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  * Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_MDC
  26
  27 #ifdef __KERNEL__
  28 # include <linux/module.h>
  29 # include <linux/pagemap.h>
  30 # include <linux/miscdevice.h>
  31 # include <linux/init.h>
  32 #else
  33 # include <liblustre.h>
  34 #endif
  35
  36 #include <linux/obd_class.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/lprocfs_status.h>
  40 #include "mdc_internal.h"
  41
  42 #define REQUEST_MINOR 244
  43
  44 static int mdc_cleanup(struct obd_device *obd, int flags);
  45
  46 extern int mds_queue_req(struct ptlrpc_request *);
  47 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  48 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  49 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
  50                           int level, int msg_flags)
  51 {
  52         struct ptlrpc_request *req;
  53         struct mds_body *body;
  54         int rc, size = sizeof(*body);
  55         ENTRY;
  56
  57         req = ptlrpc_prep_req(imp, MDS_GETSTATUS, 1, &size, NULL);
  58         if (!req)
  59                 GOTO(out, rc = -ENOMEM);
  60
  61         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
  62         req->rq_send_state = level;
  63         req->rq_replen = lustre_msg_size(1, &size);
  64
  65         mdc_pack_req_body(req);
  66         req->rq_reqmsg->flags |= msg_flags;
  67         rc = ptlrpc_queue_wait(req);
  68
  69         if (!rc) {
  70                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
  71                                            lustre_swab_mds_body);
  72                 if (body == NULL) {
  73                         CERROR ("Can't extract mds_body\n");
  74                         GOTO (out, rc = -EPROTO);
  75                 }
  76
  77                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  78
  79                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  80                        ", last_xid="LPU64"\n",
  81                        rootfid->id, req->rq_repmsg->last_committed,
  82                        req->rq_repmsg->last_xid);
  83         }
  84
  85         EXIT;
  86  out:
  87         ptlrpc_req_finished(req);
  88         return rc;
  89 }
  90
  91 /* This should be mdc_get_info("rootfid") */
  92 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
  93 {
  94         return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
  95                               0);
  96 }
  97
  98 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
  99                        struct ptlrpc_request *req)
 100 {
 101         struct mds_body *body;
 102         void            *eadata;
 103         int              rc;
 104         int              size[2] = {sizeof(*body), 0};
 105         int              bufcount = 1;
 106         ENTRY;
 107
 108         /* request message already built */
 109
 110         if (ea_size != 0) {
 111                 size[bufcount++] = ea_size;
 112                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 113                        ea_size);
 114         }
 115         req->rq_replen = lustre_msg_size(bufcount, size);
 116
 117         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 118         rc = ptlrpc_queue_wait(req);
 119         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 120         if (rc != 0)
 121                 RETURN (rc);
 122
 123         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 124                                    lustre_swab_mds_body);
 125         if (body == NULL) {
 126                 CERROR ("Can't unpack mds_body\n");
 127                 RETURN (-EPROTO);
 128         }
 129
 130         CDEBUG(D_NET, "mode: %o\n", body->mode);
 131
 132         LASSERT_REPSWAB (req, 1);
 133         if (body->eadatasize != 0) {
 134                 /* reply indicates presence of eadata; check it's there... */
 135                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
 136                 if (eadata == NULL) {
 137                         CERROR ("Missing/short eadata\n");
 138                         RETURN (-EPROTO);
 139                 }
 140         }
 141
 142         RETURN (0);
 143 }
 144
 145 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
 146                 unsigned long valid, unsigned int ea_size,
 147                 struct ptlrpc_request **request)
 148 {
 149         struct ptlrpc_request *req;
 150         struct mds_body *body;
 151         int size = sizeof(*body);
 152         int rc;
 153         ENTRY;
 154
 155         /* XXX do we need to make another request here?  We just did a getattr
 156          *     to do the lookup in the first place.
