lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_MDC
  41
  42 #ifdef __KERNEL__
  43 # include <linux/module.h>
  44 # include <linux/pagemap.h>
  45 # include <linux/miscdevice.h>
  46 # include <linux/init.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <obd_class.h>
  52 #include <lustre_dlm.h>
  53 #include <lprocfs_status.h>
  54 #include <lustre_param.h>
  55 #include "mdc_internal.h"
  56
  57 static quota_interface_t *quota_interface;
  58
  59 #define REQUEST_MINOR 244
  60
  61 static quota_interface_t *quota_interface;
  62 extern quota_interface_t mdc_quota_interface;
  63
  64 static int mdc_cleanup(struct obd_device *obd);
  65
  66 extern int mds_queue_req(struct ptlrpc_request *);
  67 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  68 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  69 static int send_getstatus(struct obd_export *exp, struct ll_fid *rootfid,
  70                           int level, int msg_flags)
  71 {
  72         struct ptlrpc_request *req;
  73         struct mds_body *body;
  74         __u32 size[3] = { sizeof(struct ptlrpc_body),
  75                           sizeof(struct mdt_body),
  76                           sizeof(struct lustre_capa) };
  77         int rc;
  78         ENTRY;
  79
  80         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
  81                               NULL);
  82         if (!req)
  83                 GOTO(out, rc = -ENOMEM);
  84
  85         req->rq_export = class_export_get(exp);
  86         req->rq_send_state = level;
  87         ptlrpc_req_set_repsize(req, 3, size);
  88
  89         mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, 0, 0);
  90         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
  91         rc = ptlrpc_queue_wait(req);
  92
  93         if (!rc) {
  94                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
  95                                           lustre_swab_mds_body);
  96                 if (body == NULL) {
  97                         CERROR ("Can't extract mds_body\n");
  98                         GOTO (out, rc = -EPROTO);
  99                 }
 100
 101                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
 102
 103                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
 104                        ", last_xid="LPU64"\n",
 105                        rootfid->id,
 106                        lustre_msg_get_last_committed(req->rq_repmsg),
 107                        lustre_msg_get_last_xid(req->rq_repmsg));
 108         }
 109
 110         EXIT;
 111  out:
 112         ptlrpc_req_finished(req);
 113         return rc;
 114 }
 115
 116 /* This should be mdc_get_info("ROOT") */
 117 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
 118 {
 119         return send_getstatus(exp, rootfid, LUSTRE_IMP_FULL, 0);
 120 }
 121
 122 static
 123 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 124                        unsigned int acl_size, struct ptlrpc_request *req)
 125 {
 126         struct obd_device *obddev = class_exp2obd(exp);
 127         struct mds_body *body;
 128         void *eadata;
 129         __u32 size[6] = { sizeof(struct ptlrpc_body),
 130                           sizeof(struct mdt_body) };
 131         int bufcount = 2, rc;
 132         ENTRY;
 133
 134         /* request message already built */
 135         if (ea_size != 0) {
 136                 size[bufcount++] = ea_size;
 137                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 138                        ea_size);
 139         }
 140         if (acl_size) {
 141                 size[bufcount++] = acl_size;
 142                 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
 143         }
 144
 145         if (mdc_exp_is_2_0_server(exp)) {
 146                 bufcount = 6;
 147         }
 148
 149         ptlrpc_req_set_repsize(req, bufcount, size);
 150
 151         mdc_enter_request(&obddev->u.cli);
 152         rc = ptlrpc_queue_wait(req);
 153         mdc_exit_request(&obddev->u.cli);
 154         if (rc != 0)
 155                 RETURN (rc);
 156
 157         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 158                                   lustre_swab_mds_body);
 159         if (body == NULL) {
 160                 CERROR ("Can't unpack mds_body\n");
 161                 RETURN (-EPROTO);
 162         }
 163
 164         CDEBUG(D_NET, "mode: %o\n", body->mode);
 165
 166         lustre_set_rep_swabbed(req, REPLY_REC_OFF + 1);
 167         if (body->eadatasize != 0) {
 168                 /* reply indicates presence of eadata; check it's there... */
 169                 eadata = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 170                                         body->eadatasize);
 171                 if (eadata == NULL) {
 172                         CERROR ("Missing/short eadata\n");
 173                         RETURN (-EPROTO);
 174                 }
 175         }
 176
 177         RETURN (0);
 178 }
 179
 180 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
 181                 obd_valid valid, unsigned int ea_size,
 182                 struct ptlrpc_request **request)
 183 {
 184         struct ptlrpc_request *req;
 185         __u32 size[2] = { sizeof(struct ptlrpc_body),
 186                           sizeof(struct mdt_body) };
 187         int acl_size = 0, rc;
 188         ENTRY;
 189
 190         /* XXX do we need to make another request here?  We just did a getattr
 191          *     to do the lookup in the first place.
 192          */
 193         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 194                               MDS_GETATTR, 2, size, NULL);
 195         if (!req)
 196                 GOTO(out, rc = -ENOMEM);
 197
 198         req->rq_export = class_export_get(exp);
 199         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
 200                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 201
 202         /* currently only root inode will call us with FLACL */
 203         if (valid & OBD_MD_FLACL)
 204                 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
 205
 206         rc = mdc_getattr_common(exp, ea_size, acl_size, req);
 207         if (rc != 0) {
 208                 ptlrpc_req_finished (req);
 209                 req = NULL;
 210         }
 211  out:
 212         *request = req;
 213         RETURN (rc);
 214 }
 215
 216 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 217                      const char *filename, int namelen, unsigned long valid,
 218                      unsigned int ea_size, struct ptlrpc_request **request)
 219 {
 220         struct ptlrpc_request *req;
 221         __u32 size[4] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
 222                           [REQ_REC_OFF] = sizeof(struct mdt_body),
 223                           [REQ_REC_OFF + 1] = namelen };
 224         int rc;
 225         int bufcount = 3;
 226         int nameoffset = REQ_REC_OFF + 1;
 227         ENTRY;
 228
 229         if (mdc_exp_is_2_0_server(exp)) {
 230                 size[REQ_REC_OFF + 1] = 0;
 231                 size[REQ_REC_OFF + 2] = namelen;
 232                 bufcount ++;
 233                 nameoffset ++;
 234         }
 235
 236         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 237                               MDS_GETATTR_NAME, bufcount, size, NULL);
 238         if (!req)
 239                 GOTO(out, rc = -ENOMEM);
 240
 241         req->rq_export = class_export_get(exp);
 242         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
 243                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 244
 245         LASSERT(strnlen(filename, namelen) == namelen - 1);
 246         memcpy(lustre_msg_buf(req->rq_reqmsg, nameoffset, namelen),
 247                filename, namelen);
 248
 249         rc = mdc_getattr_common(exp, ea_size, 0, req);
 250         if (rc != 0) {
 251                 ptlrpc_req_finished (req);
 252                 req = NULL;
 253         }
 254  out:
 255         *request = req;
