lustre/mds/handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/mds/handler.c
   5  *  Lustre Metadata Server (mds) request handler
   6  *
   7  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *   Author: Mike Shaver <shaver@clusterfs.com>
  12  *
  13  *   This file is part of Lustre, http://www.lustre.org.
  14  *
  15  *   Lustre is free software; you can redistribute it and/or
  16  *   modify it under the terms of version 2 of the GNU General Public
  17  *   License as published by the Free Software Foundation.
  18  *
  19  *   Lustre is distributed in the hope that it will be useful,
  20  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  21  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   GNU General Public License for more details.
  23  *
  24  *   You should have received a copy of the GNU General Public License
  25  *   along with Lustre; if not, write to the Free Software
  26  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  27  */
  28
  29 #ifndef EXPORT_SYMTAB
  30 # define EXPORT_SYMTAB
  31 #endif
  32 #define DEBUG_SUBSYSTEM S_MDS
  33
  34 #include <linux/module.h>
  35 #include <linux/lustre_mds.h>
  36 #include <linux/lustre_dlm.h>
  37 #include <linux/init.h>
  38 #include <linux/obd_class.h>
  39 #include <linux/random.h>
  40 #include <linux/fs.h>
  41 #include <linux/jbd.h>
  42 #include <linux/namei.h>
  43 #include <linux/ext3_fs.h>
  44 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
  45 # include <linux/smp_lock.h>
  46 # include <linux/buffer_head.h>
  47 # include <linux/workqueue.h>
  48 # include <linux/mount.h>
  49 #else
  50 # include <linux/locks.h>
  51 #endif
  52 #include <linux/obd_lov.h>
  53 #include <linux/obd_ost.h>
  54 #include <linux/lustre_mds.h>
  55 #include <linux/lustre_fsfilt.h>
  56 #include <linux/lprocfs_status.h>
  57 #include <linux/lustre_commit_confd.h>
  58 #include <linux/lustre_acl.h>
  59 #include <linux/lustre_sec.h>
  60 #include <linux/lustre_gs.h>
  61 #include "mds_internal.h"
  62
  63 static int mds_intent_policy(struct ldlm_namespace *ns,
  64                              struct ldlm_lock **lockp, void *req_cookie,
  65                              ldlm_mode_t mode, int flags, void *data);
  66 static int mds_postsetup(struct obd_device *obd);
  67 static int mds_cleanup(struct obd_device *obd, int flags);
  68
  69
  70 /* Assumes caller has already pushed into the kernel filesystem context */
  71 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  72                         loff_t offset, int count)
  73 {
  74         struct ptlrpc_bulk_desc *desc;
  75         struct l_wait_info lwi;
  76         struct page **pages;
  77         int rc = 0, npages, i, tmpcount, tmpsize = 0;
  78         ENTRY;
  79
  80         LASSERT((offset & (PAGE_SIZE - 1)) == 0); /* I'm dubious about this */
  81
  82         npages = (count + PAGE_SIZE - 1) >> PAGE_SHIFT;
  83         OBD_ALLOC(pages, sizeof(*pages) * npages);
  84         if (!pages)
  85                 GOTO(out, rc = -ENOMEM);
  86
  87         desc = ptlrpc_prep_bulk_exp(req, npages, BULK_PUT_SOURCE,
  88                                     MDS_BULK_PORTAL);
  89         if (desc == NULL)
  90                 GOTO(out_free, rc = -ENOMEM);
  91
  92         for (i = 0, tmpcount = count; i < npages; i++, tmpcount -= tmpsize) {
  93                 tmpsize = tmpcount > PAGE_SIZE ? PAGE_SIZE : tmpcount;
  94
  95                 pages[i] = alloc_pages(GFP_KERNEL, 0);
  96                 if (pages[i] == NULL)
  97                         GOTO(cleanup_buf, rc = -ENOMEM);
  98
  99                 ptlrpc_prep_bulk_page(desc, pages[i], 0, tmpsize);
 100         }
 101
 102         for (i = 0, tmpcount = count; i < npages; i++, tmpcount -= tmpsize) {
 103                 tmpsize = tmpcount > PAGE_SIZE ? PAGE_SIZE : tmpcount;
 104                 CDEBUG(D_EXT2, "reading %u@%llu from dir %lu (size %llu)\n",
 105                        tmpsize, offset, file->f_dentry->d_inode->i_ino,
 106                        file->f_dentry->d_inode->i_size);
 107
 108                 rc = fsfilt_readpage(req->rq_export->exp_obd, file,
 109                                      kmap(pages[i]), tmpsize, &offset);
 110                 kunmap(pages[i]);
 111
 112                 if (rc != tmpsize)
 113                         GOTO(cleanup_buf, rc = -EIO);
 114         }
 115
 116         LASSERT(desc->bd_nob == count);
 117
 118         rc = ptlrpc_start_bulk_transfer(desc);
 119         if (rc)
 120                 GOTO(cleanup_buf, rc);
 121
 122         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
 123                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
 124                        OBD_FAIL_MDS_SENDPAGE, rc = -EIO);
 125                 GOTO(abort_bulk, rc);
 126         }
 127
 128         lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
 129         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), &lwi);
 130         LASSERT (rc == 0 || rc == -ETIMEDOUT);
 131
 132         if (rc == 0) {
 133                 if (desc->bd_success &&
 134                     desc->bd_nob_transferred == count)
 135                         GOTO(cleanup_buf, rc);
 136
 137                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
 138         }
 139
 140         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s\n",
 141                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
 142                   desc->bd_nob_transferred, count,
 143                   req->rq_export->exp_client_uuid.uuid,
 144                   req->rq_export->exp_connection->c_remote_uuid.uuid);
 145
 146         ptlrpc_fail_export(req->rq_export);
 147
 148         EXIT;
 149  abort_bulk:
 150         ptlrpc_abort_bulk (desc);
 151  cleanup_buf:
 152         for (i = 0; i < npages; i++)
 153                 if (pages[i])
 154                         __free_pages(pages[i], 0);
 155
 156         ptlrpc_free_bulk(desc);
 157  out_free:
 158         OBD_FREE(pages, sizeof(*pages) * npages);
 159  out:
 160         return rc;
 161 }
 162
 163 extern char *ldlm_lockname[];
 164
 165 int mds_lock_mode_for_dir(struct obd_device *obd,
 166                           struct dentry *dentry, int mode)
 167 {
 168         int ret_mode = 0, split;
 169
 170         /* any dir access needs couple locks:
 171          * 1) on part of dir we gonna lookup/modify in
 172          * 2) on a whole dir to protect it from concurrent splitting
 173          *    and to flush client's cache for readdir()
 174          * so, for a given mode and dentry this routine decides what
 175          * lock mode to use for lock #2:
 176          * 1) if caller's gonna lookup in dir then we need to protect
 177          *    dir from being splitted only - LCK_CR
 178          * 2) if caller's gonna modify dir then we need to protect
 179          *    dir from being splitted and to flush cache - LCK_CW
 180          * 3) if caller's gonna modify dir and that dir seems ready
 181          *    for splitting then we need to protect it from any
 182          *    type of access (lookup/modify/split) - LCK_EX -bzzz */
 183
 184         split = mds_splitting_expected(obd, dentry);
 185
 186         /*
 187          * it is important to check here only for MDS_NO_SPLITTABLE. The reason
 188          * is that MDS_NO_SPLITTABLE means dir is not splittable in principle
 189          * and another thread will not split it on the quiet. But if we have
 190          * MDS_NO_SPLIT_EXPECTED, this means, that dir may be splitted anytime,
 191          * but not now (for current thread) and we should consider that it can
 192          * happen soon and go that branch which can yield LCK_EX to protect from
 193          * possible splitting.
 194          */
 195         if (split == MDS_NO_SPLITTABLE) {
 196                 /*
 197                  * this inode won't be splitted. so we need not to protect from
 198                  * just flush client's cache on modification.
 199                  */
 200                 if (mode == LCK_PW)
 201                         ret_mode = LCK_CW;
 202                 else
 203                         ret_mode = 0;
 204         } else {
 205                 if (mode == LCK_EX) {
 206                         ret_mode = LCK_EX;
 207                 } else if (mode == LCK_PR) {
 208                         ret_mode = LCK_CR;
 209                 } else if (mode == LCK_PW) {
 210                         /*
 211                          * caller gonna modify directory. We use concurrent
 212                          * write lock here to retract client's cache for
 213                          * readdir.
 214                          */
 215                         if (split == MDS_EXPECT_SPLIT) {
 216                                 /*
 217                                  * splitting possible. serialize any access the
 218                                  * idea is that first one seen dir is splittable
 219                                  * is given exclusive lock and split
 220                                  * directory. caller passes lock mode to
 221                                  * mds_try_to_split_dir() and splitting would be
 222                                  * done with exclusive lock only -bzzz.
 223                                  */
 224                                 CDEBUG(D_OTHER, "%s: gonna split %lu/%lu\n",
 225                                        obd->obd_name,
 226                                        (unsigned long)dentry->d_inode->i_ino,
 227                                        (unsigned long)dentry->d_inode->i_generation);
 228                                 ret_mode = LCK_EX;
 229                         } else {
 230                                 ret_mode = LCK_CW;
 231                         }
 232                 }
 233         }
 234
 235         return ret_mode;
 236 }
 237
 238 /* only valid locked dentries or errors should be returned */
 239 struct dentry *mds_id2locked_dentry(struct obd_device *obd, struct lustre_id *id,
 240                                     struct vfsmount **mnt, int lock_mode,
 241                                     struct lustre_handle *lockh, int *mode,
 242                                     char *name, int namelen, __u64 lockpart)
 243 {
 244         struct dentry *de = mds_id2dentry(obd, id, mnt), *retval = de;
 245         ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
 246         struct ldlm_res_id res_id = { .name = {0} };
 247         int flags = LDLM_FL_ATOMIC_CB, rc;
 248         ENTRY;
 249
 250         if (IS_ERR(de))
 251                 RETURN(de);
 252
 253         lockh[1].cookie = 0;
 254         res_id.name[0] = id_fid(id);
 255         res_id.name[1] = id_group(id);
 256
 257 #ifdef S_PDIROPS
 258         if (name && IS_PDIROPS(de->d_inode)) {
 259                 ldlm_policy_data_t cpolicy =
 260                         { .l_inodebits = { MDS_INODELOCK_UPDATE } };
 261                 LASSERT(mode != NULL);
 262                 *mode = mds_lock_mode_for_dir(obd, de, lock_mode);
 263                 if (*mode) {
 264                         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
 265                                               res_id, LDLM_IBITS,
 266                                               &cpolicy, *mode, &flags,
 267                                               mds_blocking_ast,
 268                                               ldlm_completion_ast, NULL, NULL,
 269                                               NULL, 0, NULL, lockh + 1);
 270                         if (rc != ELDLM_OK) {
 271                                 l_dput(de);
 272                                 RETURN(ERR_PTR(-ENOLCK));
 273                         }
 274                 }
 275                 flags = LDLM_FL_ATOMIC_CB;
 276
 277                 res_id.name[2] = full_name_hash((unsigned char *)name, namelen);
 278
 279                 CDEBUG(D_INFO, "take lock on "DLID4":"LPX64"\n",
 280                        OLID4(id), res_id.name[2]);
 281         }
 282 #else
 283 #warning "No PDIROPS support in the kernel"
 284 #endif
 285         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
 286                               LDLM_IBITS, &policy, lock_mode, &flags,
 287                               mds_blocking_ast, ldlm_completion_ast,
 288                               NULL, NULL, NULL, 0, NULL, lockh);
 289         if (rc != ELDLM_OK) {
 290                 l_dput(de);
 291                 retval = ERR_PTR(-EIO); /* XXX translate ldlm code */
 292 #ifdef S_PDIROPS
 293                 if (lockh[1].cookie)
 294                         ldlm_lock_decref(lockh + 1, *mode);
 295 #endif
 296         } else if (de->d_inode && de->d_inode->i_nlink == 0) {
 297                 /* as sometimes we lookup inode by ino/generation through
 298                    iopen mechanism, it's possible to find already unlinked
 299                    inode with nlink == 0. let's interpretate the case as
 300                    ENOENT -bzzz */
 301                 CWARN("found already unlinked inode %lu/%u\n",
 302                       de->d_inode->i_ino, de->d_inode->i_generation);
 303                 l_dput(de);
 304                 retval = ERR_PTR(-ENOENT);
 305                 ldlm_lock_decref(lockh, lock_mode);
 306 #ifdef S_PDIROPS
 307                 if (lockh[1].cookie)
 308                         ldlm_lock_decref(lockh + 1, *mode);
 309 #endif
 310         }
 311
 312         RETURN(retval);
 313 }
 314
 315 #ifndef DCACHE_DISCONNECTED
 316 #define DCACHE_DISCONNECTED DCACHE_NFSD_DISCONNECTED
 317 #endif
 318
 319 /* Look up an entry by inode number. This function ONLY returns valid dget'd
 320  * dentries with an initialized inode or errors */
 321 struct dentry *mds_id2dentry(struct obd_device *obd, struct lustre_id *id,
 322                              struct vfsmount **mnt)
 323 {
 324         struct mds_obd *mds = &obd->u.mds;
 325         struct dentry *result;
 326         struct inode *inode;
 327         unsigned long ino = 0;
 328         __u32 generation = 0;
 329         char idname[32];
 330
 331         if (!id_ino(id) && id_fid(id)) {
 332                 struct lustre_id *lid;
 333
 334                 /* if this is reint case we should use fidmap for resolving
 335                  * correct local store cookie. */
 336                 lid = mds_fidmap_lookup(obd, id);
 337                 if (!lid)
 338                         RETURN(ERR_PTR(-ENOENT));
 339
 340                 ino = (unsigned long)id_ino(lid);
 341                 generation = (__u32)id_gen(lid);
 342
 343                 CDEBUG(D_DENTRY, "fidmap resolved "DLID4"->"DLID4"\n",
 344                        OLID4(id), OLID4(lid));
 345
 346         } else if (id_ino(id)) {
 347                 ino = (unsigned long)id_ino(id);
 348                 generation = (__u32)id_gen(id);
 349         } else {
 350                 CERROR("invalid id for lookup "
 351                        DLID4"\n", OLID4(id));
 352                 LBUG();
 353         }
 354
 355         if (ino == 0)
 356                 RETURN(ERR_PTR(-ESTALE));
 357
 358         snprintf(idname, sizeof(idname), "0x%lx", ino);
 359
 360         CDEBUG(D_DENTRY, "--> mds_id2dentry: ino/gen %lu/%u, sb %p\n",
 361                ino, generation, mds->mds_sb);
 362
 363         /* under ext3 this is neither supposed to return bad inodes nor NULL
 364          * inodes. */
 365         result = ll_lookup_one_len(idname, mds->mds_id_de,
 366                                    strlen(idname));
 367         if (IS_ERR(result))
 368                 RETURN(result);
 369
 370         inode = result->d_inode;
 371         if (!inode)
 372                 RETURN(ERR_PTR(-ENOENT));
 373
 374         if (is_bad_inode(inode)) {
 375                 CERROR("bad inode returned %lu/%u\n",
 376                        inode->i_ino, inode->i_generation);
 377                 dput(result);
 378                 RETURN(ERR_PTR(-ENOENT));
 379         }
 380
 381         /* here we disabled generation check, as root inode i_generation
 382          * of cache mds and real mds are different. */
 383         if (id_fid(id) != id_fid(&mds->mds_rootid) && generation != 0 &&
 384             inode->i_generation != generation) {
 385                 /* we didn't find the right inode.. */
 386                 if (id_group(id) != mds->mds_num) {
 387                         CERROR("bad inode %lu found, link: %lu, ct: %d, generation "
 388                                "%u != %u, mds %u != %u, request to wrong MDS?\n",
 389                                inode->i_ino, (unsigned long)inode->i_nlink,
 390                                atomic_read(&inode->i_count), inode->i_generation,
 391                                generation, mds->mds_num, (unsigned)id_group(id));
 392                 } else {
 393                         CERROR("bad inode %lu found, link: %lu, ct: %d, generation "
 394                                "%u != %u, inode is recreated while request handled?\n",
 395                                inode->i_ino, (unsigned long)inode->i_nlink,
 396                                atomic_read(&inode->i_count), inode->i_generation,
 397                                generation);
 398                 }
 399                 dput(result);
 400                 RETURN(ERR_PTR(-ENOENT));
 401         }
 402
 403         if (mnt) {
 404                 *mnt = mds->mds_vfsmnt;
 405                 mntget(*mnt);
 406         }
 407
 408         RETURN(result);
 409 }
 410
 411 static
 412 int mds_req_add_idmapping(struct ptlrpc_request *req,
 413                           struct mds_export_data *med)
 414 {
 415         struct mds_req_sec_desc *rsd;
 416         struct lustre_sec_desc  *lsd;
 417         int rc;
 418
 419         if (!med->med_remote)
 420                 return 0;
 421
 422         /* maybe we should do it more completely: invalidate the gss ctxt? */
 423         if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
 424                 CWARN("didn't find mapped uid\n");
 425                 return -EPERM;
 426         }
 427
 428         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
 429         if (!rsd) {
 430                 CERROR("Can't unpack security desc\n");
 431                 return -EPROTO;
 432         }
 433
 434         lsd = mds_get_lsd(req->rq_mapped_uid);
 435         if (!lsd) {
 436                 CERROR("can't get LSD(%u), no mapping added\n",
 437                        req->rq_mapped_uid);
 438                 return -EPERM;
 439         }
 440
 441         rc = mds_idmap_add(med->med_idmap, rsd->rsd_uid, lsd->lsd_uid,
 442                            rsd->rsd_gid, lsd->lsd_gid);
 443         mds_put_lsd(lsd);
 444         return rc;
 445 }
 446
 447 static
 448 int mds_req_del_idmapping(struct ptlrpc_request *req,
 449                           struct mds_export_data *med)
 450 {
 451         struct mds_req_sec_desc *rsd;
 452         struct lustre_sec_desc  *lsd;
 453         int rc;
 454
 455         if (!med->med_remote)
 456                 return 0;
 457
 458         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
 459         if (!rsd) {
 460                 CERROR("Can't unpack security desc\n");
 461                 return -EPROTO;
 462         }
 463
 464         LASSERT(req->rq_mapped_uid != -1);
 465         lsd = mds_get_lsd(req->rq_mapped_uid);
 466         if (!lsd) {
 467                 CERROR("can't get LSD(%u), no idmapping deleted\n",
 468                        req->rq_mapped_uid);
 469                 return -EPERM;
 470         }
 471
 472         rc = mds_idmap_del(med->med_idmap, rsd->rsd_uid, lsd->lsd_uid,
 473                            rsd->rsd_gid, lsd->lsd_gid);
 474         mds_put_lsd(lsd);
 475         return rc;
 476 }
 477
 478 static int mds_init_export_data(struct ptlrpc_request *req,
 479                                 struct mds_export_data *med)
 480 {
 481         struct obd_connect_data *data, *reply;
 482         int ask_remote, ask_local;
 483         ENTRY;
 484
 485         data = lustre_msg_buf(req->rq_reqmsg, 5, sizeof(*data));
 486         reply = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*data));
 487         LASSERT(data && reply);
 488
 489         if (med->med_initialized) {
 490                 CDEBUG(D_SEC, "med already initialized, reconnect?\n");
 491                 goto reply;
 492         }
 493
 494         ask_remote = data->ocd_connect_flags & OBD_CONNECT_REMOTE;
 495         ask_local = data->ocd_connect_flags & OBD_CONNECT_LOCAL;
 496
 497         /* currently the policy is simple: satisfy client as possible
 498          * as we can.
 499          */
 500         if (req->rq_auth_uid == -1) {
 501                 if (ask_remote)
 502                         CWARN("null sec is used, force to be local\n");
 503                 med->med_remote = 0;
 504         } else {
 505                 if (ask_remote) {
 506                         if (!req->rq_remote_realm)
 507                                 CWARN("local realm asked to be remote\n");
 508                         med->med_remote = 1;
 509                 } else if (ask_local) {
 510                         if (req->rq_remote_realm)
 511                                 CWARN("remote realm asked to be local\n");
 512                         med->med_remote = 0;
 513                 } else
 514                         med->med_remote = (req->rq_remote_realm != 0);
 515         }
 516
 517         med->med_nllu = data->ocd_nllu[0];
 518         med->med_nllg = data->ocd_nllu[1];
 519
 520         med->med_initialized = 1;
 521 reply:
 522         reply->ocd_connect_flags &= ~(OBD_CONNECT_REMOTE | OBD_CONNECT_LOCAL);
 523         if (med->med_remote) {
 524                 if (!med->med_idmap)
 525                         med->med_idmap = mds_idmap_alloc();
 526
 527                 if (!med->med_idmap)
 528                         CERROR("Failed to alloc idmap, following request from "
 529                                "this client will be refused\n");
 530
 531                 reply->ocd_connect_flags |= OBD_CONNECT_REMOTE;
 532                 CDEBUG(D_SEC, "set client as remote\n");
 533         } else {
 534                 reply->ocd_connect_flags |= OBD_CONNECT_LOCAL;
 535                 CDEBUG(D_SEC, "set client as local\n");
 536         }
 537
 538         RETURN(0);
 539 }
 540
 541 static void mds_free_export_data(struct mds_export_data *med)
 542 {
 543         if (!med->med_idmap)
 544                 return;
 545
 546         LASSERT(med->med_remote);
 547         mds_idmap_free(med->med_idmap);
 548         med->med_idmap = NULL;
 549 }
 550
 551 /* Establish a connection to the MDS.
 552  *
 553  * This will set up an export structure for the client to hold state data about
 554  * that client, like open files, the last operation number it did on the server,
 555  * etc.
 556  */
 557 static int mds_connect(struct lustre_handle *conn, struct obd_device *obd,
 558                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 559                        unsigned long flags)
 560 {
 561         struct mds_export_data *med;
 562         struct mds_client_data *mcd;
 563         struct obd_export *exp;
 564         int rc;
 565         ENTRY;
 566
 567         if (!conn || !obd || !cluuid)
 568                 RETURN(-EINVAL);
 569
 570         /* XXX There is a small race between checking the list and adding a new
 571          * connection for the same UUID, but the real threat (list corruption
 572          * when multiple different clients connect) is solved.
 573          *
 574          * There is a second race between adding the export to the list, and
 575          * filling in the client data below.  Hence skipping the case of NULL
 576          * mcd above.  We should already be controlling multiple connects at the
 577          * client, and we can't hold the spinlock over memory allocations
 578          * without risk of deadlocking.