 157          */
 158         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR, 1, &size,
 159                               NULL);
 160         if (!req)
 161                 GOTO(out, rc = -ENOMEM);
 162
 163         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 164         memcpy(&body->fid1, fid, sizeof(*fid));
 165         body->valid = valid;
 166         body->eadatasize = ea_size;
 167         mdc_pack_req_body(req);
 168
 169         rc = mdc_getattr_common(exp, ea_size, req);
 170         if (rc != 0) {
 171                 ptlrpc_req_finished (req);
 172                 req = NULL;
 173         }
 174  out:
 175         *request = req;
 176         RETURN (rc);
 177 }
 178
 179 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 180                      char *filename, int namelen, unsigned long valid,
 181                      unsigned int ea_size, struct ptlrpc_request **request)
 182 {
 183         struct ptlrpc_request *req;
 184         struct mds_body *body;
 185         int rc, size[2] = {sizeof(*body), namelen};
 186         ENTRY;
 187
 188         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR_NAME, 2,
 189                               size, NULL);
 190         if (!req)
 191                 GOTO(out, rc = -ENOMEM);
 192
 193         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 194         memcpy(&body->fid1, fid, sizeof(*fid));
 195         body->valid = valid;
 196         body->eadatasize = ea_size;
 197         mdc_pack_req_body(req);
 198
 199         LASSERT (strnlen (filename, namelen) == namelen - 1);
 200         memcpy(lustre_msg_buf(req->rq_reqmsg, 1, namelen), filename, namelen);
 201
 202         rc = mdc_getattr_common(exp, ea_size, req);
 203         if (rc != 0) {
 204                 ptlrpc_req_finished (req);
 205                 req = NULL;
 206         }
 207  out:
 208         *request = req;
 209         RETURN(rc);
 210 }
 211
 212 /* This should be called with both the request and the reply still packed. */
 213 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
 214                                 int repoff)
 215 {
 216         struct mds_rec_create *rec =
 217                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
 218         struct mds_body *body =
 219                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
 220
 221         LASSERT (rec != NULL);
 222         LASSERT (body != NULL);
 223
 224         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 225         DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
 226                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 227 }
 228
 229 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
 230                       struct obd_export *exp,
 231                       struct lustre_md *md)
 232 {
 233         int rc = 0;
 234         ENTRY;
 235
 236         LASSERT(md);
 237         memset(md, 0, sizeof(*md));
 238
 239         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 240         LASSERT (md->body != NULL);
 241         LASSERT_REPSWABBED (req, offset);
 242
 243         if (md->body->valid & OBD_MD_FLEASIZE) {
 244                 int lmmsize;
 245                 struct lov_mds_md *lmm;
 246
 247                 LASSERT(S_ISREG(md->body->mode));
 248
 249                 if (md->body->eadatasize == 0) {
 250                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 251                         RETURN(-EPROTO);
 252                 }
 253                 lmmsize = md->body->eadatasize;
 254                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
 255                 LASSERT (lmm != NULL);
 256                 LASSERT_REPSWABBED (req, offset + 1);
 257
 258                 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
 259                 if (rc >= 0) {
 260                         LASSERT (rc >= sizeof (*md->lsm));
 261                         rc = 0;
 262                 }
 263         }
 264         RETURN(rc);
 265 }
 266
 267 static void mdc_commit_open(struct ptlrpc_request *req)
 268 {
 269         struct mdc_open_data *mod = req->rq_cb_data;
 270         if (mod == NULL)
 271                 return;
 272
 273         if (mod->mod_close_req != NULL)
 274                 mod->mod_close_req->rq_cb_data = NULL;
 275
 276         if (mod->mod_och != NULL)
 277                 mod->mod_och->och_mod = NULL;
 278
 279         OBD_FREE(mod, sizeof(*mod));
 280         req->rq_cb_data = NULL;
 281 }
 282
 283 static void mdc_replay_open(struct ptlrpc_request *req)
 284 {
 285         struct mdc_open_data *mod = req->rq_cb_data;
 286         struct obd_client_handle *och;
 287         struct ptlrpc_request *close_req;
 288         struct lustre_handle old;
 289         struct mds_body *body;
 290         ENTRY;
