);
2162                 GOTO(out, rc);
2163         case LL_IOC_HSM_REQUEST:
2164                 rc = mdc_ioc_hsm_request(exp, karg);
2165                 GOTO(out, rc);
2166         case OBD_IOC_CLIENT_RECOVER:
2167                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2168                 if (rc < 0)
2169                         GOTO(out, rc);
2170                 GOTO(out, rc = 0);
2171         case IOC_OSC_SET_ACTIVE:
2172                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2173                 GOTO(out, rc);
2174         /*
2175          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2176          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2177          * there'd be no LMV layer thus we might be called here. Eventually
2178          * this code should be removed.
2179          * bz20731, LU-592.
2180          */
2181         case IOC_OBD_STATFS: {
2182                 struct obd_statfs stat_buf = {0};
2183
2184                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2185                         GOTO(out, rc = -ENODEV);
2186
2187                 /* copy UUID */
2188                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2189                                  min((int)data->ioc_plen2,
2190                                      (int)sizeof(struct obd_uuid))))
2191                         GOTO(out, rc = -EFAULT);
2192
2193                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2194                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2195                                 0);
2196                 if (rc != 0)
2197                         GOTO(out, rc);
2198
2199                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2200                                      min((int) data->ioc_plen1,
2201                                          (int) sizeof(stat_buf))))
2202                         GOTO(out, rc = -EFAULT);
2203
2204                 GOTO(out, rc = 0);
2205         }
2206         case OBD_IOC_QUOTACTL: {
2207                 struct if_quotactl *qctl = karg;
2208                 struct obd_quotactl *oqctl;
2209
2210                 OBD_ALLOC_PTR(oqctl);
2211                 if (oqctl == NULL)
2212                         GOTO(out, rc = -ENOMEM);
2213
2214                 QCTL_COPY(oqctl, qctl);
2215                 rc = obd_quotactl(exp, oqctl);
2216                 if (rc == 0) {
2217                         QCTL_COPY(qctl, oqctl);
2218                         qctl->qc_valid = QC_MDTIDX;
2219                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2220                 }
2221
2222                 OBD_FREE_PTR(oqctl);
2223                 GOTO(out, rc);
2224         }
2225         case LL_IOC_GET_CONNECT_FLAGS:
2226                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2227                                  sizeof(*exp_connect_flags_ptr(exp))))
2228                         GOTO(out, rc = -EFAULT);
2229
2230                 GOTO(out, rc = 0);
2231         case LL_IOC_LOV_SWAP_LAYOUTS:
2232                 rc = mdc_ioc_swap_layouts(exp, karg);
2233                 GOTO(out, rc);
2234         default:
2235                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2236                 GOTO(out, rc = -ENOTTY);
2237         }
2238 out:
2239         module_put(THIS_MODULE);
2240
2241         return rc;
2242 }
2243
2244 static int mdc_get_info_rpc(struct obd_export *exp,
2245                             u32 keylen, void *key,
2246                             u32 vallen, void *val)
2247 {
2248         struct obd_import      *imp = class_exp2cliimp(exp);
2249         struct ptlrpc_request  *req;
2250         char                   *tmp;
2251         int                     rc = -EINVAL;
2252         ENTRY;
2253
2254         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2255         if (req == NULL)
2256                 RETURN(-ENOMEM);
2257
2258         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2259                              RCL_CLIENT, keylen);
2260         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2261                              RCL_CLIENT, sizeof(vallen));
2262
2263         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2264         if (rc) {
2265                 ptlrpc_request_free(req);
2266                 RETURN(rc);
2267         }
2268
2269         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2270         memcpy(tmp, key, keylen);
2271         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2272         memcpy(tmp, &vallen, sizeof(vallen));
2273
2274         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2275                              RCL_SERVER, vallen);
2276         ptlrpc_request_set_replen(req);
2277
2278         rc = ptlrpc_queue_wait(req);
2279         /* -EREMOTE means the get_info result is partial, and it needs to
2280          * continue on another MDT, see fid2path part in lmv_iocontrol */
2281         if (rc == 0 || rc == -EREMOTE) {
2282                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2283                 memcpy(val, tmp, vallen);
2284                 if (ptlrpc_rep_need_swab(req)) {
2285                         if (KEY_IS(KEY_FID2PATH))
2286                                 lustre_swab_fid2path(val);