 256         RETURN(rc);
 257 }
 258
 259 static
 260 int mdc_xattr_common(struct obd_export *exp, struct ll_fid *fid,
 261                      int opcode, obd_valid valid, const char *xattr_name,
 262                      const char *input, int input_size, int output_size,
 263                      int flags, struct ptlrpc_request **request)
 264 {
 265         struct obd_device *obddev = class_exp2obd(exp);
 266         struct ptlrpc_request *req;
 267         __u32 size[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
 268                         [REQ_REC_OFF] = sizeof(struct mdt_body),
 269                         [REQ_REC_OFF + 1] = 0, /* capa */
 270                         [REQ_REC_OFF + 2] = 0, /* name */
 271                         [REQ_REC_OFF + 3] = 0 };
 272         int rc, xattr_namelen = 0, bufcnt = 2, offset = REQ_REC_OFF + 1;
 273         void *tmp;
 274         ENTRY;
 275
 276         if (mdc_exp_is_2_0_server(exp)) {
 277                 bufcnt++;
 278                 offset++;
 279                 if (opcode == MDS_SETXATTR) {
 280                         size[REQ_REC_OFF] = sizeof (struct mdt_rec_setxattr);
 281                         opcode = MDS_REINT;
 282                 }
 283         }
 284
 285         if (xattr_name) {
 286                 xattr_namelen = strlen(xattr_name) + 1;
 287                 size[bufcnt++] = xattr_namelen;
 288         }
 289         if (input_size) {
 290                 LASSERT(input);
 291                 size[bufcnt++] = input_size;
 292         }
 293
 294         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 295                               opcode, bufcnt, size, NULL);
 296         if (!req)
 297                 GOTO(out, rc = -ENOMEM);
 298
 299         req->rq_export = class_export_get(exp);
 300
 301         if (opcode == MDS_REINT && mdc_exp_is_2_0_server(exp)) {
 302                 struct mdt_rec_setxattr *rec;
 303                 rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
 304                                      sizeof(struct mdt_rec_setxattr));
 305                 rec->sx_opcode = REINT_SETXATTR;
 306                 rec->sx_fsuid  = current->fsuid;
 307                 rec->sx_fsgid  = current->fsgid;
 308                 rec->sx_cap    = cfs_curproc_cap_pack();
 309                 rec->sx_suppgid1 = -1;
 310                 rec->sx_suppgid2 = -1;
 311                 rec->sx_fid    = *((struct lu_fid*)fid);
 312                 rec->sx_valid  = valid;
 313                 rec->sx_size   = output_size;
 314                 rec->sx_flags  = flags;
 315         } else {
 316                 /* request data */
 317                 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, output_size, flags);
 318         }
 319
 320         if (xattr_name) {
 321                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
 322                 memcpy(tmp, xattr_name, xattr_namelen);
 323         }
 324         if (input_size) {
 325                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
 326                 memcpy(tmp, input, input_size);
 327         }
 328
 329         size[REPLY_REC_OFF] = sizeof(struct mdt_body);
 330         if (mdc_exp_is_2_0_server(exp)) {
 331                 bufcnt = 2;
 332         } else {
 333                 /* reply buffers */
 334                 if (opcode == MDS_GETXATTR) {
 335                         bufcnt = 2;
 336                 } else {
 337                         bufcnt = 1;
 338                 }
 339
 340         }
 341
 342         /* we do this even output_size is 0, because server is doing that */
 343         size[bufcnt++] = output_size;
 344         ptlrpc_req_set_repsize(req, bufcnt, size);
 345
 346         /* make rpc */
 347         if (opcode == MDS_SETXATTR || opcode == MDS_REINT)
 348                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 349         else
 350                 mdc_enter_request(&obddev->u.cli);
 351
 352         rc = ptlrpc_queue_wait(req);
 353
 354         if (opcode == MDS_SETXATTR || opcode == MDS_REINT)
 355                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 356         else
 357                 mdc_exit_request(&obddev->u.cli);
 358
 359         if (rc != 0)
 360                 GOTO(err_out, rc);
 361
 362         if (opcode == MDS_GETXATTR) {
 363                 struct mds_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
 364                                           sizeof(*body),
 365                                           lustre_swab_mds_body);
 366                 if (body == NULL) {
 367                         CERROR ("Can't unpack mds_body\n");
 368                         GOTO(err_out, rc = -EPROTO);
 369                 }
 370         }
 371 out:
 372         *request = req;
 373         RETURN (rc);
 374 err_out:
 375         ptlrpc_req_finished(req);
 376         req = NULL;
 377         goto out;
 378 }
 379
 380 int mdc_setxattr(struct obd_export *exp, struct ll_fid *fid,
 381                  obd_valid valid, const char *xattr_name,
 382                  const char *input, int input_size,
 383                  int output_size, int flags,
 384                  struct ptlrpc_request **request)
 385 {
 386         return mdc_xattr_common(exp, fid, MDS_SETXATTR, valid, xattr_name,
 387                                 input, input_size, output_size, flags, request);
 388 }
 389
 390 int mdc_getxattr(struct obd_export *exp, struct ll_fid *fid,
 391                  obd_valid valid, const char *xattr_name,
 392                  const char *input, int input_size,
 393                  int output_size, struct ptlrpc_request **request)
 394 {
 395         return mdc_xattr_common(exp, fid, MDS_GETXATTR, valid, xattr_name,
 396                                 input, input_size, output_size, 0, request);
 397 }
 398
 399 /* For the fid-less server */
 400 static void mdc_store_inode_generation_18(struct ptlrpc_request *req,
 401                                           int reqoff, int repoff)
 402 {
 403         struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff,
 404                                                     sizeof(*rec));
 405         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff,
 406                                                sizeof(*body));
 407
 408         LASSERT (rec != NULL);
 409         LASSERT (body != NULL);
 410
 411         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 412         if (body->fid1.id == 0) {
 413                 DEBUG_REQ(D_ERROR, req, "saving replay request with id = 0 "
 414                           "gen = %u", body->fid1.generation);
 415                 LBUG();
 416         }
 417
 418         DEBUG_REQ(D_INODE, req, "storing generation %u for ino "LPU64,
 419                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 420 }
 421
 422 static void mdc_store_inode_generation_20(struct ptlrpc_request *req,
 423                                           int reqoff, int repoff)
 424 {
 425         struct mdt_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff,
 426                                                     sizeof(*rec));
 427         struct mdt_body *body = lustre_msg_buf(req->rq_repmsg, repoff,
 428                                                sizeof(*body));
 429
 430         LASSERT (rec != NULL);
 431         LASSERT (body != NULL);
 432
 433         rec->cr_fid2 = body->fid1;
 434         rec->cr_ioepoch = body->ioepoch;
 435         rec->cr_old_handle.cookie = body->handle.cookie;
 436
 437         if (!fid_is_sane(&body->fid1)) {
 438                 DEBUG_REQ(D_ERROR, req, "saving replay request with"
 439                           "insane fid");
 440                 LBUG();
 441         }
 442
 443         DEBUG_REQ(D_INODE, req, "storing generation %u for ino "LPU64,
 444                   rec->cr_fid1.f_oid, rec->cr_fid2.f_seq);
 445 }
 446
 447 /* This should be called with both the request and the reply still packed. */
 448 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
 449                                 int repoff)