 291
 292         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
 293         LASSERT (body != NULL);
 294
 295         if (mod == NULL) {
 296                 DEBUG_REQ(D_ERROR, req,
 297                           "can't properly replay without open data");
 298                 EXIT;
 299                 return;
 300         }
 301
 302         och = mod->mod_och;
 303         if (och != NULL) {
 304                 struct lustre_handle *file_fh;
 305                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 306                 file_fh = &och->och_fh;
 307                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 308                        file_fh->cookie, body->handle.cookie);
 309                 memcpy(&old, file_fh, sizeof(old));
 310                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 311         }
 312
 313         close_req = mod->mod_close_req;
 314         if (close_req != NULL) {
 315                 struct mds_body *close_body;
 316                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
 317                 close_body = lustre_msg_buf(close_req->rq_reqmsg, 0,
 318                                             sizeof(*close_body));
 319                 if (och != NULL)
 320                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 321                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 322                 memcpy(&close_body->handle, &body->handle,
 323                        sizeof(close_body->handle));
 324         }
 325
 326         EXIT;
 327 }
 328
 329 void mdc_set_open_replay_data(struct obd_client_handle *och,
 330                               struct ptlrpc_request *open_req)
 331 {
 332         struct mdc_open_data *mod;
 333         struct mds_rec_create *rec =
 334                 lustre_msg_buf(open_req->rq_reqmsg, 2, sizeof(*rec));
 335         struct mds_body *body =
 336                 lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
 337
 338         LASSERT(rec != NULL);
 339         /* outgoing messages always in my byte order */
 340         LASSERT(body != NULL);
 341         /* incoming message in my byte order (it's been swabbed) */
 342         LASSERT_REPSWABBED(open_req, 1);
 343
 344         OBD_ALLOC(mod, sizeof(*mod));
 345         if (mod == NULL) {
 346                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 347                 return;
 348         }
 349
 350         och->och_mod = mod;
 351         mod->mod_och = och;
 352         mod->mod_open_req = open_req;
 353
 354         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 355         open_req->rq_replay_cb = mdc_replay_open;
 356         open_req->rq_commit_cb = mdc_commit_open;
 357         open_req->rq_cb_data = mod;
 358         DEBUG_REQ(D_HA, open_req, "set up replay data");
 359 }
 360
 361 void mdc_clear_open_replay_data(struct obd_client_handle *och)
 362 {
 363         struct mdc_open_data *mod = och->och_mod;
 364
 365         /* Don't free the structure now (it happens in mdc_commit_open, after
 366          * we're sure we won't need to fix up the close request in the future),
 367          * but make sure that replay doesn't poke at the och, which is about to
 368          * be freed. */
 369         LASSERT(mod != (void *)0x5a5a5a5a);
 370         if (mod != NULL)
 371                 mod->mod_och = NULL;
 372         och->och_mod = NULL;
 373 }
 374
 375 static void mdc_commit_close(struct ptlrpc_request *req)
 376 {
 377         struct mdc_open_data *mod = req->rq_cb_data;
 378         struct ptlrpc_request *open_req;
 379         struct obd_import *imp = req->rq_import;
 380
 381         DEBUG_REQ(D_HA, req, "close req committed");
 382         if (mod == NULL)
 383                 return;
 384
 385         mod->mod_close_req = NULL;
 386         req->rq_cb_data = NULL;
 387         req->rq_commit_cb = NULL;
 388
 389         open_req = mod->mod_open_req;
 390         LASSERT(open_req != NULL);
 391         LASSERT(open_req != (void *)0x5a5a5a5a);
 392
 393         DEBUG_REQ(D_HA, open_req, "open req balanced");
 394         LASSERT(open_req->rq_transno != 0);
 395         LASSERT(open_req->rq_import == imp);
 396
 397         /* We no longer want to preserve this for transno-unconditional
 398          * replay. */
 399         spin_lock(&open_req->rq_lock);
 400         open_req->rq_replay = 0;
 401         spin_unlock(&open_req->rq_lock);
 402 }
 403
 404 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
 405 {
 406         union ptlrpc_async_args *aa = data;
 407         struct mdc_rpc_lock *rpc_lock = aa->pointer_arg[0];
 408
 409         mdc_put_rpc_lock(rpc_lock, NULL);
 410         wake_up(&req->rq_reply_waitq);
 411         RETURN(rc);
 412 }
 413
