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/ext2_fs.h>
  27 #include <linux/quotaops.h>
  28 #include <asm/unistd.h>
  29 #include <asm/uaccess.h>
  30
  31 #define DEBUG_SUBSYSTEM S_MDS
  32
  33 #include <linux/lustre_mds.h>
  34 #include <linux/lustre_lib.h>
  35 #include <linux/lustre_net.h>
  36
  37 struct buffer_head *ext3_bread(void *handle, struct inode *inode,
  38                                int block, int create, int *err);
  39
  40 int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  41                  __u64 offset, struct niobuf *dst)
  42 {
  43         int rc = 0;
  44         mm_segment_t oldfs = get_fs();
  45
  46         OBD_FAIL_RETURN(OBD_FAIL_MDS_SENDPAGE, -EIO);
  47
  48         if (req->rq_peer.peer_nid == 0) {
  49                 struct inode *inode = file->f_dentry->d_inode;
  50                 char *buf = (char *)(long)dst->addr;
  51
  52                 /* dst->addr is a user address, but in a different task! */
  53                 set_fs(KERNEL_DS);
  54                 /* FIXME: we need to use ext3_bread because ext3 does not
  55                  *        have the directories in page cache yet.  If we
  56                  *        just use generic_file_read() then the pages we
  57                  *        get are in a different address space than those
  58                  *        used by the filesystem == cache incoherency.
  59                  */
  60                 if (S_ISREG(inode->i_mode))
  61                         rc = file->f_op->read(file, buf, PAGE_SIZE, &offset);
  62                 else if (!strcmp(inode->i_sb->s_type->name, "ext3")) {
  63                         struct buffer_head *bh;
  64
  65                         bh = ext3_bread(NULL, inode,
  66                                         offset >> inode->i_sb->s_blocksize_bits,
  67                                         0, &rc);
  68
  69                         if (bh) {
  70                                 memcpy(buf, bh->b_data, inode->i_blksize);
  71                                 brelse(bh);
  72                                 rc = inode->i_blksize;
  73                         }
  74                 } else
  75                         rc = generic_file_read(file, buf, PAGE_SIZE, &offset);
  76
  77                 set_fs(oldfs);
  78
  79                 if (rc != PAGE_SIZE) {
  80                         rc = -EIO;
  81                         GOTO(out, rc);
  82                 }
  83                 EXIT;
  84         } else {
  85                 struct inode *inode = file->f_dentry->d_inode;
  86                 struct ptlrpc_bulk_desc *bulk;
  87                 char *buf;
  88
  89                 bulk = ptlrpc_prep_bulk(&req->rq_peer);
  90                 if (bulk == NULL) {
  91                         rc = -ENOMEM;
  92                         GOTO(out, rc);
  93                 }
  94
  95                 bulk->b_xid = req->rq_xid;
  96
  97                 OBD_ALLOC(buf, PAGE_SIZE);
  98                 if (!buf) {
  99                         rc = -ENOMEM;
 100                         GOTO(cleanup_bulk, rc);
 101                 }
 102
 103                 set_fs(KERNEL_DS);
 104                 /* FIXME: see comments above */
 105                 if (S_ISREG(inode->i_mode))
 106                         rc = file->f_op->read(file, buf, PAGE_SIZE, &offset);
 107                 else if (!strcmp(inode->i_sb->s_type->name, "ext3")) {
 108                         struct buffer_head *bh;
 109
 110                         bh = ext3_bread(NULL, inode, offset >> inode->i_blkbits,
 111                                         0, &rc);
 112
 113                         if (bh) {
 114                                 memcpy(buf, bh->b_data, inode->i_blksize);
 115                                 brelse(bh);
 116                                 rc = inode->i_blksize;
 117                         }
 118                 } else
 119                         rc = generic_file_read(file, buf, PAGE_SIZE, &offset);
 120
 121                 set_fs(oldfs);
 122
 123                 if (rc != PAGE_SIZE) {
 124                         rc = -EIO;
 125                         GOTO(cleanup_buf, rc);
 126                 }
 127
 128                 bulk->b_buf = buf;
 129                 bulk->b_buflen = PAGE_SIZE;
 130
 131                 rc = ptlrpc_send_bulk(bulk, MDS_BULK_PORTAL);
 132                 wait_event_interruptible(bulk->b_waitq,
 133                                          ptlrpc_check_bulk_sent(bulk));
 134
 135                 if (bulk->b_flags == PTL_RPC_INTR) {
 136                         rc = -EINTR;
 137                         GOTO(cleanup_buf, rc);
 138                 }
 139
 140                 EXIT;
 141         cleanup_buf:
 142                 OBD_FREE(buf, PAGE_SIZE);
 143         cleanup_bulk:
 144                 OBD_FREE(bulk, sizeof(*bulk));
 145         }
 146 out:
 147         return rc;
 148 }
 149
 150 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
 151                               struct vfsmount **mnt)
 152 {
 153         /* stolen from NFS */
 154         struct super_block *sb = mds->mds_sb;
 155         unsigned long ino = fid->id;
 156         //__u32 generation = fid->generation;
 157         __u32 generation = 0;
 158         struct inode *inode;
 159         struct list_head *lp;
 160         struct dentry *result;
 161
 162         if (ino == 0)
 163                 return ERR_PTR(-ESTALE);
 164
 165         inode = iget(sb, ino);
 166         if (inode == NULL)
 167                 return ERR_PTR(-ENOMEM);
 168
 169         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 170
 171         if (is_bad_inode(inode) ||
 172             (generation && inode->i_generation != generation)) {
 173                 /* we didn't find the right inode.. */
 174                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 175                         inode->i_ino,
 176                         inode->i_nlink, atomic_read(&inode->i_count),
 177                         inode->i_generation,
 178                         generation);
 179                 LBUG();
 180                 iput(inode);
 181                 return ERR_PTR(-ESTALE);
 182         }
 183
 184         /* now to find a dentry.