 450 {
 451         if (mdc_req_is_2_0_server(req))
 452                 mdc_store_inode_generation_20(req, reqoff, repoff);
 453         else
 454                 mdc_store_inode_generation_18(req, reqoff, repoff);
 455 }
 456
 457 #ifdef CONFIG_FS_POSIX_ACL
 458 static
 459 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
 460                    struct lustre_md *md, unsigned int offset)
 461 {
 462         struct mds_body  *body = md->body;
 463         struct posix_acl *acl;
 464         void             *buf;
 465         int               rc;
 466
 467         if (!body->aclsize)
 468                 return 0;
 469
 470         buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
 471         if (!buf) {
 472                 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
 473                        body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
 474                        (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
 475                                 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
 476                 return -EPROTO;
 477         }
 478
 479         acl = posix_acl_from_xattr(buf, body->aclsize);
 480         if (IS_ERR(acl)) {
 481                 rc = PTR_ERR(acl);
 482                 CERROR("convert xattr to acl: %d\n", rc);
 483                 return rc;
 484         }
 485
 486         rc = posix_acl_valid(acl);
 487         if (rc) {
 488                 CERROR("validate acl: %d\n", rc);
 489                 posix_acl_release(acl);
 490                 return rc;
 491         }
 492
 493         md->posix_acl = acl;
 494         return 0;
 495 }
 496 #else
 497 #define mdc_unpack_acl(exp, req, md, offset) 0
 498 #endif
 499
 500 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
 501                       struct obd_export *exp,
 502                       struct lustre_md *md)
 503 {
 504         int rc = 0;
 505         int iop = mdc_req_is_2_0_server(req);
 506         ENTRY;
 507
 508         LASSERT(md);
 509         memset(md, 0, sizeof(*md));
 510
 511         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 512         LASSERT (md->body != NULL);
 513         LASSERT(lustre_rep_swabbed(req, offset));
 514         offset++;
 515
 516         if (md->body->valid & OBD_MD_FLEASIZE) {
 517                 int lmmsize;
 518                 struct lov_mds_md *lmm;
 519
 520                 if (!S_ISREG(md->body->mode)) {
 521                         CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
 522                                "but is not\n");
 523                         GOTO(err_out, rc = -EPROTO);
 524                 }
 525
 526                 if (md->body->eadatasize == 0) {
 527                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 528                         GOTO(err_out, rc = -EPROTO);
 529                 }
 530                 lmmsize = md->body->eadatasize;
 531                 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
 532                 if (!lmm) {
 533                         CERROR ("incorrect message: lmm == 0\n");
 534                         GOTO(err_out, rc = -EPROTO);
 535                 }
 536                 LASSERT(lustre_rep_swabbed(req, offset));
 537
 538                 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
 539                 if (rc < 0)
 540                         GOTO(err_out, rc);
 541
 542                 if (rc < sizeof(*md->lsm)) {
 543                         CERROR ("lsm size too small:  rc < sizeof (*md->lsm) "
 544                                 "(%d < %d)\n", rc, (int)sizeof(*md->lsm));
 545                         GOTO(err_out, rc = -EPROTO);
 546                 }
 547                 rc = 0;
 548
 549                 if (!iop)
 550                         offset++;
 551         } else if (md->body->valid & OBD_MD_FLDIREA) {
 552                 if(!S_ISDIR(md->body->mode)) {
 553                         CERROR("OBD_MD_FLDIREA set, should be a directory, but "
 554                                "is not\n");
 555                         GOTO(err_out, rc = -EPROTO);
 556                 }
 557                 if (!iop)
 558                         offset++;
 559         }
 560         if (iop)
 561                 offset++;
 562
 563         /* for ACL, it's possible that FLACL is set but aclsize is zero.
 564          * only when aclsize != 0 there's an actual segment for ACL in
 565          * reply buffer.
 566          */
 567         if ((md->body->valid & OBD_MD_FLACL) && md->body->aclsize) {
 568                 rc = mdc_unpack_acl(exp, req, md, offset);
 569                 if (rc)
 570                         GOTO(err_out, rc);
 571                 offset++;
 572         }
 573 out:
 574         RETURN(rc);
 575
 576 err_out:
 577         if (md->lsm)
 578                 obd_free_memmd(exp, &md->lsm);
 579         goto out;
 580 }
 581
 582 void mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 583 {
 584         if (md->lsm)
 585                 obd_free_memmd(exp, &md->lsm);
 586
 587 #ifdef CONFIG_FS_POSIX_ACL
 588         if (md->posix_acl) {
 589                 posix_acl_release(md->posix_acl);
 590                 md->posix_acl = NULL;
 591         }
 592 #endif
 593 }
 594
 595 static void mdc_commit_open(struct ptlrpc_request *req)
 596 {
 597         struct mdc_open_data *mod = req->rq_cb_data;
 598         if (mod == NULL)
 599                 return;
 600
 601         if (mod->mod_close_req != NULL)
 602                 mod->mod_close_req->rq_cb_data = NULL;
 603
 604         if (mod->mod_och != NULL)
 605                 mod->mod_och->och_mod = NULL;
 606
 607         OBD_FREE_PTR(mod);
 608         req->rq_cb_data = NULL;
 609 }
 610
 611 static void mdc_replay_open(struct ptlrpc_request *req)
 612 {
 613         struct mdc_open_data *mod = req->rq_cb_data;
 614         struct obd_client_handle *och;
 615         struct ptlrpc_request *close_req;
 616         struct lustre_handle old;
 617         struct mds_body *body;
 618         ENTRY;
 619
 620         body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
 621                                   lustre_swab_mds_body);
 622         LASSERT (body != NULL);
 623
 624         if (mod == NULL) {
 625                 DEBUG_REQ(D_ERROR, req,
 626                           "can't properly replay without open data");
 627                 EXIT;
 628                 return;
 629         }
 630         DEBUG_REQ(D_HA, req, "mdc open data found");
 631
 632         och = mod->mod_och;
 633         if (och != NULL) {
 634                 struct lustre_handle *file_fh;
 635                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 636                 file_fh = &och->och_fh;
 637                 CDEBUG(D_RPCTRACE, "updating handle from "LPX64" to "LPX64"\n",
 638                        file_fh->cookie, body->handle.cookie);
 639                 old = *file_fh;
 640                 *file_fh = body->handle;
 641         }
 642
 643         close_req = mod->mod_close_req;
 644
 645         if (close_req != NULL) {
 646                 LASSERT(lustre_msg_get_opc(close_req->rq_reqmsg) == MDS_CLOSE);
 647                 if (mdc_req_is_2_0_server(close_req)) {
 648                         struct mdt_epoch *epoch = NULL;
 649
 650                         epoch = lustre_msg_buf(close_req->rq_reqmsg,
 651                                                REQ_REC_OFF, sizeof(*epoch));
 652                         LASSERT(epoch);
 653                         if (och != NULL)
 654                                 LASSERT(!memcmp(&old, &epoch->handle,
 655                                         sizeof(old)));
 656                         DEBUG_REQ(D_RPCTRACE, close_req,
 657                                   "updating close with new fh");
 658                         epoch->handle = body->handle;
 659                  } else {
 660                         struct mds_body *close_body = NULL;
 661
 662                         close_body = lustre_msg_buf(close_req->rq_reqmsg,