 414 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 415  * anything but a reply or an error. */
 416 static int mdc_close_check_reply(struct ptlrpc_request *req)
 417 {
 418         int rc = 0;
 419         unsigned long flags;
 420
 421         spin_lock_irqsave(&req->rq_lock, flags);
 422         if (PTLRPC_REQUEST_COMPLETE(req)) {
 423                 rc = 1;
 424         }
 425         spin_unlock_irqrestore (&req->rq_lock, flags);
 426         return rc;
 427 }
 428
 429 static int go_back_to_sleep(void *unused)
 430 {
 431         return 0;
 432 }
 433
 434 int mdc_close(struct obd_export *exp, struct obdo *obdo,
 435               struct obd_client_handle *och, struct ptlrpc_request **request)
 436 {
 437         struct mds_body *body;
 438         struct obd_device *obd = class_exp2obd(exp);
 439         int reqsize = sizeof(*body);
 440         int rc, repsize[3] = {sizeof(*body),
 441                               obd->u.cli.cl_max_mds_easize,
 442                               obd->u.cli.cl_max_mds_cookiesize};
 443         struct ptlrpc_request *req;
 444         struct mdc_open_data *mod;
 445         struct l_wait_info lwi;
 446         struct mdc_rpc_lock *rpc_lock = obd->u.cli.cl_rpc_lock;
 447         ENTRY;
 448
 449         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &reqsize,
 450                               NULL);
 451         if (req == NULL)
 452                 GOTO(out, rc = -ENOMEM);
 453
 454         /* Ensure that this close's handle is fixed up during replay. */
 455         LASSERT(och != NULL);
 456         mod = och->och_mod;
 457         if (likely(mod != NULL)) {
 458                 mod->mod_close_req = req;
 459                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
 460                           mod->mod_open_req);
 461         } else {
 462                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 463         }
 464
 465         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 466         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 467         memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 468         body->size = obdo->o_size;
 469         body->blocks = obdo->o_blocks;
 470         body->flags = obdo->o_flags;
 471         body->valid = obdo->o_valid;
 472
 473         req->rq_replen = lustre_msg_size(3, repsize);
 474         req->rq_commit_cb = mdc_commit_close;
 475         LASSERT(req->rq_cb_data == NULL);
 476         req->rq_cb_data = mod;
 477
 478         /* We hand a ref to the rpcd here, so we need another one of our own. */
 479         ptlrpc_request_addref(req);
 480
 481         mdc_get_rpc_lock(rpc_lock, NULL);
 482         req->rq_interpret_reply = mdc_close_interpret;
 483         req->rq_async_args.pointer_arg[0] = rpc_lock;
 484         ptlrpcd_add_req(req);
 485         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
 486                                NULL, NULL);
 487         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 488                           &lwi);
 489         if (rc == 0) {
 490                 rc = req->rq_repmsg->status;
 491                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 492                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 493                                   "= %d", rc);
 494                         if (rc > 0)
 495                                 rc = -rc;
 496                 } else if (mod == NULL) {
 497                         CERROR("Unexpected: can't find mdc_open_data, but the "
 498                                "close succeeded.  Please tell CFS.\n");
 499                 }
 500         }
 501
 502         EXIT;
 503  out:
 504         *request = req;
 505         return rc;
 506 }
 507
 508 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 509 {
 510         struct ptlrpc_request *req;
 511         struct mds_body *body;
 512         int rc, size = sizeof(*body);
 513         ENTRY;
 514
 515         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_DONE_WRITING, 1,
 516                               &size, NULL);
 517         if (req == NULL)
 518                 RETURN(-ENOMEM);
 519
 520         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 521         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 522         body->size = obdo->o_size;
 523         body->blocks = obdo->o_blocks;
 524         body->flags = obdo->o_flags;
 525         body->valid = obdo->o_valid;
 526 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 527
 528         req->rq_replen = lustre_msg_size(1, &size);
 529
 530         rc = ptlrpc_queue_wait(req);
 531         ptlrpc_req_finished(req);
 532         RETURN(rc);
 533 }
 534
 535 int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid, __u64 offset,
 536                  struct page *page, struct ptlrpc_request **request)
