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         } else if (S_ISLNK(inode->i_mode) &&
1237                    (reqbody->valid & OBD_MD_LINKNAME) != 0) {
1238                 rc = mds_pack_link(dentry, req, body, reply_off);
1239         } else if (reqbody->valid & OBD_MD_FLXATTR) {
1240                 rc = mds_pack_xattr(dentry, req, body, req_off, reply_off);
1241         } else if (reqbody->valid & OBD_MD_FLXATTRLIST) {
1242                 rc = mds_pack_xattr_list(dentry, req, body, reply_off);
1243         }
1244
1245         offset = reply_off + ((reqbody->valid & OBD_MD_FLEASIZE) ? 2 : 1);
1246         if (reqbody->valid & OBD_MD_FLACL) {
1247                 rc = mds_pack_acl(req, offset, body, inode);
1248                 offset += 2;
1249         }
1250
1251         if (reqbody->valid & OBD_MD_FLKEY) {
1252                 rc = mds_pack_gskey(obd, req->rq_repmsg, &offset,
1253                                     body, inode);
1254         }
1255
1256         mds_pack_audit(obd, inode, body);
1257
1258         if (reqbody->valid & OBD_MD_CAPA) {
1259                 struct lustre_capa *req_capa;
1260
1261                 LASSERT(!(reqbody->valid & ~OBD_MD_CAPA));
1262                 LASSERT(S_ISREG(inode->i_mode));
1263
1264                 req_capa = lustre_swab_reqbuf(req, req_off + 1,
1265                                               sizeof(*req_capa),
1266                                               lustre_swab_lustre_capa);
1267                 if (req_capa == NULL) {
1268                         CERROR("Can't unpack capa\n");
1269                         RETURN(-EFAULT);
1270                 }
1271
1272                 offset = reply_off + 1;
1273                 rc = mds_pack_capa(obd, med, reqbody, req_capa, req,
1274                                    &offset, body);
1275         }
1276
1277         if (rc == 0)
1278                 mds_body_do_reverse_map(med, body);
1279
1280         RETURN(rc);
1281 }
1282
1283 static int mds_getattr_pack_msg_cf(struct ptlrpc_request *req,
1284                                    struct dentry *dentry,
1285                                    int offset)
1286 {
1287         int rc = 0, size[1] = {sizeof(struct mds_body)};
1288         ENTRY;
1289
1290         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
1291                 CERROR("failed MDS_GETATTR_PACK test\n");
1292                 req->rq_status = -ENOMEM;
1293                 RETURN(-ENOMEM);
1294         }
1295
1296         rc = lustre_pack_reply(req, 1, size, NULL);
1297         if (rc) {
1298                 CERROR("lustre_pack_reply failed: rc %d\n", rc);
1299                 GOTO(out, req->rq_status = rc);
1300         }
1301
1302         EXIT;
1303 out:
1304         return rc;
1305 }
1306
1307 static int mds_getattr_pack_msg(struct ptlrpc_request *req, struct dentry *de,
1308                                 int offset)
1309 {
1310         struct inode *inode = de->d_inode;
1311         struct mds_obd *mds = mds_req2mds(req);
1312         struct mds_body *body;
1313         int rc = 0, size[4] = {sizeof(*body)}, bufcount = 1;
1314         ENTRY;
1315
1316         body = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*body));
1317         LASSERT(body != NULL);                 /* checked by caller */
1318         LASSERT_REQSWABBED(req, offset);       /* swabbed by caller */
1319
1320         if ((S_ISREG(inode->i_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
1321             (S_ISDIR(inode->i_mode) && (body->valid & OBD_MD_FLDIREA))) {
1322                 int rc;
1323
1324                 down(&inode->i_sem);
1325                 rc = fsfilt_get_md(req->rq_export->exp_obd, inode, NULL, 0,
1326                                    ((body->valid & OBD_MD_MEA) ? EA_MEA : EA_LOV));
1327                 up(&inode->i_sem);
1328                 if (rc < 0) {
1329                         if (rc != -ENODATA && rc != -EOPNOTSUPP)
1330                                 CERROR("error getting inode %lu MD: rc = %d\n",
1331                                        inode->i_ino, rc);
1332                         size[bufcount] = 0;
1333                 } else if (rc > mds->mds_max_mdsize) {
1334                         size[bufcount] = 0;
1335                         CERROR("MD size %d larger than maximum possible %u\n",
1336                                rc, mds->mds_max_mdsize);
1337                 } else {
1338                         size[bufcount] = rc;
1339                 }
1340                 bufcount++;
1341         } else if (S_ISLNK(inode->i_mode) && (body->valid & OBD_MD_LINKNAME)) {
1342                 if (inode->i_size + 1 != body->eadatasize)
1343                         CERROR("symlink size: %Lu, reply space: %d\n",
1344                                inode->i_size + 1, body->eadatasize);
1345                 size[bufcount] = min_t(int, inode->i_size+1, body->eadatasize);
1346                 bufcount++;
1347                 CDEBUG(D_INODE, "symlink size: %Lu, reply space: %d\n",
1348                        inode->i_size + 1, body->eadatasize);
1349         } else if ((body->valid & OBD_MD_FLXATTR)) {
1350                 char *ea_name = lustre_msg_string(req->rq_reqmsg,
1351                                                   offset + 1, 0);
1352                 rc = -EOPNOTSUPP;
1353
1354                 if (!strcmp(ea_name, XATTR_NAME_LUSTRE_ACL)) {
1355                         size[bufcount] = LUSTRE_ACL_SIZE_MAX;
1356                 } else {
1357                         if (inode->i_op && inode->i_op->getxattr)
1358                                 rc = inode->i_op->getxattr(de, ea_name,
1359                                                            NULL, 0);
1360
1361                         if (rc < 0) {
1362                                 if (rc != -ENODATA && rc != -EOPNOTSUPP)
1363                                         CERROR("error get inode %lu EA: %d\n",
1364                                                inode->i_ino, rc);
1365                                 size[bufcount] = 0;
1366                         } else {
1367                                 size[bufcount] = min_t(int,
1368                                                        body->eadatasize, rc);
1369                         }
1370                 }
1371                 bufcount++;
1372         } else if (body->valid & OBD_MD_FLXATTRLIST) {
1373                 rc = -EOPNOTSUPP;
1374                 if (inode->i_op && inode->i_op->getxattr)
