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 != LP_POISON);
 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 != LP_POISON);
 392         LASSERT(open_req->rq_type != LI_POISON);
 393
 394         DEBUG_REQ(D_HA, open_req, "open req balanced");
 395         LASSERT(open_req->rq_transno != 0);
 396         LASSERT(open_req->rq_import == imp);
 397
 398         /* We no longer want to preserve this for transno-unconditional
 399          * replay. */
 400         spin_lock(&open_req->rq_lock);
 401         open_req->rq_replay = 0;
 402         spin_unlock(&open_req->rq_lock);
 403 }
 404
 405 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
 406 {
 407         union ptlrpc_async_args *aa = data;
 408         struct mdc_rpc_lock *rpc_lock = aa->pointer_arg[0];
 409         struct obd_device *obd = aa->pointer_arg[1];
 410
 411         if (rpc_lock == NULL) {
 412                 CERROR("called with NULL rpc_lock\n");
 413         } else {
 414                 mdc_put_rpc_lock(rpc_lock, NULL);
 415                 LASSERTF(req->rq_async_args.pointer_arg[0] ==
 416                          obd->u.cli.cl_rpc_lock, "%p != %p\n",
 417                          req->rq_async_args.pointer_arg[0],
 418                          obd->u.cli.cl_rpc_lock);
 419                 aa->pointer_arg[0] = NULL;
 420         }
 421         wake_up(&req->rq_reply_waitq);
 422         RETURN(rc);
 423 }
 424
 425 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 426  * anything but a reply or an error. */
 427 static int mdc_close_check_reply(struct ptlrpc_request *req)
 428 {
 429         int rc = 0;
 430         unsigned long flags;
 431
 432         spin_lock_irqsave(&req->rq_lock, flags);
 433         if (PTLRPC_REQUEST_COMPLETE(req)) {
 434                 rc = 1;
 435         }
 436         spin_unlock_irqrestore (&req->rq_lock, flags);
 437         return rc;
 438 }
 439
 440 static int go_back_to_sleep(void *unused)
 441 {
 442         return 0;
 443 }
 444
 445 int mdc_close(struct obd_export *exp, struct obdo *obdo,
 446               struct obd_client_handle *och, struct ptlrpc_request **request)
 447 {
 448         struct mds_body *body;
 449         struct obd_device *obd = class_exp2obd(exp);
 450         int reqsize = sizeof(*body);
 451         int rc, repsize[3] = {sizeof(*body),
 452                               obd->u.cli.cl_max_mds_easize,
 453                               obd->u.cli.cl_max_mds_cookiesize};
 454         struct ptlrpc_request *req;
 455         struct mdc_open_data *mod;
 456         struct l_wait_info lwi;
 457         ENTRY;
 458
 459         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &reqsize,
 460                               NULL);
 461         if (req == NULL)
 462                 GOTO(out, rc = -ENOMEM);
 463
 464         /* Ensure that this close's handle is fixed up during replay. */
 465         LASSERT(och != NULL);
 466         mod = och->och_mod;
 467         if (likely(mod != NULL)) {
 468                 mod->mod_close_req = req;
 469                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 470                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
 471                           mod->mod_open_req);
 472         } else {
 473                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 474         }
 475
 476         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 477         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 478         memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 479         body->size = obdo->o_size;
 480         body->blocks = obdo->o_blocks;
 481         body->flags = obdo->o_flags;
 482         body->valid = obdo->o_valid;
 483
 484         req->rq_replen = lustre_msg_size(3, repsize);
 485         req->rq_commit_cb = mdc_commit_close;
 486         LASSERT(req->rq_cb_data == NULL);
 487         req->rq_cb_data = mod;
 488
 489         /* We hand a ref to the rpcd here, so we need another one of our own. */
 490         ptlrpc_request_addref(req);
 491
 492         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 493         req->rq_interpret_reply = mdc_close_interpret;
 494         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_rpc_lock;
 495         req->rq_async_args.pointer_arg[1] = obd;
 496         ptlrpcd_add_req(req);
 497         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
 498                                NULL, NULL);
 499         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 500                           &lwi);
 501         if (req->rq_repmsg == NULL) {
 502                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 503                        req->rq_status);
 504                 rc = req->rq_status ? req->rq_status : -EIO;
 505         } else if (rc == 0) {
 506                 rc = req->rq_repmsg->status;
