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_con2cl(struct lustre_handle *conn, struct ptlrpc_client **cl,
  39                struct ptlrpc_connection **connection,
  40                struct lustre_handle **rconn)
  41 {
  42         struct obd_export *export;
  43         struct client_obd *mdc;
  44
  45         export = class_conn2export(conn);
  46         if (!export)
  47                 return -ENOTCONN;
  48
  49         mdc = &export->exp_obd->u.cli;
  50
  51         *cl = mdc->cl_client;
  52         *connection = mdc->cl_conn;
  53         *rconn = &export->exp_rconnh;
  54
  55         return 0;
  56 }
  57
  58 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid,
  59                   __u64 *last_committed, __u64 *last_rcvd,
  60                   __u32 *last_xid, struct ptlrpc_request **request)
  61 {
  62         struct ptlrpc_request *req;
  63         struct mds_body *body;
  64         int rc, size = sizeof(*body);
  65         ENTRY;
  66
  67         req = ptlrpc_prep_req2(conn, MDS_GETSTATUS, 1, &size, NULL);
  68         if (!req)
  69                 GOTO(out, rc = -ENOMEM);
  70
  71         body = lustre_msg_buf(req->rq_reqmsg, 0);
  72         req->rq_level = LUSTRE_CONN_CON;
  73         req->rq_replen = lustre_msg_size(1, &size);
  74
  75         mds_pack_req_body(req);
  76         rc = ptlrpc_queue_wait(req);
  77         rc = ptlrpc_check_status(req, rc);
  78
  79         if (!rc) {
  80                 body = lustre_msg_buf(req->rq_repmsg, 0);
  81                 mds_unpack_body(body);
  82                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  83                 *last_committed = req->rq_repmsg->last_committed;
  84                 *last_rcvd = req->rq_repmsg->last_rcvd;
  85                 *last_xid = body->last_xid;
  86
  87                 CDEBUG(D_NET, "root ino=%ld, last_committed=%Lu, last_rcvd=%Lu,"
  88                        " last_xid=%d\n",
  89                        (unsigned long)rootfid->id,
  90                        (unsigned long long)*last_committed,
  91                        (unsigned long long)*last_rcvd,
  92                        body->last_xid);
  93         }
  94
  95         EXIT;
  96  out:
  97         ptlrpc_free_req(req);
  98         return rc;
  99 }
 100
 101 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
 102                    uuid_t **uuids, struct ptlrpc_request **request)
 103 {
 104         struct ptlrpc_request *req;
 105         struct mds_status_req *streq;
 106         struct lov_obd *lov = &obd->u.lov;
 107         struct client_obd *mdc = &lov->mdcobd->u.cli;
 108         struct lov_desc *desc = &lov->desc;
 109         int rc, size[2] = {sizeof(*streq)};
 110         ENTRY;
 111
 112         req = ptlrpc_prep_req2(mdc_connh, MDS_GETLOVINFO, 1, size, NULL);
 113         if (!req)
 114                 GOTO(out, rc = -ENOMEM);
 115
 116         *request = req;
 117         streq = lustre_msg_buf(req->rq_reqmsg, 0);
 118         streq->flags = HTON__u32(MDS_STATUS_LOV);
 119         streq->repbuf = HTON__u32(8000);
 120
 121         /* prepare for reply */
 122         req->rq_level = LUSTRE_CONN_CON;
 123         size[0] = sizeof(*desc);
 124         size[1] = 8000;
 125         req->rq_replen = lustre_msg_size(2, size);
 126
 127         rc = ptlrpc_queue_wait(req);
 128         rc = ptlrpc_check_status(req, rc);
 129
 130         if (!rc) {
 131                 memcpy(desc, lustre_msg_buf(req->rq_repmsg, 0), sizeof(*desc));
 132                 *uuids = lustre_msg_buf(req->rq_repmsg, 1);
 133                 lov_unpackdesc(desc);
 134         }
 135         mdc->cl_max_mdsize = sizeof(*desc) +
 136                 desc->ld_tgt_count * sizeof(uuid_t);
 137
 138         EXIT;
 139  out:
 140         return rc;
 141 }
 142
 143
 144 int mdc_getattr(struct lustre_handle *conn,
