lustre/mds/handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/mds/handler.c
   5  *
   6  *  Lustre Metadata Server (mds) request handler
   7  *
   8  *  Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   9  *
  10  *  This code is issued under the GNU General Public License.
  11  *  See the file COPYING in this distribution
  12  *
  13  *  by Peter Braam <braam@clusterfs.com>
  14  *
  15  *  This server is single threaded at present (but can easily be multi threaded)
  16  *
  17  */
  18
  19 #define EXPORT_SYMTAB
  20
  21 #include <linux/version.h>
  22 #include <linux/module.h>
  23 #include <linux/fs.h>
  24 #include <linux/stat.h>
  25 #include <linux/locks.h>
  26 #include <linux/quotaops.h>
  27 #include <asm/unistd.h>
  28 #include <asm/uaccess.h>
  29
  30 #define DEBUG_SUBSYSTEM S_MDS
  31
  32 #include <linux/lustre_mds.h>
  33 #include <linux/lustre_lib.h>
  34 #include <linux/lustre_net.h>
  35
  36 int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  37                  __u64 offset, struct niobuf *dst)
  38 {
  39         int rc = 0;
  40         mm_segment_t oldfs = get_fs();
  41         struct ptlrpc_bulk_desc *bulk;
  42         char *buf;
  43
  44         bulk = ptlrpc_prep_bulk(req->rq_connection);
  45         if (bulk == NULL) {
  46                 rc = -ENOMEM;
  47                 GOTO(out, rc);
  48         }
  49
  50         bulk->b_xid = req->rq_reqmsg->xid;
  51
  52         OBD_ALLOC(buf, PAGE_SIZE);
  53         if (!buf) {
  54                 rc = -ENOMEM;
  55                 GOTO(cleanup_bulk, rc);
  56         }
  57
  58         set_fs(KERNEL_DS);
  59         rc = mds_fs_readpage(&req->rq_obd->u.mds, file, buf, PAGE_SIZE,
  60                              &offset);
  61         set_fs(oldfs);
  62
  63         if (rc != PAGE_SIZE) {
  64                 rc = -EIO;
  65                 GOTO(cleanup_buf, rc);
  66         }
  67
  68         bulk->b_buf = buf;
  69         bulk->b_buflen = PAGE_SIZE;
  70
  71         rc = ptlrpc_send_bulk(bulk, MDS_BULK_PORTAL);
  72         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  73                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  74                        OBD_FAIL_MDS_SENDPAGE, rc);
  75                 PtlMDUnlink(bulk->b_md_h);
  76                 GOTO(cleanup_buf, rc);
  77         }
  78         wait_event_interruptible(bulk->b_waitq,
  79                                  ptlrpc_check_bulk_sent(bulk));
  80
  81         if (bulk->b_flags == PTL_RPC_INTR) {
  82                 rc = -EINTR;
  83                 GOTO(cleanup_buf, rc);
  84         }
  85
  86         EXIT;
  87  cleanup_buf:
  88         OBD_FREE(buf, PAGE_SIZE);
  89  cleanup_bulk:
  90         OBD_FREE(bulk, sizeof(*bulk));
  91  out:
  92         return rc;
  93 }
  94
  95 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
  96                               struct vfsmount **mnt)
  97 {
  98         /* stolen from NFS */
  99         struct super_block *sb = mds->mds_sb;
 100         unsigned long ino = fid->id;
 101         __u32 generation = fid->generation;
 102         struct inode *inode;
 103         struct list_head *lp;
 104         struct dentry *result;
 105
 106         if (ino == 0)
 107                 return ERR_PTR(-ESTALE);
 108
 109         inode = iget(sb, ino);
 110         if (inode == NULL)
 111                 return ERR_PTR(-ENOMEM);
 112
 113         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 114
 115         if (is_bad_inode(inode) ||
 116             (generation && inode->i_generation != generation)) {
 117                 /* we didn't find the right inode.. */
 118                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 119                         inode->i_ino,
 120                         inode->i_nlink, atomic_read(&inode->i_count),
 121                         inode->i_generation,
 122                         generation);
 123                 LBUG();
 124                 iput(inode);
 125                 return ERR_PTR(-ESTALE);
 126         }
 127
 128         /* now to find a dentry.