1375                         rc = inode->i_op->listxattr(de, NULL, 0);
1376
1377                 if (rc < 0) {
1378                         if (rc != -ENODATA && rc != -EOPNOTSUPP)
1379                                 CERROR("error getting inode %lu EA: rc = %d\n",
1380                                        inode->i_ino, rc);
1381                         size[bufcount] = 0;
1382                 } else {
1383                         size[bufcount] = min_t(int, body->eadatasize, rc);
1384                 }
1385                 bufcount++;
1386         }
1387
1388         /* may co-exist with OBD_MD_FLEASIZE */
1389         if (body->valid & OBD_MD_FLACL) {
1390                 if (req->rq_export->exp_mds_data.med_remote) {
1391                         size[bufcount++] = sizeof(int);
1392                         size[bufcount++] = sizeof(struct mds_remote_perm);
1393                 } else {
1394                         size[bufcount++] = sizeof(int);
1395                         size[bufcount++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
1396                 }
1397         }
1398
1399         if (body->valid & OBD_MD_FLKEY) {
1400                 size[bufcount++] = sizeof(int);
1401                 size[bufcount++] = sizeof(struct crypto_key);
1402         }
1403
1404         if (body->valid & OBD_MD_CAPA)
1405                 size[bufcount++] = sizeof(struct lustre_capa);
1406
1407         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
1408                 CERROR("failed MDS_GETATTR_PACK test\n");
1409                 req->rq_status = -ENOMEM;
1410                 GOTO(out, rc = -ENOMEM);
1411         }
1412
1413         rc = lustre_pack_reply(req, bufcount, size, NULL);
1414         if (rc) {
1415                 CERROR("out of memory\n");
1416                 GOTO(out, req->rq_status = rc);
1417         }
1418
1419         EXIT;
1420  out:
1421         return rc;
1422 }
1423
1424 int mds_check_mds_num(struct obd_device *obd, struct inode *inode,
1425                       char *name, int namelen)
1426 {
1427         struct mea *mea = NULL;
1428         int mea_size, rc = 0;
1429         ENTRY;
1430
1431         rc = mds_md_get_attr(obd, inode, &mea, &mea_size);
1432         if (rc)
1433                 RETURN(rc);
1434         if (mea != NULL && mea->mea_count) {
1435                 /*
1436                  * dir is already splitted, check if requested filename should
1437                  * live at this MDS or at another one.
1438                  */
1439                 int i = mea_name2idx(mea, name, namelen - 1);
1440                 if (mea->mea_master != id_group(&mea->mea_ids[i])) {
1441                         CDEBUG(D_OTHER,
1442                                "inapropriate MDS(%d) for %s. should be "
1443                                "%lu(%d)\n", mea->mea_master, name,
1444                                (unsigned long)id_group(&mea->mea_ids[i]), i);
1445                         rc = -ERESTART;
1446                 }
1447         }
1448
1449         if (mea)
1450                 OBD_FREE(mea, mea_size);
1451         RETURN(rc);
1452 }
1453
1454 int mds_getattr_size(struct obd_device *obd, struct dentry *dentry,
1455                      struct ptlrpc_request *req, struct mds_body *body)
1456 {
1457         struct inode *inode = dentry->d_inode;
1458         ENTRY;
1459
1460         LASSERT(body != NULL);
1461
1462         if (dentry->d_inode == NULL || !S_ISREG(inode->i_mode))
1463                 RETURN(0);
1464
1465         /* XXX: quite a ugly hack, need to check old code
1466          * drop FLSIZE/FLBLOCKS prior any checking to */
1467         body->valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1468
1469         if (obd->obd_recovering) {
1470                 CDEBUG(D_INODE, "size for "DLID4" is unknown yet (recovering)\n",
1471                        OLID4(&body->id1));
1472                 RETURN(0);
1473         }
1474
1475         if (atomic_read(&inode->i_writecount)) {
1476                 /* some one has opened the file for write.
1477                  * mds doesn't know actual size */
1478                 CDEBUG(D_INODE, "MDS doesn't know actual size for "DLID4"\n",
1479                        OLID4(&body->id1));
1480                 RETURN(0);
1481         }
1482         CDEBUG(D_INODE, "MDS returns "LPD64"/"LPD64" for"DLID4"\n",
1483                body->size, body->blocks, OLID4(&body->id1));
1484         body->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
1485         RETURN(0);
1486 }
1487
1488 static int mds_getattr_lock(struct ptlrpc_request *req, int offset,
1489                             struct lustre_handle *child_lockh, int child_part)
1490 {
1491         struct obd_device *obd = req->rq_export->exp_obd;
1492         struct mds_obd *mds = &obd->u.mds;
1493         struct ldlm_reply *rep = NULL;
1494         struct lvfs_run_ctxt saved;
1495         struct mds_req_sec_desc *rsd;
1496         struct mds_body *body;
1497         struct dentry *dparent = NULL, *dchild = NULL;
1498         struct lvfs_ucred uc = {NULL, NULL,};
1499         struct lustre_handle parent_lockh[2] = {{0}, {0}};
1500         unsigned int namesize = 0;
1501         int rc = 0, cleanup_phase = 0, resent_req = 0, update_mode, reply_offset;
1502         char *name = NULL;
1503         ENTRY;
1504
1505         LASSERT(!strcmp(obd->obd_type->typ_name, OBD_MDS_DEVICENAME));
1506         MD_COUNTER_INCREMENT(obd, getattr_lock);
1507
1508         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1509         if (!rsd) {
1510                 CERROR("Can't unpack security desc\n");
1511                 RETURN(-EFAULT);
1512         }
1513
1514         /* swab now, before anyone looks inside the request. */
1515         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1516                                   lustre_swab_mds_body);
1517         if (body == NULL) {
1518                 CERROR("Can't swab mds_body\n");
1519                 GOTO(cleanup, rc = -EFAULT);
1520         }
1521
1522         LASSERT_REQSWAB(req, offset + 1);
1523         name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
1524         if (name == NULL) {
1525                 CERROR("Can't unpack name\n");
1526                 GOTO(cleanup, rc = -EFAULT);
1527         }
1528         namesize = req->rq_reqmsg->buflens[offset + 1];
1529
1530         /* namesize less than 2 means we have empty name, probably came from
1531            revalidate by cfid, so no point in having name to be set */
1532         if (namesize <= 1)