 579          */
 580         rc = class_connect(conn, obd, cluuid);
 581         if (rc)
 582                 RETURN(rc);
 583         exp = class_conn2export(conn);
 584
 585         LASSERT(exp != NULL);
 586         med = &exp->exp_mds_data;
 587
 588         OBD_ALLOC(mcd, sizeof(*mcd));
 589         if (!mcd) {
 590                 CERROR("%s: out of memory for client data.\n",
 591                         obd->obd_name);
 592                 GOTO(out, rc = -ENOMEM);
 593         }
 594
 595         memcpy(mcd->mcd_uuid, cluuid, sizeof(mcd->mcd_uuid));
 596         med->med_mcd = mcd;
 597
 598         rc = mds_client_add(obd, &obd->u.mds, med, -1);
 599         if (rc)
 600                 GOTO(out, rc);
 601
 602         EXIT;
 603 out:
 604         if (rc) {
 605                 if (mcd)
 606                         OBD_FREE(mcd, sizeof(*mcd));
 607                 class_disconnect(exp, 0);
 608         } else {
 609                 class_export_put(exp);
 610         }
 611         return rc;
 612 }
 613
 614 static int mds_connect_post(struct obd_export *exp, unsigned initial,
 615                             unsigned long flags)
 616 {
 617         struct obd_device *obd = exp->exp_obd;
 618         struct mds_obd *mds = &obd->u.mds;
 619         struct mds_export_data *med;
 620         struct mds_client_data *mcd;
 621         int rc = 0;
 622         ENTRY;
 623
 624         med = &exp->exp_mds_data;
 625         mcd = med->med_mcd;
 626
 627         if (initial) {
 628                 /* some one reconnect initially, we have to reset
 629                  * data existing export can have. bug 6102 */
 630                 if (mcd->mcd_last_xid != 0)
 631                         CDEBUG(D_HA, "initial reconnect to existing export\n");
 632                 mcd->mcd_last_transno = 0;
 633                 mcd->mcd_last_xid = 0;
 634                 mcd->mcd_last_close_xid = 0;
 635                 mcd->mcd_last_result = 0;
 636                 mcd->mcd_last_data = 0;
 637         }
 638
 639         if (!(flags & OBD_OPT_MDS_CONNECTION)) {
 640                 if (!(exp->exp_flags & OBD_OPT_REAL_CLIENT)) {
 641                         atomic_inc(&mds->mds_real_clients);
 642                         CDEBUG(D_OTHER,"%s: peer from %s is real client (%d)\n",
 643                                obd->obd_name, exp->exp_client_uuid.uuid,
 644                                atomic_read(&mds->mds_real_clients));
 645                         exp->exp_flags |= OBD_OPT_REAL_CLIENT;
 646                 }
 647                 if (mds->mds_md_name)
 648                         rc = mds_md_connect(obd, mds->mds_md_name);
 649         }
 650         RETURN(rc);
 651 }
 652
 653 static int mds_init_export(struct obd_export *exp)
 654 {
 655         struct mds_export_data *med = &exp->exp_mds_data;
 656
 657         INIT_LIST_HEAD(&med->med_open_head);
 658         spin_lock_init(&med->med_open_lock);
 659         return 0;
 660 }
 661
 662 static int mds_destroy_export(struct obd_export *export)
 663 {
 664         struct obd_device *obd = export->exp_obd;
 665         struct mds_export_data *med = &export->exp_mds_data;
 666         struct lvfs_run_ctxt saved;
 667         int rc = 0;
 668         ENTRY;
 669
 670         mds_free_export_data(med);
 671         target_destroy_export(export);
 672
 673         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
 674                 GOTO(out, 0);
 675
 676         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 677
 678         /* Close any open files (which may also cause orphan unlinking). */
 679         spin_lock(&med->med_open_lock);
 680         while (!list_empty(&med->med_open_head)) {
 681                 struct list_head *tmp = med->med_open_head.next;
 682                 struct mds_file_data *mfd =
 683                         list_entry(tmp, struct mds_file_data, mfd_list);
 684                 struct lustre_id sid;
 685
 686                 BDEVNAME_DECLARE_STORAGE(btmp);
 687
 688                 /* bug 1579: fix force-closing for 2.5 */
 689                 struct dentry *dentry = mfd->mfd_dentry;
 690
 691                 list_del(&mfd->mfd_list);
 692                 spin_unlock(&med->med_open_lock);
 693
 694                 down(&dentry->d_inode->i_sem);
 695                 rc = mds_read_inode_sid(obd, dentry->d_inode, &sid);
 696                 up(&dentry->d_inode->i_sem);
 697                 if (rc) {
 698                         CERROR("Can't read inode self id, inode %lu, "
 699                                "rc %d\n", dentry->d_inode->i_ino, rc);
 700                         memset(&sid, 0, sizeof(sid));
 701                 }
 702
 703                 /* If you change this message, be sure to update
 704                  * replay_single:test_46 */
 705                 CERROR("force closing client file handle for %.*s (%s:"
 706                        DLID4")\n", dentry->d_name.len, dentry->d_name.name,
 707                        ll_bdevname(dentry->d_inode->i_sb, btmp),
 708                        OLID4(&sid));
 709
 710                 /* child inode->i_alloc_sem protects orphan_dec_test and
 711                  * is_orphan race, mds_mfd_close drops it */
 712                 DOWN_WRITE_I_ALLOC_SEM(dentry->d_inode);
 713                 rc = mds_mfd_close(NULL, 0, obd, mfd,
 714                                    !(export->exp_flags & OBD_OPT_FAILOVER));
 715                 if (rc)
 716                         CDEBUG(D_INODE, "Error closing file: %d\n", rc);
 717                 spin_lock(&med->med_open_lock);
 718         }
 719         spin_unlock(&med->med_open_lock);
 720         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 721
 722         EXIT;
 723 out:
 724         mds_client_free(export, !(export->exp_flags & OBD_OPT_FAILOVER));
 725         return rc;
 726 }
 727
 728 static int mds_disconnect(struct obd_export *exp, unsigned long flags)
 729 {
 730         unsigned long irqflags;
 731         struct obd_device *obd;
 732         struct mds_obd *mds;
 733         int rc;
 734         ENTRY;
 735
 736         LASSERT(exp != NULL);
 737         obd = class_exp2obd(exp);
 738         if (obd == NULL) {
 739                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
 740                        exp->exp_handle.h_cookie);
 741                 RETURN(-EINVAL);
 742         }
 743         mds = &obd->u.mds;
 744
 745         /*
 746          * suppress any inter-mds requests durring disconnecting lmv if this is
 747          * detected --force mode. This is needed to avoid endless recovery.
 748          */
 749         if (atomic_read(&mds->mds_real_clients) > 0 &&
 750             !(exp->exp_flags & OBD_OPT_REAL_CLIENT))
 751                 flags |= OBD_OPT_FORCE;
 752
 753         if (!(exp->exp_flags & OBD_OPT_REAL_CLIENT)
 754             && !atomic_read(&mds->mds_real_clients)) {
 755                 /* there was no client at all */
 756                 mds_md_disconnect(obd, flags);
 757         }
 758
 759         if ((exp->exp_flags & OBD_OPT_REAL_CLIENT)
 760             && atomic_dec_and_test(&mds->mds_real_clients)) {
 761                 /* time to drop LMV connections */
 762                 CDEBUG(D_OTHER, "%s: last real client %s disconnected.  "
 763                        "Disconnnect from LMV now\n",
 764                        obd->obd_name, exp->exp_client_uuid.uuid);
 765                 mds_md_disconnect(obd, flags);
 766         }
 767
 768         spin_lock_irqsave(&exp->exp_lock, irqflags);
 769         exp->exp_flags = flags;
 770         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
 771
 772         /* disconnect early so that clients can't keep using export */
 773         rc = class_disconnect(exp, flags);
 774         ldlm_cancel_locks_for_export(exp);
 775
 776         /* complete all outstanding replies */
 777         spin_lock_irqsave(&exp->exp_lock, irqflags);
 778         while (!list_empty(&exp->exp_outstanding_replies)) {
 779                 struct ptlrpc_reply_state *rs =
 780                         list_entry(exp->exp_outstanding_replies.next,
 781                                    struct ptlrpc_reply_state, rs_exp_list);
 782                 struct ptlrpc_service *svc = rs->rs_srv_ni->sni_service;
 783
 784                 spin_lock(&svc->srv_lock);
 785                 list_del_init(&rs->rs_exp_list);
 786                 ptlrpc_schedule_difficult_reply(rs);
 787                 spin_unlock(&svc->srv_lock);
 788         }
 789         spin_unlock_irqrestore(&exp->exp_lock, irqflags);
 790         RETURN(rc);
 791 }
 792
 793 static int mds_getstatus(struct ptlrpc_request *req)
 794 {
 795         struct mds_obd *mds = mds_req2mds(req);
 796         struct mds_body *body;
 797         int rc, size;
 798         ENTRY;
 799
 800         size = sizeof(*body);
 801
 802         rc = lustre_pack_reply(req, 1, &size, NULL);
 803         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
 804                 CERROR("mds: out of memory for message: size=%d\n", size);
 805                 req->rq_status = -ENOMEM;       /* superfluous? */
 806                 RETURN(-ENOMEM);
 807         }
 808
 809         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
 810         body->valid |= OBD_MD_FID;
 811
 812         memcpy(&body->id1, &mds->mds_rootid, sizeof(body->id1));
 813
 814         /*
 815          * the last_committed and last_xid fields are filled in for all replies
 816          * already - no need to do so here also.
 817          */
 818         RETURN(0);
 819 }
 820
 821 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 822                      void *data, int flag)
 823 {
 824         int do_ast;
 825         ENTRY;
 826
 827         if (flag == LDLM_CB_CANCELING) {
 828                 /* Don't need to do anything here. */
 829                 RETURN(0);
 830         }
 831
 832         /* XXX layering violation!  -phil */
 833         lock_res_and_lock(lock);
 834
 835         /*
 836          * get this: if mds_blocking_ast is racing with mds_intent_policy, such
 837          * that mds_blocking_ast is called just before l_i_p takes the ns_lock,
 838          * then by the time we get the lock, we might not be the correct
 839          * blocking function anymore.  So check, and return early, if so.
 840          */
 841         if (lock->l_blocking_ast != mds_blocking_ast) {
 842                 unlock_res_and_lock(lock);
 843                 RETURN(0);
 844         }
 845
 846         lock->l_flags |= LDLM_FL_CBPENDING;
 847         do_ast = (!lock->l_readers && !lock->l_writers);
 848         unlock_res_and_lock(lock);
 849
 850         if (do_ast) {
 851                 struct lustre_handle lockh;
 852                 int rc;
 853
 854                 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
 855                 ldlm_lock2handle(lock, &lockh);
 856                 rc = ldlm_cli_cancel(&lockh);
 857                 if (rc < 0)
 858                         CERROR("ldlm_cli_cancel: %d\n", rc);
 859         } else {
 860                 LDLM_DEBUG(lock, "Lock still has references, will be "
 861                            "cancelled later");
 862         }
 863         RETURN(0);
 864 }
 865
 866 static int mds_convert_md(struct obd_device *obd, struct inode *inode,
 867                           void *md, int size, int mea)
 868 {
 869         int rc = size;
 870
 871         if (S_ISREG(inode->i_mode)) {
 872                 rc = mds_convert_lov_ea(obd, inode, md, size);
 873         } else if (S_ISDIR(inode->i_mode)) {
 874                 if (mea) {
 875                         rc = mds_convert_mea_ea(obd, inode, md, size);
 876                 } else {
 877                         rc = mds_convert_lov_ea(obd, inode, md, size);
 878                 }
 879                 if (rc == -EINVAL) {
 880                         CERROR("Invalid EA format (nor LOV or MEA) "
 881                                "is detected. Inode %lu/%u\n",
 882                                inode->i_ino, inode->i_generation);
 883                 }
 884         }
 885         return rc;
 886 }
 887
 888 int mds_get_md(struct obd_device *obd, struct inode *inode,
 889                void *md, int *size, int lock, int mea)
 890 {
 891         int lmm_size;
 892         int rc = 0;
 893         ENTRY;
 894
 895         if (lock)
 896                 down(&inode->i_sem);
 897
 898         rc = fsfilt_get_md(obd, inode, md, *size,
 899                            (mea ? EA_MEA : EA_LOV));
 900         if (rc < 0) {
 901                 CERROR("Error %d reading eadata for ino %lu\n",
 902                        rc, inode->i_ino);
 903         } else if (rc > 0) {
 904                 lmm_size = rc;
 905                 rc = mds_convert_md(obd, inode, md,
 906                                     lmm_size, mea);
 907                 if (rc == 0) {
 908                         *size = lmm_size;
 909                         rc = lmm_size;
 910                 } else if (rc > 0) {
 911                         *size = rc;
 912                 }
 913         }
 914         if (lock)
 915                 up(&inode->i_sem);
 916
 917         RETURN(rc);
 918 }
 919
 920 /* Call with lock=1 if you want mds_pack_md to take the i_sem.
 921  * Call with lock=0 if the caller has already taken the i_sem. */
 922 int mds_pack_md(struct obd_device *obd, struct lustre_msg *msg, int offset,
 923                 struct mds_body *body, struct inode *inode, int lock, int mea)
 924 {
 925         struct mds_obd *mds = &obd->u.mds;
 926         int rc, lmm_size;
 927         void *lmm;
 928         ENTRY;
 929
 930         lmm = lustre_msg_buf(msg, offset, 0);
 931         if (lmm == NULL) {
 932                 /* Some problem with getting eadata when I sized the reply
 933                  * buffer... */
 934                 CDEBUG(D_INFO, "no space reserved for inode %lu MD\n",
 935                        inode->i_ino);
 936                 RETURN(0);
 937         }
 938         lmm_size = msg->buflens[offset];
 939
 940         /* I don't really like this, but it is a sanity check on the client
 941          * MD request.  However, if the client doesn't know how much space
 942          * to reserve for the MD, it shouldn't be bad to have too much space.
 943          */
 944         if (lmm_size > mds->mds_max_mdsize) {
 945                 CWARN("Reading MD for inode %lu of %d bytes > max %d\n",
 946                        inode->i_ino, lmm_size, mds->mds_max_mdsize);
 947                 // RETURN(-EINVAL);
 948         }
 949
 950         rc = mds_get_md(obd, inode, lmm, &lmm_size, lock, mea);
 951         if (rc > 0) {
 952                 body->valid |= S_ISDIR(inode->i_mode) ?
 953                         OBD_MD_FLDIREA : OBD_MD_FLEASIZE;
 954
 955                 if (mea)
 956                         body->valid |= OBD_MD_MEA;
 957
 958                 body->eadatasize = lmm_size;
 959                 rc = 0;
 960         }
 961
 962         RETURN(rc);
 963 }
 964
 965 int mds_pack_link(struct dentry *dentry, struct ptlrpc_request *req,
 966                   struct mds_body *repbody, int reply_off)
 967 {
 968         struct inode *inode = dentry->d_inode;
 969         char *symname;
 970         int len, rc;
 971         ENTRY;
 972
 973         symname = lustre_msg_buf(req->rq_repmsg, reply_off + 1,0);
 974         LASSERT(symname != NULL);
 975         len = req->rq_repmsg->buflens[reply_off + 1];
 976
 977         rc = inode->i_op->readlink(dentry, symname, len);
 978         if (rc < 0) {
 979                 CERROR("readlink failed: %d\n", rc);
 980         } else if (rc != len - 1) {
 981                 CERROR ("Unexpected readlink rc %d: expecting %d\n",
 982                         rc, len - 1);
 983                 rc = -EINVAL;
 984         } else {
 985                 CDEBUG(D_INODE, "read symlink dest %s\n", symname);
 986                 repbody->valid |= OBD_MD_LINKNAME;
 987                 repbody->eadatasize = rc + 1;
 988                 symname[rc] = 0;        /* NULL terminate */
 989                 rc = 0;
 990         }
 991
 992         RETURN(rc);
 993 }
 994
 995 int mds_pack_xattr(struct dentry *dentry, struct ptlrpc_request *req,
 996                    struct mds_body *repbody, int req_off, int reply_off)
 997 {
 998         struct inode *inode = dentry->d_inode;
 999         char *ea_name;
1000         void *value = NULL;
1001         int len, rc;
1002         ENTRY;
1003
1004         ea_name = lustre_msg_string(req->rq_reqmsg, req_off + 1, 0);
1005         len = req->rq_repmsg->buflens[reply_off + 1];
1006         if (len != 0)
1007                 value = lustre_msg_buf(req->rq_repmsg, reply_off + 1, len);
1008
1009         rc = -EOPNOTSUPP;
1010
1011         if (!strcmp(ea_name, XATTR_NAME_LUSTRE_ACL)) {
1012                 struct rmtacl_upcall_desc desc;
1013
1014                 if (len != LUSTRE_ACL_SIZE_MAX || !value) {
1015                         CERROR("no reply buffer prepared\n");
1016                         RETURN(-EFAULT);
1017                 }
1018
1019                 memset(&desc, 0, sizeof(desc));
1020                 desc.get = 1;
1021                 desc.cmd = lustre_msg_string(req->rq_reqmsg, req_off + 2, 0);
1022                 desc.cmdlen =  req->rq_reqmsg->buflens[req_off + 2];
1023                 desc.res = (char *) value;
1024                 desc.reslen = LUSTRE_ACL_SIZE_MAX;
1025
1026                 mds_do_remote_acl_upcall(&desc);
1027
1028                 if (desc.upcall_status)
1029                         RETURN(desc.upcall_status);
1030
1031                 if (desc.reslen > LUSTRE_ACL_SIZE_MAX) {
1032                         CERROR("downcall claim reslen %u\n", desc.reslen);
1033                         RETURN(-EINVAL);
1034                 }
1035                 /* like remote setfacl, steal "flags" in mds_body as the
1036                  * exececution status
1037                  */
1038                 repbody->flags = desc.status;
1039                 repbody->valid |= OBD_MD_FLXATTR;
1040                 repbody->eadatasize = desc.reslen;
1041
1042                 RETURN(0);
1043         }
1044
1045         if (inode->i_op && inode->i_op->getxattr)
1046                 rc = inode->i_op->getxattr(dentry, ea_name, value, len);
1047
1048         if (rc < 0) {
1049                 if (rc != -ENODATA && rc != -EOPNOTSUPP)
1050                         CERROR("getxattr failed: %d", rc);
1051         } else {
1052                 repbody->valid |= OBD_MD_FLXATTR;
1053                 repbody->eadatasize = rc;
1054                 rc = 0;
1055         }
1056
1057         RETURN(rc);
1058 }
1059
1060 int mds_pack_xattr_list(struct dentry *dentry, struct ptlrpc_request *req,
1061                         struct mds_body *repbody, int reply_off)
1062 {
1063         struct inode *inode = dentry->d_inode;
1064         void *value = NULL;
1065         int len, rc;
1066         ENTRY;
1067
1068         len = req->rq_repmsg->buflens[reply_off + 1];
1069         if (len != 0)
1070                 value = lustre_msg_buf(req->rq_repmsg, reply_off + 1, len);
1071
1072         rc = -EOPNOTSUPP;
1073         if (inode->i_op && inode->i_op->getxattr)
1074                 rc = inode->i_op->listxattr(dentry, value, len);
1075
1076         if (rc < 0) {
1077                 CERROR("listxattr failed: %d", rc);
1078         } else {
1079                 repbody->valid |= OBD_MD_FLXATTRLIST;
1080                 repbody->eadatasize = rc;
1081                 rc = 0;
1082         }
1083         RETURN(rc);
1084 }
1085
1086 static
1087 int mds_pack_posix_acl(struct lustre_msg *repmsg, int offset,
1088                        struct mds_body *body, struct inode *inode)
1089 {
1090         struct dentry de = { .d_inode = inode };
1091         __u32 buflen, *sizep;
1092         void *buf;
1093         int size;
1094         ENTRY;
1095
1096         if (!inode->i_op->getxattr)
1097                 RETURN(0);
1098
1099         sizep = lustre_msg_buf(repmsg, offset, 4);
1100         if (!sizep) {
1101                 CERROR("can't locate returned acl size buf\n");
1102                 RETURN(-EPROTO);
1103         }
1104
1105         buflen = repmsg->buflens[offset + 1];
1106         buf = lustre_msg_buf(repmsg, offset + 1, buflen);
1107
1108         size = inode->i_op->getxattr(&de, XATTR_NAME_ACL_ACCESS, buf, buflen);
1109         if (size == -ENODATA || size == -EOPNOTSUPP)
1110                 RETURN(0);
1111         if (size < 0)
1112                 RETURN(size);
1113         LASSERT(size);
1114
1115         *sizep = cpu_to_le32(size);
1116         body->valid |= OBD_MD_FLACL;
1117
1118         RETURN(0);
1119 }
1120
1121 int mds_pack_remote_perm(struct ptlrpc_request *req, int reply_off,
1122                          struct mds_body *body, struct inode *inode)
1123 {
1124         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1125         struct lustre_sec_desc *lsd;
1126         struct mds_remote_perm *perm;
1127         __u32 lsd_perms;
1128
1129         LASSERT(inode->i_op);
1130         LASSERT(inode->i_op->permission);
1131         LASSERT(req->rq_export->exp_mds_data.med_remote);
1132
1133         perm = (struct mds_remote_perm *)
1134                        lustre_msg_buf(req->rq_repmsg, reply_off, sizeof(perm));
1135         if (!perm) {
1136                 CERROR("no remote perm buf at offset %d\n", reply_off);
1137                 return -EINVAL;
1138         }
1139
1140         memset(perm, 0, sizeof(*perm));
1141
1142         /* obtain authenticated uid/gid and LSD permissions, which
1143          * might be different from current process context, from LSD
1144          */
1145         lsd = mds_get_lsd(current->uid);
1146         if (!lsd) {
1147                 CWARN("can't LSD of uid %u\n", current->uid);
1148                 RETURN(-EPERM);
1149         }
1150
1151         perm->mrp_auth_uid = lsd->lsd_uid;
1152         perm->mrp_auth_gid = lsd->lsd_gid;
1153
1154         lsd_perms = mds_lsd_get_perms(lsd, 1, 0, req->rq_peer.peer_id.nid);
1155         if (lsd_perms & LSD_PERM_SETUID)
1156                 perm->mrp_allow_setuid = 1;
1157         if (lsd_perms & LSD_PERM_SETGID)
1158                 perm->mrp_allow_setgid = 1;
1159
1160         mds_put_lsd(lsd);
1161
1162         if (mds_remote_perm_do_reverse_map(med, perm))
1163                 RETURN(-EPERM);
1164
1165         /* permission bits of current user
1166          * XXX this is low efficient, could we do it in one blow?