 129          * If possible, get a well-connected one
 130          */
 131         if (mnt)
 132                 *mnt = mds->mds_vfsmnt;
 133         spin_lock(&dcache_lock);
 134         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 135                 result = list_entry(lp,struct dentry, d_alias);
 136                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 137                         dget_locked(result);
 138                         result->d_vfs_flags |= DCACHE_REFERENCED;
 139                         spin_unlock(&dcache_lock);
 140                         iput(inode);
 141                         if (mnt)
 142                                 mntget(*mnt);
 143                         return result;
 144                 }
 145         }
 146         spin_unlock(&dcache_lock);
 147         result = d_alloc_root(inode);
 148         if (result == NULL) {
 149                 iput(inode);
 150                 return ERR_PTR(-ENOMEM);
 151         }
 152         if (mnt)
 153                 mntget(*mnt);
 154         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 155         return result;
 156 }
 157
 158 int mds_getattr(struct ptlrpc_request *req)
 159 {
 160         struct dentry *de;
 161         struct inode *inode;
 162         struct mds_body *body;
 163         struct mds_obd *mds = &req->rq_obd->u.mds;
 164         int rc, size = sizeof(*body);
 165         ENTRY;
 166
 167         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 168         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 169                 CERROR("mds: out of memory\n");
 170                 req->rq_status = -ENOMEM;
 171                 RETURN(0);
 172         }
 173
 174         body = lustre_msg_buf(req->rq_reqmsg, 0);
 175         de = mds_fid2dentry(mds, &body->fid1, NULL);
 176         if (IS_ERR(de)) {
 177                 req->rq_status = -ENOENT;
 178                 RETURN(0);
 179         }
 180
 181         body = lustre_msg_buf(req->rq_repmsg, 0);
 182         inode = de->d_inode;
 183         body->ino = inode->i_ino;
 184         body->generation = inode->i_generation;
 185         body->atime = inode->i_atime;
 186         body->ctime = inode->i_ctime;
 187         body->mtime = inode->i_mtime;
 188         body->uid = inode->i_uid;
 189         body->gid = inode->i_gid;
 190         body->size = inode->i_size;
 191         body->mode = inode->i_mode;
 192         body->nlink = inode->i_nlink;
 193         body->valid = ~0;
 194         mds_fs_get_objid(mds, inode, &body->objid);
 195         l_dput(de);
 196         RETURN(0);
 197 }
 198
 199 int mds_open(struct ptlrpc_request *req)
 200 {
 201         struct dentry *de;
 202         struct mds_body *body;
 203         struct file *file;
 204         struct vfsmount *mnt;
 205         __u32 flags;
 206         int rc, size = sizeof(*body);
 207         ENTRY;
 208
 209         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 210         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 211                 CERROR("mds: out of memory\n");
 212                 req->rq_status = -ENOMEM;
 213                 RETURN(0);
 214         }
 215
 216         body = lustre_msg_buf(req->rq_reqmsg, 0);
 217         de = mds_fid2dentry(&req->rq_obd->u.mds, &body->fid1, &mnt);
 218         if (IS_ERR(de)) {
 219                 req->rq_status = -ENOENT;
 220                 RETURN(0);
 221         }
 222         flags = body->flags;
 223         file = dentry_open(de, mnt, flags);
 224         if (!file || IS_ERR(file)) {
 225                 req->rq_status = -EINVAL;
 226                 RETURN(0);
 227         }
 228
 229         body = lustre_msg_buf(req->rq_repmsg, 0);
 230         body->objid = (__u64) (unsigned long)file;
 231         RETURN(0);
 232 }
 233
 234 int mds_close(struct ptlrpc_request *req)
 235 {
 236         struct dentry *de;
 237         struct mds_body *body;
 238         struct file *file;
 239         struct vfsmount *mnt;
 240         int rc;
 241         ENTRY;
 242
 243         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 244         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 245                 CERROR("mds: out of memory\n");
 246                 req->rq_status = -ENOMEM;
 247                 RETURN(0);
 248         }
 249
 250         body = lustre_msg_buf(req->rq_reqmsg, 0);
 251         de = mds_fid2dentry(&req->rq_obd->u.mds, &body->fid1, &mnt);