1533                 name = NULL;
1534
1535         LASSERT (offset == 1 || offset == 3);
1536         if (offset == 3) {
1537                 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
1538                 reply_offset = 1;
1539         } else {
1540                 reply_offset = 0;
1541         }
1542
1543         rc = mds_init_ucred(&uc, req, rsd);
1544         if (rc) {
1545                 if (child_lockh->cookie == 0)
1546                         mds_audit_auth(req, &uc, AUDIT_STAT, &body->id1,
1547                                        name, namesize - 1);
1548                 GOTO(cleanup, rc);
1549         }
1550
1551         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1552         cleanup_phase = 1; /* kernel context */
1553         intent_set_disposition(rep, DISP_LOOKUP_EXECD);
1554
1555         LASSERT(namesize > 0);
1556         if (child_lockh->cookie != 0) {
1557                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1558                 resent_req = 1;
1559         }
1560 #if HAVE_LOOKUP_RAW
1561         if (body->valid == OBD_MD_FLID) {
1562                 struct mds_body *mds_reply;
1563                 int size = sizeof(*mds_reply);
1564                 struct inode *dir;
1565                 ino_t inum;
1566
1567                 dparent = mds_id2dentry(obd, &body->id1, NULL);
1568                 if (IS_ERR(dparent)) {
1569                         rc = PTR_ERR(dparent);
1570                         GOTO(cleanup, rc);
1571                 }
1572                 /*
1573                  * the user requested ONLY the inode number, so do a raw lookup.
1574                  */
1575                 rc = lustre_pack_reply(req, 1, &size, NULL);
1576                 if (rc) {
1577                         CERROR("out of memory\n");
1578                         l_dput(dparent);
1579                         GOTO(cleanup, rc);
1580                 }
1581                 dir  = dparent->d_inode;
1582                 LASSERT(dir->i_op->lookup_raw != NULL);
1583                 rc = dir->i_op->lookup_raw(dir, name, namesize - 1, &inum);
1584                 l_dput(dparent);
1585                 mds_reply = lustre_msg_buf(req->rq_repmsg, 0,
1586                                            sizeof(*mds_reply));
1587
1588                 id_ino(&mds_reply->id1) = inum;
1589                 mds_reply->valid = OBD_MD_FLID;
1590                 GOTO(cleanup, rc);
1591         }
1592 #endif
1593         if (resent_req == 0) {
1594                 LASSERT(id_fid(&body->id1) != 0);
1595                 if (name) {
1596                         rc = mds_get_parent_child_locked(obd, mds, &body->id1,
1597                                                          parent_lockh, &dparent,
1598                                                          LCK_PR,
1599                                                          MDS_INODELOCK_UPDATE,
1600                                                          &update_mode,
1601                                                          name, namesize,
1602                                                          child_lockh, &dchild,
1603                                                          LCK_PR, child_part);
1604                         if (rc)
1605                                 GOTO(cleanup, rc);
1606
1607                         cleanup_phase = 2; /* dchild, dparent, locks */
1608
1609                         /*
1610                          * let's make sure this name should leave on this mds
1611                          * node.
1612                          */
1613                         rc = mds_check_mds_num(obd, dparent->d_inode, name, namesize);
1614                         if (rc)
1615                                 GOTO(cleanup, rc);
1616                 } else {
1617                         /* we have no dentry here, drop LOOKUP bit */
1618                         /* FIXME: we need MDS_INODELOCK_LOOKUP or not. */
1619                         child_part &= ~MDS_INODELOCK_LOOKUP;
1620                         CDEBUG(D_OTHER, "%s: retrieve attrs for "DLID4"\n",
1621                                obd->obd_name, OLID4(&body->id1));
1622
1623                         dchild = mds_id2locked_dentry(obd, &body->id1, NULL,
1624                                                       LCK_PR, parent_lockh,
1625                                                       &update_mode, NULL, 0,
1626                                                       MDS_INODELOCK_UPDATE);
1627                         if (IS_ERR(dchild)) {
1628                                 CERROR("can't find inode with id "DLID4", err = %d\n",
1629                                        OLID4(&body->id1), (int)PTR_ERR(dchild));
1630                                 rc = PTR_ERR(dchild);
1631                                 dchild = NULL;
1632                                 GOTO(cleanup, rc);
1633                         }
1634                         memcpy(child_lockh, parent_lockh, sizeof(parent_lockh[0]));
1635                 }
1636         } else {
1637                 struct ldlm_lock *granted_lock;
1638
1639                 DEBUG_REQ(D_DLMTRACE, req, "resent, not enqueuing new locks");
1640                 granted_lock = ldlm_handle2lock(child_lockh);
1641
1642                 LASSERTF(granted_lock != NULL, LPU64"/%lu lockh "LPX64"\n",
1643                          id_fid(&body->id1), (unsigned long)id_group(&body->id1),
1644                          child_lockh->cookie);
1645
1646                 if (name) {
1647                         /* usual named request */
1648                         dparent = mds_id2dentry(obd, &body->id1, NULL);
1649                         LASSERT(!IS_ERR(dparent));
1650                         dchild = ll_lookup_one_len(name, dparent, namesize - 1);
1651                         LASSERT(!IS_ERR(dchild));
1652                 } else {
1653                         /* client wants to get attr. by id */
1654                         dchild = mds_id2dentry(obd, &body->id1, NULL);
1655                         LASSERT(!IS_ERR(dchild));
1656                 }
1657                 LDLM_LOCK_PUT(granted_lock);
1658         }
1659
1660         cleanup_phase = 2; /* dchild, dparent, locks */
1661
1662         if (!DENTRY_VALID(dchild)) {
1663                 intent_set_disposition(rep, DISP_LOOKUP_NEG);
1664                 /*
1665                  * in the intent case, the policy clears this error: the
1666                  * disposition is enough.