2287                 }
2288         }
2289         ptlrpc_req_finished(req);
2290
2291         RETURN(rc);
2292 }
2293
2294 static void lustre_swab_hai(struct hsm_action_item *h)
2295 {
2296         __swab32s(&h->hai_len);
2297         __swab32s(&h->hai_action);
2298         lustre_swab_lu_fid(&h->hai_fid);
2299         lustre_swab_lu_fid(&h->hai_dfid);
2300         __swab64s(&h->hai_cookie);
2301         __swab64s(&h->hai_extent.offset);
2302         __swab64s(&h->hai_extent.length);
2303         __swab64s(&h->hai_gid);
2304 }
2305
2306 static void lustre_swab_hal(struct hsm_action_list *h)
2307 {
2308         struct hsm_action_item  *hai;
2309         __u32                    i;
2310
2311         __swab32s(&h->hal_version);
2312         __swab32s(&h->hal_count);
2313         __swab32s(&h->hal_archive_id);
2314         __swab64s(&h->hal_flags);
2315         hai = hai_first(h);
2316         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2317                 lustre_swab_hai(hai);
2318 }
2319
2320 static void lustre_swab_kuch(struct kuc_hdr *l)
2321 {
2322         __swab16s(&l->kuc_magic);
2323         /* __u8 l->kuc_transport */
2324         __swab16s(&l->kuc_msgtype);
2325         __swab16s(&l->kuc_msglen);
2326 }
2327
2328 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2329                                 struct lustre_kernelcomm *lk)
2330 {
2331         struct obd_import *imp = class_exp2cliimp(exp);
2332         int rc = 0;
2333
2334         if (lk->lk_group != KUC_GRP_HSM) {
2335                 CERROR("Bad copytool group %d\n", lk->lk_group);
2336                 return -EINVAL;
2337         }
2338
2339         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2340                lk->lk_uid, lk->lk_group, lk->lk_flags);
2341
2342         if (lk->lk_flags & LK_FLG_STOP) {
2343                 /* Unregister with the coordinator */
2344                 rc = mdc_ioc_hsm_ct_unregister(imp);
2345         } else {
2346                 __u32 *archives = NULL;
2347
2348                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2349                         archives = lk->lk_data;
2350
2351                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2352         }
2353
2354         return rc;
2355 }
2356
2357 /**
2358  * Send a message to any listening copytools
2359  * @param val KUC message (kuc_hdr + hsm_action_list)
2360  * @param len total length of message
2361  */
2362 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2363                                  size_t len, void *val)
2364 {
2365         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2366         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2367         int                      rc;
2368         ENTRY;
2369
2370         if (len < sizeof(*lh) + sizeof(*hal)) {
2371                 CERROR("Short HSM message %zu < %zu\n", len,
2372                        sizeof(*lh) + sizeof(*hal));
2373                 RETURN(-EPROTO);
2374         }
2375         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2376                 lustre_swab_kuch(lh);
2377                 lustre_swab_hal(hal);
2378         } else if (lh->kuc_magic != KUC_MAGIC) {
2379                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2380                 RETURN(-EPROTO);
2381         }
2382
2383         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2384                "on %s\n",
2385                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2386                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2387
2388         /* Broadcast to HSM listeners */
2389         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2390
2391         RETURN(rc);
2392 }
2393
2394 /**
2395  * callback function passed to kuc for re-registering each HSM copytool
2396  * running on MDC, after MDT shutdown/recovery.
2397  * @param data copytool registration data
2398  * @param cb_arg callback argument (obd_import)
2399  */
2400 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2401 {
2402         struct obd_import *imp = (struct obd_import *)cb_arg;
2403         struct kkuc_ct_data *kcd = data;
2404         __u32 *archives = NULL;
2405         int rc;
2406
2407         if (kcd == NULL ||
2408             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2409              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2410                 return -EPROTO;
2411
2412         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2413                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2414                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2415                        kcd->kcd_nr_archives);
2416         } else {
2417                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2418                        "(archive nr = %u)\n",
2419                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2420                 if (kcd->kcd_nr_archives != 0)
2421                         archives = kcd->kcd_archives;
2422         }
2423
2424         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2425         /* ignore error if the copytool is already registered */
2426         return (rc == -EEXIST) ? 0 : rc;
2427 }
2428
2429 /**
2430  * Re-establish all kuc contexts with MDT
2431  * after MDT shutdown/recovery.
