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_free_req(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         }
 137         req->rq_replen = lustre_msg_size(bufcount, size);
 138
 139         rc = ptlrpc_queue_wait(req);
 140         rc = ptlrpc_check_status(req, rc);
 141
 142         if (!rc) {
 143                 body = lustre_msg_buf(req->rq_repmsg, 0);
 144                 mds_unpack_body(body);
 145                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 146         }
 147
 148         EXIT;
 149  out:
 150         *request = req;
 151         return rc;
 152 }
 153
 154 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 155                             void *data, __u32 data_len)
 156 {
 157         int rc;
 158         struct inode *inode = data;
 159         struct lustre_handle lockh;
 160         ENTRY;
 161
 162         if (data_len != sizeof(*inode)) {
 163                 CERROR("data_len should be %d, but is %d\n", sizeof(*inode),
 164                        data_len);
 165                 LBUG();
 166                 RETURN(-EINVAL);
 167         }
 168
 169         /* FIXME: do something better than throwing away everything */
 170         if (inode == NULL)
 171                 LBUG();
 172         if (S_ISDIR(inode->i_mode)) {
 173                 CDEBUG(D_INODE, "invalidating inode %ld\n", inode->i_ino);
 174                 invalidate_inode_pages(inode);
 175         }
 176
 177         ldlm_lock2handle(lock, &lockh);
 178         rc = ldlm_cli_cancel(&lockh);
 179         if (rc < 0) {
 180                 CERROR("ldlm_cli_cancel: %d\n", rc);
 181                 LBUG();
 182         }
 183         RETURN(0);
 184 }
 185
 186 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 187                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 188                 struct dentry *de, struct lustre_handle *lockh,
 189                 char *tgt, int tgtlen, void *data, int datalen)
 190 {
 191         struct ptlrpc_request *req;
 192         struct obd_device *obddev = class_conn2obd(conn);
 193         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino};
 194         int size[5] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 195         int rc, flags;
 196         int repsize[3] = {sizeof(struct ldlm_reply),
 197                           sizeof(struct mds_body),
 198                           obddev->u.cli.cl_max_mds_easize};
 199         struct ldlm_reply *dlm_rep;
 200         struct ldlm_intent *lit;
 201         ENTRY;
 202
 203         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 204                           dir->i_ino);
 205
 206         switch (it->it_op) {
 207         case IT_MKDIR:
 208                 it->it_mode = (it->it_mode | S_IFDIR) & ~current->fs->umask;
 209                 break;
 210         case (IT_CREAT|IT_OPEN):
 211         case IT_CREAT:
 212                 it->it_mode |= S_IFREG; /* no break */
 213         case IT_MKNOD:
 214                 it->it_mode &= ~current->fs->umask;
 215                 break;
 216         case IT_SYMLINK:
 217                 it->it_mode = (it->it_mode | S_IFLNK) & ~current->fs->umask;
 218                 break;
 219         }
 220
 221         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 222                 size[2] = sizeof(struct mds_rec_create);
 223                 size[3] = de->d_name.len + 1;
 224                 size[4] = tgtlen + 1;
 225                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 226                                       size, NULL);
 227                 if (!req)
 228                         RETURN(-ENOMEM);
 229
 230                 /* pack the intent */
 231                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 232                 lit->opc = NTOH__u64((__u64)it->it_op);
 233
 234                 /* pack the intended request */
 235                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 236                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 237                                 de->d_name.len, tgt, tgtlen);
 238                 req->rq_replen = lustre_msg_size(3, repsize);
 239         } else if (it->it_op == IT_RENAME2) {
 240                 struct dentry *old_de = it->it_data;
 241
 242                 size[2] = sizeof(struct mds_rec_rename);
 243                 size[3] = old_de->d_name.len + 1;
 244                 size[4] = de->d_name.len + 1;
 245                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 246                                       size, NULL);
 247                 if (!req)
 248                         RETURN(-ENOMEM);
 249
 250                 /* pack the intent */
 251                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 252                 lit->opc = NTOH__u64((__u64)it->it_op);
 253
 254                 /* pack the intended request */
 255                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 256                                 old_de->d_name.name, old_de->d_name.len,
 257                                 de->d_name.name, de->d_name.len);
 258                 req->rq_replen = lustre_msg_size(3, repsize);