 145                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 146                 struct ptlrpc_request **request)
 147 {
 148         struct ptlrpc_client *cl;
 149         struct ptlrpc_connection *connection;
 150         struct lustre_handle *rconn;
 151         struct ptlrpc_request *req;
 152         struct mds_body *body;
 153         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 154         ENTRY;
 155
 156         mdc_con2cl(conn, &cl, &connection, &rconn);
 157         req = ptlrpc_prep_req2(conn, MDS_GETATTR, 1, size, NULL);
 158         if (!req)
 159                 GOTO(out, rc = -ENOMEM);
 160
 161         body = lustre_msg_buf(req->rq_reqmsg, 0);
 162         ll_ino2fid(&body->fid1, ino, 0, type);
 163         body->valid = valid;
 164
 165         if (S_ISREG(type)) {
 166                 struct client_obd *mdc = &class_conn2obd(conn)->u.cli;
 167                 bufcount = 2;
 168                 size[1] = mdc->cl_max_mdsize;
 169         } else if (valid & OBD_MD_LINKNAME) {
 170                 bufcount = 2;
 171                 size[1] = ea_size;
 172         }
 173         req->rq_replen = lustre_msg_size(bufcount, size);
 174
 175         rc = ptlrpc_queue_wait(req);
 176         rc = ptlrpc_check_status(req, rc);
 177
 178         if (!rc) {
 179                 body = lustre_msg_buf(req->rq_repmsg, 0);
 180                 mds_unpack_body(body);
 181                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 182         }
 183
 184         EXIT;
 185  out:
 186         *request = req;
 187         return rc;
 188 }
 189
 190 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 191                             void *data, __u32 data_len)
 192 {
 193         int rc;
 194         struct inode *inode = data;
 195         struct lustre_handle lockh;
 196         ENTRY;
 197
 198         if (data_len != sizeof(*inode)) {
 199                 CERROR("data_len should be %d, but is %d\n", sizeof(*inode),
 200                        data_len);
 201                 LBUG();
 202                 RETURN(-EINVAL);
 203         }
 204
 205         /* FIXME: do something better than throwing away everything */
 206         if (inode == NULL)
 207                 LBUG();
 208         if (S_ISDIR(inode->i_mode)) {
 209                 CDEBUG(D_INODE, "invalidating inode %ld\n", inode->i_ino);
 210                 invalidate_inode_pages(inode);
 211         }
 212
 213         ldlm_lock2handle(lock, &lockh);
 214         rc = ldlm_cli_cancel(&lockh);
 215         if (rc < 0) {
 216                 CERROR("ldlm_cli_cancel: %d\n", rc);
 217                 LBUG();
 218         }
 219         RETURN(0);
 220 }
 221
 222 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 223                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 224                 struct dentry *de, struct lustre_handle *lockh, __u64 id,
 225                 char *tgt, int tgtlen, void *data, int datalen)
 226 {
 227         struct ptlrpc_request *req;
 228         struct obd_device *obddev = class_conn2obd(conn);
 229         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino};
 230         int size[5] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 231         int rc, flags;
 232         int repsize[3] = {sizeof(struct ldlm_reply),
 233                           sizeof(struct mds_body),
 234                           obddev->u.cli.cl_max_mdsize};
 235         struct ldlm_reply *dlm_rep;
 236         struct ldlm_intent *lit;
 237         ENTRY;
 238
 239         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op), dir->i_ino);
 240
 241         switch (it->it_op) {
 242         case IT_MKDIR:
 243                 it->it_mode = (it->it_mode | S_IFDIR) & ~current->fs->umask;
 244                 break;
 245         case (IT_CREAT|IT_OPEN):