 663                                                     REQ_REC_OFF,
 664                                                     sizeof(*close_body));
 665                         if (och != NULL)
 666                                 LASSERT(!memcmp(&old, &close_body->handle,
 667                                         sizeof(old)));
 668                         DEBUG_REQ(D_RPCTRACE, close_req,
 669                                   "updating close with new fh");
 670                         close_body->handle = body->handle;
 671                  }
 672         }
 673
 674         EXIT;
 675 }
 676
 677 static void mdc_set_open_replay_data_20(struct obd_client_handle *och,
 678                                         struct ptlrpc_request *open_req)
 679 {
 680         struct mdc_open_data  *mod;
 681         struct obd_import     *imp = open_req->rq_import;
 682         struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
 683                                                     DLM_INTENT_REC_OFF,
 684                                                     sizeof(*rec));
 685         struct mdt_body       *body = lustre_msg_buf(open_req->rq_repmsg,
 686                                                      DLM_REPLY_REC_OFF,
 687                                                      sizeof(*body));
 688
 689         /* If request is not eligible for replay, just bail out */
 690         if (!open_req->rq_replay)
 691                 return;
 692
 693         /* incoming message in my byte order (it's been swabbed) */
 694         LASSERT(rec != NULL);
 695         LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
 696         /* outgoing messages always in my byte order */
 697         LASSERT(body != NULL);
 698
 699         /* Only if the import is replayable, we set replay_open data */
 700         if (och && imp->imp_replayable) {
 701                 OBD_ALLOC_PTR(mod);
 702                 if (mod == NULL) {
 703                         DEBUG_REQ(D_ERROR, open_req,
 704                                   "can't allocate mdc_open_data");
 705                         return;
 706                 }
 707
 708                 spin_lock(&open_req->rq_lock);
 709                 och->och_mod = mod;
 710                 mod->mod_och = och;
 711                 mod->mod_open_req = open_req;
 712                 open_req->rq_cb_data = mod;
 713                 open_req->rq_commit_cb = mdc_commit_open;
 714                 spin_unlock(&open_req->rq_lock);
 715         }
 716
 717         rec->cr_fid2 = body->fid1;
 718         rec->cr_ioepoch = body->ioepoch;
 719         rec->cr_old_handle.cookie = body->handle.cookie;
 720         open_req->rq_replay_cb = mdc_replay_open;
 721         if (!fid_is_sane(&body->fid1)) {
 722                 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
 723                           "insane fid");
 724                 LBUG();
 725         }
 726
 727         DEBUG_REQ(D_RPCTRACE, open_req, "set up replay data");
 728 }
 729
 730 static void mdc_set_open_replay_data_18(struct obd_client_handle *och,
 731                                         struct ptlrpc_request *open_req)
 732 {
 733         struct mdc_open_data *mod;
 734         struct mds_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
 735                                                     DLM_INTENT_REC_OFF,
 736                                                     sizeof(*rec));
 737         struct mds_body *body = lustre_msg_buf(open_req->rq_repmsg,
 738                                                DLM_REPLY_REC_OFF,
 739                                                sizeof(*body));
 740
 741         /* If request is not eligible for replay, just bail out */
 742         if (!open_req->rq_replay)
 743                 return;
 744
 745         /* incoming message in my byte order (it's been swabbed) */
 746         LASSERT(rec != NULL);
 747         LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
 748         /* outgoing messages always in my byte order */
 749         LASSERT(body != NULL);
 750
 751         if (och) {
 752                 OBD_ALLOC(mod, sizeof(*mod));
 753                 if (mod == NULL) {
 754                         DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 755                         return;
 756                 }
 757
 758                 spin_lock(&open_req->rq_lock);
 759                 och->och_mod = mod;
 760                 mod->mod_och = och;
 761                 mod->mod_open_req = open_req;
 762                 open_req->rq_cb_data = mod;
 763                 open_req->rq_commit_cb = mdc_commit_open;
 764                 spin_unlock(&open_req->rq_lock);
 765         }
 766
 767         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 768         open_req->rq_replay_cb = mdc_replay_open;
 769         if (body->fid1.id == 0) {
 770                 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
 771                           "id = 0 gen = %u", body->fid1.generation);
 772                 LBUG();
 773         }
 774
 775         DEBUG_REQ(D_RPCTRACE, open_req, "set up replay data");
 776 }
 777
 778 void mdc_set_open_replay_data(struct obd_client_handle *och,
 779                               struct ptlrpc_request *open_req)
 780 {
 781         if (mdc_req_is_2_0_server(open_req))
 782                 mdc_set_open_replay_data_20(och, open_req);
 783         else
 784                 mdc_set_open_replay_data_18(och, open_req);
 785 }
 786
 787 void mdc_clear_open_replay_data(struct obd_client_handle *och)
 788 {
 789         struct mdc_open_data *mod = och->och_mod;
 790
 791         /* Don't free the structure now (it happens in mdc_commit_open, after
 792          * we're sure we won't need to fix up the close request in the future),
 793          * but make sure that replay doesn't poke at the och, which is about to
 794          * be freed. */
 795         LASSERT(mod != LP_POISON);
 796         if (mod != NULL)
 797                 mod->mod_och = NULL;
 798         och->och_mod = NULL;
 799 }
 800
 801 int mdc_close(struct obd_export *exp, struct mdc_op_data *data, struct obdo *oa,
 802               struct obd_client_handle *och, struct ptlrpc_request **request)
 803 {
 804         struct obd_device *obd = class_exp2obd(exp);
 805         __u32 reqsize[4] = { sizeof(struct ptlrpc_body),
 806                              sizeof(struct mdt_body) };
 807         __u32 repsize[6] = { sizeof(struct ptlrpc_body),
 808                              sizeof(struct mdt_body),
 809                              obd->u.cli.cl_max_mds_easize,
 810                              obd->u.cli.cl_max_mds_cookiesize,
 811                              sizeof(struct lustre_capa),
 812                              sizeof(struct lustre_capa) };
 813         int rc;
 814         struct ptlrpc_request *req;
 815         int bufcount = 2;
 816         ENTRY;
 817
 818         if (mdc_exp_is_2_0_server(exp)) {
 819                 reqsize[1] = sizeof(struct mdt_epoch);
 820                 reqsize[2] = sizeof(struct mdt_rec_create);
 821                 reqsize[3] = 0; /* capa */
 822                 bufcount = 4;
 823         }
 824         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 825                               MDS_CLOSE, bufcount, reqsize, NULL);
 826         if (req == NULL)
 827                 GOTO(out, rc = -ENOMEM);
 828         req->rq_export = class_export_get(exp);
 829
 830         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 831          * portal whose threads are not taking any DLM locks and are therefore
 832          * always progressing */
 833         req->rq_request_portal = MDS_READPAGE_PORTAL;
 834         ptlrpc_at_set_req_timeout(req);
 835
 836         /* Ensure that this close's handle is fixed up during replay. */
 837         LASSERT(och != NULL);
 838         LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 839         if (likely(och->och_mod != NULL)) {
 840                 struct ptlrpc_request *open_req = och->och_mod->mod_open_req;
 841
 842                 if (open_req->rq_type == LI_POISON) {
