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                         CERROR("ldlm_cli_cancel: %d\n", rc);
 179                         LBUG();
 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                         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         switch (it->it_op) {
 220         case IT_MKDIR:
 221                 it->it_mode = (it->it_mode | S_IFDIR) & ~current->fs->umask;
 222                 break;
 223         case (IT_CREAT|IT_OPEN):
 224         case IT_CREAT:
 225                 it->it_mode |= S_IFREG; /* no break */
 226         case IT_MKNOD:
 227                 it->it_mode &= ~current->fs->umask;
 228                 break;
 229         case IT_SYMLINK:
 230                 it->it_mode = (it->it_mode | S_IFLNK) & ~current->fs->umask;
 231                 break;
 232         }
 233
 234         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 235                 size[2] = sizeof(struct mds_rec_create);
 236                 size[3] = de->d_name.len + 1;
 237                 size[4] = tgtlen + 1;
 238                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 239                                       size, NULL);
 240                 if (!req)
 241                         RETURN(-ENOMEM);
 242
 243                 /* pack the intent */
 244                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 245                 lit->opc = NTOH__u64((__u64)it->it_op);
 246
 247                 /* pack the intended request */
 248                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 249                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 250                                 de->d_name.len, tgt, tgtlen);
 251                 req->rq_replen = lustre_msg_size(3, repsize);
 252         } else if (it->it_op == IT_RENAME2) {
 253                 struct dentry *old_de = it->it_data;
 254
 255                 size[2] = sizeof(struct mds_rec_rename);
 256                 size[3] = old_de->d_name.len + 1;
 257                 size[4] = de->d_name.len + 1;
 258                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 259                                       size, NULL);
 260                 if (!req)
 261                         RETURN(-ENOMEM);
 262
 263                 /* pack the intent */
 264                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 265                 lit->opc = NTOH__u64((__u64)it->it_op);
 266
 267                 /* pack the intended request */
 268                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 269                                 old_de->d_name.name, old_de->d_name.len,
 270                                 de->d_name.name, de->d_name.len);
 271                 req->rq_replen = lustre_msg_size(3, repsize);
 272         } else if (it->it_op == IT_LINK2) {
 273                 struct dentry *old_de = it->it_data;
 274
 275                 size[2] = sizeof(struct mds_rec_link);
 276                 size[3] = de->d_name.len + 1;
 277                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 278                                       size, NULL);
 279                 if (!req)
 280                         RETURN(-ENOMEM);
 281
 282                 /* pack the intent */
 283                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 284                 lit->opc = NTOH__u64((__u64)it->it_op);
 285
 286                 /* pack the intended request */
 287                 mds_link_pack(req, 2, old_de->d_inode, dir,
 288                               de->d_name.name, de->d_name.len);
 289                 req->rq_replen = lustre_msg_size(3, repsize);
 290         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 291                 size[2] = sizeof(struct mds_rec_unlink);
 292                 size[3] = de->d_name.len + 1;
 293                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 294                                       size, NULL);
 295                 if (!req)
 296                         RETURN(-ENOMEM);
 297
 298                 /* pack the intent */
 299                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 300                 lit->opc = NTOH__u64((__u64)it->it_op);
 301
 302                 /* pack the intended request */
 303                 mds_unlink_pack(req, 2, dir, NULL,
 304                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 305                                 de->d_name.name, de->d_name.len);
 306
 307                 req->rq_replen = lustre_msg_size(3, repsize);
 308         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 309                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 310                 size[2] = sizeof(struct mds_body);