 259         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 260                 size[2] = sizeof(struct mds_rec_unlink);
 261                 size[3] = de->d_name.len + 1;
 262                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 263                                       size, NULL);
 264                 if (!req)
 265                         RETURN(-ENOMEM);
 266
 267                 /* pack the intent */
 268                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 269                 lit->opc = NTOH__u64((__u64)it->it_op);
 270
 271                 /* pack the intended request */
 272                 mds_unlink_pack(req, 2, dir, NULL,
 273                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 274                                 de->d_name.name, de->d_name.len);
 275
 276                 req->rq_replen = lustre_msg_size(3, repsize);
 277         } else if (it->it_op == IT_GETATTR || it->it_op == IT_RENAME ||
 278                    it->it_op == IT_OPEN || it->it_op == IT_SETATTR ||
 279                    it->it_op == IT_LOOKUP || it->it_op == IT_READLINK) {
 280                 size[2] = sizeof(struct mds_body);
 281                 size[3] = de->d_name.len + 1;
 282
 283                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 284                                       size, NULL);
 285                 if (!req)
 286                         RETURN(-ENOMEM);
 287
 288                 /* pack the intent */
 289                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 290                 lit->opc = NTOH__u64((__u64)it->it_op);
 291
 292                 /* pack the intended request */
 293                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 294
 295                 /* we need to replay opens */
 296                 if (it->it_op == IT_OPEN)
 297                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 298
 299                 /* get ready for the reply */
 300                 req->rq_replen = lustre_msg_size(3, repsize);
 301         } else if (it->it_op == IT_READDIR) {
 302                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 303                                       size, NULL);
 304                 if (!req)
 305                         RETURN(-ENOMEM);
 306
 307                 /* get ready for the reply */
 308                 req->rq_replen = lustre_msg_size(1, repsize);
 309         } else {
 310                 LBUG();
 311                 RETURN(-EINVAL);
 312         }
 313 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 314         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 315                               lock_type, NULL, 0, lock_mode, &flags,
 316                               ldlm_completion_ast, mdc_blocking_ast, data,
 317                               datalen, lockh);
 318         if (rc == -ENOENT) {
 319                 /* This can go when we're sure that this can never happen */
 320                 LBUG();
 321         }
 322         if (rc == ELDLM_LOCK_ABORTED) {
 323                 lock_mode = 0;
 324                 memset(lockh, 0, sizeof(*lockh));
 325                 /* rc = 0 */
 326         } else if (rc != 0) {
 327                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 328                 RETURN(rc);
 329         }
 330
 331         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 332         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 333         it->it_status = (int) dlm_rep->lock_policy_res2;
 334         it->it_lock_mode = lock_mode;
 335         it->it_data = req;
 336
 337         RETURN(0);
 338 }
 339
 340 static void mdc_replay_open(struct ptlrpc_request *req, void *data)
 341 {
 342         __u64 *fh = data;
 343         struct mds_body *body;
 344
 345         body = lustre_msg_buf(req->rq_repmsg, 0);
 346         mds_unpack_body(body);
 347         *fh = body->extra;
 348 }
 349
 350 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 351              struct lov_stripe_md *smd, __u64 cookie, __u64 *fh,
 352              struct ptlrpc_request **request)
 353 {
 354         struct mds_body *body;
 355         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 356         struct ptlrpc_request *req;
 357         ENTRY;
 358
 359         if (smd != NULL) {
 360                 bufcount = 2;
 361                 size[1] = smd->lmd_mds_easize;
 362         }
 363
 364         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 365                               NULL);
 366         if (!req)
 367                 GOTO(out, rc = -ENOMEM);
 368
 369         req->rq_flags |= PTL_RPC_FL_REPLAY;
 370         body = lustre_msg_buf(req->rq_reqmsg, 0);
 371
 372         ll_ino2fid(&body->fid1, ino, 0, type);
 373         body->flags = HTON__u32(flags);
 374         body->extra = cookie;
 375
 376         if (smd != NULL)
 377                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), smd);
 378
 379         req->rq_replen = lustre_msg_size(1, size);
 380
 381         rc = ptlrpc_queue_wait(req);
 382         rc = ptlrpc_check_status(req, rc);
 383         if (!rc) {
