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, 2002 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
  23 #define EXPORT_SYMTAB
  24 #define DEBUG_SUBSYSTEM S_MDC
  25
  26 #include <linux/module.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/lustre_mds.h>
  29 #include <linux/lustre_lite.h>
  30 #include <linux/lustre_dlm.h>
  31 #include <linux/init.h>
  32 #include <linux/obd_lov.h>
  33
  34 #define REQUEST_MINOR 244
  35
  36 extern int mds_queue_req(struct ptlrpc_request *);
  37
  38 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid,
  39                   __u64 *last_committed, __u64 *last_xid,
  40                   struct ptlrpc_request **request)
  41 {
  42         struct ptlrpc_request *req;
  43         struct mds_body *body;
  44         int rc, size = sizeof(*body);
  45         ENTRY;
  46
  47         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
  48                               NULL);
  49         if (!req)
  50                 GOTO(out, rc = -ENOMEM);
  51
  52         body = lustre_msg_buf(req->rq_reqmsg, 0);
  53         req->rq_level = LUSTRE_CONN_CON;
  54         req->rq_replen = lustre_msg_size(1, &size);
  55
  56         mds_pack_req_body(req);
  57         rc = ptlrpc_queue_wait(req);
  58         rc = ptlrpc_check_status(req, rc);
  59
  60         if (!rc) {
  61                 body = lustre_msg_buf(req->rq_repmsg, 0);
  62                 mds_unpack_body(body);
  63                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  64                 *last_committed = req->rq_repmsg->last_committed;
  65                 *last_xid = req->rq_repmsg->last_xid;
  66
  67                 CDEBUG(D_NET,"root ino=%ld, last_committed=%Lu, last_xid=%Ld\n",
  68                        (unsigned long)rootfid->id,
  69                        (unsigned long long)*last_committed,
  70                        (unsigned long long)*last_xid);
  71         }
  72
  73         EXIT;
  74  out:
  75         ptlrpc_req_finished(req);
  76         return rc;
  77 }
  78
  79 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  80                    struct ptlrpc_request **request)
  81 {
  82         struct ptlrpc_request *req;
  83         struct mds_status_req *streq;
  84         int rc, size[2] = {sizeof(*streq)};
  85         ENTRY;
  86
  87         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
  88                               size, NULL);
  89         if (!req)
  90                 GOTO(out, rc = -ENOMEM);
  91
  92         *request = req;
  93         streq = lustre_msg_buf(req->rq_reqmsg, 0);
  94         streq->flags = HTON__u32(MDS_STATUS_LOV);
  95         streq->repbuf = HTON__u32(8192);
  96
  97         /* prepare for reply */
  98         req->rq_level = LUSTRE_CONN_CON;
  99         size[0] = 512;
 100         size[1] = 8192;
 101         req->rq_replen = lustre_msg_size(2, size);
 102
 103         rc = ptlrpc_queue_wait(req);
 104         rc = ptlrpc_check_status(req, rc);
 105
 106  out:
 107         RETURN(rc);
 108 }
 109
 110
 111 int mdc_getattr(struct lustre_handle *conn,
 112                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 113                 struct ptlrpc_request **request)
 114 {
 115         struct ptlrpc_request *req;
 116         struct mds_body *body;
 117         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 118         ENTRY;
 119
 120         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
 121                               NULL);
 122         if (!req)
 123                 GOTO(out, rc = -ENOMEM);
 124
 125         body = lustre_msg_buf(req->rq_reqmsg, 0);
 126         ll_ino2fid(&body->fid1, ino, 0, type);
 127         body->valid = valid;
 128
 129         if (S_ISREG(type)) {
 130                 struct client_obd *mdc = &class_conn2obd(conn)->u.cli;
 131                 bufcount = 2;
 132                 size[1] = mdc->cl_max_mds_easize;
 133         } else if (valid & OBD_MD_LINKNAME) {
 134                 bufcount = 2;
 135                 size[1] = ea_size;
 136                 body->size = ea_size;
 137                 CDEBUG(D_INODE, "allocating %d bytes for symlink in packet\n",
 138                        ea_size);
 139         }
 140         req->rq_replen = lustre_msg_size(bufcount, size);
 141         mds_pack_req_body(req);
 142
 143         rc = ptlrpc_queue_wait(req);
 144         rc = ptlrpc_check_status(req, rc);
 145
 146         if (!rc) {
 147                 body = lustre_msg_buf(req->rq_repmsg, 0);
 148                 mds_unpack_body(body);
 149                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 150         }
 151
 152         EXIT;
 153  out:
 154         *request = req;
 155         return rc;
 156 }
 157
 158 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 159                             void *data, __u32 data_len, int flag)
 160 {
 161         int rc;
 162         struct inode *inode = data;
 163         struct lustre_handle lockh;
 164         ENTRY;
 165
 166         if (data_len != sizeof(*inode)) {
 167                 CERROR("data_len should be %d, but is %d\n", sizeof(*inode),
 168                        data_len);
 169                 LBUG();
 170                 RETURN(-EINVAL);
 171         }
 172