 311                 size[3] = de->d_name.len + 1;
 312
 313                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 314                                       size, NULL);
 315                 if (!req)
 316                         RETURN(-ENOMEM);
 317
 318                 /* pack the intent */
 319                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 320                 lit->opc = NTOH__u64((__u64)it->it_op);
 321
 322                 /* pack the intended request */
 323                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 324
 325                 /* we need to replay opens */
 326                 if (it->it_op == IT_OPEN)
 327                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 328
 329                 /* get ready for the reply */
 330                 req->rq_replen = lustre_msg_size(3, repsize);
 331         } else if (it->it_op == IT_READDIR) {
 332                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 333                                       size, NULL);
 334                 if (!req)
 335                         RETURN(-ENOMEM);
 336
 337                 /* get ready for the reply */
 338                 req->rq_replen = lustre_msg_size(1, repsize);
 339         } else {
 340                 LBUG();
 341                 RETURN(-EINVAL);
 342         }
 343 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 344         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 345                               lock_type, NULL, 0, lock_mode, &flags,
 346                               ldlm_completion_ast, mdc_blocking_ast, data,
 347                               datalen, lockh);
 348         if (rc == -ENOENT) {
 349                 /* This can go when we're sure that this can never happen */
 350                 LBUG();
 351         }
 352         if (rc == ELDLM_LOCK_ABORTED) {
 353                 lock_mode = 0;
 354                 memset(lockh, 0, sizeof(*lockh));
 355                 /* rc = 0 */
 356         } else if (rc != 0) {
 357                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 358                 RETURN(rc);
 359         }
 360
 361         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 362         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 363         it->it_status = (int) dlm_rep->lock_policy_res2;
 364         it->it_lock_mode = lock_mode;
 365         it->it_data = req;
 366
 367         RETURN(0);
 368 }
 369
 370 static void mdc_replay_open(struct ptlrpc_request *req, void *data)
 371 {
 372         struct lustre_handle *fh = data;
 373         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 374
 375         mds_unpack_body(body);
 376         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 377                fh->addr, fh->cookie, body->handle.addr, body->handle.cookie);
 378         memcpy(fh, &body->handle, sizeof(*fh));
 379 }
 380
 381 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 382              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 383              struct ptlrpc_request **request)
 384 {
 385         struct mds_body *body;
 386         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 387         struct ptlrpc_request *req;
 388         ENTRY;
 389
 390         if (lsm) {
 391                 bufcount = 2;
 392                 // size[1] = mdc->cl_max_mds_easize; soon...
 393                 size[1] = lsm->lsm_mds_easize;
 394         }
 395
 396         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 397                               NULL);
 398         if (!req)
 399                 GOTO(out, rc = -ENOMEM);
 400
 401         req->rq_flags |= PTL_RPC_FL_REPLAY;
 402         body = lustre_msg_buf(req->rq_reqmsg, 0);
 403
 404         ll_ino2fid(&body->fid1, ino, 0, type);
 405         body->flags = HTON__u32(flags);
 406         memcpy(&body->handle, fh, sizeof(body->handle));
 407
 408         if (lsm)
 409                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 410
 411         req->rq_replen = lustre_msg_size(1, size);
 412
 413         rc = ptlrpc_queue_wait(req);
 414         rc = ptlrpc_check_status(req, rc);
 415         if (!rc) {
 416                 body = lustre_msg_buf(req->rq_repmsg, 0);
 417                 mds_unpack_body(body);
 418                 memcpy(fh, &body->handle, sizeof(*fh));
 419         }
 420
 421         /* If open is replayed, we need to fix up the fh. */
 422         req->rq_replay_cb = mdc_replay_open;
 423         req->rq_replay_cb_data = fh;
 424
 425         EXIT;
 426  out:
 427         *request = req;
 428         return rc;
 429 }
 430
 431 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 432               struct lustre_handle *fh, struct ptlrpc_request **request)