1167          */
1168         if (inode->i_op->permission(inode, MAY_EXEC, NULL) == 0)
1169                 perm->mrp_perm |= MAY_EXEC;
1170         if (inode->i_op->permission(inode, MAY_WRITE, NULL) == 0)
1171                 perm->mrp_perm |= MAY_WRITE;
1172         if (inode->i_op->permission(inode, MAY_READ, NULL) == 0)
1173                 perm->mrp_perm |= MAY_READ;
1174
1175         body->valid |= (OBD_MD_FLACL | OBD_MD_FLRMTACL);
1176
1177         RETURN(0);
1178 }
1179
1180 int mds_pack_acl(struct ptlrpc_request *req, int reply_off,
1181                  struct mds_body *body, struct inode *inode)
1182 {
1183         int rc;
1184
1185         if (!req->rq_export->exp_mds_data.med_remote)
1186                 rc = mds_pack_posix_acl(req->rq_repmsg, reply_off, body, inode);
1187         else
1188                 rc = mds_pack_remote_perm(req, reply_off + 1, body, inode);
1189
1190         return rc;
1191 }
1192
1193 static int mds_getattr_internal(struct obd_device *obd, struct dentry *dentry,
1194                                 struct ptlrpc_request *req, int req_off,
1195                                 struct mds_body *reqbody, int reply_off,
1196                                 struct mds_req_sec_desc *rsd)
1197 {
1198         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1199         struct inode *inode = dentry->d_inode;
1200         struct mds_body *body;
1201         int rc = 0, offset = 0;
1202         ENTRY;
1203
1204         if (inode == NULL && !(dentry->d_flags & DCACHE_CROSS_REF))
1205                 RETURN(-ENOENT);
1206
1207         body = lustre_msg_buf(req->rq_repmsg, reply_off, sizeof(*body));
1208         LASSERT(body != NULL);                 /* caller prepped reply */
1209
1210         if (dentry->d_flags & DCACHE_CROSS_REF) {
1211                 mds_pack_dentry2body(obd, body, dentry,
1212                                      (reqbody->valid & OBD_MD_FID) ? 1 : 0);
1213                 CDEBUG(D_OTHER, "cross reference: "DLID4"\n",
1214                        OLID4(&body->id1));
1215                 RETURN(0);
1216         }
1217
1218         mds_pack_inode2body(obd, body, inode,
1219                             (reqbody->valid & OBD_MD_FID) ? 1 : 0);
1220
1221         if ((S_ISREG(inode->i_mode) && (reqbody->valid & OBD_MD_FLEASIZE)) ||
1222             (S_ISDIR(inode->i_mode) && (reqbody->valid & OBD_MD_FLDIREA))) {
1223
1224                 /* guessing what kind og attribute do we need. */
1225                 int is_mea = (S_ISDIR(inode->i_mode) &&
1226                     (reqbody->valid & OBD_MD_MEA) != 0);
1227
1228                 rc = mds_pack_md(obd, req->rq_repmsg, reply_off + 1,
1229                                  body, inode, 1, is_mea);
1230
1231                 /* if we have LOV EA data, the OST holds size, atime, mtime. */
1232                 if (!(body->valid & OBD_MD_FLEASIZE) &&
1233                     !(body->valid & OBD_MD_FLDIREA))
1234                         body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
1235                                         OBD_MD_FLATIME | OBD_MD_FLMTIME);
1236
1237         } else if (S_ISLNK(inode->i_mode) &&
1238                    (reqbody->valid & OBD_MD_LINKNAME) != 0) {
1239                 rc = mds_pack_link(dentry, req, body, reply_off);
1240         } else if (reqbody->valid & OBD_MD_FLXATTR) {
1241                 rc = mds_pack_xattr(dentry, req, body, req_off, reply_off);
1242         } else if (reqbody->valid & OBD_MD_FLXATTRLIST) {
1243                 rc = mds_pack_xattr_list(dentry, req, body, reply_off);
1244         }
1245
1246         offset = reply_off + ((reqbody->valid & OBD_MD_FLEASIZE) ? 2 : 1);
1247         if (reqbody->valid & OBD_MD_FLACL) {
1248                 rc = mds_pack_acl(req, offset, body, inode);
1249                 offset += 2;
1250         }
1251
1252         if (reqbody->valid & OBD_MD_FLKEY) {
1253                 rc = mds_pack_gskey(obd, req->rq_repmsg, &offset,
1254                                     body, inode);
1255         }
1256
1257         mds_pack_audit(obd, inode, body);
1258
1259         if (reqbody->valid & OBD_MD_CAPA) {
1260                 struct lustre_capa *req_capa;
1261
1262                 LASSERT(!(reqbody->valid & ~OBD_MD_CAPA));
1263                 LASSERT(S_ISREG(inode->i_mode));
1264
1265                 req_capa = lustre_swab_reqbuf(req, req_off + 1,
1266                                               sizeof(*req_capa),
1267                                               lustre_swab_lustre_capa);
1268                 if (req_capa == NULL) {
1269                         CERROR("Can't unpack capa\n");
1270                         RETURN(-EFAULT);
1271                 }
1272
1273                 offset = reply_off + 1;
1274                 rc = mds_pack_capa(obd, med, reqbody, req_capa, req,
1275                                    &offset, body);
1276         }
1277
1278         if (rc == 0)
1279                 mds_body_do_reverse_map(med, body);
1280
1281         RETURN(rc);
1282 }
1283
1284 static int mds_getattr_pack_msg_cf(struct ptlrpc_request *req,
1285                                    struct dentry *dentry,
1286                                    int offset)
1287 {
1288         int rc = 0, size[1] = {sizeof(struct mds_body)};
1289         ENTRY;
1290
1291         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
1292                 CERROR("failed MDS_GETATTR_PACK test\n");
1293                 req->rq_status = -ENOMEM;
1294                 RETURN(-ENOMEM);
1295         }
1296
1297         rc = lustre_pack_reply(req, 1, size, NULL);
1298         if (rc) {
1299                 CERROR("lustre_pack_reply failed: rc %d\n", rc);
1300                 GOTO(out, req->rq_status = rc);
1301         }
1302
1303         EXIT;
1304 out:
1305         return rc;
1306 }
1307
1308 static int mds_getattr_pack_msg(struct ptlrpc_request *req, struct dentry *de,
1309                                 int offset)
1310 {
1311         struct inode *inode = de->d_inode;
1312         struct mds_obd *mds = mds_req2mds(req);
1313         struct mds_body *body;
1314         int rc = 0, size[4] = {sizeof(*body)}, bufcount = 1;
1315         ENTRY;
1316
1317         body = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*body));
1318         LASSERT(body != NULL);                 /* checked by caller */
1319         LASSERT_REQSWABBED(req, offset);       /* swabbed by caller */
1320
1321         if ((S_ISREG(inode->i_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
1322             (S_ISDIR(inode->i_mode) && (body->valid & OBD_MD_FLDIREA))) {
1323                 int rc;
1324
1325                 down(&inode->i_sem);
1326                 rc = fsfilt_get_md(req->rq_export->exp_obd, inode, NULL, 0,
1327                                    ((body->valid & OBD_MD_MEA) ? EA_MEA : EA_LOV));
1328                 up(&inode->i_sem);
1329                 if (rc < 0) {
1330                         if (rc != -ENODATA && rc != -EOPNOTSUPP)
1331                                 CERROR("error getting inode %lu MD: rc = %d\n",
1332                                        inode->i_ino, rc);
1333                         size[bufcount] = 0;
1334                 } else if (rc > mds->mds_max_mdsize) {
1335                         size[bufcount] = 0;
1336                         CERROR("MD size %d larger than maximum possible %u\n",
1337                                rc, mds->mds_max_mdsize);
1338                 } else {
1339                         size[bufcount] = rc;
1340                 }
1341                 bufcount++;
1342         } else if (S_ISLNK(inode->i_mode) && (body->valid & OBD_MD_LINKNAME)) {
1343                 if (inode->i_size + 1 != body->eadatasize)
1344                         CERROR("symlink size: %Lu, reply space: %d\n",
1345                                inode->i_size + 1, body->eadatasize);
1346                 size[bufcount] = min_t(int, inode->i_size+1, body->eadatasize);
1347                 bufcount++;
1348                 CDEBUG(D_INODE, "symlink size: %Lu, reply space: %d\n",
1349                        inode->i_size + 1, body->eadatasize);
1350         } else if ((body->valid & OBD_MD_FLXATTR)) {
1351                 char *ea_name = lustre_msg_string(req->rq_reqmsg,
1352                                                   offset + 1, 0);
1353                 rc = -EOPNOTSUPP;
1354
1355                 if (!strcmp(ea_name, XATTR_NAME_LUSTRE_ACL)) {
1356                         size[bufcount] = LUSTRE_ACL_SIZE_MAX;
1357                 } else {
1358                         if (inode->i_op && inode->i_op->getxattr)
1359                                 rc = inode->i_op->getxattr(de, ea_name,
1360                                                            NULL, 0);
1361
1362                         if (rc < 0) {
1363                                 if (rc != -ENODATA && rc != -EOPNOTSUPP)
1364                                         CERROR("error get inode %lu EA: %d\n",
1365                                                inode->i_ino, rc);
1366                                 size[bufcount] = 0;
1367                         } else {
1368                                 size[bufcount] = min_t(int,
1369                                                        body->eadatasize, rc);
1370                         }
1371                 }
1372                 bufcount++;
1373         } else if (body->valid & OBD_MD_FLXATTRLIST) {
1374                 rc = -EOPNOTSUPP;
1375                 if (inode->i_op && inode->i_op->getxattr)
1376                         rc = inode->i_op->listxattr(de, NULL, 0);
1377
1378                 if (rc < 0) {
1379                         if (rc != -ENODATA && rc != -EOPNOTSUPP)
1380                                 CERROR("error getting inode %lu EA: rc = %d\n",
1381                                        inode->i_ino, rc);
1382                         size[bufcount] = 0;
1383                 } else {
1384                         size[bufcount] = min_t(int, body->eadatasize, rc);
1385                 }
1386                 bufcount++;
1387         }
1388
1389         /* may co-exist with OBD_MD_FLEASIZE */
1390         if (body->valid & OBD_MD_FLACL) {
1391                 if (req->rq_export->exp_mds_data.med_remote) {
1392                         size[bufcount++] = sizeof(int);
1393                         size[bufcount++] = sizeof(struct mds_remote_perm);
1394                 } else {
1395                         size[bufcount++] = sizeof(int);
1396                         size[bufcount++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
1397                 }
1398         }
1399
1400         if (body->valid & OBD_MD_FLKEY) {
1401                 size[bufcount++] = sizeof(int);
1402                 size[bufcount++] = sizeof(struct crypto_key);
1403         }
1404
1405         if (body->valid & OBD_MD_CAPA)
1406                 size[bufcount++] = sizeof(struct lustre_capa);
1407
1408         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
1409                 CERROR("failed MDS_GETATTR_PACK test\n");
1410                 req->rq_status = -ENOMEM;
1411                 GOTO(out, rc = -ENOMEM);
1412         }
1413
1414         rc = lustre_pack_reply(req, bufcount, size, NULL);
1415         if (rc) {
1416                 CERROR("out of memory\n");
1417                 GOTO(out, req->rq_status = rc);
1418         }
1419
1420         EXIT;
1421  out:
1422         return rc;
1423 }
1424
1425 int mds_check_mds_num(struct obd_device *obd, struct inode *inode,
1426                       char *name, int namelen)
1427 {
1428         struct mea *mea = NULL;
1429         int mea_size, rc = 0;
1430         ENTRY;
1431
1432         rc = mds_md_get_attr(obd, inode, &mea, &mea_size);
1433         if (rc)
1434                 RETURN(rc);
1435         if (mea != NULL && mea->mea_count) {
1436                 /*
1437                  * dir is already splitted, check if requested filename should
1438                  * live at this MDS or at another one.
1439                  */
1440                 int i = mea_name2idx(mea, name, namelen - 1);
1441                 if (mea->mea_master != id_group(&mea->mea_ids[i])) {
1442                         CDEBUG(D_OTHER,
1443                                "inapropriate MDS(%d) for %s. should be "
1444                                "%lu(%d)\n", mea->mea_master, name,
1445                                (unsigned long)id_group(&mea->mea_ids[i]), i);
1446                         rc = -ERESTART;
1447                 }
1448         }
1449
1450         if (mea)
1451                 OBD_FREE(mea, mea_size);
1452         RETURN(rc);
1453 }
1454
1455 int mds_getattr_size(struct obd_device *obd, struct dentry *dentry,
1456                      struct ptlrpc_request *req, struct mds_body *body)
1457 {
1458         struct inode *inode = dentry->d_inode;
1459         ENTRY;
1460
1461         LASSERT(body != NULL);
1462
1463         if (dentry->d_inode == NULL || !S_ISREG(inode->i_mode))
1464                 RETURN(0);
1465
1466         /* XXX: quite a ugly hack, need to check old code
1467          * drop FLSIZE/FLBLOCKS prior any checking to */
1468         body->valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS|OBD_MD_FLMTIME);
1469
1470         if (obd->obd_recovering) {
1471                 CDEBUG(D_INODE, "size for "DLID4" is unknown yet (recovering)\n",
1472                        OLID4(&body->id1));
1473                 RETURN(0);
1474         }
1475
1476         if (atomic_read(&inode->i_writecount)) {
1477                 /* some one has opened the file for write.
1478                  * mds doesn't know actual size */
1479                 CDEBUG(D_INODE, "MDS doesn't know actual size for "DLID4"\n",
1480                        OLID4(&body->id1));
1481                 RETURN(0);
1482         }
1483         CDEBUG(D_INODE, "MDS returns "LPD64"/"LPD64" for"DLID4"\n",
1484                body->size, body->blocks, OLID4(&body->id1));
1485         body->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLMTIME;
1486         RETURN(0);
1487 }
1488
1489 static int mds_getattr_lock(struct ptlrpc_request *req, int offset,
1490                             struct lustre_handle *child_lockh, int child_part)
1491 {
1492         struct obd_device *obd = req->rq_export->exp_obd;
1493         struct mds_obd *mds = &obd->u.mds;
1494         struct ldlm_reply *rep = NULL;
1495         struct lvfs_run_ctxt saved;
1496         struct mds_req_sec_desc *rsd;
1497         struct mds_body *body;
1498         struct dentry *dparent = NULL, *dchild = NULL;
1499         struct lvfs_ucred uc = {NULL, NULL,};
1500         struct lustre_handle parent_lockh[2] = {{0}, {0}};
1501         unsigned int namesize = 0;
1502         int rc = 0, cleanup_phase = 0, resent_req = 0, update_mode, reply_offset;
1503         char *name = NULL;
1504         ENTRY;
1505
1506         LASSERT(!strcmp(obd->obd_type->typ_name, OBD_MDS_DEVICENAME));
1507         MD_COUNTER_INCREMENT(obd, getattr_lock);
1508
1509         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1510         if (!rsd) {
1511                 CERROR("Can't unpack security desc\n");
1512                 RETURN(-EFAULT);
1513         }
1514
1515         /* swab now, before anyone looks inside the request. */
1516         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1517                                   lustre_swab_mds_body);
1518         if (body == NULL) {
1519                 CERROR("Can't swab mds_body\n");
1520                 GOTO(cleanup, rc = -EFAULT);
1521         }
1522
1523         LASSERT_REQSWAB(req, offset + 1);
1524         name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
1525         if (name == NULL) {
1526                 CERROR("Can't unpack name\n");
1527                 GOTO(cleanup, rc = -EFAULT);
1528         }
1529         namesize = req->rq_reqmsg->buflens[offset + 1];
1530
1531         /* namesize less than 2 means we have empty name, probably came from
1532            revalidate by cfid, so no point in having name to be set */
1533         if (namesize <= 1)
1534                 name = NULL;
1535
1536         LASSERT (offset == 1 || offset == 3);
1537         if (offset == 3) {
1538                 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
1539                 reply_offset = 1;
1540         } else {
1541                 reply_offset = 0;
1542         }
1543
1544         rc = mds_init_ucred(&uc, req, rsd);
1545         if (rc) {
1546                 if (child_lockh->cookie == 0)
1547                         mds_audit_auth(req, &uc, AUDIT_STAT, &body->id1,
1548                                        name, namesize - 1);
1549                 GOTO(cleanup, rc);
1550         }
1551
1552         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1553         cleanup_phase = 1; /* kernel context */
1554         intent_set_disposition(rep, DISP_LOOKUP_EXECD);
1555
1556         LASSERT(namesize > 0);
1557         if (child_lockh->cookie != 0) {
1558                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1559                 resent_req = 1;
1560         }
1561 #if HAVE_LOOKUP_RAW
1562         if (body->valid == OBD_MD_FLID) {
1563                 struct mds_body *mds_reply;
1564                 int size = sizeof(*mds_reply);
1565                 struct inode *dir;
1566                 ino_t inum;
1567
1568                 dparent = mds_id2dentry(obd, &body->id1, NULL);
1569                 if (IS_ERR(dparent)) {
1570                         rc = PTR_ERR(dparent);
1571                         GOTO(cleanup, rc);
1572                 }
1573                 /*
1574                  * the user requested ONLY the inode number, so do a raw lookup.
1575                  */
1576                 rc = lustre_pack_reply(req, 1, &size, NULL);
1577                 if (rc) {
1578                         CERROR("out of memory\n");
1579                         l_dput(dparent);
1580                         GOTO(cleanup, rc);
1581                 }
1582                 dir  = dparent->d_inode;
1583                 LASSERT(dir->i_op->lookup_raw != NULL);
1584                 rc = dir->i_op->lookup_raw(dir, name, namesize - 1, &inum);
1585                 l_dput(dparent);
1586                 mds_reply = lustre_msg_buf(req->rq_repmsg, 0,
1587                                            sizeof(*mds_reply));
1588
1589                 id_ino(&mds_reply->id1) = inum;
1590                 mds_reply->valid = OBD_MD_FLID;
1591                 GOTO(cleanup, rc);
1592         }
1593 #endif
1594         if (resent_req == 0) {
1595                 LASSERT(id_fid(&body->id1) != 0);
1596                 if (name) {
1597                         rc = mds_get_parent_child_locked(obd, mds, &body->id1,
1598                                                          parent_lockh, &dparent,
1599                                                          LCK_PR,
1600                                                          MDS_INODELOCK_UPDATE,
1601                                                          &update_mode,
1602                                                          name, namesize,
1603                                                          child_lockh, &dchild,
1604                                                          LCK_PR, child_part);
1605                         if (rc)
1606                                 GOTO(cleanup, rc);
1607
1608                         cleanup_phase = 2; /* dchild, dparent, locks */
1609
1610                         /*
1611                          * let's make sure this name should leave on this mds
1612                          * node.
1613                          */
1614                         rc = mds_check_mds_num(obd, dparent->d_inode, name, namesize);
1615                         if (rc)
1616                                 GOTO(cleanup, rc);
1617                 } else {
1618                         /* we have no dentry here, drop LOOKUP bit */
1619                         /* FIXME: we need MDS_INODELOCK_LOOKUP or not. */
1620                         child_part &= ~MDS_INODELOCK_LOOKUP;
1621                         CDEBUG(D_OTHER, "%s: retrieve attrs for "DLID4"\n",
1622                                obd->obd_name, OLID4(&body->id1));
1623
1624                         dchild = mds_id2locked_dentry(obd, &body->id1, NULL,
1625                                                       LCK_PR, parent_lockh,
1626                                                       &update_mode, NULL, 0,
1627                                                       MDS_INODELOCK_UPDATE);
1628                         if (IS_ERR(dchild)) {
1629                                 CERROR("can't find inode with id "DLID4", err = %d\n",
1630                                        OLID4(&body->id1), (int)PTR_ERR(dchild));
1631                                 rc = PTR_ERR(dchild);
1632                                 dchild = NULL;
1633                                 GOTO(cleanup, rc);
1634                         }
1635                         memcpy(child_lockh, parent_lockh, sizeof(parent_lockh[0]));
1636                 }
1637         } else {
1638                 struct ldlm_lock *granted_lock;
1639
1640                 DEBUG_REQ(D_DLMTRACE, req, "resent, not enqueuing new locks");
1641                 granted_lock = ldlm_handle2lock(child_lockh);
1642
1643                 LASSERTF(granted_lock != NULL, LPU64"/%lu lockh "LPX64"\n",
1644                          id_fid(&body->id1), (unsigned long)id_group(&body->id1),
1645                          child_lockh->cookie);
1646
1647                 if (name) {
1648                         /* usual named request */
1649                         dparent = mds_id2dentry(obd, &body->id1, NULL);
1650                         LASSERT(!IS_ERR(dparent));
1651                         dchild = ll_lookup_one_len(name, dparent, namesize - 1);
1652                         LASSERT(!IS_ERR(dchild));
1653                 } else {
1654                         /* client wants to get attr. by id */
1655                         dchild = mds_id2dentry(obd, &body->id1, NULL);
1656                         LASSERT(!IS_ERR(dchild));
1657                 }
1658                 LDLM_LOCK_PUT(granted_lock);
1659         }
1660
1661         cleanup_phase = 2; /* dchild, dparent, locks */
1662
1663         if (!DENTRY_VALID(dchild)) {
1664                 intent_set_disposition(rep, DISP_LOOKUP_NEG);
1665                 /*
1666                  * in the intent case, the policy clears this error: the
1667                  * disposition is enough.