 507                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 508                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 509                                   "= %d", rc);
 510                         if (rc > 0)
 511                                 rc = -rc;
 512                 } else if (mod == NULL) {
 513                         CERROR("Unexpected: can't find mdc_open_data, but the "
 514                                "close succeeded.  Please tell CFS.\n");
 515                 }
 516                 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 517                                         lustre_swab_mds_body)) {
 518                         CERROR("Error unpacking mds_body\n");
 519                         rc = -EPROTO;
 520                 }
 521         }
 522         if (req->rq_async_args.pointer_arg[0] != NULL) {
 523                 CERROR("returned without dropping rpc_lock: rc %d\n", rc);
 524                 mdc_close_interpret(req, &req->rq_async_args, rc);
 525                 portals_debug_dumplog();
 526         }
 527
 528         EXIT;
 529  out:
 530         *request = req;
 531         return rc;
 532 }
 533
 534 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 535 {
 536         struct ptlrpc_request *req;
 537         struct mds_body *body;
 538         int rc, size = sizeof(*body);
 539         ENTRY;
 540
 541         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_DONE_WRITING, 1,
 542                               &size, NULL);
 543         if (req == NULL)
 544                 RETURN(-ENOMEM);
 545
 546         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 547         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 548         body->size = obdo->o_size;
 549         body->blocks = obdo->o_blocks;
 550         body->flags = obdo->o_flags;
 551         body->valid = obdo->o_valid;
 552 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 553
 554         req->rq_replen = lustre_msg_size(1, &size);
 555
 556         rc = ptlrpc_queue_wait(req);
 557         ptlrpc_req_finished(req);
 558         RETURN(rc);
 559 }
 560
 561 int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid, __u64 offset,
 562                  struct page *page, struct ptlrpc_request **request)
 563 {
 564         struct obd_import *imp = class_exp2cliimp(exp);
 565         struct ptlrpc_request *req = NULL;
 566         struct ptlrpc_bulk_desc *desc = NULL;
 567         struct mds_body *body;
 568         int rc, size = sizeof(*body);
 569         ENTRY;
 570
 571         CDEBUG(D_INODE, "inode: %ld\n", (long)mdc_fid->id);
 572
 573         req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
 574         if (req == NULL)
 575                 GOTO(out, rc = -ENOMEM);
 576         /* XXX FIXME bug 249 */
 577         req->rq_request_portal = MDS_READPAGE_PORTAL;
 578
 579         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 580         if (desc == NULL)
 581                 GOTO(out, rc = -ENOMEM);
 582         /* NB req now owns desc and will free it when it gets freed */
 583
 584         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 585
 586         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE, mdc_fid);
 587
 588         req->rq_replen = lustre_msg_size(1, &size);
 589         rc = ptlrpc_queue_wait(req);
 590
 591         if (rc == 0) {
 592                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 593                                           lustre_swab_mds_body);
 594                 if (body == NULL) {
 595                         CERROR("Can't unpack mds_body\n");
 596                         GOTO(out, rc = -EPROTO);
 597                 }
 598
 599                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 600                         CERROR ("Unexpected # bytes transferred: %d"
 601                                 " (%ld expected)\n",
 602                                 req->rq_bulk->bd_nob_transferred,
 603                                 PAGE_CACHE_SIZE);
 604                         GOTO (out, rc = -EPROTO);
 605                 }
 606         }
 607
 608         EXIT;
 609  out:
 610         *request = req;
 611         return rc;
 612 }
 613
 614 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 615                          void *karg, void *uarg)
 616 {
 617         struct obd_device *obd = exp->exp_obd;
 618         struct obd_ioctl_data *data = karg;
 619         struct obd_import *imp = obd->u.cli.cl_import;
 620         struct llog_ctxt *ctxt;
 621         int rc;
 622         ENTRY;
 623
 624 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 625         MOD_INC_USE_COUNT;
 626 #else
 627         if (!try_module_get(THIS_MODULE)) {
 628                 CERROR("Can't get module. Is it alive?");
 629                 return -EINVAL;
 630         }
 631 #endif
 632         switch (cmd) {
 633         case OBD_IOC_CLIENT_RECOVER:
 634                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 635                 if (rc < 0)
 636                         GOTO(out, rc);
 637                 GOTO(out, rc = 0);
 638         case IOC_OSC_SET_ACTIVE:
 639                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 640                 GOTO(out, rc);
 641         case OBD_IOC_PARSE: {
 642                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
 643                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
 644                 GOTO(out, rc);
 645         }
 646 #ifdef __KERNEL__
 647         case OBD_IOC_LLOG_INFO:
 648         case OBD_IOC_LLOG_PRINT: {
 649                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 650                 rc = llog_ioctl(ctxt, cmd, data);
 651
 652                 GOTO(out, rc);
 653         }
 654 #endif
 655         default:
 656                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 657                 GOTO(out, rc = -ENOTTY);
 658         }
 659 out:
 660 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 661         MOD_DEC_USE_COUNT;
 662 #else
 663         module_put(THIS_MODULE);
 664 #endif
 665
 666         return rc;
 667 }
 668
 669 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 670                  void *key, obd_count vallen, void *val)
 671 {
 672         int rc = -EINVAL;
 673
 674         if (keylen == strlen("initial_recov") &&
 675             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 676                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 677                 if (vallen != sizeof(int))
 678                         RETURN(-EINVAL);
 679                 imp->imp_initial_recov = *(int *)val;
 680                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 681                        exp->exp_obd->obd_name,
 682                        imp->imp_initial_recov);
 683                 RETURN(0);
 684         }
 685
 686         RETURN(rc);
 687 }
 688
 689 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 690                       unsigned long max_age)
 691 {
 692         struct ptlrpc_request *req;
 693         struct obd_statfs *msfs;
 694         int rc, size = sizeof(*msfs);
 695         ENTRY;
 696
 697         /* We could possibly pass max_age in the request (as an absolute
 698          * timestamp or a "seconds.usec ago") so the target can avoid doing
 699          * extra calls into the filesystem if that isn't necessary (e.g.
 700          * during mount that would help a bit).  Having relative timestamps
 701          * is not so great if request processing is slow, while absolute
 702          * timestamps are not ideal because they need time synchronization. */
 703         req = ptlrpc_prep_req(obd->u.cli.cl_import, MDS_STATFS, 0, NULL, NULL);
 704         if (!req)
 705                 RETURN(-ENOMEM);
 706
 707         req->rq_replen = lustre_msg_size(1, &size);
 708
 709         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 710         rc = ptlrpc_queue_wait(req);
 711         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 712
 713         if (rc)
 714                 GOTO(out, rc);
 715
 716         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
 717         if (msfs == NULL) {
 718                 CERROR("Can't unpack obd_statfs\n");
 719                 GOTO(out, rc = -EPROTO);
 720         }
 721
 722         memcpy(osfs, msfs, sizeof (*msfs));
 723         EXIT;
 724 out:
 725         ptlrpc_req_finished(req);
 726
 727         return rc;
 728 }
 729
 730 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 731                    struct obd_client_handle *handle, int flag)
 732 {
 733         struct ptlrpc_request *req;
 734         struct mds_body *body;
 735         int rc, size = sizeof(*body);
 736         ENTRY;
 737
 738         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_PIN, 1, &size, NULL);
 739         if (req == NULL)
 740                 RETURN(-ENOMEM);
 741
 742         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 743         mdc_pack_fid(&body->fid1, ino, gen, type);
 744         body->flags = flag;
 745
 746         req->rq_replen = lustre_msg_size(1, &size);
 747
 748         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 749         rc = ptlrpc_queue_wait(req);
 750         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 751         if (rc) {
 752                 CERROR("pin failed: %d\n", rc);
 753                 ptlrpc_req_finished(req);
 754                 RETURN(rc);
 755         }
 756
 757         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 758         if (body == NULL) {
 759                 ptlrpc_req_finished(req);
 760                 RETURN(rc);
 761         }
 762
 763         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 764         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 765