 537 {
 538         struct obd_import *imp = class_exp2cliimp(exp);
 539         struct ptlrpc_request *req = NULL;
 540         struct ptlrpc_bulk_desc *desc = NULL;
 541         struct mds_body *body;
 542         int rc, size = sizeof(*body);
 543         ENTRY;
 544
 545         CDEBUG(D_INODE, "inode: %ld\n", (long)mdc_fid->id);
 546
 547         req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
 548         if (req == NULL)
 549                 GOTO(out, rc = -ENOMEM);
 550         /* XXX FIXME bug 249 */
 551         req->rq_request_portal = MDS_READPAGE_PORTAL;
 552
 553         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 554         if (desc == NULL)
 555                 GOTO(out, rc = -ENOMEM);
 556         /* NB req now owns desc and will free it when it gets freed */
 557
 558         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 559
 560         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE, mdc_fid);
 561
 562         req->rq_replen = lustre_msg_size(1, &size);
 563         rc = ptlrpc_queue_wait(req);
 564
 565         if (rc == 0) {
 566                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 567                                           lustre_swab_mds_body);
 568                 if (body == NULL) {
 569                         CERROR("Can't unpack mds_body\n");
 570                         GOTO(out, rc = -EPROTO);
 571                 }
 572
 573                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 574                         CERROR ("Unexpected # bytes transferred: %d"
 575                                 " (%ld expected)\n",
 576                                 req->rq_bulk->bd_nob_transferred,
 577                                 PAGE_CACHE_SIZE);
 578                         GOTO (out, rc = -EPROTO);
 579                 }
 580         }
 581
 582         EXIT;
 583  out:
 584         *request = req;
 585         return rc;
 586 }
 587
 588 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 589                          void *karg, void *uarg)
 590 {
 591         struct obd_device *obd = exp->exp_obd;
 592         struct obd_ioctl_data *data = karg;
 593         struct obd_import *imp = obd->u.cli.cl_import;
 594         struct llog_ctxt *ctxt;
 595         int rc;
 596         ENTRY;
 597
 598         MOD_INC_USE_COUNT;
 599
 600         switch (cmd) {
 601         case OBD_IOC_CLIENT_RECOVER:
 602                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 603                 if (rc < 0)
 604                         GOTO(out, rc);
 605                 GOTO(out, rc = 0);
 606         case IOC_OSC_SET_ACTIVE:
 607                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 608                 GOTO(out, rc);
 609         case OBD_IOC_PARSE: {
 610                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
 611                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
 612                 GOTO(out, rc);
 613         }
 614 #ifdef __KERNEL__
 615         case OBD_IOC_LLOG_INFO:
 616         case OBD_IOC_LLOG_PRINT: {
 617                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 618                 rc = llog_ioctl(ctxt, cmd, data);
 619
 620                 GOTO(out, rc);
 621         }
 622 #endif
 623         default:
 624                 CERROR("osc_ioctl(): unrecognised ioctl %#x\n", cmd);
 625                 GOTO(out, rc = -ENOTTY);
 626         }
 627 out:
 628         MOD_DEC_USE_COUNT;
 629         return rc;
 630 }
 631
 632 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 633                  void *key, obd_count vallen, void *val)
 634 {
 635         int rc = -EINVAL;
 636
 637         if (keylen == strlen("initial_recov") &&
 638             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 639                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 640                 if (vallen != sizeof(int))
 641                         RETURN(-EINVAL);
 642                 imp->imp_initial_recov = *(int *)val;
 643                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 644                        exp->exp_obd->obd_name,
 645                        imp->imp_initial_recov);
 646                 RETURN(0);
 647         }
 648
 649         RETURN(rc);
 650 }
 651
 652 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 653                       unsigned long max_age)
 654 {
 655         struct ptlrpc_request *req;
 656         struct obd_statfs *msfs;
 657         int rc, size = sizeof(*msfs);
 658         ENTRY;
 659
 660         /* We could possibly pass max_age in the request (as an absolute
 661          * timestamp or a "seconds.usec ago") so the target can avoid doing
 662          * extra calls into the filesystem if that isn't necessary (e.g.