 185          * If possible, get a well-connected one
 186          */
 187         if (mnt)
 188                 *mnt = mds->mds_vfsmnt;
 189         spin_lock(&dcache_lock);
 190         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 191                 result = list_entry(lp,struct dentry, d_alias);
 192                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 193                         dget_locked(result);
 194                         result->d_vfs_flags |= DCACHE_REFERENCED;
 195                         spin_unlock(&dcache_lock);
 196                         iput(inode);
 197                         if (mnt)
 198                                 mntget(*mnt);
 199                         return result;
 200                 }
 201         }
 202         spin_unlock(&dcache_lock);
 203         result = d_alloc_root(inode);
 204         if (result == NULL) {
 205                 iput(inode);
 206                 return ERR_PTR(-ENOMEM);
 207         }
 208         if (mnt)
 209                 mntget(*mnt);
 210         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 211         return result;
 212 }
 213
 214 static inline void mds_get_objid(struct inode *inode, __u64 *id)
 215 {
 216         /* FIXME: it is only by luck that this works on ext3 */
 217         memcpy(id, &inode->u.ext2_i.i_data, sizeof(*id));
 218 }
 219
 220 int mds_getattr(struct ptlrpc_request *req)
 221 {
 222         struct dentry *de;
 223         struct inode *inode;
 224         struct mds_rep *rep;
 225         int rc;
 226
 227         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 228                           &req->rq_replen, &req->rq_repbuf);
 229         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 230                 CERROR("mds: out of memory\n");
 231                 req->rq_status = -ENOMEM;
 232                 RETURN(0);
 233         }
 234
 235         req->rq_rephdr->xid = req->rq_reqhdr->xid;
 236         rep = req->rq_rep.mds;
 237
 238         de = mds_fid2dentry(&req->rq_obd->u.mds, &req->rq_req.mds->fid1, NULL);
 239         if (IS_ERR(de)) {
 240                 req->rq_rephdr->status = -ENOENT;
 241                 RETURN(0);
 242         }
 243
 244         inode = de->d_inode;
 245         rep->ino = inode->i_ino;
 246         rep->atime = inode->i_atime;
 247         rep->ctime = inode->i_ctime;
 248         rep->mtime = inode->i_mtime;
 249         rep->uid = inode->i_uid;
 250         rep->gid = inode->i_gid;
 251         rep->size = inode->i_size;
 252         rep->mode = inode->i_mode;
 253         rep->nlink = inode->i_nlink;
 254         rep->valid = ~0;
 255         mds_get_objid(inode, &rep->objid);
 256         dput(de);
 257         return 0;
 258 }
 259
 260 int mds_open(struct ptlrpc_request *req)
 261 {
 262         struct dentry *de;
 263         struct mds_rep *rep;
 264         struct file *file;
 265         struct vfsmount *mnt;
 266         __u32 flags;
 267         int rc;
 268
 269         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 270                           &req->rq_replen, &req->rq_repbuf);
 271         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 272                 CERROR("mds: out of memory\n");
 273                 req->rq_status = -ENOMEM;
 274                 RETURN(0);
 275         }
 276
 277         req->rq_rephdr->xid = req->rq_reqhdr->xid;
 278         rep = req->rq_rep.mds;
 279
 280         de = mds_fid2dentry(&req->rq_obd->u.mds, &req->rq_req.mds->fid1, &mnt);
 281         if (IS_ERR(de)) {
 282                 req->rq_rephdr->status = -ENOENT;
 283                 RETURN(0);
 284         }
 285         flags = req->rq_req.mds->flags;
 286         file = dentry_open(de, mnt, flags);
 287         if (!file || IS_ERR(file)) {
 288                 req->rq_rephdr->status = -EINVAL;
 289                 RETURN(0);
 290         }
 291
 292         rep->objid = (__u64) (unsigned long)file;
 293         return 0;
 294 }
 295
 296 int mds_close(struct ptlrpc_request *req)
 297 {
 298         struct dentry *de;
 299         struct mds_rep *rep;
 300         struct file *file;
 301         struct vfsmount *mnt;
 302         int rc;
 303
 304         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 305                           &req->rq_replen, &req->rq_repbuf);
 306         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 307                 CERROR("mds: out of memory\n");