 173         switch (flag) {
 174         case LDLM_CB_BLOCKING:
 175                 ldlm_lock2handle(lock, &lockh);
 176                 rc = ldlm_cli_cancel(&lockh);
 177                 if (rc < 0) {
 178                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 179                         RETURN(rc);
 180                 }
 181                 break;
 182         case LDLM_CB_CANCELING:
 183                 /* FIXME: do something better than throwing away everything */
 184                 if (inode == NULL)
 185                         LBUG();
 186                 if (S_ISDIR(inode->i_mode)) {
 187                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 188                                inode->i_ino);
 189                         ll_invalidate_inode_pages(inode);
 190                 }
 191                 break;
 192         default:
 193                 LBUG();
 194         }
 195
 196         RETURN(0);
 197 }
 198
 199 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 200                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 201                 struct dentry *de, struct lustre_handle *lockh,
 202                 char *tgt, int tgtlen, void *data, int datalen)
 203 {
 204         struct ptlrpc_request *req;
 205         struct obd_device *obddev = class_conn2obd(conn);
 206         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino};
 207         int size[5] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 208         int rc, flags;
 209         int repsize[3] = {sizeof(struct ldlm_reply),
 210                           sizeof(struct mds_body),
 211                           obddev->u.cli.cl_max_mds_easize};
 212         struct ldlm_reply *dlm_rep;
 213         struct ldlm_intent *lit;
 214         ENTRY;
 215
 216         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 217                           dir->i_ino);
 218
 219         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 220                 switch (it->it_op) {
 221                 case IT_MKDIR:
 222                         it->it_mode |= S_IFDIR;
 223                         break;
 224                 case (IT_CREAT|IT_OPEN):
 225                 case IT_CREAT:
 226                         it->it_mode |= S_IFREG;
 227                         break;
 228                 case IT_SYMLINK:
 229                         it->it_mode |= S_IFLNK;
 230                         break;
 231                 }
 232                 it->it_mode &= ~current->fs->umask;
 233
 234                 size[2] = sizeof(struct mds_rec_create);
 235                 size[3] = de->d_name.len + 1;
 236                 size[4] = tgtlen + 1;
 237                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 238                                       size, NULL);
 239                 if (!req)
 240                         RETURN(-ENOMEM);
 241
 242                 /* pack the intent */
 243                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 244                 lit->opc = NTOH__u64((__u64)it->it_op);
 245
 246                 /* pack the intended request */
 247                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 248                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 249                                 de->d_name.len, tgt, tgtlen);
 250                 req->rq_replen = lustre_msg_size(3, repsize);
 251         } else if (it->it_op == IT_RENAME2) {
 252                 struct dentry *old_de = it->it_data;
 253
 254                 size[2] = sizeof(struct mds_rec_rename);
 255                 size[3] = old_de->d_name.len + 1;
 256                 size[4] = de->d_name.len + 1;
 257                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 258                                       size, NULL);
 259                 if (!req)
 260                         RETURN(-ENOMEM);
 261
 262                 /* pack the intent */
 263                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 264                 lit->opc = NTOH__u64((__u64)it->it_op);
 265
 266                 /* pack the intended request */
 267                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 268                                 old_de->d_name.name, old_de->d_name.len,
 269                                 de->d_name.name, de->d_name.len);
 270                 req->rq_replen = lustre_msg_size(3, repsize);
 271         } else if (it->it_op == IT_LINK2) {
 272                 struct dentry *old_de = it->it_data;
 273
 274                 size[2] = sizeof(struct mds_rec_link);
 275                 size[3] = de->d_name.len + 1;
 276                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 277                                       size, NULL);
 278                 if (!req)
 279                         RETURN(-ENOMEM);
 280
 281                 /* pack the intent */
 282                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 283                 lit->opc = NTOH__u64((__u64)it->it_op);
 284
 285                 /* pack the intended request */
 286                 mds_link_pack(req, 2, old_de->d_inode, dir,
 287                               de->d_name.name, de->d_name.len);
 288                 req->rq_replen = lustre_msg_size(3, repsize);