1668                  */
1669                 rc = -ENOENT;
1670                 GOTO(cleanup, rc);
1671         } else {
1672                 intent_set_disposition(rep, DISP_LOOKUP_POS);
1673         }
1674
1675         if (req->rq_repmsg == NULL) {
1676                 if (dchild->d_flags & DCACHE_CROSS_REF)
1677                         rc = mds_getattr_pack_msg_cf(req, dchild, offset);
1678                 else
1679                         rc = mds_getattr_pack_msg(req, dchild, offset);
1680                 if (rc != 0) {
1681                         CERROR ("mds_getattr_pack_msg: %d\n", rc);
1682                         GOTO (cleanup, rc);
1683                 }
1684         }
1685
1686         rc = mds_getattr_internal(obd, dchild, req, offset, body,
1687                                   reply_offset, rsd);
1688         if (rc)
1689                 GOTO(cleanup, rc); /* returns the lock to the client */
1690
1691         /* probably MDS knows actual size? */
1692         body = lustre_msg_buf(req->rq_repmsg, reply_offset, sizeof(*body));
1693         LASSERT(body != NULL);
1694         mds_getattr_size(obd, dchild, req, body);
1695
1696         GOTO(cleanup, rc);
1697
1698  cleanup:
1699         switch (cleanup_phase) {
1700         case 2:
1701                 if (resent_req == 0) {
1702                         mds_audit(req, dchild, name, namesize - 1,
1703                                   AUDIT_STAT, rc);
1704                         if (rc && DENTRY_VALID(dchild))
1705                                 ldlm_lock_decref(child_lockh, LCK_PR);
1706                         if (name)
1707                                 ldlm_lock_decref(parent_lockh, LCK_PR);
1708 #ifdef S_PDIROPS
1709                         if (parent_lockh[1].cookie != 0)
1710                                 ldlm_lock_decref(parent_lockh + 1, update_mode);
1711 #endif
1712                 }
1713                 if (dparent)
1714                         l_dput(dparent);
1715
1716                 l_dput(dchild);
1717         case 1:
1718                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1719         default:
1720                 mds_exit_ucred(&uc);
1721         }
1722         return rc;
1723 }
1724
1725 static int mds_getattr(struct ptlrpc_request *req, int offset)
1726 {
1727         struct obd_device *obd = req->rq_export->exp_obd;
1728         struct lvfs_run_ctxt saved;
1729         struct dentry *de;
1730         struct mds_req_sec_desc *rsd;
1731         struct mds_body *body;
1732         struct lvfs_ucred uc = {NULL, NULL,};
1733         int rc = 0;
1734         ENTRY;
1735
1736         MD_COUNTER_INCREMENT(obd, getattr);
1737
1738         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1739         if (!rsd) {
1740                 CERROR("Can't unpack security desc\n");
1741                 RETURN(-EFAULT);
1742         }
1743
1744         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1745                                   lustre_swab_mds_body);
1746         if (body == NULL) {
1747                 CERROR ("Can't unpack body\n");
1748                 RETURN (-EFAULT);
1749         }
1750
1751         rc = mds_init_ucred(&uc, req, rsd);
1752         if (rc) {
1753                 mds_exit_ucred(&uc);
1754                 RETURN(rc);
1755         }
1756
1757         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1758         de = mds_id2dentry(obd, &body->id1, NULL);
1759         if (IS_ERR(de)) {
1760                 rc = req->rq_status = PTR_ERR(de);
1761                 GOTO(out_pop, rc);
1762         }
1763
1764         rc = mds_getattr_pack_msg(req, de, offset);
1765         if (rc != 0) {
1766                 CERROR("mds_getattr_pack_msg: %d\n", rc);
1767                 GOTO(out_pop, rc);
1768         }
1769
1770         req->rq_status = mds_getattr_internal(obd, de, req, offset, body,
1771                                               0, rsd);
1772         l_dput(de);
1773
1774         EXIT;
1775 out_pop:
1776         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1777         mds_exit_ucred(&uc);
1778         return rc;
1779 }
1780
1781 static int mds_access_check(struct ptlrpc_request *req, int offset)
1782 {
1783         struct obd_device *obd = req->rq_export->exp_obd;
1784         struct lvfs_run_ctxt saved;
1785         struct dentry *de;
1786         struct mds_req_sec_desc *rsd;
1787         struct mds_body *body;
1788         struct lvfs_ucred uc;
1789         int rep_size[2] = {sizeof(*body),
1790                            sizeof(struct mds_remote_perm)};
1791         int rc = 0;
1792         ENTRY;
1793
1794         if (!req->rq_export->exp_mds_data.med_remote) {
1795                 CERROR("from local client "LPU64"\n", req->rq_peer.peer_id.nid);
1796                 RETURN(-EINVAL);
1797         }
1798
1799         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1800         if (!rsd) {
1801                 CERROR("Can't unpack security desc\n");
1802                 RETURN(-EFAULT);
1803         }
1804
1805         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1806                                   lustre_swab_mds_body);
1807         if (body == NULL) {
1808                 CERROR ("Can't unpack body\n");
1809                 RETURN (-EFAULT);
1810         }
1811
1812         MD_COUNTER_INCREMENT(obd, access_check);
1813
1814         rc = mds_init_ucred(&uc, req, rsd);
1815         if (rc) {
1816                 CERROR("init ucred error: %d\n", rc);
1817                 RETURN(rc);
1818         }
1819         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1820
1821         de = mds_id2dentry(obd, &body->id1, NULL);
1822         if (IS_ERR(de)) {
1823                 CERROR("grab ino "LPU64": err %ld\n",
1824                        body->id1.li_stc.u.e3s.l3s_ino, PTR_ERR(de));
1825                 GOTO(out_pop, rc = PTR_ERR(de));
1826         }
1827
1828         rc = lustre_pack_reply(req, 2, rep_size, NULL);
1829         if (rc) {
1830                 CERROR("pack reply error: %d\n", rc);
1831                 GOTO(out_dput, rc = -EINVAL);
1832         }
1833
1834         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1835         LASSERT(body);
1836
1837         rc = mds_pack_remote_perm(req, 1, body, de->d_inode);
1838
1839         EXIT;
1840
1841 out_dput:
1842         l_dput(de);
1843 out_pop:
1844         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1845         mds_exit_ucred(&uc);
1846         return rc;
1847 }
1848
1849 static int mds_obd_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1850                           unsigned long max_age)
1851 {
1852         int rc;
1853         ENTRY;
1854
1855         spin_lock(&obd->obd_osfs_lock);
1856         rc = fsfilt_statfs(obd, obd->u.mds.mds_sb, max_age);
1857         if (rc == 0)
1858                 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
1859         spin_unlock(&obd->obd_osfs_lock);
1860
1861         RETURN(rc);
1862 }
1863
1864 static int mds_statfs(struct ptlrpc_request *req)
1865 {
1866         struct obd_device *obd = req->rq_export->exp_obd;
1867         int rc, size = sizeof(struct obd_statfs);
1868         ENTRY;
1869
1870         /* This will trigger a watchdog timeout */
1871         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
1872                          (MDS_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
1873
1874         rc = lustre_pack_reply(req, 1, &size, NULL);
1875         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
1876                 CERROR("mds: statfs lustre_pack_reply failed: rc = %d\n", rc);
1877                 GOTO(out, rc);
1878         }
1879
1880         OBD_COUNTER_INCREMENT(obd, statfs);
1881
1882         /* We call this so that we can cache a bit - 1 jiffie worth */
1883         rc = mds_obd_statfs(obd, lustre_msg_buf(req->rq_repmsg, 0, size),
1884                             jiffies - HZ);
1885         if (rc) {
1886                 CERROR("mds_obd_statfs failed: rc %d\n", rc);
1887                 GOTO(out, rc);
1888         }
1889
1890         EXIT;
1891 out:
1892         req->rq_status = rc;
1893         return rc;
1894 }
1895
1896 static int mds_sync(struct ptlrpc_request *req, int offset)
1897 {
1898         struct obd_device *obd = req->rq_export->exp_obd;
1899         struct mds_obd *mds = &obd->u.mds;
1900         struct mds_body *body;
1901         int rc, size = sizeof(*body);
1902         ENTRY;
1903
1904         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1905                                   lustre_swab_mds_body);
1906         if (body == NULL)
1907                 GOTO(out, rc = -EPROTO);
1908
1909         rc = lustre_pack_reply(req, 1, &size, NULL);
1910         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) {
1911                 CERROR("fsync lustre_pack_reply failed: rc = %d\n", rc);
1912                 GOTO(out, rc);
1913         }
1914
1915         if (id_ino(&body->id1) == 0) {
1916                 /* an id of zero is taken to mean "sync whole filesystem" */
1917                 rc = fsfilt_sync(obd, mds->mds_sb);
1918                 if (rc)
1919                         GOTO(out, rc);
1920         } else {
1921                 /* just any file to grab fsync method - "file" arg unused */
1922                 struct file *file = mds->mds_rcvd_filp;
1923                 struct mds_body *rep_body;
1924                 struct dentry *de;
1925
1926                 de = mds_id2dentry(obd, &body->id1, NULL);
1927                 if (IS_ERR(de))
1928                         GOTO(out, rc = PTR_ERR(de));
1929
1930                 rc = file->f_op->fsync(NULL, de, 1);
1931                 if (rc)
1932                         GOTO(out, rc);
1933
1934                 rep_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep_body));
1935                 mds_pack_inode2body(obd, rep_body, de->d_inode,
1936                                     (body->valid & OBD_MD_FID) ? 1 : 0);
1937                 l_dput(de);
1938         }
1939
1940         EXIT;
1941 out:
1942         req->rq_status = rc;
1943         return rc;
1944 }
1945
1946 /* mds_readpage does not take a DLM lock on the inode, because the client must
1947  * already have a PR lock.
1948  *
1949  * If we were to take another one here, a deadlock will result, if another
1950  * thread is already waiting for a PW lock. */
1951 static int mds_readpage(struct ptlrpc_request *req, int offset)
1952 {
1953         struct obd_device *obd = req->rq_export->exp_obd;
1954         struct vfsmount *mnt;
1955         struct dentry *de;
1956         struct file *file;
1957         struct mds_req_sec_desc *rsd;
1958         struct mds_body *body, *repbody;
1959         struct lvfs_run_ctxt saved;
1960         int rc, size = sizeof(*repbody);
1961         struct lvfs_ucred uc = {NULL, NULL,};
1962         ENTRY;
1963
1964         rc = lustre_pack_reply(req, 1, &size, NULL);
1965         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
1966                 CERROR("mds: out of memory\n");
1967                 GOTO(out, rc = -ENOMEM);
1968         }
1969
1970         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1971         if (!rsd) {
1972                 CERROR("Can't unpack security desc\n");
1973                 GOTO (out, rc = -EFAULT);
1974         }
1975
1976         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1977                                   lustre_swab_mds_body);
1978         if (body == NULL) {
1979                 CERROR("Can't unpack body\n");
1980                 GOTO (out, rc = -EFAULT);
1981         }
1982
1983         rc = mds_init_ucred(&uc, req, rsd);
1984         if (rc) {
1985                 mds_audit_auth(req, &uc, AUDIT_READDIR, &body->id1,
1986                                NULL, 0);
1987                 GOTO(out, rc);
1988         }
1989
1990         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1991         de = mds_id2dentry(obd, &body->id1, &mnt);
1992         if (IS_ERR(de))
1993                 GOTO(out_pop, rc = PTR_ERR(de));
1994
1995         CDEBUG(D_INODE, "ino %lu\n", de->d_inode->i_ino);
1996
1997         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
1998         /* note: in case of an error, dentry_open puts dentry */
1999         if (IS_ERR(file))
2000                 GOTO(out_pop, rc = PTR_ERR(file));
2001
2002         /* body->size is actually the offset -eeb */
2003         if ((body->size & (de->d_inode->i_blksize - 1)) != 0) {
2004                 CERROR("offset "LPU64" not on a block boundary of %lu\n",
2005                        body->size, de->d_inode->i_blksize);
2006                 GOTO(out_file, rc = -EFAULT);
2007         }
2008
2009         /* body->nlink is actually the #bytes to read -eeb */
2010         if (body->nlink & (de->d_inode->i_blksize - 1)) {
2011                 CERROR("size %u is not multiple of blocksize %lu\n",
2012                        body->nlink, de->d_inode->i_blksize);
2013                 GOTO(out_file, rc = -EFAULT);
2014         }
2015
2016         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*repbody));
2017         repbody->size = file->f_dentry->d_inode->i_size;
2018         repbody->valid = OBD_MD_FLSIZE;
2019
2020         /* to make this asynchronous make sure that the handling function
2021            doesn't send a reply when this function completes. Instead a
2022            callback function would send the reply */
2023         /* body->size is actually the offset -eeb */
2024         rc = mds_sendpage(req, file, body->size, body->nlink);
2025
2026         EXIT;
2027 out_file:
2028         filp_close(file, 0);
2029 out_pop:
2030         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2031 out:
2032         mds_exit_ucred(&uc);
2033         req->rq_status = rc;
2034         return 0;
2035 }
2036
2037 int mds_read_md(struct obd_device *obd, struct lustre_id *id,
2038                 char **data, int *datalen)
2039 {
2040         struct dentry *dentry;
2041         struct mds_obd *mds = &obd->u.mds;
2042         int rc = 0, mea = 0;
2043         char *ea;
2044         ENTRY;
2045
2046         LASSERT(id);
2047         LASSERT(obd);
2048
2049         dentry = mds_id2dentry(obd, id, NULL);
2050         if (IS_ERR(dentry))
2051                 GOTO(out, rc = PTR_ERR(dentry));
2052
2053         if (!dentry->d_inode) {
2054                 CERROR("Can't find object "DLID4".\n",
2055                        OLID4(id));
2056                 GOTO(out_dentry, rc = -EINVAL);
2057         }
2058         if (S_ISDIR(dentry->d_inode->i_mode)) {
2059                 *datalen = obd_packmd(mds->mds_md_exp, NULL, NULL);
2060                 mea = 1;
2061         } else {
2062                 *datalen = obd_packmd(mds->mds_dt_exp, NULL, NULL);
2063                 mea = 0;
2064         }
2065         OBD_ALLOC(ea, *datalen);
2066         if (!ea) {
2067                 *datalen = 0;
2068                 GOTO(out_dentry, rc = PTR_ERR(dentry));
2069         }
2070         *data = ea;
2071         down(&dentry->d_inode->i_sem);
2072         rc = fsfilt_get_md(obd, dentry->d_inode, *data, *datalen,
2073                            (mea ? EA_MEA : EA_LOV));
2074         up(&dentry->d_inode->i_sem);
2075
2076         if (rc < 0)
2077                 CERROR("Error %d reading eadata for ino %lu\n",
2078                         rc, dentry->d_inode->i_ino);
2079 out_dentry:
2080         l_dput(dentry);
2081 out:
2082         RETURN(rc);
2083 }
2084 EXPORT_SYMBOL(mds_read_md);
2085
2086 int mds_reint(struct ptlrpc_request *req, int offset,
2087               struct lustre_handle *lockh)
2088 {
2089         struct mds_update_record *rec;
2090         struct mds_req_sec_desc *rsd;
2091         int rc;
2092         ENTRY;
2093
2094         OBD_ALLOC(rec, sizeof(*rec));
2095         if (rec == NULL)
2096                 RETURN(-ENOMEM);
2097
2098         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
2099         if (!rsd) {
2100                 CERROR("Can't unpack security desc\n");
2101                 GOTO(out, rc = -EFAULT);
2102         }
2103
2104         rc = mds_update_unpack(req, offset, rec);
2105         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
2106                 CERROR("invalid record\n");
2107                 GOTO(out, req->rq_status = -EINVAL);
2108         }
2109
2110         rc = mds_init_ucred(&rec->ur_uc, req, rsd);
2111         if (rc) {
2112                 mds_audit_reint(req, rec);
2113                 GOTO(out, rc);
2114         }
2115
2116         /* rc will be used to interrupt a for loop over multiple records */
2117         rc = mds_reint_rec(rec, offset, req, lockh);
2118
2119 out:
2120         mds_exit_ucred(&rec->ur_uc);
2121         OBD_FREE(rec, sizeof(*rec));
2122         RETURN(rc);
2123 }
2124
2125 static int mds_filter_recovery_request(struct ptlrpc_request *req,
2126                                        struct obd_device *obd, int *process)
2127 {
2128         switch (req->rq_reqmsg->opc) {
2129         case MDS_CONNECT: /* This will never get here, but for completeness. */
2130         case OST_CONNECT: /* This will never get here, but for completeness. */
2131         case MDS_DISCONNECT:
2132         case OST_DISCONNECT:
2133                *process = 1;
2134                RETURN(0);
2135
2136         case MDS_CLOSE:
2137         case MDS_SYNC: /* used in unmounting */
2138         case OBD_PING:
2139         case MDS_REINT:
2140         case LDLM_ENQUEUE:
2141         case OST_CREATE:
2142                 *process = target_queue_recovery_request(req, obd);
2143                 RETURN(0);
2144
2145         default:
2146                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2147                 *process = 0;
2148                 /* XXX what should we set rq_status to here? */
2149                 req->rq_status = -EAGAIN;
2150                 RETURN(ptlrpc_error(req));
2151         }
2152 }
2153
2154 static char *reint_names[] = {
2155         [REINT_SETATTR] "setattr",
2156         [REINT_CREATE]  "create",
2157         [REINT_LINK]    "link",
2158         [REINT_UNLINK]  "unlink",
2159         [REINT_RENAME]  "rename",
2160         [REINT_OPEN]    "open",
2161 };
2162
2163 #define FILTER_VALID_FLAGS (OBD_MD_FLTYPE | OBD_MD_FLMODE | OBD_MD_FLGENER  | \
2164                             OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ| \
2165                             OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME| \
2166                             OBD_MD_FLID)
2167
2168 static void reconstruct_create(struct ptlrpc_request *req)
2169 {
2170         struct mds_export_data *med = &req->rq_export->exp_mds_data;
2171         struct mds_client_data *mcd = med->med_mcd;
2172         struct dentry *dentry;
2173         struct ost_body *body;
2174         struct lustre_id id;
2175         int rc;
2176         ENTRY;
2177
2178         /* copy rc, transno and disp; steal locks */
2179         mds_req_from_mcd(req, mcd);
2180         if (req->rq_status) {
2181                 EXIT;
2182                 return;
2183         }
2184
2185         id_gen(&id) = 0;
2186         id_group(&id) = 0;
2187
2188         id_ino(&id) = mcd->mcd_last_data;
2189         LASSERT(id_ino(&id) != 0);
2190
2191         dentry = mds_id2dentry(req2obd(req), &id, NULL);
2192         if (IS_ERR(dentry)) {
2193                 CERROR("can't find inode "LPU64"\n", id_ino(&id));
2194                 req->rq_status = PTR_ERR(dentry);
2195                 EXIT;
2196                 return;
2197         }
2198
2199         CWARN("reconstruct reply for x"LPU64" (remote ino) "LPU64" -> %lu/%u\n",
2200               req->rq_xid, id_ino(&id), dentry->d_inode->i_ino,
2201               dentry->d_inode->i_generation);
2202
2203         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
2204         obdo_from_inode(&body->oa, dentry->d_inode, FILTER_VALID_FLAGS);
2205         body->oa.o_id = dentry->d_inode->i_ino;
2206         body->oa.o_generation = dentry->d_inode->i_generation;
2207         body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2208
2209         down(&dentry->d_inode->i_sem);
2210         rc = mds_read_inode_sid(req2obd(req), dentry->d_inode, &id);
2211         up(&dentry->d_inode->i_sem);
2212         if (rc) {
2213                 CERROR("Can't read inode self id, inode %lu, "
2214                        "rc %d\n", dentry->d_inode->i_ino, rc);
2215                 id_fid(&id) = 0;
2216         }
2217
2218         body->oa.o_fid = id_fid(&id);
2219         body->oa.o_mds = id_group(&id);
2220         l_dput(dentry);
2221
2222         EXIT;
2223 }
2224
2225 static int mds_inode_init_acl(struct obd_device *obd, void *handle,
2226                               struct dentry *de, void *xattr, int xattr_size)
2227 {
2228         struct inode *inode = de->d_inode;
2229         struct posix_acl *acl;
2230         mode_t mode;
2231         int rc = 0;
2232
2233         LASSERT(handle);
2234         LASSERT(inode);
2235         LASSERT(xattr);
2236         LASSERT(xattr_size > 0);
2237
2238         if (!inode->i_op->getxattr || !inode->i_op->setxattr) {
2239                 CERROR("backend fs dosen't support xattr\n");
2240                 return -EOPNOTSUPP;
2241         }
2242
2243         /* set default acl */
2244         if (S_ISDIR(inode->i_mode)) {
2245                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_DEFAULT,
2246                                            xattr, xattr_size, 0);
2247                 if (rc) {
2248                         CERROR("set default acl err: %d\n", rc);
2249                         return rc;
2250                 }
2251         }
2252
2253         /* set access acl */
2254         acl = posix_acl_from_xattr(xattr, xattr_size);
2255         if (acl == NULL || IS_ERR(acl)) {
2256                 CERROR("insane attr data\n");
2257                 return PTR_ERR(acl);
2258         }
2259
2260         if (posix_acl_valid(acl)) {
2261                 CERROR("default acl not valid: %d\n", rc);
2262                 rc = -EFAULT;
2263                 goto out;
2264         }
2265
2266         mode = inode->i_mode;
2267         rc = posix_acl_create_masq(acl, &mode);
2268         if (rc < 0) {
2269                 CERROR("create masq err %d\n", rc);
2270                 goto out;
2271         }
2272
2273         if (inode->i_mode != mode) {
2274                 struct iattr iattr = { .ia_valid = ATTR_MODE,
2275                                        .ia_mode = mode };
2276                 int rc2;
2277
2278                 rc2 = fsfilt_setattr(obd, de, handle, &iattr, 0);
2279                 if (rc2) {
2280                         CERROR("setattr mode err: %d\n", rc2);
2281                         rc = rc2;
2282                         goto out;
2283                 }
2284         }
2285
2286         if (rc > 0) {
2287                 /* we didn't change acl except mode bits of some
2288                  * entries, so should be fit into original size.