 663          * during mount that would help a bit).  Having relative timestamps
 664          * is not so great if request processing is slow, while absolute
 665          * timestamps are not ideal because they need time synchronization. */
 666         req = ptlrpc_prep_req(obd->u.cli.cl_import, MDS_STATFS, 0, NULL, NULL);
 667         if (!req)
 668                 RETURN(-ENOMEM);
 669
 670         req->rq_replen = lustre_msg_size(1, &size);
 671
 672         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 673         rc = ptlrpc_queue_wait(req);
 674         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 675
 676         if (rc)
 677                 GOTO(out, rc);
 678
 679         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
 680         if (msfs == NULL) {
 681                 CERROR("Can't unpack obd_statfs\n");
 682                 GOTO(out, rc = -EPROTO);
 683         }
 684
 685         memcpy(osfs, msfs, sizeof (*msfs));
 686         EXIT;
 687 out:
 688         ptlrpc_req_finished(req);
 689
 690         return rc;
 691 }
 692
 693 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 694                    struct obd_client_handle *handle, int flag)
 695 {
 696         struct ptlrpc_request *req;
 697         struct mds_body *body;
 698         int rc, size = sizeof(*body);
 699         ENTRY;
 700
 701         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_PIN, 1, &size, NULL);
 702         if (req == NULL)
 703                 RETURN(-ENOMEM);
 704
 705         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 706         mdc_pack_fid(&body->fid1, ino, gen, type);
 707         body->flags = flag;
 708
 709         req->rq_replen = lustre_msg_size(1, &size);
 710
 711         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 712         rc = ptlrpc_queue_wait(req);
 713         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 714         if (rc) {
 715                 CERROR("pin failed: %d\n", rc);
 716                 ptlrpc_req_finished(req);
 717                 RETURN(rc);
 718         }
 719
 720         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 721         if (body == NULL) {
 722                 ptlrpc_req_finished(req);
 723                 RETURN(rc);
 724         }
 725
 726         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 727         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 728
 729         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 730         if (handle->och_mod == NULL) {
 731                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 732                 RETURN(-ENOMEM);
 733         }
 734         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 735
 736         RETURN(rc);
 737 }
 738
 739 static int mdc_unpin(struct obd_export *exp,
 740                      struct obd_client_handle *handle, int flag)
 741 {
 742         struct ptlrpc_request *req;
 743         struct mds_body *body;
 744         int rc, size = sizeof(*body);
 745         ENTRY;
 746
 747         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 748                 RETURN(0);
 749
 750         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &size, NULL);
 751         if (req == NULL)
 752                 RETURN(-ENOMEM);
 753
 754         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 755         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 756         body->flags = flag;
 757
 758         req->rq_replen = lustre_msg_size(0, NULL);
 759         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 760         rc = ptlrpc_queue_wait(req);
 761         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 762