1667                  */
1668                 rc = -ENOENT;
1669                 GOTO(cleanup, rc);
1670         } else {
1671                 intent_set_disposition(rep, DISP_LOOKUP_POS);
1672         }
1673
1674         if (req->rq_repmsg == NULL) {
1675                 if (dchild->d_flags & DCACHE_CROSS_REF)
1676                         rc = mds_getattr_pack_msg_cf(req, dchild, offset);
1677                 else
1678                         rc = mds_getattr_pack_msg(req, dchild, offset);
1679                 if (rc != 0) {
1680                         CERROR ("mds_getattr_pack_msg: %d\n", rc);
1681                         GOTO (cleanup, rc);
1682                 }
1683         }
1684
1685         rc = mds_getattr_internal(obd, dchild, req, offset, body,
1686                                   reply_offset, rsd);
1687         if (rc)
1688                 GOTO(cleanup, rc); /* returns the lock to the client */
1689
1690         /* probably MDS knows actual size? */
1691         body = lustre_msg_buf(req->rq_repmsg, reply_offset, sizeof(*body));
1692         LASSERT(body != NULL);
1693         mds_getattr_size(obd, dchild, req, body);
1694
1695         GOTO(cleanup, rc);
1696
1697  cleanup:
1698         switch (cleanup_phase) {
1699         case 2:
1700                 if (resent_req == 0) {
1701                         mds_audit(req, dchild, name, namesize - 1,
1702                                   AUDIT_STAT, rc);
1703                         if (rc && DENTRY_VALID(dchild))
1704                                 ldlm_lock_decref(child_lockh, LCK_PR);
1705                         if (name)
1706                                 ldlm_lock_decref(parent_lockh, LCK_PR);
1707 #ifdef S_PDIROPS
1708                         if (parent_lockh[1].cookie != 0)
1709                                 ldlm_lock_decref(parent_lockh + 1, update_mode);
1710 #endif
1711                 }
1712                 if (dparent)
1713                         l_dput(dparent);
1714
1715                 l_dput(dchild);
1716         case 1:
1717                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1718         default:
1719                 mds_exit_ucred(&uc);
1720         }
1721         return rc;
1722 }
1723
1724 static int mds_getattr(struct ptlrpc_request *req, int offset)
1725 {
1726         struct obd_device *obd = req->rq_export->exp_obd;
1727         struct lvfs_run_ctxt saved;
1728         struct dentry *de;
1729         struct mds_req_sec_desc *rsd;
1730         struct mds_body *body;
1731         struct lvfs_ucred uc = {NULL, NULL,};
1732         int rc = 0;
1733         ENTRY;
1734
1735         MD_COUNTER_INCREMENT(obd, getattr);
1736
1737         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1738         if (!rsd) {
1739                 CERROR("Can't unpack security desc\n");
1740                 RETURN(-EFAULT);
1741         }
1742
1743         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1744                                   lustre_swab_mds_body);
1745         if (body == NULL) {
1746                 CERROR ("Can't unpack body\n");
1747                 RETURN (-EFAULT);
1748         }
1749
1750         rc = mds_init_ucred(&uc, req, rsd);
1751         if (rc) {
1752                 mds_exit_ucred(&uc);
1753                 RETURN(rc);
1754         }
1755
1756         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1757         de = mds_id2dentry(obd, &body->id1, NULL);
1758         if (IS_ERR(de)) {
1759                 rc = req->rq_status = PTR_ERR(de);
1760                 GOTO(out_pop, rc);
1761         }
1762
1763         rc = mds_getattr_pack_msg(req, de, offset);
1764         if (rc != 0) {
1765                 CERROR("mds_getattr_pack_msg: %d\n", rc);
1766                 GOTO(out_pop, rc);
1767         }
1768
1769         req->rq_status = mds_getattr_internal(obd, de, req, offset, body,
1770                                               0, rsd);
1771         l_dput(de);
1772
1773         EXIT;
1774 out_pop:
1775         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1776         mds_exit_ucred(&uc);
1777         return rc;
1778 }
1779
1780 static int mds_access_check(struct ptlrpc_request *req, int offset)
1781 {
1782         struct obd_device *obd = req->rq_export->exp_obd;
1783         struct lvfs_run_ctxt saved;
1784         struct dentry *de;
1785         struct mds_req_sec_desc *rsd;
1786         struct mds_body *body;
1787         struct lvfs_ucred uc;
1788         int rep_size[2] = {sizeof(*body),
1789                            sizeof(struct mds_remote_perm)};
1790         int rc = 0;
1791         ENTRY;
1792
1793         if (!req->rq_export->exp_mds_data.med_remote) {
1794                 CERROR("from local client "LPU64"\n", req->rq_peer.peer_id.nid);
1795                 RETURN(-EINVAL);
1796         }
1797
1798         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1799         if (!rsd) {
1800                 CERROR("Can't unpack security desc\n");
1801                 RETURN(-EFAULT);
1802         }
1803
1804         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1805                                   lustre_swab_mds_body);
1806         if (body == NULL) {
1807                 CERROR ("Can't unpack body\n");
1808                 RETURN (-EFAULT);
1809         }
1810
1811         MD_COUNTER_INCREMENT(obd, access_check);
1812
1813         rc = mds_init_ucred(&uc, req, rsd);
1814         if (rc) {
1815                 CERROR("init ucred error: %d\n", rc);
1816                 RETURN(rc);
1817         }
1818         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1819
1820         de = mds_id2dentry(obd, &body->id1, NULL);
1821         if (IS_ERR(de)) {
1822                 CERROR("grab ino "LPU64": err %ld\n",