2289                  */
2290                 rc = posix_acl_to_xattr(acl, xattr, xattr_size);
2291                 LASSERT(rc > 0);
2292
2293                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_ACCESS,
2294                                            xattr, xattr_size, 0);
2295                 if (rc)
2296                         CERROR("set access acl err: %d\n", rc);
2297         }
2298 out:
2299         posix_acl_release(acl);
2300         return rc;
2301 }
2302
2303 static int mdt_obj_create(struct ptlrpc_request *req)
2304 {
2305         struct obd_device *obd = req->rq_export->exp_obd;
2306         struct mds_obd *mds = &obd->u.mds;
2307         struct ost_body *body, *repbody;
2308         void *acl = NULL;
2309         int acl_size;
2310         char idname[LL_ID_NAMELEN];
2311         int size = sizeof(*repbody);
2312         struct inode *parent_inode;
2313         struct lvfs_run_ctxt saved;
2314         int rc, cleanup_phase = 0;
2315         struct dentry *new = NULL;
2316         struct dentry_params dp;
2317         int mealen, flags = 0;
2318         struct lvfs_ucred uc;
2319         struct lustre_id id;
2320         struct mea *mea;
2321         void *handle = NULL;
2322         unsigned long cr_inum = 0;
2323         __u64 fid = 0;
2324         ENTRY;
2325
2326         DEBUG_REQ(D_HA, req, "create remote object");
2327         parent_inode = mds->mds_unnamed_dir->d_inode;
2328
2329         body = lustre_swab_reqbuf(req, 0, sizeof(*body),
2330                                   lustre_swab_ost_body);
2331         if (body == NULL)
2332                 RETURN(-EFAULT);
2333
2334         /* acl data is packed transparently, no swab here */
2335         LASSERT(req->rq_reqmsg->bufcount >= 2);
2336         acl_size = req->rq_reqmsg->buflens[1];
2337         if (acl_size) {
2338                 acl = lustre_msg_buf(req->rq_reqmsg, 1, acl_size);
2339                 if (!acl) {
2340                         CERROR("No default acl buf?\n");
2341                         RETURN(-EFAULT);
2342                 }
2343         }
2344
2345         rc = lustre_pack_reply(req, 1, &size, NULL);
2346         if (rc)
2347                 RETURN(rc);
2348
2349         MDS_CHECK_RESENT(req, reconstruct_create(req));
2350
2351         uc.luc_lsd = NULL;
2352         uc.luc_ginfo = NULL;
2353         uc.luc_uid = body->oa.o_uid;
2354         uc.luc_gid = body->oa.o_gid;
2355         uc.luc_fsuid = body->oa.o_uid;
2356         uc.luc_fsgid = body->oa.o_gid;
2357
2358         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2359         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
2360
2361         /* in REPLAY case inum should be given (client or other MDS fills it) */
2362         if (body->oa.o_id && ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2363             (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))) {
2364                 /*
2365                  * this is re-create request from MDS holding directory name.
2366                  * we have to lookup given ino/gen first. if it exists (good
2367                  * case) then there is nothing to do. if it does not then we
2368                  * have to recreate it.
2369                  */
2370                 id_ino(&id) = body->oa.o_id;
2371                 id_gen(&id) = body->oa.o_generation;
2372
2373                 new = mds_id2dentry(obd, &id, NULL);
2374                 if (!IS_ERR(new) && new->d_inode) {
2375                         struct lustre_id sid;
2376
2377                         CDEBUG(D_OTHER, "mkdir repairing %lu/%lu\n",
2378                                (unsigned long)id_ino(&id),
2379                                (unsigned long)id_gen(&id));
2380
2381                         obdo_from_inode(&repbody->oa, new->d_inode,
2382                                         FILTER_VALID_FLAGS);
2383
2384                         repbody->oa.o_id = new->d_inode->i_ino;
2385                         repbody->oa.o_generation = new->d_inode->i_generation;
2386                         repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2387                         cleanup_phase = 1;
2388
2389                         down(&new->d_inode->i_sem);
2390                         rc = mds_read_inode_sid(obd, new->d_inode, &sid);
2391                         up(&new->d_inode->i_sem);
2392                         if (rc) {
2393                                 CERROR("Can't read inode self id "
2394                                        "inode %lu, rc %d.\n",
2395                                        new->d_inode->i_ino, rc);
2396                                 GOTO(cleanup, rc);
2397                         }
2398
2399                         repbody->oa.o_fid = id_fid(&sid);
2400                         repbody->oa.o_mds = id_group(&sid);
2401                         LASSERT(id_fid(&sid) != 0);
2402
2403                         /*
2404                          * here we could use fid passed in body->oa.o_fid and
2405                          * thus avoid mds_read_inode_sid().
2406                          */
2407                         cr_inum = new->d_inode->i_ino;
2408                         GOTO(cleanup, rc = 0);
2409                 }
2410         }
2411
2412         down(&parent_inode->i_sem);
2413         handle = fsfilt_start(obd, parent_inode, FSFILT_OP_MKDIR, NULL);
2414         if (IS_ERR(handle)) {
2415                 up(&parent_inode->i_sem);
2416                 CERROR("fsfilt_start() failed, rc = %d\n",
2417                        (int)PTR_ERR(handle));
2418                 GOTO(cleanup, rc = PTR_ERR(handle));
2419         }
2420         cleanup_phase = 1; /* transaction */
2421
2422 repeat:
2423         rc = sprintf(idname, "%u.%u", ll_insecure_random_int(), current->pid);
2424         new = lookup_one_len(idname, mds->mds_unnamed_dir, rc);
2425         if (IS_ERR(new)) {
2426                 CERROR("%s: can't lookup new inode (%s) for mkdir: %d\n",
2427                        obd->obd_name, idname, (int) PTR_ERR(new));
2428                 fsfilt_commit(obd, mds->mds_sb, new->d_inode, handle, 0);
2429                 up(&parent_inode->i_sem);
2430                 RETURN(PTR_ERR(new));
2431         } else if (new->d_inode) {
2432                 CERROR("%s: name exists. repeat\n", obd->obd_name);
2433                 goto repeat;
2434         }
2435         if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2436              lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2437                 fid = body->oa.o_fid;
2438         } else {
2439                 fid = mds_alloc_fid(obd);
2440         }
2441         new->d_fsdata = (void *)&dp;
2442         dp.p_inum = 0;
2443         dp.p_ptr = req;
2444
2445         dp.p_fid = fid;
2446         dp.p_group = mds->mds_num;
2447
2448         if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) ||
2449             (body->oa.o_flags & OBD_FL_RECREATE_OBJS)) {
2450                 LASSERT(body->oa.o_id != 0);
2451                 dp.p_inum = body->oa.o_id;
2452                 DEBUG_REQ(D_HA, req, "replay create obj %lu/%lu",
2453                           (unsigned long)body->oa.o_id,
2454                           (unsigned long)body->oa.o_generation);
2455         }
2456
2457         rc = vfs_mkdir(parent_inode, new, body->oa.o_mode);
2458         if (rc == 0) {
2459                 if (acl) {
2460                         rc = mds_inode_init_acl(obd, handle, new,
2461                                                 acl, acl_size);
2462                         if (rc) {
2463                                 up(&parent_inode->i_sem);
2464                                 GOTO(cleanup, rc);
2465                         }
2466                 }
2467                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2468                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2469                         new->d_inode->i_generation = body->oa.o_generation;
2470                         mark_inode_dirty(new->d_inode);
2471
2472                         /*
2473                          * avoiding asserts in cache flush case, as
2474                          * @body->oa.o_id should be zero.
2475                          */
2476                         if (body->oa.o_id) {
2477                                 LASSERTF(body->oa.o_id == new->d_inode->i_ino,
2478                                          "BUG 3550: failed to recreate obj "
2479                                          LPU64" -> %lu\n", body->oa.o_id,
2480                                          new->d_inode->i_ino);
2481
2482                                 LASSERTF(body->oa.o_generation ==
2483                                          new->d_inode->i_generation,
2484                                          "BUG 3550: failed to recreate obj/gen "
2485                                          LPU64"/%u -> %lu/%u\n", body->oa.o_id,
2486                                          body->oa.o_generation,
2487                                          new->d_inode->i_ino,
2488                                          new->d_inode->i_generation);
2489                         }
2490                 }
2491
2492                 obdo_from_inode(&repbody->oa, new->d_inode, FILTER_VALID_FLAGS);
2493                 repbody->oa.o_id = new->d_inode->i_ino;
2494                 repbody->oa.o_generation = new->d_inode->i_generation;
2495                 repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FID;
2496
2497                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2498                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2499                         LASSERT(body->oa.o_id != 0);
2500                         LASSERT(body->oa.o_fid != 0);
2501                 }
2502
2503                 mds_inode2id(obd, &id, new->d_inode, fid);
2504                 mds_update_inode_ids(obd, new->d_inode, handle, &id,
2505                                      body->oa.o_valid & OBD_MD_FLID ?
2506                                      NULL : obdo_id(&body->oa));
2507
2508                 /* initializing o_fid after it is allocated. */
2509                 repbody->oa.o_fid = id_fid(&id);
2510                 repbody->oa.o_mds = id_group(&id);
2511
2512                 rc = fsfilt_del_dir_entry(obd, new);
2513                 up(&parent_inode->i_sem);
2514                 if (rc) {
2515                         CERROR("can't remove name for object: %d\n", rc);
2516                         GOTO(cleanup, rc);
2517                 }
2518
2519                 cleanup_phase = 2; /* created directory object */
2520
2521                 CDEBUG(D_OTHER, "created dirobj: %lu/%lu mode %o\n",
2522                        (unsigned long)new->d_inode->i_ino,
2523                        (unsigned long)new->d_inode->i_generation,
2524                        (unsigned)new->d_inode->i_mode);
2525                 cr_inum = new->d_inode->i_ino;
2526         } else {
2527                 up(&parent_inode->i_sem);
2528                 CERROR("%s: can't create dirobj: %d\n", obd->obd_name, rc);
2529                 GOTO(cleanup, rc);
2530         }
2531
2532         if (body->oa.o_valid & OBD_MD_FLID) {
2533                 /* this is new object for splitted dir. We have to prevent
2534                  * recursive splitting on it -bzzz */
2535                 mealen = obd_size_diskmd(mds->mds_md_exp, NULL);
2536
2537                 OBD_ALLOC(mea, mealen);
2538                 if (mea == NULL)
2539                         GOTO(cleanup, rc = -ENOMEM);
2540
2541                 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
2542                 mea->mea_master = body->oa.o_mds;      /* master mds num */
2543                 mea->mea_count = 0;
2544
2545                 obdo2id(&mea->mea_ids[body->oa.o_mds], &body->oa);
2546
2547                 down(&new->d_inode->i_sem);
2548                 rc = fsfilt_set_md(obd, new->d_inode, handle,
2549                                    mea, mealen, EA_MEA);
2550                 up(&new->d_inode->i_sem);
2551                 if (rc)
2552                         CERROR("fsfilt_set_md() failed, "
2553                                "rc = %d\n", rc);
2554
2555                 OBD_FREE(mea, mealen);
2556
2557                 CDEBUG(D_OTHER, "%s: mark non-splittable %lu/%u - %d\n",
2558                        obd->obd_name, new->d_inode->i_ino,
2559                        new->d_inode->i_generation, flags);
2560         } else if (body->oa.o_easize) {
2561                 /* we pass LCK_EX to split routine to signal that we have
2562                  * exclusive access to the directory. simple because nobody
2563                  * knows it already exists -bzzz */
2564                 rc = mds_try_to_split_dir(obd, new, NULL,
2565                                           body->oa.o_easize, LCK_EX);
2566                 if (rc < 0) {
2567                         CERROR("Can't split directory %lu, error = %d.\n",
2568                                new->d_inode->i_ino, rc);
2569                 } else {
2570                         rc = 0;
2571                 }
2572         }
2573
2574         EXIT;
2575 cleanup:
2576         if (rc == 0 && (body->oa.o_flags & OBD_FL_REINT)) {
2577                 rc = mds_fidmap_add(obd, &id);
2578                 if (rc < 0) {
2579                         CERROR("can't create fid->ino mapping, "
2580                                "err %d\n", rc);
2581                 } else {
2582                         rc = 0;
2583                 }
2584         }
2585
2586         switch (cleanup_phase) {
2587         case 2: /* object has been created, but we'll may want to replay it later */
2588                 if (rc == 0)
2589                         ptlrpc_require_repack(req);
2590         case 1: /* transaction */
2591                 rc = mds_finish_transno(mds, parent_inode, handle,
2592                                         req, rc, cr_inum);
2593         }
2594
2595         l_dput(new);
2596         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2597         return rc;
2598 }
2599
2600 static int mdt_get_info(struct ptlrpc_request *req)
2601 {
2602         struct obd_export *exp = req->rq_export;
2603         int keylen, rc = 0;
2604         char *key;
2605         ENTRY;
2606
2607         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2608         if (key == NULL) {
2609                 DEBUG_REQ(D_HA, req, "no get_info key");
2610                 RETURN(-EFAULT);
2611         }
2612         keylen = req->rq_reqmsg->buflens[0];
2613
2614         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
2615             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
2616             (keylen < strlen("lovdesc") || strcmp(key, "lovdesc") != 0) &&
2617             (keylen < strlen("getext") || strcmp(key, "getext") != 0) &&
2618             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
2619                 RETURN(-EPROTO);
2620
2621         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
2622                 struct lustre_id *reply;
2623                 int size = sizeof(*reply);
2624
2625                 rc = lustre_pack_reply(req, 1, &size, NULL);
2626                 if (rc)
2627                         RETURN(rc);
2628
2629                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2630                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2631         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2632                 struct lov_desc *reply;
2633                 int size = sizeof(*reply);
2634
2635                 rc = lustre_pack_reply(req, 1, &size, NULL);
2636                 if (rc)
2637                         RETURN(rc);
2638
2639                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2640                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2641         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2642                 struct fid_extent *reply;
2643                 int size = sizeof(*reply);
2644
2645                 rc = lustre_pack_reply(req, 1, &size, NULL);
2646                 if (rc)
2647                         RETURN(rc);
2648
2649                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2650                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2651         } else {
2652                 obd_id *reply;
2653                 int size = sizeof(*reply);
2654
2655                 rc = lustre_pack_reply(req, 1, &size, NULL);
2656                 if (rc)
2657                         RETURN(rc);
2658
2659                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2660                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2661         }
2662
2663         req->rq_repmsg->status = 0;
2664         RETURN(rc);
2665 }
2666
2667 static int mds_set_info(struct obd_export *exp, __u32 keylen,
2668                         void *key, __u32 vallen, void *val)
2669 {
2670         struct obd_device *obd;
2671         struct mds_obd *mds;
2672         int rc = 0;
2673         ENTRY;
2674
2675         obd = class_exp2obd(exp);
2676         if (obd == NULL) {
2677                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2678                        exp->exp_handle.h_cookie);
2679                 RETURN(-EINVAL);
2680         }
2681
2682         mds = &obd->u.mds;
2683         if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2684                 rc = mds_set_audit(obd, val);
2685                 RETURN(rc);
2686         } else if (keylen >= strlen("ids") && memcmp(key, "ids", keylen) == 0) {
2687                 struct lustre_id *ids = (struct lustre_id *)val;
2688                 struct dentry *de;
2689                 struct inode *inode;
2690                 void *handle;
2691                 int err;
2692
2693                 de = mds_id2dentry(obd, ids, NULL);
2694                 if (IS_ERR(de)) {
2695                         rc = PTR_ERR(de);
2696                         CERROR("lookup by an id error rc=%d\n ", rc);
2697                         RETURN(rc);
2698                 }
2699                 inode = de->d_inode;
2700                 if (inode == NULL)
2701                         GOTO(out_put, rc = -ENOENT);
2702
2703                 down(&inode->i_sem);
2704                 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
2705                 if (IS_ERR(handle)) {
2706                         up(&inode->i_sem);
2707                         GOTO(out_put, rc = PTR_ERR(handle));
2708                 }
2709
2710                 rc = mds_update_inode_ids(obd, inode, handle, NULL, ids + 1);
2711
2712                 err = fsfilt_commit(obd, mds->mds_sb, inode, handle,
2713                                     exp->exp_sync);
2714                 if (err) {
2715                         CERROR("error committing transaction: %d\n", err);
2716                         if (!rc) rc = err;
2717                 }
2718                 up(&inode->i_sem);
2719 out_put:
2720                 l_dput(de);
2721                 RETURN(rc);
2722         }
2723
2724         if (keylen >= strlen("crypto_type") &&
2725              memcmp(key, "crypto_type", keylen) == 0) {
2726                 rc = mds_set_crypto_type(obd, val, vallen);
2727                 RETURN(rc);
2728         }
2729
2730         if (keylen >= strlen("setext") && !memcmp(key, "setext", keylen)) {
2731                 struct fid_extent *ext = val;
2732
2733                 CDEBUG(D_IOCTL, "set last fid to extent ["LPD64"-"LPD64"]\n",
2734                        ext->fe_start, ext->fe_width);
2735
2736                 /* set lastfid into fid extent start. All next object creates
2737                  * will use that fid. */
2738                 mds_set_last_fid(obd, ext->fe_start);
2739
2740                 /* setting the same extent to OSC to avoid ids intersecting in
2741                  * object ids, as all cache MDSs have the same group 0. */
2742                 rc = obd_set_info(mds->mds_dt_exp, strlen("setext"),
2743                                   "setext", sizeof(*ext), ext);
2744                 if (rc) {
2745                         CERROR("can't set extent ["LPD64"-"LPD64"] to %s, "
2746                                "err %d\n", ext->fe_start, ext->fe_width,
2747                                mds->mds_dt_exp->exp_obd->obd_name, rc);
2748                 }
2749                 RETURN(rc);
2750         }
2751         CDEBUG(D_IOCTL, "invalid key\n");
2752         RETURN(-EINVAL);
2753 }
2754
2755 static int mdt_set_info(struct ptlrpc_request *req)
2756 {
2757         char *key, *val;
2758         struct obd_export *exp = req->rq_export;
2759         int keylen, rc = 0, vallen = 0;
2760         ENTRY;
2761
2762         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2763         if (key == NULL) {
2764                 DEBUG_REQ(D_HA, req, "no set_info key");
2765                 RETURN(-EFAULT);
2766         }
2767         keylen = req->rq_reqmsg->buflens[0];
2768
2769         if ((keylen == strlen("crypto_type") &&
2770             memcmp(key, "crypto_type", keylen) == 0)) {
2771                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2772                 if (rc)
2773                         RETURN(rc);
2774
2775                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2776                 vallen = req->rq_reqmsg->buflens[1];
2777
2778                 rc = obd_set_info(exp, keylen, key, vallen, val);
2779                 req->rq_repmsg->status = 0;
2780                 RETURN(rc);
2781         } else if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2782                 struct audit_attr_msg msg, *p;
2783                 int rc = 0;
2784
2785                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2786                 if (rc)
2787                         RETURN(rc);
2788
2789                 p = lustre_swab_reqbuf(req, 1, sizeof(msg),
2790                                        lustre_swab_audit_attr);
2791
2792                 msg = *p;
2793                 //CDEBUG(D_INFO, "Get new audit setting 0x%x\n", (__u32)msg.attr);
2794                 rc = obd_set_info(exp, keylen, key, sizeof(msg), &msg);
2795
2796                 req->rq_repmsg->status = rc;
2797                 RETURN(rc);
2798         } else if (keylen == strlen("ids") &&
2799                    memcmp(key, "ids", keylen) == 0) {
2800                 struct lustre_id *id, ids[2];
2801
2802                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2803                 if (rc)
2804                         RETURN(rc);
2805                 id = lustre_swab_reqbuf(req, 1, sizeof(struct lustre_id),
2806                                         lustre_swab_lustre_id);
2807                 ids[0] = *id;
2808                 id = lustre_swab_reqbuf(req, 2, sizeof(struct lustre_id),
2809                                         lustre_swab_lustre_id);
2810                 ids[1] = *id;
2811
2812                 rc = obd_set_info(exp, keylen, key, vallen, ids);
2813                 req->rq_repmsg->status = rc;
2814                 RETURN(rc);
2815         } else if (keylen == strlen("setext") &&
2816                    memcmp(key, "setext", keylen) == 0) {
2817                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2818                 if (rc)
2819                         RETURN(rc);
2820
2821                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2822                 vallen = req->rq_reqmsg->buflens[1];
2823
2824                 rc = obd_set_info(exp, keylen, key, vallen, val);
2825                 req->rq_repmsg->status = 0;
2826                 RETURN(rc);
2827         }
2828
2829         CDEBUG(D_IOCTL, "invalid key\n");
2830         RETURN(-EINVAL);
2831 }
2832
2833 static void mds_revoke_export_locks(struct obd_export *exp)
2834 {
2835         struct list_head *locklist = &exp->exp_ldlm_data.led_held_locks;
2836         struct list_head  rpc_list;
2837         struct ldlm_lock *lock, *next;
2838         struct ldlm_lock_desc desc;
2839
2840         /* don't do this for local client */
2841         if (!exp->u.eu_mds_data.med_remote)
2842                 return;
2843
2844         /* don't revoke locks during recovery */
2845         if (exp->exp_obd->obd_recovering)
2846                 return;
2847
2848         ENTRY;
2849         INIT_LIST_HEAD(&rpc_list);
2850
2851         spin_lock(&exp->exp_ldlm_data.led_lock);
2852         list_for_each_entry_safe(lock, next, locklist, l_export_chain) {
2853
2854                 lock_res_and_lock(lock);
2855                 if (lock->l_req_mode != lock->l_granted_mode) {
2856                         unlock_res_and_lock(lock);
2857                         continue;
2858                 }
2859
2860                 LASSERT(lock->l_resource);
2861                 if (lock->l_resource->lr_type != LDLM_IBITS &&
2862                     lock->l_resource->lr_type != LDLM_PLAIN) {
2863                         unlock_res_and_lock(lock);
2864                         continue;
2865                 }
2866
2867                 if (lock->l_flags & LDLM_FL_AST_SENT) {