 289         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 290                 size[2] = sizeof(struct mds_rec_unlink);
 291                 size[3] = de->d_name.len + 1;
 292                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 293                                       size, NULL);
 294                 if (!req)
 295                         RETURN(-ENOMEM);
 296
 297                 /* pack the intent */
 298                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 299                 lit->opc = NTOH__u64((__u64)it->it_op);
 300
 301                 /* pack the intended request */
 302                 mds_unlink_pack(req, 2, dir, NULL,
 303                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 304                                 de->d_name.name, de->d_name.len);
 305
 306                 req->rq_replen = lustre_msg_size(3, repsize);
 307         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 308                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 309                 size[2] = sizeof(struct mds_body);
 310                 size[3] = de->d_name.len + 1;
 311
 312                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 313                                       size, NULL);
 314                 if (!req)
 315                         RETURN(-ENOMEM);
 316
 317                 /* pack the intent */
 318                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 319                 lit->opc = NTOH__u64((__u64)it->it_op);
 320
 321                 /* pack the intended request */
 322                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 323
 324                 /* we need to replay opens */
 325                 if (it->it_op == IT_OPEN)
 326                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 327
 328                 /* get ready for the reply */
 329                 req->rq_replen = lustre_msg_size(3, repsize);
 330         } else if (it->it_op == IT_READDIR) {
 331                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 332                                       size, NULL);
 333                 if (!req)
 334                         RETURN(-ENOMEM);
 335
 336                 /* get ready for the reply */
 337                 req->rq_replen = lustre_msg_size(1, repsize);
 338         } else {
 339                 LBUG();
 340                 RETURN(-EINVAL);
 341         }
 342 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 343         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 344                               lock_type, NULL, 0, lock_mode, &flags,
 345                               ldlm_completion_ast, mdc_blocking_ast, data,
 346                               datalen, lockh);
 347         if (rc == -ENOENT) {
 348                 /* This can go when we're sure that this can never happen */
 349                 LBUG();
 350         }
 351         if (rc == ELDLM_LOCK_ABORTED) {
 352                 lock_mode = 0;
 353                 memset(lockh, 0, sizeof(*lockh));
 354                 /* rc = 0 */
 355         } else if (rc != 0) {
 356                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 357                 RETURN(rc);
 358         }
 359
 360         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 361         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 362         it->it_status = (int) dlm_rep->lock_policy_res2;
 363         it->it_lock_mode = lock_mode;
 364         it->it_data = req;
 365
 366         RETURN(0);
 367 }
 368
 369 static void mdc_replay_open(struct ptlrpc_request *req, void *data)
 370 {
 371         struct lustre_handle *fh = data;
 372         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 373
 374         mds_unpack_body(body);
 375         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 376                fh->addr, fh->cookie, body->handle.addr, body->handle.cookie);
 377         memcpy(fh, &body->handle, sizeof(*fh));
 378 }
 379
 380 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 381              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 382              struct ptlrpc_request **request)
 383 {
 384         struct mds_body *body;
 385         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 386         struct ptlrpc_request *req;
 387         ENTRY;
 388
 389         if (lsm) {
 390                 bufcount = 2;
 391                 // size[1] = mdc->cl_max_mds_easize; soon...
 392                 size[1] = lsm->lsm_mds_easize;
 393         }
 394
 395         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 396                               NULL);
 397         if (!req)
 398                 GOTO(out, rc = -ENOMEM);
 399
 400         req->rq_flags |= PTL_RPC_FL_REPLAY;
 401         body = lustre_msg_buf(req->rq_reqmsg, 0);
 402
 403         ll_ino2fid(&body->fid1, ino, 0, type);
 404         body->flags = HTON__u32(flags);
 405         memcpy(&body->handle, fh, sizeof(body->handle));
 406
 407         if (lsm)
 408                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 409
 410         req->rq_replen = lustre_msg_size(1, size);
 411
 412         rc = ptlrpc_queue_wait(req);
 413         rc = ptlrpc_check_status(req, rc);