1823                        body->id1.li_stc.u.e3s.l3s_ino, PTR_ERR(de));
1824                 GOTO(out_pop, rc = PTR_ERR(de));
1825         }
1826
1827         rc = lustre_pack_reply(req, 2, rep_size, NULL);
1828         if (rc) {
1829                 CERROR("pack reply error: %d\n", rc);
1830                 GOTO(out_dput, rc = -EINVAL);
1831         }
1832
1833         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1834         LASSERT(body);
1835
1836         rc = mds_pack_remote_perm(req, 1, body, de->d_inode);
1837
1838         EXIT;
1839
1840 out_dput:
1841         l_dput(de);
1842 out_pop:
1843         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1844         mds_exit_ucred(&uc);
1845         return rc;
1846 }
1847
1848 static int mds_obd_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1849                           unsigned long max_age)
1850 {
1851         int rc;
1852         ENTRY;
1853
1854         spin_lock(&obd->obd_osfs_lock);
1855         rc = fsfilt_statfs(obd, obd->u.mds.mds_sb, max_age);
1856         if (rc == 0)
1857                 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
1858         spin_unlock(&obd->obd_osfs_lock);
1859
1860         RETURN(rc);
1861 }
1862
1863 static int mds_statfs(struct ptlrpc_request *req)
1864 {
1865         struct obd_device *obd = req->rq_export->exp_obd;
1866         int rc, size = sizeof(struct obd_statfs);
1867         ENTRY;
1868
1869         /* This will trigger a watchdog timeout */
1870         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
1871                          (MDS_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
1872
1873         rc = lustre_pack_reply(req, 1, &size, NULL);
1874         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
1875                 CERROR("mds: statfs lustre_pack_reply failed: rc = %d\n", rc);
1876                 GOTO(out, rc);
1877         }
1878
1879         OBD_COUNTER_INCREMENT(obd, statfs);
1880
1881         /* We call this so that we can cache a bit - 1 jiffie worth */
1882         rc = mds_obd_statfs(obd, lustre_msg_buf(req->rq_repmsg, 0, size),
1883                             jiffies - HZ);
1884         if (rc) {
1885                 CERROR("mds_obd_statfs failed: rc %d\n", rc);
1886                 GOTO(out, rc);
1887         }
1888
1889         EXIT;
1890 out:
1891         req->rq_status = rc;
1892         return rc;
1893 }
1894
1895 static int mds_sync(struct ptlrpc_request *req, int offset)
1896 {
1897         struct obd_device *obd = req->rq_export->exp_obd;
1898         struct mds_obd *mds = &obd->u.mds;
1899         struct mds_body *body;
1900         int rc, size = sizeof(*body);
1901         ENTRY;
1902
1903         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1904                                   lustre_swab_mds_body);
1905         if (body == NULL)
1906                 GOTO(out, rc = -EPROTO);
1907
1908         rc = lustre_pack_reply(req, 1, &size, NULL);
1909         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) {
1910                 CERROR("fsync lustre_pack_reply failed: rc = %d\n", rc);
1911                 GOTO(out, rc);
1912         }
1913
1914         if (id_ino(&body->id1) == 0) {
1915                 /* an id of zero is taken to mean "sync whole filesystem" */
1916                 rc = fsfilt_sync(obd, mds->mds_sb);
1917                 if (rc)
1918                         GOTO(out, rc);
1919         } else {
1920                 /* just any file to grab fsync method - "file" arg unused */
1921                 struct file *file = mds->mds_rcvd_filp;
1922                 struct mds_body *rep_body;
1923                 struct dentry *de;
1924
1925                 de = mds_id2dentry(obd, &body->id1, NULL);
1926                 if (IS_ERR(de))
1927                         GOTO(out, rc = PTR_ERR(de));
1928
1929                 rc = file->f_op->fsync(NULL, de, 1);
1930                 if (rc)
1931                         GOTO(out, rc);
1932
1933                 rep_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep_body));
1934                 mds_pack_inode2body(obd, rep_body, de->d_inode,
1935                                     (body->valid & OBD_MD_FID) ? 1 : 0);
1936                 l_dput(de);
1937         }
1938
1939         EXIT;
1940 out:
1941         req->rq_status = rc;
1942         return rc;
1943 }
1944
1945 /* mds_readpage does not take a DLM lock on the inode, because the client must
1946  * already have a PR lock.
1947  *
1948  * If we were to take another one here, a deadlock will result, if another
1949  * thread is already waiting for a PW lock. */
1950 static int mds_readpage(struct ptlrpc_request *req, int offset)
1951 {
1952         struct obd_device *obd = req->rq_export->exp_obd;
1953         struct vfsmount *mnt;
1954         struct dentry *de;
1955         struct file *file;
1956         struct mds_req_sec_desc *rsd;
1957         struct mds_body *body, *repbody;
1958         struct lvfs_run_ctxt saved;
1959         int rc, size = sizeof(*repbody);
1960         struct lvfs_ucred uc = {NULL, NULL,};
1961         ENTRY;
1962
1963         rc = lustre_pack_reply(req, 1, &size, NULL);
1964         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
1965                 CERROR("mds: out of memory\n");
1966                 GOTO(out, rc = -ENOMEM);
1967         }
1968
1969         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1970         if (!rsd) {
1971                 CERROR("Can't unpack security desc\n");
1972                 GOTO (out, rc = -EFAULT);
1973         }
1974
1975         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1976                                   lustre_swab_mds_body);
1977         if (body == NULL) {