2868                         unlock_res_and_lock(lock);
2869                         continue;
2870                 }
2871
2872                 LASSERT(lock->l_blocking_ast);
2873                 LASSERT(!lock->l_blocking_lock);
2874
2875                 lock->l_flags |= LDLM_FL_AST_SENT;
2876                 unlock_res_and_lock(lock);
2877
2878                 list_move(&lock->l_export_chain, &rpc_list);
2879         }
2880         spin_unlock(&exp->exp_ldlm_data.led_lock);
2881
2882         while (!list_empty(&rpc_list)) {
2883                 lock = list_entry(rpc_list.next, struct ldlm_lock,
2884                                   l_export_chain);
2885                 list_del_init(&lock->l_export_chain);
2886
2887                 /* the desc just pretend to exclusive */
2888                 ldlm_lock2desc(lock, &desc);
2889                 desc.l_req_mode = LCK_EX;
2890                 desc.l_granted_mode = 0;
2891
2892                 lock->l_blocking_ast(lock, &desc, NULL, LDLM_CB_BLOCKING);
2893         }
2894         EXIT;
2895 }
2896
2897 static int mds_msg_check_version(struct lustre_msg *msg)
2898 {
2899         int rc;
2900
2901         switch (msg->opc) {
2902         case MDS_CONNECT:
2903         case MDS_DISCONNECT:
2904         case OBD_PING:
2905                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2906                 if (rc)
2907                         CERROR("bad opc %u version %08x, expecting %08x\n",
2908                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2909                 break;
2910         case MDS_STATFS:
2911         case MDS_GETSTATUS:
2912         case MDS_GETATTR:
2913         case MDS_GETATTR_LOCK:
2914         case MDS_ACCESS_CHECK:
2915         case MDS_READPAGE:
2916         case MDS_REINT:
2917         case MDS_CLOSE:
2918         case MDS_DONE_WRITING:
2919         case MDS_PIN:
2920         case MDS_SYNC:
2921         case MDS_PARSE_ID:
2922                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2923                 if (rc)
2924                         CERROR("bad opc %u version %08x, expecting %08x\n",
2925                                msg->opc, msg->version, LUSTRE_MDS_VERSION);
2926                 break;
2927         case LDLM_ENQUEUE:
2928         case LDLM_CONVERT:
2929         case LDLM_BL_CALLBACK:
2930         case LDLM_CP_CALLBACK:
2931         case LDLM_FLK_DEADLOCK_CHK:
2932                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2933                 if (rc)
2934                         CERROR("bad opc %u version %08x, expecting %08x\n",
2935                                msg->opc, msg->version, LUSTRE_DLM_VERSION);
2936                 break;
2937         case OBD_LOG_CANCEL:
2938         case LLOG_ORIGIN_HANDLE_OPEN:
2939         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2940         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2941         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2942         case LLOG_ORIGIN_HANDLE_CLOSE:
2943         case LLOG_CATINFO:
2944                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2945                 if (rc)
2946                         CERROR("bad opc %u version %08x, expecting %08x\n",
2947                                msg->opc, msg->version, LUSTRE_LOG_VERSION);
2948                 break;
2949         case OST_CREATE:
2950         case OST_WRITE:
2951         case OST_GET_INFO:
2952         case OST_SET_INFO:
2953                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2954                 if (rc)
2955                         CERROR("bad opc %u version %08x, expecting %08x\n",
2956                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2957                 break;
2958         case SEC_INIT:
2959         case SEC_INIT_CONTINUE:
2960         case SEC_FINI:
2961                 rc = 0;
2962                 break;
2963         default:
2964                 CERROR("MDS unknown opcode %d\n", msg->opc);
2965                 rc = -ENOTSUPP;
2966                 break;
2967         }
2968
2969         return rc;
2970 }
2971
2972 int mds_handle(struct ptlrpc_request *req)
2973 {
2974         int should_process, fail = OBD_FAIL_MDS_ALL_REPLY_NET;
2975         struct obd_device *obd = NULL;
2976         struct mds_obd *mds = NULL; /* quell gcc overwarning */
2977         int rc = 0;
2978         ENTRY;
2979
2980         OBD_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0);
2981
2982         rc = mds_msg_check_version(req->rq_reqmsg);
2983         if (rc) {
2984                 CERROR("MDS drop mal-formed request\n");
2985                 RETURN(rc);
2986         }
2987
2988         /* Security opc should NOT trigger any recovery events */
2989         if (req->rq_reqmsg->opc == SEC_INIT ||
2990             req->rq_reqmsg->opc == SEC_INIT_CONTINUE) {
2991                 if (req->rq_export) {
2992                         mds_req_add_idmapping(req,
2993                                               &req->rq_export->exp_mds_data);
2994                         mds_revoke_export_locks(req->rq_export);
2995                 }
2996                 GOTO(out, rc = 0);
2997         } else if (req->rq_reqmsg->opc == SEC_FINI) {
2998                 if (req->rq_export) {
2999                         mds_req_del_idmapping(req,
3000                                               &req->rq_export->exp_mds_data);
3001                         mds_revoke_export_locks(req->rq_export);
3002                 }
3003                 GOTO(out, rc = 0);
3004         }
3005
3006         LASSERT(current->journal_info == NULL);
3007         /* XXX identical to OST */
3008         if (req->rq_reqmsg->opc != MDS_CONNECT) {
3009                 struct mds_export_data *med;
3010                 int recovering;
3011
3012                 if (req->rq_export == NULL) {
3013                         CERROR("operation %d on unconnected MDS from %s\n",
3014                                req->rq_reqmsg->opc,
3015                                req->rq_peerstr);
3016                         req->rq_status = -ENOTCONN;
3017                         GOTO(out, rc = -ENOTCONN);
3018                 }
3019
3020                 med = &req->rq_export->exp_mds_data;
3021                 obd = req->rq_export->exp_obd;
3022                 mds = &obd->u.mds;
3023
3024                 /* sanity check: if the xid matches, the request must
3025                  * be marked as a resent or replayed */
3026                 if (req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_xid) ||
3027                    req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_close_xid)) {
3028                         LASSERTF(lustre_msg_get_flags(req->rq_reqmsg) &
3029                                  (MSG_RESENT | MSG_REPLAY),
3030                                  "rq_xid "LPU64" matches last_xid, "
3031                                  "expected RESENT flag\n",
3032                                  req->rq_xid);
3033                 }
3034                 /* else: note the opposite is not always true; a
3035                  * RESENT req after a failover will usually not match
3036                  * the last_xid, since it was likely never
3037                  * committed. A REPLAYed request will almost never
3038                  * match the last xid, however it could for a
3039                  * committed, but still retained, open. */
3040
3041                 spin_lock_bh(&obd->obd_processing_task_lock);
3042                 recovering = obd->obd_recovering;
3043                 spin_unlock_bh(&obd->obd_processing_task_lock);
3044                 if (recovering) {
3045                         rc = mds_filter_recovery_request(req, obd,
3046                                                          &should_process);
3047                         if (rc || should_process == 0) {
3048                                 RETURN(rc);
3049                         } else if (should_process < 0) {
3050                                 req->rq_status = should_process;
3051                                 rc = ptlrpc_error(req);
3052                                 RETURN(rc);
3053                         }
3054                 }
3055         }
3056
3057         switch (req->rq_reqmsg->opc) {
3058         case MDS_CONNECT:
3059                 DEBUG_REQ(D_INODE, req, "connect");
3060                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
3061                 rc = target_handle_connect(req);
3062                 if (!rc) {
3063                         struct mds_export_data *med;
3064
3065                         LASSERT(req->rq_export);
3066                         med = &req->rq_export->u.eu_mds_data;
3067                         mds_init_export_data(req, med);
3068                         mds_req_add_idmapping(req, med);
3069
3070                         /* Now that we have an export, set mds. */
3071                         obd = req->rq_export->exp_obd;
3072                         mds = mds_req2mds(req);
3073                 }
3074                 break;
3075
3076         case MDS_DISCONNECT:
3077                 DEBUG_REQ(D_INODE, req, "disconnect");
3078                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
3079                 rc = target_handle_disconnect(req);
3080                 req->rq_status = rc;            /* superfluous? */
3081                 break;
3082
3083         case MDS_GETSTATUS:
3084                 DEBUG_REQ(D_INODE, req, "getstatus");
3085                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
3086                 rc = mds_getstatus(req);
3087                 break;
3088
3089         case MDS_GETATTR:
3090                 DEBUG_REQ(D_INODE, req, "getattr");
3091                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
3092                 rc = mds_getattr(req, MDS_REQ_REC_OFF);
3093                 break;
3094
3095         case MDS_ACCESS_CHECK:
3096                 DEBUG_REQ(D_INODE, req, "access_check");
3097                 OBD_FAIL_RETURN(OBD_FAIL_MDS_ACCESS_CHECK_NET, 0);
3098                 rc = mds_access_check(req, MDS_REQ_REC_OFF);
3099                 break;
3100
3101         case MDS_GETATTR_LOCK: {
3102                 struct lustre_handle lockh;
3103                 DEBUG_REQ(D_INODE, req, "getattr_lock");
3104                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_LOCK_NET, 0);
3105
3106                 /* If this request gets a reconstructed reply, we won't be
3107                  * acquiring any new locks in mds_getattr_lock, so we don't
3108                  * want to cancel.
3109                  */
3110                 lockh.cookie = 0;
3111                 rc = mds_getattr_lock(req, MDS_REQ_REC_OFF, &lockh,
3112                                       MDS_INODELOCK_UPDATE);
3113                 /* this non-intent call (from an ioctl) is special */
3114                 req->rq_status = rc;
3115                 if (rc == 0 && lockh.cookie)
3116                         ldlm_lock_decref(&lockh, LCK_PR);
3117                 break;
3118         }
3119         case MDS_STATFS:
3120                 DEBUG_REQ(D_INODE, req, "statfs");
3121                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
3122                 rc = mds_statfs(req);
3123                 break;
3124
3125         case MDS_READPAGE:
3126                 DEBUG_REQ(D_INODE, req, "readpage");
3127                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
3128                 rc = mds_readpage(req, MDS_REQ_REC_OFF);
3129
3130                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_SENDPAGE)) {
3131                         if (req->rq_reply_state) {
3132                                 lustre_free_reply_state (req->rq_reply_state);
3133                                 req->rq_reply_state = NULL;
3134                         }
3135                         RETURN(0);
3136                 }
3137
3138                 break;
3139         case MDS_REINT: {
3140                 __u32 *opcp = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
3141                                              sizeof (*opcp));
3142                 __u32  opc;
3143                 int size[3] = {sizeof(struct mds_body), mds->mds_max_mdsize,
3144                                mds->mds_max_cookiesize};
3145                 int bufcount;
3146
3147                 /* NB only peek inside req now; mds_reint() will swab it */
3148                 if (opcp == NULL) {
3149                         CERROR ("Can't inspect opcode\n");
3150                         rc = -EINVAL;
3151                         break;
3152                 }
3153                 opc = *opcp;
3154                 if (lustre_msg_swabbed (req->rq_reqmsg))
3155                         __swab32s(&opc);
3156
3157                 DEBUG_REQ(D_INODE, req, "reint %d (%s)", opc,
3158                           (opc < sizeof(reint_names) / sizeof(reint_names[0]) ||
3159                            reint_names[opc] == NULL) ? reint_names[opc] :
3160                                                        "unknown opcode");
3161
3162                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
3163
3164                 if (opc == REINT_UNLINK || opc == REINT_RENAME)
3165                         bufcount = 3;
3166                 else if (opc == REINT_OPEN)
3167                         bufcount = 2;
3168                 else
3169                         bufcount = 1;
3170
3171                 /* for SETATTR: I have different reply setting for
3172                  * remote setfacl, so delay the reply buffer allocation.
3173                  */
3174                 if (opc != REINT_SETATTR) {
3175                         rc = lustre_pack_reply(req, bufcount, size, NULL);
3176                         if (rc)
3177                                 break;
3178                 }
3179
3180                 rc = mds_reint(req, MDS_REQ_REC_OFF, NULL);
3181                 fail = OBD_FAIL_MDS_REINT_NET_REP;
3182                 break;
3183         }
3184
3185         case MDS_CLOSE:
3186                 DEBUG_REQ(D_INODE, req, "close");
3187                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
3188                 rc = mds_close(req, MDS_REQ_REC_OFF);
3189                 break;
3190
3191         case MDS_DONE_WRITING:
3192                 DEBUG_REQ(D_INODE, req, "done_writing");
3193                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DONE_WRITING_NET, 0);
3194                 rc = mds_done_writing(req, MDS_REQ_REC_OFF);
3195                 break;
3196
3197         case MDS_PIN:
3198                 DEBUG_REQ(D_INODE, req, "pin");
3199                 OBD_FAIL_RETURN(OBD_FAIL_MDS_PIN_NET, 0);
3200                 rc = mds_pin(req, MDS_REQ_REC_OFF);
3201                 break;
3202
3203         case MDS_SYNC:
3204                 DEBUG_REQ(D_INODE, req, "sync");
3205                 OBD_FAIL_RETURN(OBD_FAIL_MDS_SYNC_NET, 0);
3206                 rc = mds_sync(req, MDS_REQ_REC_OFF);
3207                 break;
3208         case MDS_PARSE_ID:
3209                 DEBUG_REQ(D_INODE, req, "parseid");
3210                 rc = mds_parse_id(req);
3211                 break;
3212         case OBD_PING:
3213                 DEBUG_REQ(D_INODE, req, "ping");
3214                 rc = target_handle_ping(req);
3215                 break;
3216
3217         case OBD_LOG_CANCEL:
3218                 CDEBUG(D_INODE, "log cancel\n");
3219                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
3220                 rc = -ENOTSUPP; /* la la la */
3221                 break;
3222
3223         case LDLM_ENQUEUE:
3224                 DEBUG_REQ(D_INODE, req, "enqueue");
3225                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
3226                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
3227                                          ldlm_server_blocking_ast, NULL);
3228                 fail = OBD_FAIL_LDLM_REPLY;
3229                 break;
3230         case LDLM_CONVERT:
3231                 DEBUG_REQ(D_INODE, req, "convert");
3232                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
3233                 rc = ldlm_handle_convert(req);
3234                 break;
3235         case LDLM_BL_CALLBACK:
3236         case LDLM_CP_CALLBACK:
3237                 DEBUG_REQ(D_INODE, req, "callback");
3238                 CERROR("callbacks should not happen on MDS\n");
3239                 LBUG();
3240                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
3241                 break;
3242         case LDLM_FLK_DEADLOCK_CHK:
3243                 DEBUG_REQ(D_INODE, req, "flock deadlock check");
3244                 rc = ldlm_handle_flock_deadlock_check(req);
3245                 break;
3246         case LLOG_ORIGIN_HANDLE_OPEN:
3247                 DEBUG_REQ(D_INODE, req, "llog_init");
3248                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3249                 rc = llog_origin_handle_open(req);
3250                 break;
3251         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3252                 DEBUG_REQ(D_INODE, req, "llog next block");
3253                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3254                 rc = llog_origin_handle_next_block(req);
3255                 break;
3256         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3257                 DEBUG_REQ(D_INODE, req, "llog prev block");
3258                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3259                 rc = llog_origin_handle_prev_block(req);
3260                 break;
3261         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3262                 DEBUG_REQ(D_INODE, req, "llog read header");
3263                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3264                 rc = llog_origin_handle_read_header(req);
3265                 break;
3266         case LLOG_ORIGIN_HANDLE_CLOSE:
3267                 DEBUG_REQ(D_INODE, req, "llog close");
3268                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3269                 rc = llog_origin_handle_close(req);
3270                 break;
3271         case OST_CREATE:
3272                 DEBUG_REQ(D_INODE, req, "ost_create");
3273                 rc = mdt_obj_create(req);
3274                 break;
3275         case OST_GET_INFO:
3276                 DEBUG_REQ(D_INODE, req, "get_info");
3277                 rc = mdt_get_info(req);
3278                 break;
3279         case OST_SET_INFO:
3280                 DEBUG_REQ(D_INODE, req, "set_info");
3281                 rc = mdt_set_info(req);
3282                 break;
3283         case OST_WRITE:
3284                 CDEBUG(D_INODE, "write\n");
3285                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
3286                 rc = ost_brw_write(req, NULL);
3287                 LASSERT(current->journal_info == NULL);
3288                 /* mdt_brw sends its own replies */
3289                 RETURN(rc);
3290                 break;
3291         case LLOG_CATINFO:
3292                 DEBUG_REQ(D_INODE, req, "llog catinfo");
3293                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3294                 rc = llog_catinfo(req);
3295                 break;
3296         default:
3297                 req->rq_status = -ENOTSUPP;
3298                 rc = ptlrpc_error(req);
3299                 RETURN(rc);
3300         }
3301
3302         LASSERT(current->journal_info == NULL);
3303
3304         /* If we're DISCONNECTing, the mds_export_data is already freed */
3305         if (!rc && req->rq_reqmsg->opc != MDS_DISCONNECT) {
3306                 struct mds_export_data *med = &req->rq_export->exp_mds_data;
3307                 struct obd_device *obd = list_entry(mds, struct obd_device,
3308                                                     u.mds);
3309                 req->rq_repmsg->last_xid =
3310                         le64_to_cpu(med->med_mcd->mcd_last_xid);
3311
3312                 if (!obd->obd_no_transno) {
3313                         req->rq_repmsg->last_committed =
3314                                 obd->obd_last_committed;
3315                 } else {
3316                         DEBUG_REQ(D_IOCTL, req,
3317                                   "not sending last_committed update");
3318                 }
3319                 CDEBUG(D_INFO, "last_transno "LPU64", last_committed "LPU64
3320                        ", xid "LPU64"\n",
3321                        mds->mds_last_transno, obd->obd_last_committed,
3322                        req->rq_xid);
3323         }
3324  out:
3325
3326
3327         target_send_reply(req, rc, fail);
3328         RETURN(0);
3329 }
3330
3331 /* Update the server data on disk.  This stores the new mount_count and also the
3332  * last_rcvd value to disk.  If we don't have a clean shutdown, then the server
3333  * last_rcvd value may be less than that of the clients.  This will alert us
3334  * that we may need to do client recovery.
3335  *
3336  * Also assumes for mds_last_transno that we are not modifying it (no locking).
3337  */
3338 int mds_update_server_data(struct obd_device *obd, int force_sync)
3339 {
3340         struct mds_obd *mds = &obd->u.mds;
3341         struct mds_server_data *msd = mds->mds_server_data;
3342         struct file *filp = mds->mds_rcvd_filp;
3343         struct lvfs_run_ctxt saved;
3344         loff_t off = 0;
3345         int rc;
3346         ENTRY;
3347
3348         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3349         msd->msd_last_transno = cpu_to_le64(mds->mds_last_transno);
3350
3351         CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
3352                mds->mds_mount_count, mds->mds_last_transno);
3353         rc = fsfilt_write_record(obd, filp, msd, sizeof(*msd), &off, force_sync);
3354         if (rc)
3355                 CERROR("error writing MDS server data: rc = %d\n", rc);
3356         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3357
3358         RETURN(rc);
3359 }
3360
3361 /* saves last allocated fid counter to file. */
3362 int mds_update_last_fid(struct obd_device *obd, void *handle,
3363                         int force_sync)
3364 {
3365         struct mds_obd *mds = &obd->u.mds;
3366         struct file *filp = mds->mds_fid_filp;
3367         struct lvfs_run_ctxt saved;
3368         loff_t off = 0;
3369         __u64 last_fid;
3370         int rc = 0;
3371         ENTRY;
3372
3373         spin_lock(&mds->mds_last_fid_lock);
3374         last_fid = mds->mds_last_fid;
3375         spin_unlock(&mds->mds_last_fid_lock);
3376
3377         CDEBUG(D_SUPER, "MDS last_fid is #"LPU64"\n",
3378                last_fid);
3379
3380         if (handle) {
3381                 fsfilt_add_journal_cb(obd, mds->mds_sb, last_fid,
3382                                       handle, mds_commit_last_fid_cb,
3383                                       NULL);
3384         }
3385
3386         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3387         rc = fsfilt_write_record(obd, filp, &last_fid, sizeof(last_fid),
3388                                  &off, force_sync);
3389         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3390
3391         if (rc) {
3392                 CERROR("error writing MDS last_fid #"LPU64
3393                        ", err = %d\n", last_fid, rc);
3394                 RETURN(rc);
3395         }
3396
3397         CDEBUG(D_SUPER, "wrote fid #"LPU64" at idx "
3398                "%llu: err = %d\n", last_fid, off, rc);
3399
3400         RETURN(rc);
3401 }
3402
3403 void mds_set_last_fid(struct obd_device *obd, __u64 fid)
3404 {
3405         struct mds_obd *mds = &obd->u.mds;
3406
3407         spin_lock(&mds->mds_last_fid_lock);
3408         if (fid > mds->mds_last_fid)
3409                 mds->mds_last_fid = fid;
3410         spin_unlock(&mds->mds_last_fid_lock);
3411 }
3412
3413 void mds_commit_last_transno_cb(struct obd_device *obd,
3414                                 __u64 transno, void *data,
3415                                 int error)
3416 {
3417         obd_transno_commit_cb(obd, transno, error);
3418 }
3419
3420 void mds_commit_last_fid_cb(struct obd_device *obd,
3421                             __u64 fid, void *data,
3422                             int error)
3423 {
3424         if (error) {
3425                 CERROR("%s: fid "LPD64" commit error: %d\n",
3426                        obd->obd_name, fid, error);
3427                 return;
3428         }
3429
3430         CDEBUG(D_HA, "%s: fid "LPD64" committed\n",
3431                obd->obd_name, fid);
3432 }
3433
3434 __u64 mds_alloc_fid(struct obd_device *obd)
3435 {
3436         struct mds_obd *mds = &obd->u.mds;
3437         __u64 fid;
3438
3439         spin_lock(&mds->mds_last_fid_lock);
3440         fid = ++mds->mds_last_fid;
3441         spin_unlock(&mds->mds_last_fid_lock);
3442
3443         return fid;
3444 }
3445
3446 /*
3447  * reads inode self id from inode EA. Probably later this should be replaced by
3448  * caching inode self id to avoid raeding it every time it is needed.