 246         case IT_CREAT:
 247                 it->it_mode |= S_IFREG; /* no break */
 248         case IT_MKNOD:
 249                 it->it_mode &= ~current->fs->umask;
 250                 break;
 251         case IT_SYMLINK:
 252                 it->it_mode = (it->it_mode | S_IFLNK) & ~current->fs->umask;
 253                 break;
 254         }
 255
 256         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 257                 size[2] = sizeof(struct mds_rec_create);
 258                 size[3] = de->d_name.len + 1;
 259                 size[4] = tgtlen + 1;
 260                 req = ptlrpc_prep_req2(conn, LDLM_ENQUEUE, 5, size, NULL);
 261                 if (!req)
 262                         RETURN(-ENOMEM);
 263
 264                 /* pack the intent */
 265                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 266                 lit->opc = NTOH__u64((__u64)it->it_op);
 267
 268                 /* pack the intended request */
 269                 mds_create_pack(req, 2, dir, it->it_mode, id, current->fsuid,
 270                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 271                                 de->d_name.len, tgt, tgtlen);
 272                 req->rq_replen = lustre_msg_size(3, repsize);
 273         } else if (it->it_op == IT_RENAME2) {
 274                 struct dentry *old_de = it->it_data;
 275
 276                 size[2] = sizeof(struct mds_rec_rename);
 277                 size[3] = old_de->d_name.len + 1;
 278                 size[4] = de->d_name.len + 1;
 279                 req = ptlrpc_prep_req2(conn, LDLM_ENQUEUE, 5, size, NULL);
 280                 if (!req)
 281                         RETURN(-ENOMEM);
 282
 283                 /* pack the intent */
 284                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 285                 lit->opc = NTOH__u64((__u64)it->it_op);
 286
 287                 /* pack the intended request */
 288                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 289                                 old_de->d_name.name, old_de->d_name.len,
 290                                 de->d_name.name, de->d_name.len);
 291                 req->rq_replen = lustre_msg_size(1, repsize);
 292         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 293                 size[2] = sizeof(struct mds_rec_unlink);
 294                 size[3] = de->d_name.len + 1;
 295                 req = ptlrpc_prep_req2(conn, LDLM_ENQUEUE, 4, size, NULL);
 296                 if (!req)
 297                         RETURN(-ENOMEM);
 298
 299                 /* pack the intent */
 300                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 301                 lit->opc = NTOH__u64((__u64)it->it_op);
 302
 303                 /* pack the intended request */
 304                 mds_unlink_pack(req, 2, dir, NULL, de->d_name.name,
 305                                 de->d_name.len);
 306
 307                 req->rq_replen = lustre_msg_size(3, repsize);
 308         } else if (it->it_op == IT_GETATTR || it->it_op == IT_RENAME ||
 309                    it->it_op == IT_OPEN || it->it_op == IT_SETATTR ||
 310                    it->it_op == IT_LOOKUP || it->it_op == IT_READLINK) {
 311                 size[2] = sizeof(struct mds_body);
 312                 size[3] = de->d_name.len + 1;
 313
 314                 req = ptlrpc_prep_req2(conn, LDLM_ENQUEUE, 4, 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                 /* get ready for the reply */
 326                 req->rq_replen = lustre_msg_size(3, repsize);
 327         } else if (it->it_op == IT_READDIR) {
 328                 req = ptlrpc_prep_req2(conn, LDLM_ENQUEUE, 1, size, NULL);
 329                 if (!req)
 330                         RETURN(-ENOMEM);
 331
 332                 /* get ready for the reply */