1978                 CERROR("Can't unpack body\n");
1979                 GOTO (out, rc = -EFAULT);
1980         }
1981
1982         rc = mds_init_ucred(&uc, req, rsd);
1983         if (rc) {
1984                 mds_audit_auth(req, &uc, AUDIT_READDIR, &body->id1,
1985                                NULL, 0);
1986                 GOTO(out, rc);
1987         }
1988
1989         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1990         de = mds_id2dentry(obd, &body->id1, &mnt);
1991         if (IS_ERR(de))
1992                 GOTO(out_pop, rc = PTR_ERR(de));
1993
1994         CDEBUG(D_INODE, "ino %lu\n", de->d_inode->i_ino);
1995
1996         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
1997         /* note: in case of an error, dentry_open puts dentry */
1998         if (IS_ERR(file))
1999                 GOTO(out_pop, rc = PTR_ERR(file));
2000
2001         /* body->size is actually the offset -eeb */
2002         if ((body->size & (de->d_inode->i_blksize - 1)) != 0) {
2003                 CERROR("offset "LPU64" not on a block boundary of %lu\n",
2004                        body->size, de->d_inode->i_blksize);
2005                 GOTO(out_file, rc = -EFAULT);
2006         }
2007
2008         /* body->nlink is actually the #bytes to read -eeb */
2009         if (body->nlink & (de->d_inode->i_blksize - 1)) {
2010                 CERROR("size %u is not multiple of blocksize %lu\n",
2011                        body->nlink, de->d_inode->i_blksize);
2012                 GOTO(out_file, rc = -EFAULT);
2013         }
2014
2015         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*repbody));
2016         repbody->size = file->f_dentry->d_inode->i_size;
2017         repbody->valid = OBD_MD_FLSIZE;
2018
2019         /* to make this asynchronous make sure that the handling function
2020            doesn't send a reply when this function completes. Instead a
2021            callback function would send the reply */
2022         /* body->size is actually the offset -eeb */
2023         rc = mds_sendpage(req, file, body->size, body->nlink);
2024
2025         EXIT;
2026 out_file:
2027         filp_close(file, 0);
2028 out_pop:
2029         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2030 out:
2031         mds_exit_ucred(&uc);
2032         req->rq_status = rc;
2033         return 0;
2034 }
2035
2036 int mds_read_md(struct obd_device *obd, struct lustre_id *id,
2037                 char **data, int *datalen)
2038 {
2039         struct dentry *dentry;
2040         struct mds_obd *mds = &obd->u.mds;
2041         int rc = 0, mea = 0;
2042         char *ea;
2043         ENTRY;
2044
2045         LASSERT(id);
2046         LASSERT(obd);
2047
2048         dentry = mds_id2dentry(obd, id, NULL);
2049         if (IS_ERR(dentry))
2050                 GOTO(out, rc = PTR_ERR(dentry));
2051
2052         if (!dentry->d_inode) {
2053                 CERROR("Can't find object "DLID4".\n",
2054                        OLID4(id));
2055                 GOTO(out_dentry, rc = -EINVAL);
2056         }
2057         if (S_ISDIR(dentry->d_inode->i_mode)) {
2058                 *datalen = obd_packmd(mds->mds_md_exp, NULL, NULL);
2059                 mea = 1;
2060         } else {
2061                 *datalen = obd_packmd(mds->mds_dt_exp, NULL, NULL);
2062                 mea = 0;
2063         }
2064         OBD_ALLOC(ea, *datalen);
2065         if (!ea) {
2066                 *datalen = 0;
2067                 GOTO(out_dentry, rc = PTR_ERR(dentry));
2068         }
2069         *data = ea;
2070         down(&dentry->d_inode->i_sem);
2071         rc = fsfilt_get_md(obd, dentry->d_inode, *data, *datalen,
2072                            (mea ? EA_MEA : EA_LOV));
2073         up(&dentry->d_inode->i_sem);
2074
2075         if (rc < 0)
2076                 CERROR("Error %d reading eadata for ino %lu\n",
2077                         rc, dentry->d_inode->i_ino);
2078 out_dentry:
2079         l_dput(dentry);
2080 out:
2081         RETURN(rc);
2082 }
2083 EXPORT_SYMBOL(mds_read_md);
2084
2085 int mds_reint(struct ptlrpc_request *req, int offset,
2086               struct lustre_handle *lockh)
2087 {
2088         struct mds_update_record *rec;
2089         struct mds_req_sec_desc *rsd;
2090         int rc;
2091         ENTRY;
2092
2093         OBD_ALLOC(rec, sizeof(*rec));
2094         if (rec == NULL)
2095                 RETURN(-ENOMEM);
2096
2097         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
2098         if (!rsd) {
2099                 CERROR("Can't unpack security desc\n");
2100                 GOTO(out, rc = -EFAULT);
2101         }
2102
2103         rc = mds_update_unpack(req, offset, rec);
2104         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
2105                 CERROR("invalid record\n");
2106                 GOTO(out, req->rq_status = -EINVAL);
2107         }
2108
2109         rc = mds_init_ucred(&rec->ur_uc, req, rsd);
2110         if (rc) {
2111                 mds_audit_reint(req, rec);
2112                 GOTO(out, rc);
2113         }
2114
2115         /* rc will be used to interrupt a for loop over multiple records */
2116         rc = mds_reint_rec(rec, offset, req, lockh);
2117
2118 out:
2119         mds_exit_ucred(&rec->ur_uc);
2120         OBD_FREE(rec, sizeof(*rec));
2121         RETURN(rc);
2122 }
2123
2124 static int mds_filter_recovery_request(struct ptlrpc_request *req,
2125                                        struct obd_device *obd, int *process)
2126 {
2127         switch (req->rq_reqmsg->opc) {
2128         case MDS_CONNECT: /* This will never get here, but for completeness. */
2129         case OST_CONNECT: /* This will never get here, but for completeness. */
2130         case MDS_DISCONNECT:
2131         case OST_DISCONNECT:
2132                *process = 1;
2133                RETURN(0);
2134
2135         case MDS_CLOSE:
2136         case MDS_SYNC: /* used in unmounting */
2137         case OBD_PING:
2138         case MDS_REINT:
2139         case LDLM_ENQUEUE:
2140         case OST_CREATE:
2141                 *process = target_queue_recovery_request(req, obd);
2142                 RETURN(0);
2143
2144         default:
2145                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2146                 *process = 0;
2147                 /* XXX what should we set rq_status to here? */
2148                 req->rq_status = -EAGAIN;
2149                 RETURN(ptlrpc_error(req));
2150         }
2151 }
2152
2153 static char *reint_names[] = {
2154         [REINT_SETATTR] "setattr",
2155         [REINT_CREATE]  "create",
2156         [REINT_LINK]    "link",
2157         [REINT_UNLINK]  "unlink",
2158         [REINT_RENAME]  "rename",
2159         [REINT_OPEN]    "open",
2160 };
2161
2162 #define FILTER_VALID_FLAGS (OBD_MD_FLTYPE | OBD_MD_FLMODE | OBD_MD_FLGENER  | \
2163                             OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ| \
2164                             OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME| \
2165                             OBD_MD_FLID)
2166
2167 static void reconstruct_create(struct ptlrpc_request *req)
2168 {
2169         struct mds_export_data *med = &req->rq_export->exp_mds_data;
2170         struct mds_client_data *mcd = med->med_mcd;
2171         struct dentry *dentry;
2172         struct ost_body *body;
2173         struct lustre_id id;
2174         int rc;
2175         ENTRY;
2176
2177         /* copy rc, transno and disp; steal locks */
2178         mds_req_from_mcd(req, mcd);
2179         if (req->rq_status) {
2180                 EXIT;
2181                 return;
2182         }
2183
2184         id_gen(&id) = 0;
2185         id_group(&id) = 0;
2186
2187         id_ino(&id) = mcd->mcd_last_data;
2188         LASSERT(id_ino(&id) != 0);
2189
2190         dentry = mds_id2dentry(req2obd(req), &id, NULL);
2191         if (IS_ERR(dentry)) {
2192                 CERROR("can't find inode "LPU64"\n", id_ino(&id));
2193                 req->rq_status = PTR_ERR(dentry);
2194                 EXIT;
2195                 return;
2196         }
2197
2198         CWARN("reconstruct reply for x"LPU64" (remote ino) "LPU64" -> %lu/%u\n",
2199               req->rq_xid, id_ino(&id), dentry->d_inode->i_ino,
2200               dentry->d_inode->i_generation);
2201
2202         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
2203         obdo_from_inode(&body->oa, dentry->d_inode, FILTER_VALID_FLAGS);
2204         body->oa.o_id = dentry->d_inode->i_ino;
2205         body->oa.o_generation = dentry->d_inode->i_generation;
2206         body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2207
2208         down(&dentry->d_inode->i_sem);
2209         rc = mds_read_inode_sid(req2obd(req), dentry->d_inode, &id);
2210         up(&dentry->d_inode->i_sem);
2211         if (rc) {
2212                 CERROR("Can't read inode self id, inode %lu, "
2213                        "rc %d\n", dentry->d_inode->i_ino, rc);
2214                 id_fid(&id) = 0;
2215         }
2216
2217         body->oa.o_fid = id_fid(&id);
2218         body->oa.o_mds = id_group(&id);
2219         l_dput(dentry);
2220
2221         EXIT;
2222 }
2223
2224 static int mds_inode_init_acl(struct obd_device *obd, void *handle,
2225                               struct dentry *de, void *xattr, int xattr_size)
2226 {
2227         struct inode *inode = de->d_inode;
2228         struct posix_acl *acl;
2229         mode_t mode;
2230         int rc = 0;
2231
2232         LASSERT(handle);
2233         LASSERT(inode);
2234         LASSERT(xattr);
2235         LASSERT(xattr_size > 0);
2236
2237         if (!inode->i_op->getxattr || !inode->i_op->setxattr) {
2238                 CERROR("backend fs dosen't support xattr\n");
2239                 return -EOPNOTSUPP;
2240         }
2241
2242         /* set default acl */
2243         if (S_ISDIR(inode->i_mode)) {
2244                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_DEFAULT,
2245                                            xattr, xattr_size, 0);
2246                 if (rc) {
2247                         CERROR("set default acl err: %d\n", rc);
2248                         return rc;
2249                 }
2250         }
2251
2252         /* set access acl */
2253         acl = posix_acl_from_xattr(xattr, xattr_size);
2254         if (acl == NULL || IS_ERR(acl)) {
2255                 CERROR("insane attr data\n");
2256                 return PTR_ERR(acl);
2257         }
2258
2259         if (posix_acl_valid(acl)) {
2260                 CERROR("default acl not valid: %d\n", rc);
2261                 rc = -EFAULT;
2262                 goto out;
2263         }
2264
2265         mode = inode->i_mode;
2266         rc = posix_acl_create_masq(acl, &mode);
2267         if (rc < 0) {
2268                 CERROR("create masq err %d\n", rc);
2269                 goto out;
2270         }
2271
2272         if (inode->i_mode != mode) {
2273                 struct iattr iattr = { .ia_valid = ATTR_MODE,
2274                                        .ia_mode = mode };
2275                 int rc2;
2276
2277                 rc2 = fsfilt_setattr(obd, de, handle, &iattr, 0);
2278                 if (rc2) {
2279                         CERROR("setattr mode err: %d\n", rc2);
2280                         rc = rc2;
2281                         goto out;
2282                 }
2283         }
2284
2285         if (rc > 0) {
2286                 /* we didn't change acl except mode bits of some
2287                  * entries, so should be fit into original size.