 414         if (!rc) {
 415                 body = lustre_msg_buf(req->rq_repmsg, 0);
 416                 mds_unpack_body(body);
 417                 memcpy(fh, &body->handle, sizeof(*fh));
 418         }
 419
 420         /* If open is replayed, we need to fix up the fh. */
 421         req->rq_replay_cb = mdc_replay_open;
 422         req->rq_replay_cb_data = fh;
 423
 424         EXIT;
 425  out:
 426         *request = req;
 427         return rc;
 428 }
 429
 430 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 431               struct lustre_handle *fh, struct ptlrpc_request **request)
 432 {
 433         struct mds_body *body;
 434         int rc, size = sizeof(*body);
 435         struct ptlrpc_request *req;
 436
 437         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 438                               NULL);
 439         if (!req)
 440                 GOTO(out, rc = -ENOMEM);
 441
 442         body = lustre_msg_buf(req->rq_reqmsg, 0);
 443         ll_ino2fid(&body->fid1, ino, 0, type);
 444         memcpy(&body->handle, fh, sizeof(body->handle));
 445
 446         req->rq_replen = lustre_msg_size(0, NULL);
 447
 448         rc = ptlrpc_queue_wait(req);
 449         rc = ptlrpc_check_status(req, rc);
 450
 451         EXIT;
 452  out:
 453         *request = req;
 454         return rc;
 455 }
 456
 457 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 458                  char *addr, struct ptlrpc_request **request)
 459 {
 460         struct ptlrpc_connection *connection =
 461                 client_conn2cli(conn)->cl_import.imp_connection;
 462         struct ptlrpc_request *req = NULL;
 463         struct ptlrpc_bulk_desc *desc = NULL;
 464         struct ptlrpc_bulk_page *bulk = NULL;
 465         struct mds_body *body;
 466         int rc, size = sizeof(*body);
 467         ENTRY;
 468
 469         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 470
 471         desc = ptlrpc_prep_bulk(connection);
 472         if (desc == NULL)
 473                 GOTO(out, rc = -ENOMEM);
 474
 475         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 476                               NULL);
 477         if (!req)
 478                 GOTO(out2, rc = -ENOMEM);
 479
 480         bulk = ptlrpc_prep_bulk_page(desc);
 481         bulk->bp_buflen = PAGE_SIZE;
 482         bulk->bp_buf = addr;
 483         bulk->bp_xid = req->rq_xid;
 484         desc->bd_portal = MDS_BULK_PORTAL;
 485
 486         rc = ptlrpc_register_bulk(desc);
 487         if (rc) {
 488                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 489                 GOTO(out2, rc);
 490         }
 491
 492         body = lustre_msg_buf(req->rq_reqmsg, 0);
 493         body->fid1.id = ino;
 494         body->fid1.f_type = type;
 495         body->size = offset;
 496
 497         req->rq_replen = lustre_msg_size(1, &size);
 498         rc = ptlrpc_queue_wait(req);
 499         rc = ptlrpc_check_status(req, rc);
 500         if (rc) {
 501                 ptlrpc_abort_bulk(desc);
 502                 GOTO(out2, rc);
 503         } else {
 504                 body = lustre_msg_buf(req->rq_repmsg, 0);
 505                 mds_unpack_body(body);
 506         }
 507
 508         EXIT;
 509  out2:
 510         ptlrpc_free_bulk(desc);
 511  out:
 512         *request = req;
 513         return rc;
 514 }
 515
 516 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 517                struct ptlrpc_request **request)
 518 {
 519         struct ptlrpc_request *req;
 520         int rc, size = sizeof(*osfs);
 521         ENTRY;
 522
 523         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 524                               NULL);
 525         if (!req)
 526                 GOTO(out, rc = -ENOMEM);
 527         req->rq_replen = lustre_msg_size(1, &size);
 528
 529         rc = ptlrpc_queue_wait(req);
 530         rc = ptlrpc_check_status(req, rc);
 531
 532         if (rc)
 533                 GOTO(out, rc);
 534
 535         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 536
 537         EXIT;
 538 out:
 539         *request = req;
 540
 541         return rc;
 542 }
 543
 544 struct obd_ops mdc_obd_ops = {
 545         o_setup:   client_obd_setup,
 546         o_cleanup: client_obd_cleanup,
 547         o_connect: client_obd_connect,
 548         o_disconnect: client_obd_disconnect,
 549 };
 550
 551 static int __init ptlrpc_request_init(void)
 552 {
 553         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 554 }
 555
 556 static void __exit ptlrpc_request_exit(void)
 557 {
 558         class_unregister_type(LUSTRE_MDC_NAME);
 559 }
 560
 561 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 562 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 563 MODULE_LICENSE("GPL");
 564
 565 EXPORT_SYMBOL(mdc_getstatus);
 566 EXPORT_SYMBOL(mdc_getlovinfo);
 567 EXPORT_SYMBOL(mdc_enqueue);
 568 EXPORT_SYMBOL(mdc_getattr);
 569 EXPORT_SYMBOL(mdc_statfs);
 570 EXPORT_SYMBOL(mdc_create);
 571 EXPORT_SYMBOL(mdc_unlink);
 572 EXPORT_SYMBOL(mdc_rename);
 573 EXPORT_SYMBOL(mdc_link);
 574 EXPORT_SYMBOL(mdc_readpage);
 575 EXPORT_SYMBOL(mdc_setattr);
 576 EXPORT_SYMBOL(mdc_close);
 577 EXPORT_SYMBOL(mdc_open);
 578
 579 module_init(ptlrpc_request_init);
 580 module_exit(ptlrpc_request_exit);