 843                         CERROR("LBUG POISONED open %p!\n", open_req);
 844                         LBUG();
 845                         ptlrpc_req_finished(req);
 846                         req = NULL;
 847                         GOTO(out, rc = -EIO);
 848                 }
 849                 och->och_mod->mod_close_req = req;
 850                 DEBUG_REQ(D_RPCTRACE, req, "close req");
 851                 DEBUG_REQ(D_RPCTRACE, open_req, "clear open replay");
 852
 853                 /* We no longer want to preserve this open for replay even
 854                  * though the open was committed. b=3632, b=3633 */
 855                 spin_lock(&open_req->rq_lock);
 856                 open_req->rq_replay = 0;
 857                 spin_unlock(&open_req->rq_lock);
 858         } else {
 859                 CDEBUG(D_RPCTRACE, "couldn't find open req; expecting error\n");
 860         }
 861
 862         mdc_close_pack(req, REQ_REC_OFF, data, oa, oa->o_valid, och);
 863
 864         ptlrpc_req_set_repsize(req, 6, repsize);
 865
 866         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 867         rc = ptlrpc_queue_wait(req);
 868         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 869
 870         if (req->rq_repmsg == NULL) {
 871                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 872                        req->rq_status);
 873                 if (rc == 0)
 874                         rc = req->rq_status ? req->rq_status : -EIO;
 875         } else if (rc == 0) {
 876                 rc = lustre_msg_get_status(req->rq_repmsg);
 877                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 878                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 879                                   "= %d", rc);
 880                         if (rc > 0)
 881                                 rc = -rc;
 882                 } else if (och->och_mod == NULL) {
 883                         CERROR("Unexpected: can't find mdc_open_data, but the "
 884                                "close succeeded.  Please tell <http://bugzilla.lustre.org/>.\n");
 885                 }
 886                 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
 887                                         sizeof(struct mds_body),
 888                                         lustre_swab_mds_body)) {
 889                         CERROR("Error unpacking mds_body\n");
 890                         rc = -EPROTO;
 891                 }
 892         }
 893
 894         EXIT;
 895         *request = req;
 896  out:
 897         if (rc != 0 && req && req->rq_commit_cb)
 898                 req->rq_commit_cb(req);
 899
 900         return rc;
 901 }
 902
 903 int mdc_done_writing(struct obd_export *exp, struct mdc_op_data *data,
 904                      struct obdo *obdo)
 905 {
 906         struct ptlrpc_request *req;
 907         struct mds_body *body;
 908         __u32 size[2] = { sizeof(struct ptlrpc_body),
 909                           sizeof(struct mdt_body) };
 910         int rc;
 911         ENTRY;
 912
 913         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 914                               MDS_DONE_WRITING, 2, size, NULL);
 915         if (req == NULL)
 916                 RETURN(-ENOMEM);
 917
 918         req->rq_export = class_export_get(exp);
 919         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 920         body->fid1 = data->fid1;
 921         body->size = obdo->o_size;
 922         body->blocks = obdo->o_blocks;
 923         body->flags = obdo->o_flags;
 924         body->valid = obdo->o_valid;
 925 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 926
 927         ptlrpc_req_set_repsize(req, 2, size);
 928
 929         rc = ptlrpc_queue_wait(req);
 930         ptlrpc_req_finished(req);
 931         RETURN(rc);
 932 }
 933
 934 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
 935                  struct page *page, struct ptlrpc_request **request)
 936 {
 937         struct obd_import *imp = class_exp2cliimp(exp);
 938         struct ptlrpc_request *req = NULL;
 939         struct ptlrpc_bulk_desc *desc = NULL;
 940         struct mds_body *body;
 941         __u32 size[2] = { sizeof(struct ptlrpc_body),
 942                           sizeof(struct mdt_body) };
 943         int rc;
 944         ENTRY;
 945
 946         CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
 947
 948         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
 949                               NULL);
 950         if (req == NULL)
 951                 GOTO(out, rc = -ENOMEM);
 952
 953         req->rq_export = class_export_get(exp);
 954         req->rq_request_portal = MDS_READPAGE_PORTAL;
 955         ptlrpc_at_set_req_timeout(req);
 956
 957         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 958         if (desc == NULL)
 959                 GOTO(out, rc = -ENOMEM);
 960         /* NB req now owns desc and will free it when it gets freed */
 961
 962         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
 963
 964         mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
 965
 966         ptlrpc_req_set_repsize(req, 2, size);
 967         rc = ptlrpc_queue_wait(req);
 968
 969         if (rc == 0) {
 970                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 971                                           lustre_swab_mds_body);
 972                 if (body == NULL) {
 973                         CERROR("Can't unpack mds_body\n");
 974                         GOTO(out, rc = -EPROTO);
 975                 }
 976
 977                 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
 978                         CERROR ("Unexpected # bytes transferred: %d"
 979                                 " (%lu expected)\n",
 980                                 req->rq_bulk->bd_nob_transferred,
 981                                 CFS_PAGE_SIZE);
 982                         GOTO (out, rc = -EPROTO);
 983                 }
 984         }
 985
 986         EXIT;
 987  out:
 988         *request = req;
 989         return rc;
 990 }
 991
 992
 993 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 994                          void *karg, void *uarg)
 995 {
 996         struct obd_device *obd = exp->exp_obd;
 997         struct obd_ioctl_data *data = karg;
 998         struct obd_import *imp = obd->u.cli.cl_import;
 999         struct llog_ctxt *ctxt;
1000         int rc;
1001         ENTRY;
1002
1003         if (!try_module_get(THIS_MODULE)) {
1004                 CERROR("Can't get module. Is it alive?");
1005                 return -EINVAL;
1006         }
1007         switch (cmd) {
1008         case OBD_IOC_CLIENT_RECOVER:
1009                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1010                 if (rc < 0)
1011                         GOTO(out, rc);
1012                 GOTO(out, rc = 0);
1013         case IOC_OSC_SET_ACTIVE:
1014                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1015                 GOTO(out, rc);
1016         case OBD_IOC_PARSE: {
1017                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1018                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1019                 llog_ctxt_put(ctxt);
1020                 GOTO(out, rc);
1021         }
1022 #ifdef __KERNEL__
1023         case OBD_IOC_LLOG_INFO:
1024         case OBD_IOC_LLOG_PRINT: {
1025                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1026                 rc = llog_ioctl(ctxt, cmd, data);
1027                 llog_ctxt_put(ctxt);
1028                 GOTO(out, rc);
1029         }
1030 #endif
1031         case OBD_IOC_POLL_QUOTACHECK:
1032                 rc = lquota_poll_check(quota_interface, exp,
1033                                        (struct if_quotacheck *)karg);
1034                 GOTO(out, rc);
1035         case OBD_IOC_PING_TARGET:
1036                 rc = ptlrpc_obd_ping(obd);
1037                 GOTO(out, rc);
1038         default:
1039                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1040                 GOTO(out, rc = -ENOTTY);