 433 {
 434         struct mds_body *body;
 435         int rc, size = sizeof(*body);
 436         struct ptlrpc_request *req;
 437
 438         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 439                               NULL);
 440         if (!req)
 441                 GOTO(out, rc = -ENOMEM);
 442
 443         body = lustre_msg_buf(req->rq_reqmsg, 0);
 444         ll_ino2fid(&body->fid1, ino, 0, type);
 445         memcpy(&body->handle, fh, sizeof(body->handle));
 446
 447         req->rq_replen = lustre_msg_size(0, NULL);
 448
 449         rc = ptlrpc_queue_wait(req);
 450         rc = ptlrpc_check_status(req, rc);
 451
 452         EXIT;
 453  out:
 454         *request = req;
 455         return rc;
 456 }
 457
 458 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 459                  char *addr, struct ptlrpc_request **request)
 460 {
 461         struct ptlrpc_connection *connection =
 462                 client_conn2cli(conn)->cl_import.imp_connection;
 463         struct ptlrpc_request *req = NULL;
 464         struct ptlrpc_bulk_desc *desc = NULL;
 465         struct ptlrpc_bulk_page *bulk = NULL;
 466         struct mds_body *body;
 467         int rc, size = sizeof(*body);
 468         ENTRY;
 469
 470         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 471
 472         desc = ptlrpc_prep_bulk(connection);
 473         if (desc == NULL)
 474                 GOTO(out, rc = -ENOMEM);
 475
 476         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 477                               NULL);
 478         if (!req)
 479                 GOTO(out2, rc = -ENOMEM);
 480
 481         bulk = ptlrpc_prep_bulk_page(desc);
 482         bulk->bp_buflen = PAGE_SIZE;
 483         bulk->bp_buf = addr;
 484         bulk->bp_xid = req->rq_xid;
 485         desc->bd_portal = MDS_BULK_PORTAL;
 486
 487         rc = ptlrpc_register_bulk(desc);
 488         if (rc) {
 489                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 490                 GOTO(out2, rc);
 491         }
 492
 493         body = lustre_msg_buf(req->rq_reqmsg, 0);
 494         body->fid1.id = ino;
 495         body->fid1.f_type = type;
 496         body->size = offset;
 497
 498         req->rq_replen = lustre_msg_size(1, &size);
 499         rc = ptlrpc_queue_wait(req);
 500         rc = ptlrpc_check_status(req, rc);
 501         if (rc) {
 502                 ptlrpc_abort_bulk(desc);
 503                 GOTO(out2, rc);
 504         } else {
 505                 body = lustre_msg_buf(req->rq_repmsg, 0);
 506                 mds_unpack_body(body);
 507         }
 508
 509         EXIT;
 510  out2:
 511         ptlrpc_free_bulk(desc);
 512  out:
 513         *request = req;
 514         return rc;
 515 }
 516
 517 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 518                struct ptlrpc_request **request)
 519 {
 520         struct ptlrpc_request *req;
 521         int rc, size = sizeof(*osfs);
 522         ENTRY;
 523
 524         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 525                               NULL);
 526         if (!req)
 527                 GOTO(out, rc = -ENOMEM);
 528         req->rq_replen = lustre_msg_size(1, &size);
 529
 530         rc = ptlrpc_queue_wait(req);
 531         rc = ptlrpc_check_status(req, rc);
 532
 533         if (rc)
 534                 GOTO(out, rc);
 535
 536         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 537
 538         EXIT;
 539 out:
 540         *request = req;
 541
 542         return rc;
 543 }
 544
 545 struct obd_ops mdc_obd_ops = {
 546         o_setup:   client_obd_setup,
 547         o_cleanup: client_obd_cleanup,
 548         o_connect: client_obd_connect,
 549         o_disconnect: client_obd_disconnect,
 550 };
 551
 552 static int __init ptlrpc_request_init(void)
 553 {
 554         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 555 }
 556
 557 static void __exit ptlrpc_request_exit(void)
 558 {
 559         class_unregister_type(LUSTRE_MDC_NAME);
 560 }
 561
 562 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 563 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 564 MODULE_LICENSE("GPL");
 565
 566 EXPORT_SYMBOL(mdc_getstatus);
 567 EXPORT_SYMBOL(mdc_getlovinfo);
 568 EXPORT_SYMBOL(mdc_enqueue);
 569 EXPORT_SYMBOL(mdc_getattr);
 570 EXPORT_SYMBOL(mdc_statfs);
 571 EXPORT_SYMBOL(mdc_create);
 572 EXPORT_SYMBOL(mdc_unlink);
 573 EXPORT_SYMBOL(mdc_rename);
 574 EXPORT_SYMBOL(mdc_link);
 575 EXPORT_SYMBOL(mdc_readpage);
 576 EXPORT_SYMBOL(mdc_setattr);
 577 EXPORT_SYMBOL(mdc_close);
 578 EXPORT_SYMBOL(mdc_open);
 579
 580 module_init(ptlrpc_request_init);
 581 module_exit(ptlrpc_request_exit);