 252         if (IS_ERR(de)) {
 253                 req->rq_status = -ENOENT;
 254                 RETURN(0);
 255         }
 256
 257         file = (struct file *)(unsigned long)body->objid;
 258         req->rq_status = filp_close(file, 0);
 259         l_dput(de);
 260         mntput(mnt);
 261
 262         RETURN(0);
 263 }
 264
 265 int mds_readpage(struct ptlrpc_request *req)
 266 {
 267         struct vfsmount *mnt;
 268         struct dentry *de;
 269         struct file *file;
 270         struct niobuf *niobuf;
 271         struct mds_body *body;
 272         int rc, size = sizeof(*body);
 273         ENTRY;
 274
 275         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 276         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 277                 CERROR("mds: out of memory\n");
 278                 req->rq_status = -ENOMEM;
 279                 RETURN(0);
 280         }
 281
 282         body = lustre_msg_buf(req->rq_reqmsg, 0);
 283         de = mds_fid2dentry(&req->rq_obd->u.mds, &body->fid1, &mnt);
 284         if (IS_ERR(de)) {
 285                 req->rq_status = PTR_ERR(de);
 286                 RETURN(0);
 287         }
 288
 289         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 290
 291         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 292         /* note: in case of an error, dentry_open puts dentry */
 293         if (IS_ERR(file)) {
 294                 req->rq_status = PTR_ERR(file);
 295                 RETURN(0);
 296         }
 297
 298         niobuf = lustre_msg_buf(req->rq_reqmsg, 1);
 299         if (!niobuf) {
 300                 req->rq_status = -EINVAL;
 301                 LBUG();
 302                 RETURN(0);
 303         }
 304
 305         /* to make this asynchronous make sure that the handling function
 306            doesn't send a reply when this function completes. Instead a
 307            callback function would send the reply */
 308         rc = mds_sendpage(req, file, body->size, niobuf);
 309
 310         filp_close(file, 0);
 311         req->rq_status = rc;
 312         RETURN(0);
 313 }
 314
 315 int mds_reint(struct ptlrpc_request *req)
 316 {
 317         int rc;
 318         struct mds_update_record rec;
 319
 320         rc = mds_update_unpack(req, &rec);
 321         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 322                 CERROR("invalid record\n");
 323                 req->rq_status = -EINVAL;
 324                 RETURN(0);
 325         }
 326         /* rc will be used to interrupt a for loop over multiple records */
 327         rc = mds_reint_rec(&rec, req);
 328         return 0;
 329 }
 330
 331 int mds_handle(struct obd_device *dev, struct ptlrpc_service *svc,
 332                struct ptlrpc_request *req)
 333 {
 334         int rc;
 335         ENTRY;
 336
 337         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 338         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 339                 CERROR("lustre_mds: Invalid request\n");
 340                 GOTO(out, rc);
 341         }
 342
 343         if (req->rq_reqmsg->type != PTL_RPC_REQUEST) {
 344                 CERROR("lustre_mds: wrong packet type sent %d\n",
 345                        req->rq_reqmsg->type);
 346                 GOTO(out, rc = -EINVAL);
 347         }
 348
 349         switch (req->rq_reqmsg->opc) {
 350         case MDS_GETATTR:
 351                 CDEBUG(D_INODE, "getattr\n");
 352                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 353                 rc = mds_getattr(req);
 354                 break;
 355
 356         case MDS_READPAGE:
 357                 CDEBUG(D_INODE, "readpage\n");
 358                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 359                 rc = mds_readpage(req);
 360
 361                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 362                         return 0;
 363                 break;
 364
 365         case MDS_REINT:
 366                 CDEBUG(D_INODE, "reint\n");
 367                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 368                 rc = mds_reint(req);
 369                 break;
 370
 371         case MDS_OPEN:
 372                 CDEBUG(D_INODE, "open\n");
 373                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 374                 rc = mds_open(req);