1041         }
1042 out:
1043         module_put(THIS_MODULE);
1044         return rc;
1045 }
1046
1047 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1048                        void *key, obd_count vallen, void *val,
1049                        struct ptlrpc_request_set *set)
1050 {
1051         struct obd_import *imp = class_exp2cliimp(exp);
1052         int rc = -EINVAL;
1053
1054         if (KEY_IS(KEY_INIT_RECOV)) {
1055                 if (vallen != sizeof(int))
1056                         RETURN(-EINVAL);
1057                 spin_lock(&imp->imp_lock);
1058                 imp->imp_initial_recov = *(int *)val;
1059                 spin_unlock(&imp->imp_lock);
1060
1061                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1062                        exp->exp_obd->obd_name, imp->imp_initial_recov);
1063                 RETURN(0);
1064         }
1065         /* Turn off initial_recov after we try all backup servers once */
1066         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1067                 if (vallen != sizeof(int))
1068                         RETURN(-EINVAL);
1069
1070                 spin_lock(&imp->imp_lock);
1071                 imp->imp_initial_recov_bk = *(int *)val;
1072                 if (imp->imp_initial_recov_bk)
1073                         imp->imp_initial_recov = 1;
1074                 spin_unlock(&imp->imp_lock);
1075
1076                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1077                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1078                 RETURN(0);
1079         }
1080         /* Accept the broken "read-only" key for 1.6.6 servers. b=17493 */
1081         if (KEY_IS(KEY_READONLY) || KEY_IS(KEY_READONLY_166COMPAT)) {
1082                 struct ptlrpc_request *req;
1083                 __u32 size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1084                 char *bufs[3] = { NULL, key, val };
1085
1086                 if (vallen != sizeof(int))
1087                         RETURN(-EINVAL);
1088
1089                 if (*((int *)val)) {
1090                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1091                         imp->imp_connect_data.ocd_connect_flags |=
1092                                 OBD_CONNECT_RDONLY;
1093                 } else {
1094                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1095                         imp->imp_connect_data.ocd_connect_flags &=
1096                                 ~OBD_CONNECT_RDONLY;
1097                 }
1098
1099                 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1100                                       3, size, bufs);
1101                 if (req == NULL)
1102                         RETURN(-ENOMEM);
1103
1104                 req->rq_export = class_export_get(exp);
1105                 ptlrpc_req_set_repsize(req, 1, NULL);
1106                 if (set) {
1107                         rc = 0;
1108                         ptlrpc_set_add_req(set, req);
1109                         ptlrpc_check_set(set);
1110                 } else {
1111                         rc = ptlrpc_queue_wait(req);
1112                         ptlrpc_req_finished(req);
1113                 }
1114
1115                 RETURN(rc);
1116         }
1117
1118         RETURN(rc);
1119 }
1120
1121 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1122                  __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1123 {
1124         int rc = -EINVAL;
1125
1126         if (KEY_IS(KEY_MAX_EASIZE)) {
1127                 int mdsize, *max_easize;
1128
1129                 if (*vallen != sizeof(int))
1130                         RETURN(-EINVAL);
1131                 mdsize = *(int*)val;
1132                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1133                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1134                 max_easize = val;
1135                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1136                 RETURN(0);
1137         }
1138         RETURN(rc);
1139 }
1140
1141 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1142                       __u64 max_age, __u32 flags)
1143 {
1144         struct ptlrpc_request *req;
1145         struct obd_statfs *msfs;
1146         struct obd_import     *imp = NULL;
1147         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1148         int rc;
1149         ENTRY;
1150
1151         /*Since the request might also come from lprocfs, so we need
1152          *sync this with client_disconnect_export Bug15684*/
1153         down_read(&obd->u.cli.cl_sem);
1154         if (obd->u.cli.cl_import)
1155                 imp = class_import_get(obd->u.cli.cl_import);
1156         up_read(&obd->u.cli.cl_sem);
1157         if (!imp)
1158                 RETURN(-ENODEV);
1159
1160
1161         /* We could possibly pass max_age in the request (as an absolute
1162          * timestamp or a "seconds.usec ago") so the target can avoid doing
1163          * extra calls into the filesystem if that isn't necessary (e.g.
1164          * during mount that would help a bit).  Having relative timestamps
1165          * is not so great if request processing is slow, while absolute
1166          * timestamps are not ideal because they need time synchronization. */
1167         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_STATFS, 1, NULL,
1168                               NULL);
1169         if (!req)
1170                 GOTO(output, rc = -ENOMEM);
1171
1172         ptlrpc_req_set_repsize(req, 2, size);
1173
1174         if (flags & OBD_STATFS_NODELAY) {
1175                 /* procfs requests not want stay in wait for avoid deadlock */
1176                 req->rq_no_resend = 1;
1177                 req->rq_no_delay = 1;
1178         }
1179
1180         rc = ptlrpc_queue_wait(req);
1181
1182         if (rc)
1183                 GOTO(out, rc);
1184
1185         msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1186                                   lustre_swab_obd_statfs);
1187         if (msfs == NULL) {
1188                 CERROR("Can't unpack obd_statfs\n");
1189                 GOTO(out, rc = -EPROTO);
1190         }
1191
1192         memcpy(osfs, msfs, sizeof(*msfs));
1193         EXIT;
1194 out:
1195         ptlrpc_req_finished(req);
1196 output:
1197         class_import_put(imp);
1198         return rc;
1199 }
1200
1201 static int mdc_pin(struct obd_export *exp, struct ll_fid *fid,
1202                    struct obd_client_handle *handle, int flag)
1203 {
1204         struct ptlrpc_request *req;
1205         struct mds_body *body;
1206         __u32 size[3] = { sizeof(struct ptlrpc_body),
1207                           sizeof(struct mdt_body), 0 };
1208         int rc;
1209         int bufcount = 2;
1210         ENTRY;
1211
1212         if (mdc_exp_is_2_0_server(exp))
1213                 bufcount = 3;
1214         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1215                               MDS_PIN, bufcount, size, NULL);
1216         if (req == NULL)
1217                 RETURN(-ENOMEM);
1218
1219         req->rq_export = class_export_get(exp);
1220         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1221         body->fid1 = *fid;
1222         body->flags = flag;
1223
1224         ptlrpc_req_set_repsize(req, 2, size);
1225
1226         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1227         rc = ptlrpc_queue_wait(req);
1228         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1229         if (rc) {
1230                 CERROR("pin failed: %d\n", rc);
1231                 ptlrpc_req_finished(req);
1232                 RETURN(rc);
1233         }
1234
1235         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1236                                   lustre_swab_mds_body);
1237         if (body == NULL) {
1238                 ptlrpc_req_finished(req);
1239                 RETURN(rc);
1240         }
1241
1242         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1243         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1244