2432  */
2433 static int mdc_kuc_reregister(struct obd_import *imp)
2434 {
2435         /* re-register HSM agents */
2436         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2437                                          mdc_hsm_ct_reregister, imp);
2438 }
2439
2440 static int mdc_set_info_async(const struct lu_env *env,
2441                               struct obd_export *exp,
2442                               u32 keylen, void *key,
2443                               u32 vallen, void *val,
2444                               struct ptlrpc_request_set *set)
2445 {
2446         struct obd_import       *imp = class_exp2cliimp(exp);
2447         int                      rc;
2448         ENTRY;
2449
2450         if (KEY_IS(KEY_READ_ONLY)) {
2451                 if (vallen != sizeof(int))
2452                         RETURN(-EINVAL);
2453
2454                 spin_lock(&imp->imp_lock);
2455                 if (*((int *)val)) {
2456                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2457                         imp->imp_connect_data.ocd_connect_flags |=
2458                                                         OBD_CONNECT_RDONLY;
2459                 } else {
2460                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2461                         imp->imp_connect_data.ocd_connect_flags &=
2462                                                         ~OBD_CONNECT_RDONLY;
2463                 }
2464                 spin_unlock(&imp->imp_lock);
2465
2466                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2467                                        keylen, key, vallen, val, set);
2468                 RETURN(rc);
2469         }
2470         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2471                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2472                                        keylen, key, vallen, val, set);
2473                 RETURN(rc);
2474         }
2475         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2476                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2477                                            val);
2478                 RETURN(rc);
2479         }
2480
2481         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2482                 __u32 *default_easize = val;
2483
2484                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2485                 RETURN(0);
2486         }
2487
2488         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2489         RETURN(rc);
2490 }
2491
2492 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2493                         __u32 keylen, void *key, __u32 *vallen, void *val)
2494 {
2495         int rc = -EINVAL;
2496
2497         if (KEY_IS(KEY_MAX_EASIZE)) {
2498                 __u32 mdsize, *max_easize;
2499
2500                 if (*vallen != sizeof(int))
2501                         RETURN(-EINVAL);
2502                 mdsize = *(__u32 *)val;
2503                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2504                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2505                 max_easize = val;
2506                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2507                 RETURN(0);
2508         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2509                 __u32 *default_easize;
2510
2511                 if (*vallen != sizeof(int))
2512                         RETURN(-EINVAL);
2513                 default_easize = val;
2514                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2515                 RETURN(0);
2516         } else if (KEY_IS(KEY_CONN_DATA)) {
2517                 struct obd_import *imp = class_exp2cliimp(exp);
2518                 struct obd_connect_data *data = val;
2519
2520                 if (*vallen != sizeof(*data))
2521                         RETURN(-EINVAL);
2522
2523                 *data = imp->imp_connect_data;
2524                 RETURN(0);
2525         } else if (KEY_IS(KEY_TGT_COUNT)) {
2526                 *((__u32 *)val) = 1;
2527                 RETURN(0);
2528         }
2529
2530         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2531
2532         RETURN(rc);
2533 }
2534
2535 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2536                      struct ptlrpc_request **request)
2537 {
2538         struct ptlrpc_request *req;
2539         int                    rc;
2540         ENTRY;
2541
2542         *request = NULL;
2543         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2544         if (req == NULL)
2545                 RETURN(-ENOMEM);
2546
2547         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2548         if (rc) {