 333                 req->rq_replen = lustre_msg_size(1, repsize);
 334         } else {
 335                 LBUG();
 336                 RETURN(-EINVAL);
 337         }
 338 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 339         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 340                               lock_type, NULL, 0, lock_mode, &flags,
 341                               ldlm_completion_ast, mdc_blocking_ast, data,
 342                               datalen, lockh);
 343         if (rc == -ENOENT) {
 344                 /* This can go when we're sure that this can never happen */
 345                 LBUG();
 346         }
 347         if (rc == ELDLM_LOCK_ABORTED) {
 348                 lock_mode = 0;
 349                 memset(lockh, 0, sizeof(*lockh));
 350                 /* rc = 0 */
 351         } else if (rc != 0) {
 352                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 353                 RETURN(rc);
 354         }
 355
 356         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 357         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 358         it->it_status = (int) dlm_rep->lock_policy_res2;
 359         it->it_lock_mode = lock_mode;
 360         it->it_data = req;
 361
 362         RETURN(0);
 363 }
 364
 365 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 366              struct lov_stripe_md *smd, __u64 cookie, __u64 *fh,
 367              struct ptlrpc_request **request)
 368 {
 369         struct ptlrpc_client *cl;
 370         struct ptlrpc_connection *connection;
 371         struct lustre_handle *rconn;
 372         struct mds_body *body;
 373         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 374         struct ptlrpc_request *req;
 375         ENTRY;
 376
 377         if (smd != NULL) {
 378                 bufcount = 2;
 379                 size[1] = smd->lmd_easize;
 380         }
 381
 382         mdc_con2cl(conn, &cl, &connection, &rconn);
 383         req = ptlrpc_prep_req2(conn, MDS_OPEN, bufcount, size, NULL);
 384         if (!req)
 385                 GOTO(out, rc = -ENOMEM);
 386
 387         req->rq_flags |= PTL_RPC_FL_REPLAY;
 388         body = lustre_msg_buf(req->rq_reqmsg, 0);
 389
 390         ll_ino2fid(&body->fid1, ino, 0, type);
 391         body->flags = HTON__u32(flags);
 392         body->extra = cookie;
 393
 394         if (smd != NULL)
 395                 memcpy(lustre_msg_buf(req->rq_reqmsg, 1), smd, smd->lmd_easize);
 396
 397         req->rq_replen = lustre_msg_size(1, size);
 398
 399         rc = ptlrpc_queue_wait(req);
 400         rc = ptlrpc_check_status(req, rc);
 401         if (!rc) {
 402                 body = lustre_msg_buf(req->rq_repmsg, 0);
 403                 mds_unpack_body(body);
 404                 *fh = body->extra;
 405         }
 406
 407         EXIT;
 408  out:
 409         *request = req;
 410         return rc;
 411 }
 412
 413 int mdc_close(struct lustre_handle *conn,
 414               obd_id ino, int type, __u64 fh, struct ptlrpc_request **request)
 415 {
 416         struct mds_body *body;
 417         int rc, size = sizeof(*body);
 418         struct ptlrpc_request *req;
 419
 420         req = ptlrpc_prep_req2(conn, MDS_CLOSE, 1, &size, NULL);
 421         if (!req)
 422                 GOTO(out, rc = -ENOMEM);
 423
 424         body = lustre_msg_buf(req->rq_reqmsg, 0);
 425         ll_ino2fid(&body->fid1, ino, 0, type);
 426         body->extra = fh;
 427
 428         req->rq_replen = lustre_msg_size(0, NULL);
 429
 430         rc = ptlrpc_queue_wait(req);
 431         rc = ptlrpc_check_status(req, rc);
 432
 433         EXIT;
 434  out:
 435         *request = req;
 436         return rc;
 437 }
 438
 439 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 440                  char *addr, struct ptlrpc_request **request)