3449  */
3450 int mds_read_inode_sid(struct obd_device *obd, struct inode *inode,
3451                        struct lustre_id *id)
3452 {
3453         int rc;
3454         ENTRY;
3455
3456         LASSERT(id != NULL);
3457         LASSERT(obd != NULL);
3458         LASSERT(inode != NULL);
3459
3460         rc = fsfilt_get_md(obd, inode, &id->li_fid,
3461                            sizeof(id->li_fid), EA_SID);
3462         if (rc < 0) {
3463                 CERROR("fsfilt_get_md() failed, "
3464                        "rc = %d\n", rc);
3465                 RETURN(rc);
3466         } else if (!rc) {
3467                 rc = -ENODATA;
3468                 RETURN(rc);
3469         } else {
3470                 rc = 0;
3471         }
3472         id_ino(id) = inode->i_ino;
3473         id_gen(id) = inode->i_generation;
3474         id_type(id) = S_IFMT & inode->i_mode;
3475
3476         RETURN(rc);
3477 }
3478
3479 int mds_read_inode_pid(struct obd_device *obd, struct inode *inode,
3480                        struct lustre_id *id)
3481 {
3482         int rc;
3483         ENTRY;
3484
3485         LASSERT(inode && id);
3486
3487         rc = fsfilt_get_md(obd, inode, id, sizeof(*id), EA_PID);
3488         if (rc < 0)
3489                CERROR("get parent id from EA failed, rc=%d\n", rc);
3490         else if (!rc)
3491                 rc = -ENODATA;
3492         else
3493                 rc = 0;
3494
3495         RETURN(rc);
3496 }
3497
3498 /* updates inode self id in EA. */
3499 int mds_update_inode_ids(struct obd_device *obd, struct inode *inode,
3500                          void *handle, struct lustre_id *id,
3501                          struct lustre_id *pid)
3502 {
3503         int rc = 0;
3504         ENTRY;
3505
3506         LASSERT(id || pid);
3507         LASSERT(id == NULL || id_fid(id) != 0);
3508         LASSERT(pid == NULL || id_fid(pid) != 0);
3509         LASSERT(obd != NULL);
3510         LASSERT(inode != NULL);
3511
3512         if (id) {
3513                 mds_set_last_fid(obd, id_fid(id));
3514                 rc = fsfilt_set_md(obd, inode, handle, &id->li_fid,
3515                                    sizeof(id->li_fid), EA_SID);
3516                 LASSERTF(rc == 0, "failed to update fid:  %d\n", rc);
3517         }
3518         if (pid) {
3519                 rc = fsfilt_set_md(obd, inode, handle, pid,
3520                                    sizeof(*pid), EA_PID);
3521                 LASSERTF(rc == 0, "failed to update parent fid:  %d\n", rc);
3522         }
3523
3524         RETURN(rc);
3525 }
3526
3527 /* mount the file system (secretly) */
3528 static int mds_setup(struct obd_device *obd, obd_count len, void *buf)
3529 {
3530         struct lustre_cfg* lcfg = buf;
3531         struct mds_obd *mds = &obd->u.mds;
3532         struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
3533         char *options = NULL;
3534         struct vfsmount *mnt;
3535         char ns_name[48];
3536         unsigned long page;
3537         struct crypto_tfm *tfm = NULL;
3538         int rc = 0;
3539         ENTRY;
3540
3541         if (lcfg->lcfg_bufcount < 3)
3542                 RETURN(rc = -EINVAL);
3543
3544         if (LUSTRE_CFG_BUFLEN(lcfg, 1) == 0 || LUSTRE_CFG_BUFLEN(lcfg, 2) == 0)
3545                 RETURN(rc = -EINVAL);
3546
3547         obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
3548         if (IS_ERR(obd->obd_fsops))
3549                 RETURN(rc = PTR_ERR(obd->obd_fsops));
3550
3551         mds->mds_max_mdsize = sizeof(struct lov_mds_md);
3552
3553         page = get_zeroed_page(GFP_KERNEL);
3554         if (!page)
3555                 RETURN(-ENOMEM);
3556
3557         options = (char *)page;
3558
3559         /*
3560          * here we use "iopen_nopriv" hardcoded, because it affects MDS utility
3561          * and the rest of options are passed by mount options. Probably this
3562          * should be moved to somewhere else like startup scripts or lconf. */
3563         sprintf(options, "iopen_nopriv");
3564
3565         if (LUSTRE_CFG_BUFLEN(lcfg, 4) > 0 && lustre_cfg_buf(lcfg, 4))
3566                 sprintf(options + strlen(options), ",%s",
3567                         lustre_cfg_string(lcfg, 4));
3568
3569         /* we have to know mdsnum before touching underlying fs -bzzz */
3570         atomic_set(&mds->mds_open_count, 0);
3571         sema_init(&mds->mds_md_sem, 1);
3572         mds->mds_md_connected = 0;
3573         mds->mds_md_name = NULL;
3574
3575         if (LUSTRE_CFG_BUFLEN(lcfg, 5) > 0 && lustre_cfg_buf(lcfg, 5) &&
3576             strncmp(lustre_cfg_string(lcfg, 5), "dumb", LUSTRE_CFG_BUFLEN(lcfg, 5))) {
3577                 class_uuid_t uuid;
3578
3579                 generate_random_uuid(uuid);
3580                 class_uuid_unparse(uuid, &mds->mds_md_uuid);
3581
3582                 OBD_ALLOC(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3583                 if (mds->mds_md_name == NULL)
3584                         RETURN(rc = -ENOMEM);
3585
3586                 memcpy(mds->mds_md_name, lustre_cfg_buf(lcfg, 5),
3587                        LUSTRE_CFG_BUFLEN(lcfg, 5));
3588
3589                 CDEBUG(D_OTHER, "MDS: %s is master for %s\n",
3590                        obd->obd_name, mds->mds_md_name);
3591
3592                 rc = mds_md_connect(obd, mds->mds_md_name);
3593                 if (rc) {
3594                         OBD_FREE(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3595                         GOTO(err_ops, rc);
3596                 }
3597         }
3598
3599         mds->mds_obd_type = MDS_MASTER_OBD;
3600
3601         if (LUSTRE_CFG_BUFLEN(lcfg, 6) > 0 && lustre_cfg_buf(lcfg, 6) &&
3602             strncmp(lustre_cfg_string(lcfg, 6), "dumb",
3603                     LUSTRE_CFG_BUFLEN(lcfg, 6))) {
3604                 if (!memcmp(lustre_cfg_string(lcfg, 6), "master",
3605                             strlen("master"))) {
3606                         mds->mds_obd_type = MDS_MASTER_OBD;
3607                 } else if (!memcmp(lustre_cfg_string(lcfg, 6), "cache",
3608                                    strlen("cache"))) {
3609                         mds->mds_obd_type = MDS_CACHE_OBD;
3610                 }
3611         }
3612
3613         rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
3614                            lustre_cfg_string(lcfg, 2),
3615                            options, 0, &lvfs_ctxt);
3616
3617         free_page(page);
3618
3619         if (rc || !lvfs_ctxt) {
3620                 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
3621                 GOTO(err_ops, rc);
3622         }
3623
3624         mnt = lvfs_ctxt->loc_mnt;
3625         mds->mds_lvfs_ctxt = lvfs_ctxt;
3626         ll_clear_rdonly(ll_sbdev(mnt->mnt_sb));
3627
3628         CDEBUG(D_SUPER, "%s: mnt = %p\n", lustre_cfg_string(lcfg, 1), mnt);
3629
3630         mds->mds_fidext_thumb = 0;
3631         sema_init(&mds->mds_epoch_sem, 1);
3632         atomic_set(&mds->mds_real_clients, 0);
3633         spin_lock_init(&mds->mds_fidext_lock);
3634         spin_lock_init(&mds->mds_fidmap_lock);
3635         spin_lock_init(&mds->mds_transno_lock);
3636         spin_lock_init(&mds->mds_last_fid_lock);
3637         sema_init(&mds->mds_orphan_recovery_sem, 1);
3638         mds->mds_max_cookiesize = sizeof(struct llog_cookie);
3639
3640         sprintf(ns_name, "mds-%s", obd->obd_uuid.uuid);
3641         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
3642
3643         if (obd->obd_namespace == NULL) {
3644                 mds_cleanup(obd, 0);
3645                 GOTO(err_put, rc = -ENOMEM);
3646         }
3647         ldlm_register_intent(obd->obd_namespace, mds_intent_policy);
3648
3649         tfm = crypto_alloc_tfm(CAPA_HMAC_ALG, 0);
3650         if (!tfm)
3651                 GOTO(err_ns, rc = -ENOSYS);
3652
3653         mds->mds_capa_hmac = tfm;
3654         mds->mds_capa_timeout = CAPA_TIMEOUT;
3655         mds->mds_capa_key_timeout = CAPA_KEY_TIMEOUT;
3656
3657         rc = mds_fs_setup(obd, mnt);
3658         if (rc) {
3659                 CERROR("%s: MDS filesystem method init failed: rc = %d\n",
3660                        obd->obd_name, rc);
3661                 GOTO(err_capa, rc);
3662         }
3663
3664         rc = llog_start_commit_thread();
3665         if (rc < 0)
3666
3667                 GOTO(err_fs, rc);
3668
3669         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3) &&
3670             strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3671                     LUSTRE_CFG_BUFLEN(lcfg, 3))) {
3672                 class_uuid_t uuid;
3673
3674                 generate_random_uuid(uuid);
3675                 class_uuid_unparse(uuid, &mds->mds_dt_uuid);
3676
3677                 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
3678                 if (mds->mds_profile == NULL)
3679                         GOTO(err_fs, rc = -ENOMEM);
3680
3681                 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
3682                         LUSTRE_CFG_BUFLEN(lcfg, 3));
3683         }
3684
3685         /*
3686          * setup root dir and files ID dir if lmv already connected, or there is
3687          * not lmv at all.
3688          */
3689         if (mds->mds_md_exp || (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 &&
3690                                 lustre_cfg_buf(lcfg, 3) &&
3691                                 strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3692                                         LUSTRE_CFG_BUFLEN(lcfg, 3)))) {
3693                 rc = mds_fs_setup_rootid(obd);
3694                 if (rc)
3695                         GOTO(err_fs, rc);
3696
3697                 rc = mds_fs_setup_virtid(obd);
3698                 if (rc)
3699                         GOTO(err_fs, rc);
3700
3701         }
3702
3703         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3704                            "mds_ldlm_client", &obd->obd_ldlm_client);
3705         obd->obd_replayable = 1;
3706
3707         mds->mds_crypto_type = NO_CRYPTO;
3708
3709         rc = mds_postsetup(obd);
3710         if (rc)
3711                 GOTO(err_fs, rc);
3712
3713         RETURN(0);
3714
3715 err_fs:
3716         /* No extra cleanup needed for llog_init_commit_thread() */
3717         mds_fs_cleanup(obd, 0);
3718 err_capa:
3719         crypto_free_tfm(mds->mds_capa_hmac);
3720 err_ns:
3721         ldlm_namespace_free(obd->obd_namespace, 0);
3722         obd->obd_namespace = NULL;
3723 err_put:
3724         unlock_kernel();
3725         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3726         mds->mds_sb = 0;
3727         lock_kernel();
3728 err_ops:
3729         fsfilt_put_ops(obd->obd_fsops);
3730         return rc;
3731 }
3732
3733 static int mds_fs_post_setup(struct obd_device *obd)
3734 {
3735         struct mds_obd *mds = &obd->u.mds;
3736         struct dentry *dentry;
3737         int rc = 0;
3738         ENTRY;
3739
3740         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
3741         if (IS_ERR(dentry)) {
3742                 CERROR("Can't find ROOT, err = %d\n",
3743                        (int)PTR_ERR(dentry));
3744                 RETURN(PTR_ERR(dentry));
3745         }
3746         rc = fsfilt_post_setup(obd, dentry);
3747         //set id2name function handler
3748         fsfilt_set_info(obd, mds->mds_sb, NULL, 7, "id2name",
3749                         sizeof(mds_audit_id2name), mds_audit_id2name);
3750
3751         l_dput(dentry);
3752         RETURN(rc);
3753 }
3754
3755 static int mds_postsetup(struct obd_device *obd)
3756 {
3757         struct mds_obd *mds = &obd->u.mds;
3758         int rc = 0;
3759         ENTRY;
3760
3761         rc = obd_llog_setup(obd, &obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT,
3762                             obd, 0, NULL, &llog_lvfs_ops);
3763         if (rc)
3764                 RETURN(rc);
3765
3766         if (mds->mds_profile) {
3767                 struct llog_ctxt *lgctxt;
3768                 struct lvfs_run_ctxt saved;
3769                 struct lustre_profile *lprof;
3770                 struct config_llog_instance cfg;
3771
3772                 cfg.cfg_instance = NULL;
3773                 cfg.cfg_uuid = mds->mds_dt_uuid;
3774                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3775
3776                 lgctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
3777                 if (!lgctxt) {
3778                         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3779                         GOTO(err_llog, rc = -EINVAL);
3780                 }
3781
3782                 rc = class_config_process_llog(lgctxt, mds->mds_profile, &cfg);
3783                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3784
3785                 if (rc)
3786                         GOTO(err_llog, rc);
3787
3788                 lprof = class_get_profile(mds->mds_profile);
3789                 if (lprof == NULL) {
3790                         CERROR("No profile found: %s\n", mds->mds_profile);
3791                         GOTO(err_cleanup, rc = -ENOENT);
3792                 }
3793                 rc = mds_dt_connect(obd, lprof->lp_lov);
3794                 if (rc)
3795                         GOTO(err_cleanup, rc);
3796
3797                 rc = mds_md_postsetup(obd);
3798                 if (rc)
3799                         GOTO(err_cleanup, rc);
3800         }
3801         rc = mds_fs_post_setup(obd);
3802         if (rc)
3803                 CERROR("can not post setup fsfilt\n");
3804         RETURN(rc);
3805 err_cleanup:
3806         mds_dt_clean(obd);
3807 err_llog:
3808         obd_llog_cleanup(llog_get_context(&obd->obd_llogs,
3809                                           LLOG_CONFIG_ORIG_CTXT));
3810         return rc;
3811 }
3812
3813 int mds_postrecov_common(struct obd_device *obd)
3814 {
3815         struct mds_obd *mds = &obd->u.mds;
3816         struct llog_ctxt *ctxt;
3817         int rc, item = 0, valsize;
3818         struct timeval tstart, now;
3819          __u32 group;
3820         ENTRY;
3821
3822         CDEBUG(D_ERROR, "%s: post-recovery\n", obd->obd_name);
3823         LASSERT(!obd->obd_recovering);
3824         ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT);
3825         LASSERT(ctxt != NULL);
3826
3827         /* clean PENDING dir */
3828         do_gettimeofday(&tstart);
3829         rc = mds_cleanup_orphans(obd);
3830         do_gettimeofday(&now);
3831         CDEBUG(D_ERROR, "%s: cleaned %d MDS orphans in %us\n",
3832                obd->obd_name, rc, (unsigned) (now.tv_sec - tstart.tv_sec));
3833         if (rc < 0)
3834                 GOTO(out, rc);
3835         item = rc;
3836
3837         group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
3838         valsize = sizeof(group);
3839         rc = obd_set_info(mds->mds_dt_exp, strlen("mds_conn"),
3840                           "mds_conn", valsize, &group);
3841         if (rc) {
3842                 CERROR("%s: failed to obd_set_info(): %d\n",
3843                        obd->obd_name, rc);
3844                 GOTO(out, rc);
3845         }
3846
3847         rc = llog_connect(ctxt, obd->u.mds.mds_dt_desc.ld_tgt_count,
3848                           NULL, NULL, NULL);
3849         if (rc) {
3850                 CERROR("%s: failed at llog_origin_connect: %d\n",
3851                        obd->obd_name, rc);
3852                 GOTO(out, rc);
3853         }
3854
3855         /* remove the orphaned precreated objects */
3856         do_gettimeofday(&tstart);
3857         rc = mds_dt_clear_orphans(mds, NULL /* all OSTs */);
3858         do_gettimeofday(&now);
3859         CDEBUG(D_ERROR, "%s: cleaned OSS orphans in %us: %d\n",
3860                obd->obd_name, (unsigned) (now.tv_sec - tstart.tv_sec), rc);
3861         if (rc) {
3862                 CERROR("%s: can't clear OSS orphans: %d\n",
3863                        obd->obd_name, rc);
3864                 GOTO(err_llog, rc);
3865         }
3866
3867 out:
3868         RETURN(rc < 0 ? rc : item);
3869
3870 err_llog:
3871         /* cleanup all llogging subsystems */
3872         rc = obd_llog_finish(obd, &obd->obd_llogs,
3873                              mds->mds_dt_desc.ld_tgt_count);
3874         if (rc)
3875                 CERROR("%s: failed to cleanup llogging subsystems\n",
3876                         obd->obd_name);
3877         goto out;
3878 }
3879
3880 int mds_postrecov(struct obd_device *obd)
3881 {
3882         int rc;
3883         ENTRY;
3884         rc = mds_postrecov_common(obd);
3885         if (rc >= 0)
3886                 rc = mds_md_reconnect(obd);
3887         RETURN(rc);
3888 }
3889
3890 int mds_dt_clean(struct obd_device *obd)
3891 {
3892         struct mds_obd *mds = &obd->u.mds;
3893         ENTRY;
3894
3895         if (mds->mds_profile) {
3896                 char * cln_prof;
3897                 struct llog_ctxt *llctx;
3898                 struct lvfs_run_ctxt saved;
3899                 struct config_llog_instance cfg;
3900                 int len = strlen(mds->mds_profile) + sizeof("-clean") + 1;
3901
3902                 OBD_ALLOC(cln_prof, len);
3903                 if (!cln_prof) {
3904                         CERROR("can't allocate memory, processing cleanup "
3905                                "profile is skipped\n");
3906                         goto out;
3907                 }
3908
3909                 sprintf(cln_prof, "%s-clean", mds->mds_profile);
3910
3911                 cfg.cfg_instance = NULL;
3912                 cfg.cfg_uuid = mds->mds_dt_uuid;
3913
3914                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3915                 llctx = llog_get_context(&obd->obd_llogs,
3916                                          LLOG_CONFIG_ORIG_CTXT);
3917                 class_config_process_llog(llctx, cln_prof, &cfg);
3918                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3919
3920                 OBD_FREE(cln_prof, len);
3921         out:
3922                 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
3923                 mds->mds_profile = NULL;
3924         }
3925         RETURN(0);
3926 }
3927
3928 int mds_md_clean(struct obd_device *obd)
3929 {
3930         struct mds_obd *mds = &obd->u.mds;
3931         ENTRY;
3932
3933         if (mds->mds_md_name) {
3934                 OBD_FREE(mds->mds_md_name, strlen(mds->mds_md_name) + 1);
3935                 mds->mds_md_name = NULL;
3936         }
3937         RETURN(0);
3938 }
3939
3940 static int mds_precleanup(struct obd_device *obd, int flags)
3941 {
3942         int rc = 0;
3943         ENTRY;
3944
3945         mds_md_clean(obd);
3946         mds_dt_disconnect(obd, flags);
3947         mds_dt_clean(obd);
3948         obd_llog_cleanup(llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT));
3949         RETURN(rc);
3950 }
3951
3952 extern void lgss_svc_cache_purge_all(void);
3953 static int mds_cleanup(struct obd_device *obd, int flags)
3954 {
3955         struct mds_obd *mds = &obd->u.mds;
3956         ENTRY;
3957
3958         if (mds->mds_sb == NULL)
3959                 RETURN(0);
3960
3961         mds_update_server_data(obd, 1);
3962         mds_update_last_fid(obd, NULL, 1);
3963
3964         if (mds->mds_dt_objids != NULL) {
3965                 int size = mds->mds_dt_desc.ld_tgt_count *
3966                         sizeof(obd_id);
3967                 OBD_FREE(mds->mds_dt_objids, size);
3968         }
3969         mds_fs_cleanup(obd, flags);
3970
3971         unlock_kernel();
3972
3973         /* 2 seems normal on mds, (may_umount() also expects 2
3974           fwiw), but we only see 1 at this point in obdfilter. */
3975         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3976
3977         mds->mds_sb = 0;
3978
3979         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
3980
3981         spin_lock_bh(&obd->obd_processing_task_lock);
3982         if (obd->obd_recovering) {
3983                 target_cancel_recovery_timer(obd);
3984                 obd->obd_recovering = 0;
3985         }
3986         spin_unlock_bh(&obd->obd_processing_task_lock);
3987
3988         lock_kernel();
3989         fsfilt_put_ops(obd->obd_fsops);
3990
3991 #ifdef ENABLE_GSS
3992         /* XXX */
3993         lgss_svc_cache_purge_all();
3994 #endif
3995
3996         spin_lock(&mds->mds_denylist_lock);
3997         while (!list_empty( &mds->mds_denylist ) ) {
3998                 deny_sec_t *p_deny_sec = list_entry(mds->mds_denylist.next,
3999                                                     deny_sec_t, list);
4000                 list_del(&p_deny_sec->list);
4001                 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
4002         }
4003         spin_unlock(&mds->mds_denylist_lock);
4004
4005         mds_capa_keys_cleanup(obd);
4006
4007         if (mds->mds_capa_hmac)
4008                 crypto_free_tfm(mds->mds_capa_hmac);
4009         RETURN(0);
4010 }
4011
4012 static int set_security(const char *value, char **sec)
4013 {
4014         if (!strcmp(value, "null"))
4015                 *sec = "null";
4016         else if (!strcmp(value, "krb5i"))
4017                 *sec = "krb5i";
4018         else if (!strcmp(value, "krb5p"))
4019                 *sec = "krb5p";
4020         else {
4021                 CERROR("Unrecognized security flavor %s\n", value);
4022                 return -EINVAL;
4023         }
4024
4025         return 0;
4026 }
4027
4028 static int mds_process_config(struct obd_device *obd, obd_count len, void *buf)
4029 {
4030         struct lustre_cfg *lcfg = buf;
4031         struct mds_obd *mds = &obd->u.mds;
4032         int rc = 0;
4033         ENTRY;
4034
4035         switch(lcfg->lcfg_command) {
4036         case LCFG_SET_SECURITY: {
4037                 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
4038                     (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
4039                         GOTO(out, rc = -EINVAL);
4040
4041                 if (!strcmp(lustre_cfg_string(lcfg, 1), "mds_sec"))
4042                         rc = set_security(lustre_cfg_string(lcfg, 2),
4043                                           &mds->mds_mds_sec);
4044                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "oss_sec"))
4045                         rc = set_security(lustre_cfg_string(lcfg, 2),
4046                                           &mds->mds_ost_sec);
4047                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
4048                         spin_lock(&mds->mds_denylist_lock);
4049                         rc = add_deny_security(lustre_cfg_string(lcfg, 2),
4050                                                &mds->mds_denylist);
4051                         spin_unlock(&mds->mds_denylist_lock);
4052                 } else {
4053                         CERROR("Unrecognized key\n");
4054                         rc = -EINVAL;
4055                 }
4056                 break;
4057         }
4058         default:
4059                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
4060                 GOTO(out, rc = -EINVAL);
4061         }
4062 out:
4063         RETURN(rc);
4064 }
4065
4066 static void fixup_handle_for_resent_req(struct ptlrpc_request *req,
4067                                         int offset,
4068                                         struct ldlm_lock *new_lock,
4069                                         struct ldlm_lock **old_lock,
4070                                         struct lustre_handle *lockh)
4071 {
4072         struct obd_export *exp = req->rq_export;
4073         struct obd_device *obd = exp->exp_obd;
4074         struct ldlm_request *dlmreq =
4075                 lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*dlmreq));
4076         struct lustre_handle remote_hdl = dlmreq->lock_handle1;
4077         struct list_head *iter;
4078
4079         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4080                 return;
4081
4082         spin_lock(&obd->obd_namespace->ns_hash_lock);
4083         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
4084                 struct ldlm_lock *lock;
4085                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
4086                 if (lock == new_lock)
4087                         continue;
4088                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
4089                         lockh->cookie = lock->l_handle.h_cookie;
4090                         LDLM_DEBUG(lock, "restoring lock cookie");
4091                         DEBUG_REQ(D_HA, req, "restoring lock cookie "LPX64,
4092                                   lockh->cookie);
4093                         if (old_lock)
4094                                 *old_lock = LDLM_LOCK_GET(lock);
4095                         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4096                         return;
4097                 }
4098         }
4099         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4100
4101         /* If the xid matches, then we know this is a resent request,
4102          * and allow it. (It's probably an OPEN, for which we don't
4103          * send a lock */
4104         if (req->rq_xid ==
4105             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_xid))
4106                 return;
4107
4108         if (req->rq_xid ==
4109             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_close_xid))
4110                 return;
4111
4112         /* This remote handle isn't enqueued, so we never received or
4113          * processed this request.  Clear MSG_RESENT, because it can
4114          * be handled like any normal request now. */
4115
4116         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4117
4118         DEBUG_REQ(D_HA, req, "no existing lock with rhandle "LPX64,
4119                   remote_hdl.cookie);
4120 }
4121
4122 int intent_disposition(struct ldlm_reply *rep, int flag)
4123 {
4124         if (!rep)
4125                 return 0;
4126         return (rep->lock_policy_res1 & flag);
4127 }
4128
4129 void intent_set_disposition(struct ldlm_reply *rep, int flag)
4130 {
4131         if (!rep)
4132                 return;
4133         rep->lock_policy_res1 |= flag;
4134 }
4135
4136 static int mds_intent_prepare_reply_buffers(struct ptlrpc_request *req,
4137                                             struct ldlm_intent *it)
4138 {
4139         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
4140         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
4141         int rc, reply_buffers;
4142         int repsize[8] = {sizeof(struct ldlm_reply),
4143                           sizeof(struct mds_body),
4144                           mds->mds_max_mdsize};
4145         ENTRY;
4146
4147         reply_buffers = 3;
4148         if (it->opc & ( IT_OPEN | IT_GETATTR | IT_LOOKUP | IT_CHDIR )) {
4149                 /*Ugly here, Actually, we should prepare the reply buffer
4150                  *after we know whether these stuff exist or not, which should
4151                  * be fixed in future, Now each item is in the fix position,
4152                  * the sequence is lsm, acl, crypto ea, capa.*/
4153                 repsize[reply_buffers++] = sizeof(int);
4154                 if (med->med_remote)
4155                         repsize[reply_buffers++] =
4156                                 sizeof(struct mds_remote_perm);
4157                 else
4158                         repsize[reply_buffers++] =
4159                                 xattr_acl_size(LL_ACL_MAX_ENTRIES);
4160
4161                 repsize[reply_buffers++] = sizeof(int);
4162                 repsize[reply_buffers++] = sizeof(struct crypto_key);
4163
4164                 /* XXX: if new buffer is to be added, capability reply
4165                  *      buffer should always been reserved. */
4166                 if (it->opc & IT_OPEN)
4167                         repsize[reply_buffers++] = sizeof(struct lustre_capa);
4168         }
4169
4170         rc = lustre_pack_reply(req, reply_buffers, repsize, NULL);
4171
4172         RETURN(rc);
4173 }
4174
4175 static int mds_intent_policy(struct ldlm_namespace *ns,
4176                              struct ldlm_lock **lockp, void *req_cookie,
4177                              ldlm_mode_t mode, int flags, void *data)
4178 {
4179         struct ptlrpc_request *req = req_cookie;
4180         struct ldlm_lock *lock = *lockp;
4181         struct ldlm_intent *it;
4182         struct ldlm_reply *rep;
4183         struct lustre_handle lockh[2] = {{0}, {0}};
4184         struct ldlm_lock *new_lock = NULL;
4185         int getattr_part = MDS_INODELOCK_UPDATE;
4186         int rc;
4187
4188         int offset = MDS_REQ_INTENT_REC_OFF;
4189         ENTRY;
4190
4191         LASSERT(req != NULL);
4192         MD_COUNTER_INCREMENT(req->rq_export->exp_obd, intent_lock);
4193
4194         if (req->rq_reqmsg->bufcount <= MDS_REQ_INTENT_IT_OFF) {
4195                 /* No intent was provided */
4196                 int size = sizeof(struct ldlm_reply);
4197                 rc = lustre_pack_reply(req, 1, &size, NULL);
4198                 LASSERT(rc == 0);
4199                 RETURN(0);
4200         }
4201
4202         it = lustre_swab_reqbuf(req, MDS_REQ_INTENT_IT_OFF, sizeof(*it),
4203                                 lustre_swab_ldlm_intent);
4204         if (it == NULL) {
4205                 CERROR("Intent missing\n");
4206                 RETURN(req->rq_status = -EFAULT);
4207         }
4208
4209         LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
4210
4211         rc = mds_intent_prepare_reply_buffers(req, it);
4212
4213         if (rc)
4214                 RETURN(req->rq_status = rc);
4215
4216         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
4217         LASSERT(rep != NULL);
4218
4219         intent_set_disposition(rep, DISP_IT_EXECD);
4220
4221         /* execute policy */
4222         switch ((long)it->opc) {
4223         case IT_OPEN:
4224         case IT_CREAT|IT_OPEN:
4225                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4226                                             lock, NULL, lockh);
4227                 /* XXX swab here to assert that an mds_open reint
4228                  * packet is following */
4229                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4230                                             lock, NULL, lockh);
4231                 rep->lock_policy_res2 = mds_reint(req, offset, lockh);
4232
4233                 if (rep->lock_policy_res2) {
4234                         /*
4235                          * mds_open() returns ENOLCK where it should return
4236                          * zero, but it has no lock to return.