2549                 ptlrpc_request_free(req);
2550                 RETURN(rc);
2551         }
2552
2553         mdc_pack_body(req, fid, 0, 0, -1, 0);
2554
2555         ptlrpc_request_set_replen(req);
2556
2557         rc = ptlrpc_queue_wait(req);
2558         if (rc)
2559                 ptlrpc_req_finished(req);
2560         else
2561                 *request = req;
2562         RETURN(rc);
2563 }
2564
2565 struct mdc_rmfid_args {
2566         int *mra_rcs;
2567         int mra_nr;
2568 };
2569
2570 int mdc_rmfid_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2571                           void *args, int rc)
2572 {
2573         struct mdc_rmfid_args *aa;
2574         int *rcs, size;
2575         ENTRY;
2576
2577         if (!rc) {
2578                 aa = ptlrpc_req_async_args(aa, req);
2579
2580                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2581                                             RCL_SERVER);
2582                 LASSERT(size == sizeof(int) * aa->mra_nr);
2583                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2584                 LASSERT(rcs);
2585                 LASSERT(aa->mra_rcs);
2586                 LASSERT(aa->mra_nr);
2587                 memcpy(aa->mra_rcs, rcs, size);
2588         }
2589
2590         RETURN(rc);
2591 }
2592
2593 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2594                      int *rcs, struct ptlrpc_request_set *set)
2595 {
2596         struct ptlrpc_request *req;
2597         struct mdc_rmfid_args *aa;
2598         struct mdt_body *b;
2599         struct lu_fid *tmp;
2600         int rc, flen;
2601         ENTRY;
2602
2603         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2604         if (req == NULL)
2605                 RETURN(-ENOMEM);
2606
2607         flen = fa->fa_nr * sizeof(struct lu_fid);
2608         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2609                              RCL_CLIENT, flen);
2610         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2611                              RCL_SERVER, flen);
2612         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2613                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2614         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2615         if (rc) {
2616                 ptlrpc_request_free(req);
2617                 RETURN(rc);
2618         }
2619         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2620         memcpy(tmp, fa->fa_fids, flen);
2621
2622         mdc_pack_body(req, NULL, 0, 0, -1, 0);
2623         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2624         b->mbo_ctime = ktime_get_real_seconds();
2625
2626         ptlrpc_request_set_replen(req);
2627
2628         LASSERT(rcs);
2629         aa = ptlrpc_req_async_args(aa, req);
2630         aa->mra_rcs = rcs;
2631         aa->mra_nr = fa->fa_nr;
2632         req->rq_interpret_reply = mdc_rmfid_interpret;
2633
2634         ptlrpc_set_add_req(set, req);
2635         ptlrpc_check_set(NULL, set);
2636
2637         RETURN(rc);
2638 }
2639
2640 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2641                             enum obd_import_event event)
2642 {
2643         struct client_obd *cli = &obd->u.cli;
2644         int rc = 0;
2645
2646         LASSERT(imp->imp_obd == obd);
2647
2648         switch (event) {
2649         case IMP_EVENT_DISCON:
2650                 spin_lock(&cli->cl_loi_list_lock);
2651                 cli->cl_avail_grant = 0;
2652                 cli->cl_lost_grant = 0;
2653                 spin_unlock(&cli->cl_loi_list_lock);
2654                 break;
2655         case IMP_EVENT_INACTIVE:
2656                 /*
2657                  * Flush current sequence to make client obtain new one
2658                  * from server in case of disconnect/reconnect.
2659                  */
2660                 down_read(&cli->cl_seq_rwsem);
2661                 if (cli->cl_seq)
2662                         seq_client_flush(cli->cl_seq);
2663                 up_read(&cli->cl_seq_rwsem);
2664
2665                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2666                 break;
2667         case IMP_EVENT_INVALIDATE: {
2668                 struct ldlm_namespace *ns = obd->obd_namespace;
2669                 struct lu_env *env;
2670                 __u16 refcheck;
2671
2672                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2673
2674                 env = cl_env_get(&refcheck);
2675                 if (!IS_ERR(env)) {
2676                         /* Reset grants. All pages go to failing rpcs due to
2677                          * the invalid import.