 763         if (rc != 0)
 764                 CERROR("unpin failed: %d\n", rc);
 765
 766         ptlrpc_req_finished(req);
 767         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 768         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 769         RETURN(rc);
 770 }
 771
 772 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
 773              struct ptlrpc_request **request)
 774 {
 775         struct ptlrpc_request *req;
 776         struct mds_body *body;
 777         int size = sizeof(*body);
 778         int rc;
 779         ENTRY;
 780
 781         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_SYNC, 1,&size,NULL);
 782         if (!req)
 783                 RETURN(rc = -ENOMEM);
 784
 785         if (fid) {
 786                 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 787                 memcpy(&body->fid1, fid, sizeof(*fid));
 788                 mdc_pack_req_body(req);
 789         }
 790
 791         req->rq_replen = lustre_msg_size(1, &size);
 792
 793         rc = ptlrpc_queue_wait(req);
 794         if (rc || request == NULL)
 795                 ptlrpc_req_finished(req);
 796         else
 797                 *request = req;
 798
 799         RETURN(rc);
 800 }
 801
 802 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
 803 {
 804         struct lprocfs_static_vars lvars;
 805
 806         lprocfs_init_vars(mdc, &lvars);
 807         return lprocfs_obd_attach(dev, lvars.obd_vars);
 808 }
 809
 810 static int mdc_detach(struct obd_device *dev)
 811 {
 812         return lprocfs_obd_detach(dev);
 813 }
 814
 815 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
 816 {
 817         struct client_obd *cli = &obd->u.cli;
 818         int rc;
 819         ENTRY;
 820
 821         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 822         if (!cli->cl_rpc_lock)
 823                 RETURN(-ENOMEM);
 824         mdc_init_rpc_lock(cli->cl_rpc_lock);
 825
 826         ptlrpcd_addref();
 827
 828         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 829         if (!cli->cl_setattr_lock)
 830                 GOTO(out_free_rpc, rc = -ENOMEM);
 831         mdc_init_rpc_lock(cli->cl_setattr_lock);
 832
 833         rc = client_obd_setup(obd, len, buf);
 834         if (rc) {
 835                 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 836  out_free_rpc:
 837                 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 838         }
 839
 840         rc = obd_llog_init(obd, obd, 0, NULL);
 841         if (rc) {
 842                 mdc_cleanup(obd, 0);
 843                 CERROR("failed to setup llogging subsystems\n");
 844         }
 845
 846         RETURN(rc);
 847 }
 848
 849 int mdc_init_ea_size(struct obd_device *obd, char *lov_name)
 850 {
 851         struct client_obd *cli = &obd->u.cli;
 852         struct obd_device *lov_obd;
 853         struct obd_export *exp;
 854         struct lustre_handle conn;
 855         struct lov_desc desc;
 856         int valsize;
 857         int rc;
 858
 859         lov_obd = class_name2obd(lov_name);
 860         if (!lov_obd) {
 861                 CERROR("MDC cannot locate LOV %s!\n", lov_name);
 862                 RETURN(-ENOTCONN);
 863         }
 864
 865         rc = obd_connect(&conn, lov_obd, &obd->obd_uuid);
 866         if (rc) {
 867                 CERROR("MDC failed connect to LOV %s (%d)\n", lov_name, rc);
 868                 RETURN(rc);
 869         }
 870         exp = class_conn2export(&conn);
 871
 872         valsize = sizeof(desc);
 873         rc = obd_get_info(exp, strlen("lovdesc") + 1, "lovdesc",
 874                           &valsize, &desc);
 875         if (rc == 0) {