 375                 break;
 376
 377         case MDS_CLOSE:
 378                 CDEBUG(D_INODE, "close\n");
 379                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 380                 rc = mds_close(req);
 381                 break;
 382
 383         default:
 384                 rc = ptlrpc_error(svc, req);
 385                 RETURN(rc);
 386         }
 387
 388         EXIT;
 389 out:
 390         if (rc) {
 391                 ptlrpc_error(svc, req);
 392         } else {
 393                 CDEBUG(D_NET, "sending reply\n");
 394                 ptlrpc_reply(svc, req);
 395         }
 396
 397         return 0;
 398 }
 399
 400 static int mds_prep(struct obd_device *obddev)
 401 {
 402         struct obd_run_ctxt saved;
 403         struct mds_obd *mds = &obddev->u.mds;
 404         struct super_operations *s_ops;
 405         struct file *f;
 406         loff_t off = 0;
 407         __u64 mount_count;
 408         int rc;
 409
 410         mds->mds_service = ptlrpc_init_svc(128 * 1024,
 411                                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
 412                                            "self", mds_handle);
 413
 414         if (!mds->mds_service) {
 415                 CERROR("failed to start service\n");
 416                 RETURN(-EINVAL);
 417         }
 418
 419         rc = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
 420         if (rc) {
 421                 CERROR("cannot start thread\n");
 422                 GOTO(err_svc, rc);
 423         }
 424
 425         push_ctxt(&saved, &mds->mds_ctxt);
 426         rc = simple_mkdir(current->fs->pwd, "ROOT", 0700);
 427         if (rc && rc != -EEXIST) {
 428                 CERROR("cannot create ROOT directory\n");
 429                 GOTO(err_svc, rc);
 430         }
 431         rc = simple_mkdir(current->fs->pwd, "FH", 0700);
 432         if (rc && rc != -EEXIST) {
 433                 CERROR("cannot create FH directory\n");
 434                 GOTO(err_svc, rc);
 435         }
 436
 437         f = filp_open("mount_count", O_RDWR | O_CREAT, 0644);
 438         if (IS_ERR(f)) {
 439                 CERROR("cannot open/create mount_count file, rc = %ld\n",
 440                        PTR_ERR(f));
 441                 GOTO(err_svc, rc = PTR_ERR(f));
 442         }
 443         rc = lustre_fread(f, (char *)&mount_count, sizeof(mount_count), &off);
 444         if (rc == 0) {
 445                 CERROR("empty MDS mount_count, new MDS?\n");
 446                 /* XXX maybe this should just be a random number? */
 447                 mds->mds_mount_count = 0;
 448         } else if (rc != sizeof(mount_count)) {
 449                 CERROR("error reading mount_count: rc = %d\n", rc);
 450                 /* XXX maybe this should just be a random number? */
 451                 mds->mds_mount_count = 0;
 452         } else {
 453                 mds->mds_mount_count = le64_to_cpu(mount_count);
 454         }
 455
 456         mds->mds_mount_count++;
 457         CDEBUG(D_SUPER, "MDS mount_count is %Ld\n", mds->mds_mount_count);
 458         off = 0;
 459         mount_count = cpu_to_le64(mds->mds_mount_count);
 460         rc = lustre_fwrite(f, (char *)&mount_count, sizeof(mount_count), &off);
 461         if (rc != sizeof(mount_count))
 462                 CERROR("error writing mount_count: rc = %d\n", rc);
 463         rc = filp_close(f, 0);
 464         if (rc)
 465                 CERROR("error closing mount_count: rc = %d\n", rc);
 466
 467         f = filp_open("last_rcvd", O_RDWR | O_CREAT, 0644);
 468         if (IS_ERR(f)) {
 469                 CERROR("cannot open/create last_rcvd file\n");
 470                 GOTO(err_svc, rc = PTR_ERR(f));
 471         }
 472         mds->mds_last_rcvd = f;
 473         pop_ctxt(&saved);
 474
 475         /*
 476          * Replace the client filesystem delete_inode method with our own,
 477          * so that we can clear the object ID before the inode is deleted.
 478          * The fs_delete_inode method will call cl_delete_inode for us.
 479          *
 480          * We need to do this for the MDS superblock only, hence we install
 481          * a modified copy of the original superblock method table.
 482          *
 483          * We still assume that there is only a single MDS client filesystem
 484          * type, as we don't have access to the mds struct in * delete_inode.