1245         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1246         if (handle->och_mod == NULL) {
1247                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1248                 RETURN(-ENOMEM);
1249         }
1250         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1251
1252         RETURN(rc);
1253 }
1254
1255 static int mdc_unpin(struct obd_export *exp,
1256                      struct obd_client_handle *handle, int flag)
1257 {
1258         struct ptlrpc_request *req;
1259         struct mds_body *body;
1260         __u32 size[2] = { sizeof(struct ptlrpc_body),
1261                           sizeof(struct mdt_body) };
1262         int rc;
1263         ENTRY;
1264
1265         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1266                 RETURN(0);
1267
1268         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1269                               MDS_CLOSE, 2, size, NULL);
1270         if (req == NULL)
1271                 RETURN(-ENOMEM);
1272
1273         req->rq_export = class_export_get(exp);
1274         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1275         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1276         body->flags = flag;
1277
1278         ptlrpc_req_set_repsize(req, 1, NULL);
1279         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1280         rc = ptlrpc_queue_wait(req);
1281         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1282
1283         if (rc != 0)
1284                 CERROR("unpin failed: %d\n", rc);
1285
1286         ptlrpc_req_finished(req);
1287         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1288         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1289         RETURN(rc);
1290 }
1291
1292 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1293              struct ptlrpc_request **request)
1294 {
1295         struct ptlrpc_request *req;
1296         __u32 size[3] = { sizeof(struct ptlrpc_body),
1297                           sizeof(struct mdt_body), 0 };
1298         int bufcount = 2;
1299         int rc;
1300         ENTRY;
1301
1302
1303         if (mdc_exp_is_2_0_server(exp))
1304                 bufcount = 3;
1305         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1306                               MDS_SYNC, bufcount, size, NULL);
1307         if (!req)
1308                 RETURN(rc = -ENOMEM);
1309
1310         req->rq_export = class_export_get(exp);
1311         mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1312
1313         ptlrpc_req_set_repsize(req, 2, size);
1314
1315         rc = ptlrpc_queue_wait(req);
1316         if (rc || request == NULL)
1317                 ptlrpc_req_finished(req);
1318         else
1319                 *request = req;
1320
1321         RETURN(rc);
1322 }
1323
1324 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1325                             enum obd_import_event event)
1326 {
1327         int rc = 0;
1328
1329         LASSERT(imp->imp_obd == obd);
1330
1331         switch (event) {
1332         case IMP_EVENT_DISCON: {
1333                 ptlrpc_import_setasync(imp, -obd->obd_namespace->ns_max_unused);
1334                 break;
1335         }
1336         case IMP_EVENT_INACTIVE: {
1337                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1338                 break;
1339         }
1340         case IMP_EVENT_INVALIDATE: {
1341                 struct ldlm_namespace *ns = obd->obd_namespace;
1342
1343                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1344
1345                 break;
1346         }
1347         case IMP_EVENT_ACTIVE: {
1348                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1349                 break;
1350         }
1351         case IMP_EVENT_OCD:
1352                 ptlrpc_import_setasync(imp, obd->obd_namespace->ns_max_unused);
1353                 break;
1354
1355         default:
1356                 CERROR("Unknown import event %x\n", event);
1357                 LBUG();
1358         }
1359         RETURN(rc);
1360 }
1361
1362 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1363 {
1364         struct client_obd *cli = &obd->u.cli;
1365         struct lprocfs_static_vars lvars = { 0 };
1366         int rc;
1367         ENTRY;
1368
1369         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1370         if (!cli->cl_rpc_lock)
1371                 RETURN(-ENOMEM);
1372         mdc_init_rpc_lock(cli->cl_rpc_lock);
1373
1374         ptlrpcd_addref();
1375
1376         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1377         if (!cli->cl_setattr_lock)
1378                 GOTO(err_rpc_lock, rc = -ENOMEM);
1379         mdc_init_rpc_lock(cli->cl_setattr_lock);
1380
1381         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1382         if (!cli->cl_close_lock)
1383                 GOTO(err_setattr_lock, rc = -ENOMEM);
1384         mdc_init_rpc_lock(cli->cl_close_lock);
1385
1386         rc = client_obd_setup(obd, len, buf);
1387         if (rc)
1388                 GOTO(err_close_lock, rc);
1389         lprocfs_mdc_init_vars(&lvars);
1390         if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0)
1391                 ptlrpc_lprocfs_register_obd(obd);
1392
1393         rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1394         if (rc) {
1395                 mdc_cleanup(obd);
1396                 CERROR("failed to setup llogging subsystems\n");
1397         }
1398
1399         RETURN(rc);
1400
1401 err_close_lock:
1402         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1403 err_setattr_lock:
1404         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1405 err_rpc_lock:
1406         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1407         ptlrpcd_decref();
1408         RETURN(rc);
1409 }
1410
1411 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1412  * us to make MDS RPCs with large enough reply buffers to hold the
1413  * maximum-sized (= maximum striped) EA and cookie without having to
1414  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1415 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1416 {
1417         struct obd_device *obd = mdc_exp->exp_obd;
1418         struct client_obd *cli = &obd->u.cli;
1419         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
1420         struct lov_desc desc;
1421         __u32 valsize = sizeof(desc);
1422         __u32 stripes;
1423         int rc, size;
1424         ENTRY;
1425
1426         rc = obd_get_info(lov_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
1427                           &valsize, &desc, NULL);
1428         if (rc)
1429                 RETURN(rc);
1430
1431         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1432         lsm.lsm_stripe_count = stripes;
1433         size = obd_size_diskmd(lov_exp, &lsm);
1434
1435         if (cli->cl_max_mds_easize < size)
1436                 cli->cl_max_mds_easize = size;
1437
1438         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1439         size = obd_size_diskmd(lov_exp, &lsm);
1440
1441         if (cli->cl_default_mds_easize < size)
1442                 cli->cl_default_mds_easize = size;
1443
1444         size = stripes * sizeof(struct llog_cookie);
1445         if (cli->cl_max_mds_cookiesize < size)
1446                 cli->cl_max_mds_cookiesize = size;
1447
1448         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1449                cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1450
1451         RETURN(0);
1452 }
1453
1454 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1455 {
1456         int rc = 0;
1457         ENTRY;
1458
1459         switch (stage) {
1460         case OBD_CLEANUP_EARLY:
1461         case OBD_CLEANUP_EXPORTS:
1462                 /* If we set up but never connected, the
1463                    client import will not have been cleaned. */
1464                 if (obd->u.cli.cl_import) {
1465                         struct obd_import *imp;