 308                 req->rq_status = -ENOMEM;
 309                 RETURN(0);
 310         }
 311
 312         req->rq_rephdr->xid = req->rq_reqhdr->xid;
 313         rep = req->rq_rep.mds;
 314
 315         de = mds_fid2dentry(&req->rq_obd->u.mds, &req->rq_req.mds->fid1, &mnt);
 316         if (IS_ERR(de)) {
 317                 req->rq_rephdr->status = -ENOENT;
 318                 RETURN(0);
 319         }
 320
 321         file = (struct file *)(unsigned long) req->rq_req.mds->objid;
 322
 323         req->rq_rephdr->status = filp_close(file, 0);
 324         dput(de);
 325         mntput(mnt);
 326         return 0;
 327 }
 328
 329
 330 int mds_readpage(struct ptlrpc_request *req)
 331 {
 332         struct vfsmount *mnt;
 333         struct dentry *de;
 334         struct file *file;
 335         struct niobuf *niobuf;
 336         struct mds_rep *rep;
 337         int rc;
 338
 339         ENTRY;
 340
 341         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 342                           &req->rq_replen, &req->rq_repbuf);
 343         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 344                 CERROR("mds: out of memory\n");
 345                 req->rq_status = -ENOMEM;
 346                 RETURN(0);
 347         }
 348
 349         req->rq_rephdr->xid = req->rq_reqhdr->xid;
 350         rep = req->rq_rep.mds;
 351
 352         de = mds_fid2dentry(&req->rq_obd->u.mds, &req->rq_req.mds->fid1, &mnt);
 353         if (IS_ERR(de)) {
 354                 req->rq_rephdr->status = PTR_ERR(de);
 355                 RETURN(0);
 356         }
 357
 358         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 359
 360         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 361         /* note: in case of an error, dentry_open puts dentry */
 362         if (IS_ERR(file)) {
 363                 req->rq_rephdr->status = PTR_ERR(file);
 364                 RETURN(0);
 365         }
 366
 367         niobuf = mds_req_tgt(req->rq_req.mds);
 368
 369         /* to make this asynchronous make sure that the handling function
 370            doesn't send a reply when this function completes. Instead a
 371            callback function would send the reply */
 372         rc = mds_sendpage(req, file, req->rq_req.mds->size, niobuf);
 373
 374         filp_close(file, 0);
 375         req->rq_rephdr->status = rc;
 376         RETURN(0);
 377 }
 378
 379 int mds_reint(struct ptlrpc_request *req)
 380 {
 381         char *buf;
 382         int rc, len;
 383         struct mds_update_record rec;
 384
 385         buf = mds_req_tgt(req->rq_req.mds);
 386         len = req->rq_req.mds->tgtlen;
 387
 388         rc = mds_update_unpack(buf, len, &rec);
 389         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 390                 CERROR("invalid record\n");
 391                 req->rq_status = -EINVAL;
 392                 RETURN(0);
 393         }
 394         /* rc will be used to interrupt a for loop over multiple records */
 395         rc = mds_reint_rec(&rec, req);
 396         return 0;
 397 }
 398
 399 int mds_handle(struct obd_device *dev, struct ptlrpc_service *svc,
 400                struct ptlrpc_request *req)
 401 {
 402         int rc;
 403         struct ptlreq_hdr *hdr;
 404
 405         ENTRY;
 406
 407         hdr = (struct ptlreq_hdr *)req->rq_reqbuf;
 408
 409         if (NTOH__u32(hdr->type) != PTL_RPC_REQUEST) {
 410                 CERROR("lustre_mds: wrong packet type sent %d\n",
 411                        NTOH__u32(hdr->type));
 412                 rc = -EINVAL;
 413                 GOTO(out, rc);
 414         }
 415
 416         rc = mds_unpack_req(req->rq_reqbuf, req->rq_reqlen,
 417                             &req->rq_reqhdr, &req->rq_req);
 418         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 419                 CERROR("lustre_mds: Invalid request\n");
 420                 GOTO(out, rc);
 421         }
 422
 423         switch (req->rq_reqhdr->opc) {
 424
 425         case MDS_GETATTR:
 426                 CDEBUG(D_INODE, "getattr\n");
 427                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 428                 rc = mds_getattr(req);
 429                 break;
 430
 431         case MDS_READPAGE:
 432                 CDEBUG(D_INODE, "readpage\n");
 433                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 434                 rc = mds_readpage(req);
 435                 break;