 876                 cli->cl_max_mds_easize = obd_size_diskmd(exp, NULL);
 877                 cli->cl_max_mds_cookiesize = desc.ld_tgt_count *
 878                         sizeof(struct llog_cookie);
 879         }
 880         obd_disconnect(exp, 0);
 881         RETURN(rc);
 882 }
 883
 884 static int mdc_precleanup(struct obd_device *obd, int flags)
 885 {
 886         int rc = 0;
 887
 888         rc = obd_llog_finish(obd, 0);
 889         if (rc != 0)
 890                 CERROR("failed to cleanup llogging subsystems\n");
 891
 892         RETURN(rc);
 893 }
 894
 895 static int mdc_cleanup(struct obd_device *obd, int flags)
 896 {
 897         struct client_obd *cli = &obd->u.cli;
 898
 899         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 900         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 901
 902         ptlrpcd_decref();
 903
 904         return client_obd_cleanup(obd, flags);
 905 }
 906
 907
 908 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
 909                          int count, struct llog_logid *logid)
 910 {
 911         struct llog_ctxt *ctxt;
 912         int rc;
 913         ENTRY;
 914
 915         rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
 916                         &llog_client_ops);
 917         if (rc == 0) {
 918                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 919                 ctxt->loc_imp = obd->u.cli.cl_import;
 920         }
 921
 922         RETURN(rc);
 923 }
 924
 925 static int mdc_llog_finish(struct obd_device *obd, int count)
 926 {
 927         int rc;
 928         ENTRY;
 929
 930         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
 931         RETURN(rc);
 932 }
 933
 934 struct obd_ops mdc_obd_ops = {
 935         o_owner:       THIS_MODULE,
 936         o_attach:      mdc_attach,
 937         o_detach:      mdc_detach,
 938         o_setup:       mdc_setup,
 939         o_precleanup:  mdc_precleanup,
 940         o_cleanup:     mdc_cleanup,
 941         o_connect:     client_connect_import,
 942         o_disconnect:  client_disconnect_export,
 943         o_iocontrol:   mdc_iocontrol,
 944         o_set_info:    mdc_set_info,
 945         o_statfs:      mdc_statfs,
 946         o_pin:         mdc_pin,
 947         o_unpin:       mdc_unpin,
 948         o_llog_init:   mdc_llog_init,
 949         o_llog_finish: mdc_llog_finish,
 950 };
 951
 952 int __init mdc_init(void)
 953 {
 954         struct lprocfs_static_vars lvars;
 955         lprocfs_init_vars(mdc, &lvars);
 956         return class_register_type(&mdc_obd_ops, lvars.module_vars,
 957                                    LUSTRE_MDC_NAME);
 958 }
 959
 960 static void /*__exit*/ mdc_exit(void)
 961 {
 962         class_unregister_type(LUSTRE_MDC_NAME);
 963 }
 964
 965 #ifdef __KERNEL__
 966 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 967 MODULE_DESCRIPTION("Lustre Metadata Client");
 968 MODULE_LICENSE("GPL");
 969
 970 EXPORT_SYMBOL(mdc_req2lustre_md);
 971 EXPORT_SYMBOL(mdc_change_cbdata);
 972 EXPORT_SYMBOL(mdc_getstatus);
 973 EXPORT_SYMBOL(mdc_getattr);
 974 EXPORT_SYMBOL(mdc_getattr_name);
 975 EXPORT_SYMBOL(mdc_create);
 976 EXPORT_SYMBOL(mdc_unlink);
 977 EXPORT_SYMBOL(mdc_rename);
 978 EXPORT_SYMBOL(mdc_link);
 979 EXPORT_SYMBOL(mdc_readpage);
 980 EXPORT_SYMBOL(mdc_setattr);
 981 EXPORT_SYMBOL(mdc_close);
 982 EXPORT_SYMBOL(mdc_done_writing);
 983 EXPORT_SYMBOL(mdc_sync);
 984 EXPORT_SYMBOL(mdc_set_open_replay_data);
 985 EXPORT_SYMBOL(mdc_clear_open_replay_data);
 986 EXPORT_SYMBOL(mdc_store_inode_generation);
 987 EXPORT_SYMBOL(mdc_init_ea_size);
 988
 989 module_init(mdc_init);
 990 module_exit(mdc_exit);
 991 #endif