2678                          */
2679                         osc_io_unplug(env, cli, NULL);
2680
2681                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2682                                                  osc_ldlm_resource_invalidate,
2683                                                  env, 0);
2684                         cl_env_put(env, &refcheck);
2685                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2686                 } else {
2687                         rc = PTR_ERR(env);
2688                 }
2689                 break;
2690         }
2691         case IMP_EVENT_ACTIVE:
2692                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2693                 /* redo the kuc registration after reconnecting */
2694                 if (rc == 0)
2695                         rc = mdc_kuc_reregister(imp);
2696                 break;
2697         case IMP_EVENT_OCD: {
2698                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2699
2700                 if (OCD_HAS_FLAG(ocd, GRANT))
2701                         osc_init_grant(cli, ocd);
2702
2703                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2704                 break;
2705         }
2706         case IMP_EVENT_DEACTIVATE:
2707         case IMP_EVENT_ACTIVATE:
2708                 break;
2709         default:
2710                 CERROR("Unknown import event %x\n", event);
2711                 LBUG();
2712         }
2713         RETURN(rc);
2714 }
2715
2716 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2717                   struct lu_fid *fid, struct md_op_data *op_data)
2718 {
2719         struct client_obd *cli = &exp->exp_obd->u.cli;
2720         int rc = -EIO;
2721
2722         ENTRY;
2723
2724         down_read(&cli->cl_seq_rwsem);
2725         if (cli->cl_seq)
2726                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2727         up_read(&cli->cl_seq_rwsem);
2728
2729         RETURN(rc);
2730 }
2731
2732 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2733 {
2734         struct client_obd *cli = &exp->exp_obd->u.cli;
2735         return &cli->cl_target_uuid;
2736 }
2737
2738 /**
2739  * Determine whether the lock can be canceled before replaying it during
2740  * recovery, non zero value will be return if the lock can be canceled,
2741  * or zero returned for not
2742  */
2743 static int mdc_cancel_weight(struct ldlm_lock *lock)
2744 {
2745         if (lock->l_resource->lr_type != LDLM_IBITS)
2746                 RETURN(0);
2747
2748         /* FIXME: if we ever get into a situation where there are too many
2749          * opened files with open locks on a single node, then we really
2750          * should replay these open locks to reget it */
2751         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2752                 RETURN(0);
2753
2754         /* Special case for DoM locks, cancel only unused and granted locks */
2755         if (ldlm_has_dom(lock) &&
2756             (lock->l_granted_mode != lock->l_req_mode ||
2757              osc_ldlm_weigh_ast(lock) != 0))
2758                 RETURN(0);
2759
2760         RETURN(1);
2761 }
2762
2763 static int mdc_resource_inode_free(struct ldlm_resource *res)
2764 {
2765         if (res->lr_lvb_inode)
2766                 res->lr_lvb_inode = NULL;
2767
2768         return 0;
2769 }
2770
2771 static struct ldlm_valblock_ops inode_lvbo = {
2772         .lvbo_free = mdc_resource_inode_free
2773 };
2774
2775 static int mdc_llog_init(struct obd_device *obd)
2776 {
2777         struct obd_llog_group   *olg = &obd->obd_olg;
2778         struct llog_ctxt        *ctxt;
2779         int                      rc;
2780
2781         ENTRY;
2782
2783         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2784                         &llog_client_ops);
2785         if (rc < 0)
2786                 RETURN(rc);
2787
2788         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2789         llog_initiator_connect(ctxt);
2790         llog_ctxt_put(ctxt);
2791
2792         RETURN(0);
2793 }
2794
2795 static void mdc_llog_finish(struct obd_device *obd)
2796 {
2797         struct llog_ctxt *ctxt;
2798
2799         ENTRY;
2800
2801         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2802         if (ctxt != NULL)
2803                 llog_cleanup(NULL, ctxt);
2804
2805         EXIT;
2806 }
2807
2808 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2809 {
2810         int rc;
2811
2812         ENTRY;
2813
2814         rc = osc_setup_common(obd, cfg);
2815         if (rc < 0)
2816                 RETURN(rc);
2817
2818         rc = mdc_tunables_init(obd);
2819         if (rc)
2820                 GOTO(err_osc_cleanup, rc);
2821
2822         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2823
2824         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2825
2826         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2827
2828         rc = mdc_llog_init(obd);
2829         if (rc) {
2830                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2831                        obd->obd_name, rc);
2832                 GOTO(err_llog_cleanup, rc);
2833         }
2834
2835         rc = mdc_changelog_cdev_init(obd);
2836         if (rc) {
2837                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2838                        obd->obd_name, rc);
2839                 GOTO(err_changelog_cleanup, rc);
2840         }
2841
2842         RETURN(rc);
2843
2844 err_changelog_cleanup:
2845         mdc_llog_finish(obd);
2846 err_llog_cleanup:
2847         lprocfs_free_md_stats(obd);
2848         ptlrpc_lprocfs_unregister_obd(obd);
2849 err_osc_cleanup:
2850         osc_cleanup_common(obd);
2851         return rc;
2852 }
2853
2854 /* Initialize the default and maximum LOV EA sizes.  This allows
2855  * us to make MDS RPCs with large enough reply buffers to hold a default
2856  * sized EA without having to calculate this (via a call into the
2857  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2858  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2859  * a large number of stripes is possible.  If a larger reply buffer is
2860  * required it will be reallocated in the ptlrpc layer due to overflow.