 384                 body = lustre_msg_buf(req->rq_repmsg, 0);
 385                 mds_unpack_body(body);
 386                 *fh = body->extra;
 387         }
 388
 389         /* If open is replayed, we need to fix up the fh. */
 390         req->rq_replay_cb = mdc_replay_open;
 391         req->rq_replay_cb_data = fh;
 392
 393         EXIT;
 394  out:
 395         *request = req;
 396         return rc;
 397 }
 398
 399 int mdc_close(struct lustre_handle *conn,
 400               obd_id ino, int type, __u64 fh, struct ptlrpc_request **request)
 401 {
 402         struct mds_body *body;
 403         int rc, size = sizeof(*body);
 404         struct ptlrpc_request *req;
 405
 406         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 407                               NULL);
 408         if (!req)
 409                 GOTO(out, rc = -ENOMEM);
 410
 411         body = lustre_msg_buf(req->rq_reqmsg, 0);
 412         ll_ino2fid(&body->fid1, ino, 0, type);
 413         body->extra = fh;
 414
 415         req->rq_replen = lustre_msg_size(0, NULL);
 416
 417         rc = ptlrpc_queue_wait(req);
 418         rc = ptlrpc_check_status(req, rc);
 419
 420         EXIT;
 421  out:
 422         *request = req;
 423         return rc;
 424 }
 425
 426 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 427                  char *addr, struct ptlrpc_request **request)
 428 {
 429         struct ptlrpc_connection *connection =
 430                 client_conn2cli(conn)->cl_import.imp_connection;
 431         struct ptlrpc_request *req = NULL;
 432         struct ptlrpc_bulk_desc *desc = NULL;
 433         struct ptlrpc_bulk_page *bulk = NULL;
 434         struct mds_body *body;
 435         int rc, size = sizeof(*body);
 436         ENTRY;
 437
 438         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 439
 440         desc = ptlrpc_prep_bulk(connection);
 441         if (desc == NULL)
 442                 GOTO(out, rc = -ENOMEM);
 443
 444         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 445                               NULL);
 446         if (!req)
 447                 GOTO(out2, rc = -ENOMEM);
 448
 449         bulk = ptlrpc_prep_bulk_page(desc);
 450         bulk->bp_buflen = PAGE_SIZE;
 451         bulk->bp_buf = addr;
 452         bulk->bp_xid = req->rq_xid;
 453         desc->bd_portal = MDS_BULK_PORTAL;
 454
 455         rc = ptlrpc_register_bulk(desc);
 456         if (rc) {
 457                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 458                 GOTO(out2, rc);
 459         }
 460
 461         body = lustre_msg_buf(req->rq_reqmsg, 0);
 462         body->fid1.id = ino;
 463         body->fid1.f_type = type;
 464         body->size = offset;
 465
 466         req->rq_replen = lustre_msg_size(1, &size);
 467         rc = ptlrpc_queue_wait(req);
 468         rc = ptlrpc_check_status(req, rc);
 469         if (rc) {
 470                 ptlrpc_abort_bulk(desc);
 471                 GOTO(out2, rc);
 472         } else {
 473                 body = lustre_msg_buf(req->rq_repmsg, 0);
 474                 mds_unpack_body(body);
 475         }
 476
 477         EXIT;
 478  out2:
 479         ptlrpc_free_bulk(desc);
 480  out:
 481         *request = req;
 482         return rc;
 483 }
 484
 485 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 486                struct ptlrpc_request **request)
 487 {
 488         struct ptlrpc_request *req;
 489         int rc, size = sizeof(*osfs);
 490         ENTRY;
 491
 492         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 493                               NULL);
 494         if (!req)
 495                 GOTO(out, rc = -ENOMEM);
 496         req->rq_replen = lustre_msg_size(1, &size);
 497
 498         rc = ptlrpc_queue_wait(req);
 499         rc = ptlrpc_check_status(req, rc);
 500
 501         if (rc)
 502                 GOTO(out, rc);
 503
 504         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 505
 506         EXIT;
 507 out:
 508         *request = req;
 509
 510         return rc;
 511 }
 512
 513 struct obd_ops mdc_obd_ops = {
 514         o_setup:   client_obd_setup,
 515         o_cleanup: client_obd_cleanup,
 516         o_connect: client_obd_connect,
 517         o_disconnect: client_obd_disconnect,
 518 };
 519
 520 static int __init ptlrpc_request_init(void)
 521 {
 522         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 523 }
 524
 525 static void __exit ptlrpc_request_exit(void)
 526 {
 527         class_unregister_type(LUSTRE_MDC_NAME);
 528 }
 529
 530 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 531 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 532 MODULE_LICENSE("GPL");
 533
 534 EXPORT_SYMBOL(mdc_getstatus);
 535 EXPORT_SYMBOL(mdc_getlovinfo);
 536 EXPORT_SYMBOL(mdc_enqueue);
 537 EXPORT_SYMBOL(mdc_getattr);
 538 EXPORT_SYMBOL(mdc_statfs);
 539 EXPORT_SYMBOL(mdc_create);
 540 EXPORT_SYMBOL(mdc_unlink);
 541 EXPORT_SYMBOL(mdc_rename);
 542 EXPORT_SYMBOL(mdc_link);
 543 EXPORT_SYMBOL(mdc_readpage);
 544 EXPORT_SYMBOL(mdc_setattr);
 545 EXPORT_SYMBOL(mdc_close);
 546 EXPORT_SYMBOL(mdc_open);
 547
 548 module_init(ptlrpc_request_init);
 549 module_exit(ptlrpc_request_exit);