 441 {
 442         struct ptlrpc_connection *connection = client_conn2cli(conn)->cl_conn;
 443         struct ptlrpc_request *req = NULL;
 444         struct ptlrpc_bulk_desc *desc = NULL;
 445         struct ptlrpc_bulk_page *bulk = NULL;
 446         struct mds_body *body;
 447         int rc, size = sizeof(*body);
 448         ENTRY;
 449
 450         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 451
 452         desc = ptlrpc_prep_bulk(connection);
 453         if (desc == NULL)
 454                 GOTO(out, rc = -ENOMEM);
 455
 456         req = ptlrpc_prep_req2(conn, MDS_READPAGE, 1, &size, NULL);
 457         if (!req)
 458                 GOTO(out2, rc = -ENOMEM);
 459
 460         bulk = ptlrpc_prep_bulk_page(desc);
 461         bulk->b_buflen = PAGE_SIZE;
 462         bulk->b_buf = addr;
 463         bulk->b_xid = req->rq_xid;
 464         desc->b_portal = MDS_BULK_PORTAL;
 465
 466         rc = ptlrpc_register_bulk(desc);
 467         if (rc) {
 468                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 469                 GOTO(out2, rc);
 470         }
 471
 472         body = lustre_msg_buf(req->rq_reqmsg, 0);
 473         body->fid1.id = ino;
 474         body->fid1.f_type = type;
 475         body->size = offset;
 476
 477         req->rq_replen = lustre_msg_size(1, &size);
 478         rc = ptlrpc_queue_wait(req);
 479         rc = ptlrpc_check_status(req, rc);
 480         if (rc) {
 481                 ptlrpc_abort_bulk(desc);
 482                 GOTO(out2, rc);
 483         } else {
 484                 body = lustre_msg_buf(req->rq_repmsg, 0);
 485                 mds_unpack_body(body);
 486         }
 487
 488         EXIT;
 489  out2:
 490         ptlrpc_free_bulk(desc);
 491  out:
 492         *request = req;
 493         return rc;
 494 }
 495
 496 int mdc_statfs(struct lustre_handle *conn, struct statfs *sfs,
 497                struct ptlrpc_request **request)
 498 {
 499         struct obd_statfs *osfs;
 500         struct ptlrpc_request *req;
 501         int rc, size = sizeof(*osfs);
 502         ENTRY;
 503
 504         req = ptlrpc_prep_req2(conn, MDS_STATFS, 0, NULL, NULL);
 505         if (!req)
 506                 GOTO(out, rc = -ENOMEM);
 507         req->rq_replen = lustre_msg_size(1, &size);
 508
 509         rc = ptlrpc_queue_wait(req);
 510         rc = ptlrpc_check_status(req, rc);
 511
 512         if (rc)
 513                 GOTO(out, rc);
 514
 515         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 516         obd_statfs_unpack(osfs, sfs);
 517
 518         EXIT;
 519 out:
 520         *request = req;
 521
 522         return rc;
 523 }
 524
 525 struct obd_ops mdc_obd_ops = {
 526         o_setup:   client_obd_setup,
 527         o_cleanup: client_obd_cleanup,
 528         o_connect: client_obd_connect,
 529         o_disconnect: client_obd_disconnect,
 530 };
 531
 532 static int __init ptlrpc_request_init(void)
 533 {
 534         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 535 }
 536
 537 static void __exit ptlrpc_request_exit(void)
 538 {
 539         class_unregister_type(LUSTRE_MDC_NAME);
 540 }
 541
 542 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 543 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 544 MODULE_LICENSE("GPL");
 545
 546 EXPORT_SYMBOL(mdc_getstatus);
 547 EXPORT_SYMBOL(mdc_getlovinfo);
 548 EXPORT_SYMBOL(mdc_enqueue);
 549 EXPORT_SYMBOL(mdc_getattr);
 550 EXPORT_SYMBOL(mdc_statfs);
 551 EXPORT_SYMBOL(mdc_create);
 552 EXPORT_SYMBOL(mdc_unlink);
 553 EXPORT_SYMBOL(mdc_rename);
 554 EXPORT_SYMBOL(mdc_link);
 555 EXPORT_SYMBOL(mdc_readpage);
 556 EXPORT_SYMBOL(mdc_setattr);
 557 EXPORT_SYMBOL(mdc_close);
 558 EXPORT_SYMBOL(mdc_open);
 559
 560 module_init(ptlrpc_request_init);
 561 module_exit(ptlrpc_request_exit);