2861  */
2862 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2863                             __u32 def_easize)
2864 {
2865         struct obd_device *obd = exp->exp_obd;
2866         struct client_obd *cli = &obd->u.cli;
2867         ENTRY;
2868
2869         if (cli->cl_max_mds_easize < easize)
2870                 cli->cl_max_mds_easize = easize;
2871
2872         if (cli->cl_default_mds_easize < def_easize)
2873                 cli->cl_default_mds_easize = def_easize;
2874
2875         RETURN(0);
2876 }
2877
2878 static int mdc_precleanup(struct obd_device *obd)
2879 {
2880         ENTRY;
2881
2882         osc_precleanup_common(obd);
2883         mdc_changelog_cdev_finish(obd);
2884
2885         obd_cleanup_client_import(obd);
2886         ptlrpc_lprocfs_unregister_obd(obd);
2887         lprocfs_free_md_stats(obd);
2888         mdc_llog_finish(obd);
2889         RETURN(0);
2890 }
2891
2892 static int mdc_cleanup(struct obd_device *obd)
2893 {
2894         return osc_cleanup_common(obd);
2895 }
2896
2897 static const struct obd_ops mdc_obd_ops = {
2898         .o_owner            = THIS_MODULE,
2899         .o_setup            = mdc_setup,
2900         .o_precleanup       = mdc_precleanup,
2901         .o_cleanup          = mdc_cleanup,
2902         .o_add_conn         = client_import_add_conn,
2903         .o_del_conn         = client_import_del_conn,
2904         .o_connect          = client_connect_import,
2905         .o_reconnect        = osc_reconnect,
2906         .o_disconnect       = osc_disconnect,
2907         .o_iocontrol        = mdc_iocontrol,
2908         .o_set_info_async   = mdc_set_info_async,
2909         .o_statfs           = mdc_statfs,
2910         .o_statfs_async     = mdc_statfs_async,
2911         .o_fid_init         = client_fid_init,
2912         .o_fid_fini         = client_fid_fini,
2913         .o_fid_alloc        = mdc_fid_alloc,
2914         .o_import_event     = mdc_import_event,
2915         .o_get_info         = mdc_get_info,
2916         .o_get_uuid         = mdc_get_uuid,
2917         .o_quotactl         = mdc_quotactl,
2918 };
2919
2920 static const struct md_ops mdc_md_ops = {
2921         .m_get_root         = mdc_get_root,
2922         .m_null_inode       = mdc_null_inode,
2923         .m_close            = mdc_close,
2924         .m_create           = mdc_create,
2925         .m_enqueue          = mdc_enqueue,
2926         .m_getattr          = mdc_getattr,
2927         .m_getattr_name     = mdc_getattr_name,
2928         .m_intent_lock      = mdc_intent_lock,
2929         .m_link             = mdc_link,
2930         .m_rename           = mdc_rename,
2931         .m_setattr          = mdc_setattr,
2932         .m_setxattr         = mdc_setxattr,
2933         .m_getxattr         = mdc_getxattr,
2934         .m_fsync                = mdc_fsync,
2935         .m_file_resync          = mdc_file_resync,
2936         .m_read_page            = mdc_read_page,
2937         .m_unlink           = mdc_unlink,
2938         .m_cancel_unused    = mdc_cancel_unused,
2939         .m_init_ea_size     = mdc_init_ea_size,
2940         .m_set_lock_data    = mdc_set_lock_data,
2941         .m_lock_match       = mdc_lock_match,
2942         .m_get_lustre_md    = mdc_get_lustre_md,
2943         .m_free_lustre_md   = mdc_free_lustre_md,
2944         .m_set_open_replay_data = mdc_set_open_replay_data,
2945         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2946         .m_intent_getattr_async = mdc_intent_getattr_async,
2947         .m_revalidate_lock      = mdc_revalidate_lock,
2948         .m_rmfid                = mdc_rmfid,
2949 };
2950
2951 static int __init mdc_init(void)
2952 {
2953         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2954                                    LUSTRE_MDC_NAME, &mdc_device_type);
2955 }
2956
2957 static void __exit mdc_exit(void)
2958 {
2959         class_unregister_type(LUSTRE_MDC_NAME);
2960 }
2961
2962 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2963 MODULE_DESCRIPTION("Lustre Metadata Client");
2964 MODULE_VERSION(LUSTRE_VERSION_STRING);
2965 MODULE_LICENSE("GPL");
2966
2967 module_init(mdc_init);
2968 module_exit(mdc_exit);