4237                          */
4238                         if (rep->lock_policy_res2 == ENOLCK)
4239                                 rep->lock_policy_res2 = 0;
4240
4241                         RETURN(ELDLM_LOCK_ABORTED);
4242                 }
4243
4244                 /*
4245                  * IT_OPEN may return lock on cross-node dentry that we want to
4246                  * hold during attr retrival -bzzz
4247                  */
4248                 if (lockh[0].cookie == 0)
4249                         RETURN(ELDLM_LOCK_ABORTED);
4250
4251                 break;
4252         case IT_LOOKUP:
4253                 getattr_part = MDS_INODELOCK_LOOKUP;
4254         case IT_CHDIR:
4255         case IT_GETATTR:
4256                 getattr_part |= MDS_INODELOCK_LOOKUP;
4257         case IT_READDIR:
4258                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4259                                             lock, &new_lock, lockh);
4260                 rep->lock_policy_res2 = mds_getattr_lock(req, offset, lockh,
4261                                                          getattr_part);
4262                 /* FIXME: LDLM can set req->rq_status. MDS sets
4263                    policy_res{1,2} with disposition and status.
4264                    - replay: returns 0 & req->status is old status
4265                    - otherwise: returns req->status */
4266                 if (intent_disposition(rep, DISP_LOOKUP_NEG))
4267                         rep->lock_policy_res2 = 0;
4268                 if (!intent_disposition(rep, DISP_LOOKUP_POS) ||
4269                     rep->lock_policy_res2)
4270                         RETURN(ELDLM_LOCK_ABORTED);
4271                 if (req->rq_status != 0) {
4272                         LBUG();
4273                         rep->lock_policy_res2 = req->rq_status;
4274                         RETURN(ELDLM_LOCK_ABORTED);
4275                 }
4276                 break;
4277         case IT_UNLINK:
4278                 rc = mds_lock_and_check_slave(offset, req, lockh);
4279                 if ((rep->lock_policy_res2 = rc)) {
4280                         if (rc == ENOLCK)
4281                                 rep->lock_policy_res2 = 0;
4282                         RETURN(ELDLM_LOCK_ABORTED);
4283                 }
4284                 break;
4285         default:
4286                 CERROR("Unhandled intent "LPD64"\n", it->opc);
4287                 LBUG();
4288         }
4289
4290         /* By this point, whatever function we called above must have either
4291          * filled in 'lockh', been an intent replay, or returned an error.  We
4292          * want to allow replayed RPCs to not get a lock, since we would just
4293          * drop it below anyways because lock replay is done separately by the
4294          * client afterwards.  For regular RPCs we want to give the new lock to
4295          * the client instead of whatever lock it was about to get. */
4296         if (new_lock == NULL)
4297                 new_lock = ldlm_handle2lock(&lockh[0]);
4298         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY))
4299                 RETURN(0);
4300
4301         LASSERTF(new_lock != NULL, "op "LPX64" lockh "LPX64"\n",
4302                  it->opc, lockh[0].cookie);
4303
4304         /* If we've already given this lock to a client once, then we should
4305          * have no readers or writers.  Otherwise, we should have one reader
4306          * _or_ writer ref (which will be zeroed below) before returning the
4307          * lock to a client. */
4308         if (new_lock->l_export == req->rq_export) {
4309                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4310         } else {
4311                 LASSERT(new_lock->l_export == NULL);
4312                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4313         }
4314
4315         *lockp = new_lock;
4316
4317         if (new_lock->l_export == req->rq_export) {
4318                 /* Already gave this to the client, which means that we
4319                  * reconstructed a reply. */
4320                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4321                         MSG_RESENT);
4322                 RETURN(ELDLM_LOCK_REPLACED);
4323         }
4324
4325         /* Fixup the lock to be given to the client */
4326         lock_res_and_lock(new_lock);
4327         new_lock->l_readers = 0;
4328         new_lock->l_writers = 0;
4329
4330         new_lock->l_export = class_export_get(req->rq_export);
4331
4332         spin_lock(&new_lock->l_export->exp_ldlm_data.led_lock);
4333         list_add(&new_lock->l_export_chain,
4334                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
4335         spin_unlock(&new_lock->l_export->exp_ldlm_data.led_lock);
4336
4337         new_lock->l_blocking_ast = lock->l_blocking_ast;
4338         new_lock->l_completion_ast = lock->l_completion_ast;
4339
4340         memcpy(&new_lock->l_remote_handle, &lock->l_remote_handle,
4341                sizeof(lock->l_remote_handle));
4342
4343         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4344
4345         unlock_res_and_lock(new_lock);
4346         LDLM_LOCK_PUT(new_lock);
4347
4348         RETURN(ELDLM_LOCK_REPLACED);
4349 }
4350
4351 int mds_attach(struct obd_device *dev, obd_count len, void *data)
4352 {
4353         struct lprocfs_static_vars lvars;
4354         int rc = 0;
4355         struct mds_obd *mds = &dev->u.mds;
4356
4357         spin_lock_init(&mds->mds_denylist_lock);
4358         INIT_LIST_HEAD(&mds->mds_denylist);
4359
4360         lprocfs_init_multi_vars(0, &lvars);
4361
4362         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
4363         if (rc)
4364                 return rc;
4365
4366         return lprocfs_alloc_md_stats(dev, 0);
4367 }
4368
4369 int mds_detach(struct obd_device *dev)
4370 {
4371         lprocfs_free_md_stats(dev);
4372         return lprocfs_obd_detach(dev);
4373 }
4374
4375 int mdt_attach(struct obd_device *dev, obd_count len, void *data)
4376 {
4377         struct lprocfs_static_vars lvars;
4378
4379         lprocfs_init_multi_vars(1, &lvars);
4380         return lprocfs_obd_attach(dev, lvars.obd_vars);
4381 }
4382
4383 int mdt_detach(struct obd_device *dev)
4384 {
4385         return lprocfs_obd_detach(dev);
4386 }
4387
4388 static int mdt_setup(struct obd_device *obd, obd_count len, void *buf)
4389 {
4390         struct mds_obd *mds = &obd->u.mds;
4391         int rc = 0;
4392         ENTRY;
4393
4394         mds->mds_service =
4395                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4396                                 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
4397                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4398                                 mds_handle, "mds", obd->obd_proc_entry);
4399
4400         if (!mds->mds_service) {
4401                 CERROR("failed to start service\n");
4402                 RETURN(-ENOMEM);
4403         }
4404
4405         rc = ptlrpc_start_n_threads(obd, mds->mds_service, MDT_NUM_THREADS,
4406                                     "ll_mdt");
4407         if (rc)
4408                 GOTO(err_thread, rc);
4409
4410         mds->mds_setattr_service =
4411                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4412                                 MDS_SETATTR_PORTAL, MDC_REPLY_PORTAL,
4413                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4414                                 mds_handle, "mds_setattr",
4415                                 obd->obd_proc_entry);
4416         if (!mds->mds_setattr_service) {
4417                 CERROR("failed to start getattr service\n");
4418                 GOTO(err_thread, rc = -ENOMEM);
4419         }
4420
4421         rc = ptlrpc_start_n_threads(obd, mds->mds_setattr_service,
4422                                     MDT_NUM_THREADS, "ll_mdt_attr");
4423         if (rc)
4424                 GOTO(err_thread2, rc);
4425
4426         mds->mds_readpage_service =
4427                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4428                                 MDS_READPAGE_PORTAL, MDC_REPLY_PORTAL,
4429                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4430                                 mds_handle, "mds_readpage",
4431                                 obd->obd_proc_entry);
4432         if (!mds->mds_readpage_service) {
4433                 CERROR("failed to start readpage service\n");
4434                 GOTO(err_thread2, rc = -ENOMEM);
4435         }
4436
4437         rc = ptlrpc_start_n_threads(obd, mds->mds_readpage_service,
4438                                     MDT_NUM_THREADS, "ll_mdt_rdpg");
4439
4440         if (rc)
4441                 GOTO(err_thread3, rc);
4442
4443         mds->mds_close_service =
4444                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4445                                 MDS_CLOSE_PORTAL, MDC_REPLY_PORTAL,
4446                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4447                                 mds_handle, "mds_close",
4448                                 obd->obd_proc_entry);
4449         if (!mds->mds_close_service) {
4450                 CERROR("failed to start close service\n");
4451                 GOTO(err_thread3, rc = -ENOMEM);
4452         }
4453
4454         rc = ptlrpc_start_n_threads(obd, mds->mds_close_service,
4455                                     MDT_NUM_THREADS, "ll_mdt_clos");
4456
4457         if (rc)
4458                 GOTO(err_thread4, rc);
4459         RETURN(0);
4460
4461 err_thread4:
4462         ptlrpc_unregister_service(mds->mds_close_service);
4463 err_thread3:
4464         ptlrpc_unregister_service(mds->mds_readpage_service);
4465 err_thread2:
4466         ptlrpc_unregister_service(mds->mds_setattr_service);
4467 err_thread:
4468         ptlrpc_unregister_service(mds->mds_service);
4469         return rc;
4470 }
4471
4472 static int mdt_cleanup(struct obd_device *obd, int flags)
4473 {
4474         struct mds_obd *mds = &obd->u.mds;
4475         ENTRY;
4476
4477         ptlrpc_stop_all_threads(mds->mds_close_service);
4478         ptlrpc_unregister_service(mds->mds_close_service);
4479
4480         ptlrpc_stop_all_threads(mds->mds_readpage_service);
4481         ptlrpc_unregister_service(mds->mds_readpage_service);
4482
4483         ptlrpc_stop_all_threads(mds->mds_setattr_service);
4484         ptlrpc_unregister_service(mds->mds_setattr_service);
4485
4486         ptlrpc_stop_all_threads(mds->mds_service);
4487         ptlrpc_unregister_service(mds->mds_service);
4488
4489         RETURN(0);
4490 }
4491
4492 static struct dentry *mds_lvfs_id2dentry(__u64 ino, __u32 gen,
4493                                          __u64 gr, void *data)
4494 {
4495         struct lustre_id id;
4496         struct obd_device *obd = data;
4497
4498         id_ino(&id) = ino;
4499         id_gen(&id) = gen;
4500         return mds_id2dentry(obd, &id, NULL);
4501 }
4502
4503 static int mds_get_info(struct obd_export *exp, __u32 keylen,
4504                         void *key, __u32 *valsize, void *val)
4505 {
4506         struct obd_device *obd;
4507         struct mds_obd *mds;
4508         int rc = 0;
4509         ENTRY;
4510
4511         obd = class_exp2obd(exp);
4512         mds = &obd->u.mds;
4513
4514         if (obd == NULL) {
4515                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
4516                        exp->exp_handle.h_cookie);
4517                 RETURN(-EINVAL);
4518         }
4519
4520         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
4521                 /* get log_context handle. */
4522                 struct llog_ctxt *ctxt;
4523                 unsigned long *llh_handle = val;
4524                 *valsize = sizeof(unsigned long);
4525                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_REINT_ORIG_CTXT);
4526                 if (!ctxt) {
4527                         CERROR("Cannot get REINT llog context\n");
4528                         RETURN(-ENOENT);
4529                 }
4530                 *llh_handle = (unsigned long)ctxt;
4531                 RETURN(0);
4532         }
4533         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
4534                 /* get log_context handle. */
4535                 unsigned long *sb = val;
4536                 *valsize = sizeof(unsigned long);
4537                 *sb = (unsigned long)obd->u.mds.mds_sb;
4538                 RETURN(0);
4539         }
4540
4541         if (keylen >= strlen("mdsize") && memcmp(key, "mdsize", keylen) == 0) {
4542                 __u32 *mdsize = val;
4543                 *valsize = sizeof(*mdsize);
4544                 *mdsize = mds->mds_max_mdsize;
4545                 RETURN(0);
4546         }
4547
4548         if (keylen >= strlen("mdsnum") && strcmp(key, "mdsnum") == 0) {
4549                 __u32 *mdsnum = val;
4550                 *valsize = sizeof(*mdsnum);
4551                 *mdsnum = mds->mds_num;
4552                 RETURN(0);
4553         }
4554
4555         if (keylen >= strlen("rootid") && strcmp(key, "rootid") == 0) {
4556                 struct lustre_id *rootid = val;
4557                 *valsize = sizeof(*rootid);
4558                 *rootid = mds->mds_rootid;
4559                 RETURN(0);
4560         }
4561
4562         if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
4563                 struct lov_desc *desc = val;
4564                 *valsize = sizeof(*desc);
4565                 *desc = mds->mds_dt_desc;
4566                 RETURN(0);
4567         }
4568
4569         if (keylen >= strlen("getext") && strcmp(key, "getext") == 0) {
4570                 struct fid_extent *ext = val;
4571                 *valsize = sizeof(*ext);
4572
4573                 spin_lock(&mds->mds_fidext_lock);
4574                 ext->fe_width = MDS_FIDEXT_SIZE;
4575                 ext->fe_start = mds->mds_fidext_thumb + 1;
4576                 mds->mds_fidext_thumb += MDS_FIDEXT_SIZE;
4577                 spin_unlock(&mds->mds_fidext_lock);
4578
4579                 RETURN(0);
4580         }
4581
4582         rc = fsfilt_get_info(obd, mds->mds_sb, NULL, keylen, key, valsize, val);
4583         if (rc)
4584                 CDEBUG(D_IOCTL, "invalid key\n");
4585
4586         RETURN(rc);
4587 }
4588
4589 struct lvfs_callback_ops mds_lvfs_ops = {
4590         l_id2dentry:     mds_lvfs_id2dentry,
4591 };
4592
4593 int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
4594                 int objcount, struct obd_ioobj *obj,
4595                 int niocount, struct niobuf_remote *nb,
4596                 struct niobuf_local *res,
4597                 struct obd_trans_info *oti, struct lustre_capa *capa);
4598
4599 int mds_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
4600                  int objcount, struct obd_ioobj *obj, int niocount,
4601                  struct niobuf_local *res, struct obd_trans_info *oti,
4602                  int rc);
4603
4604 /* use obd ops to offer management infrastructure */
4605 static struct obd_ops mds_obd_ops = {
4606         .o_owner           = THIS_MODULE,
4607         .o_attach          = mds_attach,
4608         .o_detach          = mds_detach,
4609         .o_connect         = mds_connect,
4610         .o_connect_post    = mds_connect_post,
4611         .o_init_export     = mds_init_export,
4612         .o_destroy_export  = mds_destroy_export,
4613         .o_disconnect      = mds_disconnect,
4614         .o_setup           = mds_setup,
4615         .o_precleanup      = mds_precleanup,
4616         .o_cleanup         = mds_cleanup,
4617         .o_process_config  = mds_process_config,
4618         .o_postrecov       = mds_postrecov,
4619         .o_statfs          = mds_obd_statfs,
4620         .o_iocontrol       = mds_iocontrol,
4621         .o_create          = mds_obd_create,
4622         .o_destroy         = mds_obd_destroy,
4623         .o_llog_init       = mds_llog_init,
4624         .o_llog_finish     = mds_llog_finish,
4625         .o_notify          = mds_notify,
4626         .o_get_info        = mds_get_info,
4627         .o_set_info        = mds_set_info,
4628         .o_preprw          = mds_preprw,
4629         .o_commitrw        = mds_commitrw,
4630 };
4631
4632 static struct obd_ops mdt_obd_ops = {
4633         .o_owner           = THIS_MODULE,
4634         .o_attach          = mdt_attach,
4635         .o_detach          = mdt_detach,
4636         .o_setup           = mdt_setup,
4637         .o_cleanup         = mdt_cleanup,
4638 };
4639
4640 static int __init mds_init(void)
4641 {
4642         struct lprocfs_static_vars lvars;
4643         int rc = 0;
4644         ENTRY;
4645
4646         mds_init_lsd_cache();
4647         mds_init_rmtacl_upcall_cache();
4648
4649         lprocfs_init_multi_vars(0, &lvars);
4650         class_register_type(&mds_obd_ops, NULL, lvars.module_vars,
4651                             OBD_MDS_DEVICENAME);
4652         lprocfs_init_multi_vars(1, &lvars);
4653         class_register_type(&mdt_obd_ops, NULL, lvars.module_vars,
4654                             OBD_MDT_DEVICENAME);
4655
4656         rc = mds_capa_key_start_thread();
4657         if (rc) {
4658                 class_unregister_type(OBD_MDT_DEVICENAME);
4659                 class_unregister_type(OBD_MDS_DEVICENAME);
4660                 mds_cleanup_lsd_cache();
4661         }
4662
4663         mds_eck_timer.function = mds_capa_key_timer_callback;
4664         mds_eck_timer.data = 0;
4665         init_timer(&mds_eck_timer);
4666
4667         RETURN(rc);
4668 }
4669
4670 static void /*__exit*/ mds_exit(void)
4671 {
4672         mds_capa_key_stop_thread();
4673         mds_cleanup_rmtacl_upcall_cache();
4674         mds_cleanup_lsd_cache();
4675
4676         class_unregister_type(OBD_MDS_DEVICENAME);
4677         class_unregister_type(OBD_MDT_DEVICENAME);
4678 }
4679
4680 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4681 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
4682 MODULE_LICENSE("GPL");
4683
4684 module_init(mds_init);
4685 module_exit(mds_exit);