 436
 437         case MDS_REINT:
 438                 CDEBUG(D_INODE, "reint\n");
 439                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 440                 rc = mds_reint(req);
 441                 break;
 442
 443         case MDS_OPEN:
 444                 CDEBUG(D_INODE, "open\n");
 445                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 446                 rc = mds_open(req);
 447                 break;
 448
 449         case MDS_CLOSE:
 450                 CDEBUG(D_INODE, "close\n");
 451                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 452                 rc = mds_close(req);
 453                 break;
 454
 455         default:
 456                 rc = ptlrpc_error(dev, svc, req);
 457                 RETURN(rc);
 458         }
 459
 460         EXIT;
 461 out:
 462         if (rc) {
 463                 CERROR("no header\n");
 464                 return 0;
 465         }
 466
 467         if( req->rq_status) {
 468                 ptlrpc_error(dev, svc, req);
 469         } else {
 470                 CDEBUG(D_NET, "sending reply\n");
 471                 ptlrpc_reply(dev, svc, req);
 472         }
 473
 474         return 0;
 475 }
 476
 477
 478 /* mount the file system (secretly) */
 479 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 480 {
 481         struct obd_ioctl_data* data = buf;
 482         struct mds_obd *mds = &obddev->u.mds;
 483         struct vfsmount *mnt;
 484         int err;
 485         ENTRY;
 486
 487         mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, NULL);
 488         err = PTR_ERR(mnt);
 489         if (IS_ERR(mnt)) {
 490                 CERROR("do_kern_mount failed: %d\n", err);
 491                 RETURN(err);
 492         }
 493
 494         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 495         if (!mds->mds_sb)
 496                 RETURN(-ENODEV);
 497
 498         mds->mds_vfsmnt = mnt;
 499         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 500
 501         mds->mds_ctxt.pwdmnt = mnt;
 502         mds->mds_ctxt.pwd = mnt->mnt_root;
 503         mds->mds_ctxt.fs = KERNEL_DS;
 504
 505         mds->mds_service = ptlrpc_init_svc(64 * 1024,
 506                                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
 507                                            "self", mds_handle);
 508
 509         rpc_register_service(mds->mds_service, "self");
 510
 511         err = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
 512         if (err)
 513                 CERROR("cannot start thread\n");
 514
 515         MOD_INC_USE_COUNT;
 516         RETURN(0);
 517 }
 518
 519 static int mds_cleanup(struct obd_device * obddev)
 520 {
 521         struct super_block *sb;
 522         struct mds_obd *mds = &obddev->u.mds;
 523
 524         ENTRY;
 525
 526         if ( !list_empty(&obddev->obd_gen_clients) ) {
 527                 CERROR("still has clients!\n");
 528                 RETURN(-EBUSY);
 529         }
 530
 531         ptlrpc_stop_thread(mds->mds_service);
 532         rpc_unregister_service(mds->mds_service);
 533
 534         if (!list_empty(&mds->mds_service->srv_reqs)) {
 535                 // XXX reply with errors and clean up
 536                 CERROR("Request list not empty!\n");
 537         }
 538
 539         rpc_unregister_service(mds->mds_service);
 540         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 541
 542         sb = mds->mds_sb;
 543         if (!mds->mds_sb)
 544                 RETURN(0);
 545
 546         unlock_kernel();
 547         mntput(mds->mds_vfsmnt);
 548         mds->mds_sb = 0;
 549         kfree(mds->mds_fstype);
 550         lock_kernel();
 551
 552         MOD_DEC_USE_COUNT;
 553         RETURN(0);
 554 }
 555
 556 /* use obd ops to offer management infrastructure */
 557 static struct obd_ops mds_obd_ops = {
 558         o_setup:       mds_setup,
 559         o_cleanup:     mds_cleanup,
 560 };
 561
 562 static int __init mds_init(void)
 563 {
 564         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
 565         return 0;
 566 }
 567
 568 static void __exit mds_exit(void)
 569 {
 570         obd_unregister_type(LUSTRE_MDS_NAME);
 571 }
 572
 573 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 574 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
 575 MODULE_LICENSE("GPL");
 576
 577 module_init(mds_init);
 578 module_exit(mds_exit);