 485          */
 486         OBD_ALLOC(s_ops, sizeof(*s_ops));
 487         memcpy(s_ops, mds->mds_sb->s_op, sizeof(*s_ops));
 488         mds->mds_fsops->cl_delete_inode = s_ops->delete_inode;
 489         s_ops->delete_inode = mds->mds_fsops->fs_delete_inode;
 490         mds->mds_sb->s_op = s_ops;
 491
 492         RETURN(0);
 493
 494 err_svc:
 495         rpc_unregister_service(mds->mds_service);
 496         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 497
 498         return rc;
 499 }
 500
 501 /* mount the file system (secretly) */
 502 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 503 {
 504         struct obd_ioctl_data* data = buf;
 505         struct mds_obd *mds = &obddev->u.mds;
 506         struct vfsmount *mnt;
 507         int err = 0;
 508         ENTRY;
 509
 510 #ifdef CONFIG_DEV_RDONLY
 511         dev_clear_rdonly(2);
 512 #endif
 513         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 514
 515         if (!strcmp(mds->mds_fstype, "ext3"))
 516                 mds->mds_fsops = &mds_ext3_fs_ops;
 517         else if (!strcmp(mds->mds_fstype, "ext2"))
 518                 mds->mds_fsops = &mds_ext2_fs_ops;
 519         else {
 520                 CERROR("unsupported MDS filesystem type %s\n", mds->mds_fstype);
 521                 GOTO(err_kfree, (err = -EPERM));
 522         }
 523
 524         MOD_INC_USE_COUNT;
 525         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
 526         if (IS_ERR(mnt)) {
 527                 CERROR("do_kern_mount failed: %d\n", err);
 528                 GOTO(err_dec, err = PTR_ERR(mnt));
 529         }
 530
 531         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 532         if (!mds->mds_sb)
 533                 GOTO(err_put, (err = -ENODEV));
 534
 535         mds->mds_vfsmnt = mnt;
 536         mds->mds_ctxt.pwdmnt = mnt;
 537         mds->mds_ctxt.pwd = mnt->mnt_root;
 538         mds->mds_ctxt.fs = KERNEL_DS;
 539
 540         err = mds_prep(obddev);
 541         if (err)
 542                 GOTO(err_put, err);
 543
 544         RETURN(0);
 545
 546 err_put:
 547         unlock_kernel();
 548         mntput(mds->mds_vfsmnt);
 549         mds->mds_sb = 0;
 550         lock_kernel();
 551 err_dec:
 552         MOD_DEC_USE_COUNT;
 553 err_kfree:
 554         kfree(mds->mds_fstype);
 555         return err;
 556 }
 557
 558 static int mds_cleanup(struct obd_device * obddev)
 559 {
 560         struct super_operations *s_ops = NULL;
 561         struct super_block *sb;
 562         struct mds_obd *mds = &obddev->u.mds;
 563
 564         ENTRY;
 565
 566         if ( !list_empty(&obddev->obd_gen_clients) ) {
 567                 CERROR("still has clients!\n");
 568                 RETURN(-EBUSY);
 569         }
 570
 571         ptlrpc_stop_thread(mds->mds_service);
 572         rpc_unregister_service(mds->mds_service);
 573         if (!list_empty(&mds->mds_service->srv_reqs)) {
 574                 // XXX reply with errors and clean up
 575                 CERROR("Request list not empty!\n");
 576         }
 577         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 578
 579         sb = mds->mds_sb;
 580         if (!mds->mds_sb)
 581                 RETURN(0);
 582
 583         if (mds->mds_last_rcvd) {
 584                 int rc = filp_close(mds->mds_last_rcvd, 0);
 585                 mds->mds_last_rcvd = NULL;
 586
 587                 if (rc)
 588                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
 589         }
 590         s_ops = sb->s_op;
 591
 592         unlock_kernel();
 593         mntput(mds->mds_vfsmnt);
 594         mds->mds_sb = 0;
 595         kfree(mds->mds_fstype);
 596         lock_kernel();
 597 #ifdef CONFIG_DEV_RDONLY
 598         dev_clear_rdonly(2);
 599 #endif
 600         OBD_FREE(s_ops, sizeof(*s_ops));
 601
 602         MOD_DEC_USE_COUNT;
 603         RETURN(0);
 604 }
 605
 606 /* use obd ops to offer management infrastructure */
 607 static struct obd_ops mds_obd_ops = {
 608         o_setup:       mds_setup,
 609         o_cleanup:     mds_cleanup,
 610 };
 611
 612 static int __init mds_init(void)
 613 {
 614         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
 615         return 0;
 616 }
 617
 618 static void __exit mds_exit(void)
 619 {
 620         obd_unregister_type(LUSTRE_MDS_NAME);
 621 }
 622
 623 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 624 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
 625 MODULE_LICENSE("GPL");
 626
 627 module_init(mds_init);
 628 module_exit(mds_exit);