 766         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 767         if (handle->och_mod == NULL) {
 768                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 769                 RETURN(-ENOMEM);
 770         }
 771         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 772
 773         RETURN(rc);
 774 }
 775
 776 static int mdc_unpin(struct obd_export *exp,
 777                      struct obd_client_handle *handle, int flag)
 778 {
 779         struct ptlrpc_request *req;
 780         struct mds_body *body;
 781         int rc, size = sizeof(*body);
 782         ENTRY;
 783
 784         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 785                 RETURN(0);
 786
 787         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &size, NULL);
 788         if (req == NULL)
 789                 RETURN(-ENOMEM);
 790
 791         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 792         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 793         body->flags = flag;
 794
 795         req->rq_replen = lustre_msg_size(0, NULL);
 796         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 797         rc = ptlrpc_queue_wait(req);
 798         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 799
 800         if (rc != 0)
 801                 CERROR("unpin failed: %d\n", rc);
 802
 803         ptlrpc_req_finished(req);
 804         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 805         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 806         RETURN(rc);
 807 }
 808
 809 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
 810              struct ptlrpc_request **request)
 811 {
 812         struct ptlrpc_request *req;
 813         struct mds_body *body;
 814         int size = sizeof(*body);
 815         int rc;
 816         ENTRY;
 817
 818         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_SYNC, 1,&size,NULL);
 819         if (!req)
 820                 RETURN(rc = -ENOMEM);
 821
 822         if (fid) {
 823                 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 824                 memcpy(&body->fid1, fid, sizeof(*fid));
 825                 mdc_pack_req_body(req);
 826         }
 827
 828         req->rq_replen = lustre_msg_size(1, &size);
 829
 830         rc = ptlrpc_queue_wait(req);
 831         if (rc || request == NULL)
 832                 ptlrpc_req_finished(req);
 833         else
 834                 *request = req;
 835
 836         RETURN(rc);
 837 }
 838
 839 static int mdc_import_event(struct obd_device *obd,
 840                             struct obd_import *imp,
 841                             enum obd_import_event event)
 842 {
 843         int rc = 0;
 844
 845         LASSERT(imp->imp_obd == obd);
 846
 847         switch (event) {
 848         case IMP_EVENT_DISCON: {
 849                 break;
 850         }
 851         case IMP_EVENT_INACTIVE: {
 852                 if (obd->obd_observer)
 853                         rc = obd_notify(obd->obd_observer, obd, 0);
 854                 break;
 855         }
 856         case IMP_EVENT_INVALIDATE: {
 857                 struct ldlm_namespace *ns = obd->obd_namespace;
 858
 859                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
 860
 861                 break;
 862         }
 863         case IMP_EVENT_ACTIVE: {
 864                 if (obd->obd_observer)
 865                         rc = obd_notify(obd->obd_observer, obd, 1);
 866                 break;
 867         }
 868         default:
 869                 CERROR("Unknown import event %d\n", event);
 870                 LBUG();
 871         }
 872         RETURN(rc);
 873 }
 874
 875 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
 876 {
 877         struct client_obd *cli = &obd->u.cli;
 878         struct lprocfs_static_vars lvars;
 879         int rc;
 880         ENTRY;
 881
 882         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 883         if (!cli->cl_rpc_lock)
 884                 RETURN(-ENOMEM);
 885         mdc_init_rpc_lock(cli->cl_rpc_lock);
 886
 887         ptlrpcd_addref();
 888
 889         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 890         if (!cli->cl_setattr_lock)
 891                 GOTO(err_rpc_lock, rc = -ENOMEM);
 892         mdc_init_rpc_lock(cli->cl_setattr_lock);
 893
 894         rc = client_obd_setup(obd, len, buf);
 895         if (rc)
 896                 GOTO(err_setattr_lock, rc);
 897         lprocfs_init_vars(mdc, &lvars);
 898         lprocfs_obd_setup(obd, lvars.obd_vars);
 899
 900         rc = obd_llog_init(obd, obd, 0, NULL);
 901         if (rc) {
 902                 mdc_cleanup(obd, 0);
 903                 CERROR("failed to setup llogging subsystems\n");
 904         }
 905
 906         RETURN(rc);
 907
 908 err_setattr_lock:
 909         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 910 err_rpc_lock:
 911         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 912         ptlrpcd_decref();
 913         RETURN(rc);
 914 }
 915
 916 int mdc_init_ea_size(struct obd_device *obd, char *lov_name)
 917 {
 918         struct client_obd *cli = &obd->u.cli;
 919         struct obd_device *lov_obd;
 920         struct obd_export *exp;
 921         struct lustre_handle conn;
 922         struct lov_desc desc;
 923         int valsize;
 924         int rc;
 925
 926         lov_obd = class_name2obd(lov_name);
 927         if (!lov_obd) {
 928                 CERROR("MDC cannot locate LOV %s!\n", lov_name);
 929                 RETURN(-ENOTCONN);
 930         }
 931
 932         rc = obd_connect(&conn, lov_obd, &obd->obd_uuid);
 933         if (rc) {
 934                 CERROR("MDC failed connect to LOV %s (%d)\n", lov_name, rc);
 935                 RETURN(rc);
 936         }
 937         exp = class_conn2export(&conn);
 938
 939         valsize = sizeof(desc);
 940         rc = obd_get_info(exp, strlen("lovdesc") + 1, "lovdesc",
 941                           &valsize, &desc);
 942         if (rc == 0) {
 943                 cli->cl_max_mds_easize = obd_size_diskmd(exp, NULL);
 944                 cli->cl_max_mds_cookiesize = desc.ld_tgt_count *
 945                         sizeof(struct llog_cookie);
 946         }
 947         obd_disconnect(exp, 0);
 948         RETURN(rc);
 949 }
 950
 951 static int mdc_precleanup(struct obd_device *obd, int flags)
 952 {
 953         int rc = 0;
 954
 955         rc = obd_llog_finish(obd, 0);
 956         if (rc != 0)
 957                 CERROR("failed to cleanup llogging subsystems\n");
 958
 959         RETURN(rc);
 960 }
 961
 962 static int mdc_cleanup(struct obd_device *obd, int flags)
 963 {
 964         struct client_obd *cli = &obd->u.cli;
 965
 966         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 967         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 968
 969         lprocfs_obd_cleanup(obd);
 970         ptlrpcd_decref();
 971
 972         return client_obd_cleanup(obd, flags);
 973 }
 974
 975
 976 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
 977                          int count, struct llog_catid *logid)
 978 {
 979         struct llog_ctxt *ctxt;
 980         int rc;
 981         ENTRY;
 982
 983         rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
 984                         &llog_client_ops);
 985         if (rc == 0) {
 986                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 987                 ctxt->loc_imp = obd->u.cli.cl_import;
 988         }
 989
 990         RETURN(rc);
 991 }
 992
 993 static int mdc_llog_finish(struct obd_device *obd, int count)
 994 {
 995         int rc;
 996         ENTRY;
 997
 998         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
 999         RETURN(rc);
1000 }
1001
1002 struct obd_ops mdc_obd_ops = {
1003         .o_owner        = THIS_MODULE,
1004         .o_setup        = mdc_setup,
1005         .o_precleanup   = mdc_precleanup,
1006         .o_cleanup      = mdc_cleanup,
1007         .o_connect      = client_connect_import,
1008         .o_disconnect   = client_disconnect_export,
1009         .o_iocontrol    = mdc_iocontrol,
1010         .o_set_info     = mdc_set_info,
1011         .o_statfs       = mdc_statfs,
1012         .o_pin          = mdc_pin,
1013         .o_unpin        = mdc_unpin,
1014         .o_import_event = mdc_import_event,
1015         .o_llog_init    = mdc_llog_init,
1016         .o_llog_finish  = mdc_llog_finish,
1017 };
1018
1019 int __init mdc_init(void)
1020 {
1021         struct lprocfs_static_vars lvars;
1022         lprocfs_init_vars(mdc, &lvars);
1023         return class_register_type(&mdc_obd_ops, lvars.module_vars,
1024                                    LUSTRE_MDC_NAME);
1025 }
1026
1027 #ifdef __KERNEL__
1028 static void /*__exit*/ mdc_exit(void)
1029 {
1030         class_unregister_type(LUSTRE_MDC_NAME);
1031 }
1032
1033 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1034 MODULE_DESCRIPTION("Lustre Metadata Client");
1035 MODULE_LICENSE("GPL");
1036
1037 EXPORT_SYMBOL(mdc_req2lustre_md);
1038 EXPORT_SYMBOL(mdc_change_cbdata);
1039 EXPORT_SYMBOL(mdc_getstatus);
1040 EXPORT_SYMBOL(mdc_getattr);
1041 EXPORT_SYMBOL(mdc_getattr_name);
1042 EXPORT_SYMBOL(mdc_create);
1043 EXPORT_SYMBOL(mdc_unlink);
1044 EXPORT_SYMBOL(mdc_rename);
1045 EXPORT_SYMBOL(mdc_link);
1046 EXPORT_SYMBOL(mdc_readpage);
1047 EXPORT_SYMBOL(mdc_setattr);
1048 EXPORT_SYMBOL(mdc_close);
1049 EXPORT_SYMBOL(mdc_done_writing);
1050 EXPORT_SYMBOL(mdc_sync);
1051 EXPORT_SYMBOL(mdc_set_open_replay_data);
1052 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1053 EXPORT_SYMBOL(mdc_store_inode_generation);
1054 EXPORT_SYMBOL(mdc_init_ea_size);
1055
1056 module_init(mdc_init);
1057 module_exit(mdc_exit);
1058 #endif