1466                         down_write(&obd->u.cli.cl_sem);
1467                         imp = obd->u.cli.cl_import;
1468                         CERROR("client import never connected\n");
1469                         ptlrpc_invalidate_import(imp);
1470                         class_destroy_import(imp);
1471                         up_write(&obd->u.cli.cl_sem);
1472                         obd->u.cli.cl_import = NULL;
1473                 }
1474                 rc = obd_llog_finish(obd, 0);
1475                 if (rc != 0)
1476                         CERROR("failed to cleanup llogging subsystems\n");
1477                 break;
1478         case OBD_CLEANUP_SELF_EXP:
1479                 break;
1480         case OBD_CLEANUP_OBD:
1481                 break;
1482         }
1483         RETURN(rc);
1484 }
1485
1486 static int mdc_cleanup(struct obd_device *obd)
1487 {
1488         struct client_obd *cli = &obd->u.cli;
1489
1490         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1491         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1492         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1493
1494         ptlrpc_lprocfs_unregister_obd(obd);
1495         lprocfs_obd_cleanup(obd);
1496         ptlrpcd_decref();
1497
1498         return client_obd_cleanup(obd);
1499 }
1500
1501
1502 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1503                          int count, struct llog_catid *logid,
1504                          struct obd_uuid *uuid)
1505 {
1506         struct llog_ctxt *ctxt;
1507         int rc;
1508         ENTRY;
1509
1510         rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1511                         &llog_client_ops);
1512         if (rc == 0) {
1513                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1514                 llog_initiator_connect(ctxt);
1515                 llog_ctxt_put(ctxt);
1516         }
1517
1518         rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1519                        &llog_client_ops);
1520         if (rc == 0) {
1521                 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1522                 llog_initiator_connect(ctxt);
1523                 llog_ctxt_put(ctxt);
1524         } else {
1525                 GOTO(err_cleanup, rc);
1526         }
1527
1528         RETURN(rc);
1529 err_cleanup:
1530         ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1531         if (ctxt)
1532                 llog_cleanup(ctxt);
1533         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1534         if (ctxt)
1535                 llog_cleanup(ctxt);
1536         return rc;
1537 }
1538
1539 static int mdc_llog_finish(struct obd_device *obd, int count)
1540 {
1541         struct llog_ctxt *ctxt;
1542         int rc = 0;
1543         ENTRY;
1544
1545         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1546         if (ctxt) {
1547                 rc = llog_cleanup(ctxt);
1548                 if (rc) {
1549                         CERROR("Can not cleanup LLOG_CONFIG_REPL_CTXT "
1550                                "rc %d\n", rc);
1551                 }
1552         }
1553         ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1554         if (ctxt)
1555                 rc = llog_cleanup(ctxt);
1556         RETURN(rc);
1557 }
1558
1559 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1560 {
1561         struct lustre_cfg *lcfg = buf;
1562         struct lprocfs_static_vars lvars = { 0 };
1563         int rc = 0;
1564
1565         lprocfs_mdc_init_vars(&lvars);
1566
1567         rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1568         return(rc);
1569 }
1570
1571 static int mdc_fid_init(struct obd_export *exp)
1572 {
1573         struct client_obd *cli;
1574         char              *prefix;
1575         int                rc;
1576         ENTRY;
1577
1578         cli = &exp->exp_obd->u.cli;
1579
1580         OBD_ALLOC_PTR(cli->cl_seq);
1581         if (cli->cl_seq == NULL)
1582                 RETURN(-ENOMEM);
1583
1584         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1585         if (prefix == NULL)
1586                 GOTO(out_free_seq, rc = -ENOMEM);
1587
1588         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", exp->exp_obd->obd_name);
1589
1590         /* Init client side sequence-manager */
1591         rc = seq_client_init(cli->cl_seq, exp,
1592                              LUSTRE_SEQ_METADATA,
1593                              LUSTRE_SEQ_MAX_WIDTH,
1594                              prefix);
1595         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1596         if (rc)
1597                 GOTO(out_free_seq, rc);
1598
1599         RETURN(rc);
1600
1601 out_free_seq:
1602         OBD_FREE_PTR(cli->cl_seq);
1603         cli->cl_seq = NULL;
1604         return rc;
1605 }
1606
1607 static int mdc_fid_fini(struct obd_export *exp)
1608 {
1609         struct client_obd *cli = &exp->exp_obd->u.cli;
1610         ENTRY;
1611
1612         if (cli->cl_seq != NULL) {
1613                 LASSERT(cli->cl_seq->lcs_exp == exp);
1614                 seq_client_fini(cli->cl_seq);
1615                 OBD_FREE_PTR(cli->cl_seq);
1616                 cli->cl_seq = NULL;
1617         }
1618
1619         RETURN(0);
1620 }
1621
1622 struct obd_ops mdc_obd_ops = {
1623         .o_owner        = THIS_MODULE,
1624         .o_setup        = mdc_setup,
1625         .o_precleanup   = mdc_precleanup,
1626         .o_cleanup      = mdc_cleanup,
1627         .o_add_conn     = client_import_add_conn,
1628         .o_del_conn     = client_import_del_conn,
1629         .o_connect      = client_connect_import,
1630         .o_disconnect   = client_disconnect_export,
1631         .o_fid_init     = mdc_fid_init,
1632         .o_fid_fini     = mdc_fid_fini,
1633         .o_iocontrol    = mdc_iocontrol,
1634         .o_set_info_async = mdc_set_info_async,
1635         .o_get_info     = mdc_get_info,
1636         .o_statfs       = mdc_statfs,
1637         .o_pin          = mdc_pin,
1638         .o_unpin        = mdc_unpin,
1639         .o_import_event = mdc_import_event,
1640         .o_llog_init    = mdc_llog_init,
1641         .o_llog_finish  = mdc_llog_finish,
1642         .o_process_config = mdc_process_config,
1643 };
1644
1645 int __init mdc_init(void)
1646 {
1647         int rc;
1648         struct lprocfs_static_vars lvars = { 0 };
1649         lprocfs_mdc_init_vars(&lvars);
1650         request_module("lquota");
1651         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1652         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1653
1654         rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1655                                  LUSTRE_MDC_NAME);
1656         if (rc && quota_interface)
1657                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1658
1659         RETURN(rc);
1660 }
1661
1662 #ifdef __KERNEL__
1663 static void /*__exit*/ mdc_exit(void)
1664 {
1665         if (quota_interface)
1666                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1667
1668         class_unregister_type(LUSTRE_MDC_NAME);
1669 }
1670
1671 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1672 MODULE_DESCRIPTION("Lustre Metadata Client");
1673 MODULE_LICENSE("GPL");
1674
1675 EXPORT_SYMBOL(mdc_req2lustre_md);
1676 EXPORT_SYMBOL(mdc_free_lustre_md);
1677 EXPORT_SYMBOL(mdc_change_cbdata);
1678 EXPORT_SYMBOL(mdc_getstatus);
1679 EXPORT_SYMBOL(mdc_getattr);
1680 EXPORT_SYMBOL(mdc_getattr_name);
1681 EXPORT_SYMBOL(mdc_create);
1682 EXPORT_SYMBOL(mdc_unlink);
1683 EXPORT_SYMBOL(mdc_rename);
1684 EXPORT_SYMBOL(mdc_link);
1685 EXPORT_SYMBOL(mdc_readpage);
1686 EXPORT_SYMBOL(mdc_setattr);
1687 EXPORT_SYMBOL(mdc_close);
1688 EXPORT_SYMBOL(mdc_done_writing);
1689 EXPORT_SYMBOL(mdc_sync);
1690 EXPORT_SYMBOL(mdc_set_open_replay_data);
1691 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1692 EXPORT_SYMBOL(mdc_store_inode_generation);
1693 EXPORT_SYMBOL(mdc_init_ea_size);
1694 EXPORT_SYMBOL(mdc_getxattr);
1695 EXPORT_SYMBOL(mdc_setxattr);
1696
1697 module_init(mdc_init);
1698 module_exit(mdc_exit);
1699 #endif