2288                  */
2289                 rc = posix_acl_to_xattr(acl, xattr, xattr_size);
2290                 LASSERT(rc > 0);
2291
2292                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_ACCESS,
2293                                            xattr, xattr_size, 0);
2294                 if (rc)
2295                         CERROR("set access acl err: %d\n", rc);
2296         }
2297 out:
2298         posix_acl_release(acl);
2299         return rc;
2300 }
2301
2302 static int mdt_obj_create(struct ptlrpc_request *req)
2303 {
2304         struct obd_device *obd = req->rq_export->exp_obd;
2305         struct mds_obd *mds = &obd->u.mds;
2306         struct ost_body *body, *repbody;
2307         void *acl = NULL;
2308         int acl_size;
2309         char idname[LL_ID_NAMELEN];
2310         int size = sizeof(*repbody);
2311         struct inode *parent_inode;
2312         struct lvfs_run_ctxt saved;
2313         int rc, cleanup_phase = 0;
2314         struct dentry *new = NULL;
2315         struct dentry_params dp;
2316         int mealen, flags = 0;
2317         struct lvfs_ucred uc;
2318         struct lustre_id id;
2319         struct mea *mea;
2320         void *handle = NULL;
2321         unsigned long cr_inum = 0;
2322         __u64 fid = 0;
2323         ENTRY;
2324
2325         DEBUG_REQ(D_HA, req, "create remote object");
2326         parent_inode = mds->mds_unnamed_dir->d_inode;
2327
2328         body = lustre_swab_reqbuf(req, 0, sizeof(*body),
2329                                   lustre_swab_ost_body);
2330         if (body == NULL)
2331                 RETURN(-EFAULT);
2332
2333         /* acl data is packed transparently, no swab here */
2334         LASSERT(req->rq_reqmsg->bufcount >= 2);
2335         acl_size = req->rq_reqmsg->buflens[1];
2336         if (acl_size) {
2337                 acl = lustre_msg_buf(req->rq_reqmsg, 1, acl_size);
2338                 if (!acl) {
2339                         CERROR("No default acl buf?\n");
2340                         RETURN(-EFAULT);
2341                 }
2342         }
2343
2344         rc = lustre_pack_reply(req, 1, &size, NULL);
2345         if (rc)
2346                 RETURN(rc);
2347
2348         MDS_CHECK_RESENT(req, reconstruct_create(req));
2349
2350         uc.luc_lsd = NULL;
2351         uc.luc_ginfo = NULL;
2352         uc.luc_uid = body->oa.o_uid;
2353         uc.luc_gid = body->oa.o_gid;
2354         uc.luc_fsuid = body->oa.o_uid;
2355         uc.luc_fsgid = body->oa.o_gid;
2356
2357         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2358         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
2359
2360         /* in REPLAY case inum should be given (client or other MDS fills it) */
2361         if (body->oa.o_id && ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2362             (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))) {
2363                 /*
2364                  * this is re-create request from MDS holding directory name.
2365                  * we have to lookup given ino/gen first. if it exists (good
2366                  * case) then there is nothing to do. if it does not then we
2367                  * have to recreate it.
2368                  */
2369                 id_ino(&id) = body->oa.o_id;
2370                 id_gen(&id) = body->oa.o_generation;
2371
2372                 new = mds_id2dentry(obd, &id, NULL);
2373                 if (!IS_ERR(new) && new->d_inode) {
2374                         struct lustre_id sid;
2375
2376                         CDEBUG(D_OTHER, "mkdir repairing %lu/%lu\n",
2377                                (unsigned long)id_ino(&id),
2378                                (unsigned long)id_gen(&id));
2379
2380                         obdo_from_inode(&repbody->oa, new->d_inode,
2381                                         FILTER_VALID_FLAGS);
2382
2383                         repbody->oa.o_id = new->d_inode->i_ino;
2384                         repbody->oa.o_generation = new->d_inode->i_generation;
2385                         repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2386                         cleanup_phase = 1;
2387
2388                         down(&new->d_inode->i_sem);
2389                         rc = mds_read_inode_sid(obd, new->d_inode, &sid);
2390                         up(&new->d_inode->i_sem);
2391                         if (rc) {
2392                                 CERROR("Can't read inode self id "
2393                                        "inode %lu, rc %d.\n",
2394                                        new->d_inode->i_ino, rc);
2395                                 GOTO(cleanup, rc);
2396                         }
2397
2398                         repbody->oa.o_fid = id_fid(&sid);
2399                         repbody->oa.o_mds = id_group(&sid);
2400                         LASSERT(id_fid(&sid) != 0);
2401
2402                         /*
2403                          * here we could use fid passed in body->oa.o_fid and
2404                          * thus avoid mds_read_inode_sid().
2405                          */
2406                         cr_inum = new->d_inode->i_ino;
2407                         GOTO(cleanup, rc = 0);
2408                 }
2409         }
2410
2411         down(&parent_inode->i_sem);
2412         handle = fsfilt_start(obd, parent_inode, FSFILT_OP_MKDIR, NULL);
2413         if (IS_ERR(handle)) {
2414                 up(&parent_inode->i_sem);
2415                 CERROR("fsfilt_start() failed, rc = %d\n",
2416                        (int)PTR_ERR(handle));
2417                 GOTO(cleanup, rc = PTR_ERR(handle));
2418         }
2419         cleanup_phase = 1; /* transaction */
2420
2421 repeat:
2422         rc = sprintf(idname, "%u.%u", ll_insecure_random_int(), current->pid);
2423         new = lookup_one_len(idname, mds->mds_unnamed_dir, rc);
2424         if (IS_ERR(new)) {
2425                 CERROR("%s: can't lookup new inode (%s) for mkdir: %d\n",
2426                        obd->obd_name, idname, (int) PTR_ERR(new));
2427                 fsfilt_commit(obd, mds->mds_sb, new->d_inode, handle, 0);
2428                 up(&parent_inode->i_sem);
2429                 RETURN(PTR_ERR(new));
2430         } else if (new->d_inode) {
2431                 CERROR("%s: name exists. repeat\n", obd->obd_name);
2432                 goto repeat;
2433         }
2434         if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2435              lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2436                 fid = body->oa.o_fid;
2437         } else {
2438                 fid = mds_alloc_fid(obd);
2439         }
2440         new->d_fsdata = (void *)&dp;
2441         dp.p_inum = 0;
2442         dp.p_ptr = req;
2443
2444         dp.p_fid = fid;
2445         dp.p_group = mds->mds_num;
2446
2447         if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) ||
2448             (body->oa.o_flags & OBD_FL_RECREATE_OBJS)) {
2449                 LASSERT(body->oa.o_id != 0);
2450                 dp.p_inum = body->oa.o_id;
2451                 DEBUG_REQ(D_HA, req, "replay create obj %lu/%lu",
2452                           (unsigned long)body->oa.o_id,
2453                           (unsigned long)body->oa.o_generation);
2454         }
2455
2456         rc = vfs_mkdir(parent_inode, new, body->oa.o_mode);
2457         if (rc == 0) {
2458                 if (acl) {
2459                         rc = mds_inode_init_acl(obd, handle, new,
2460                                                 acl, acl_size);
2461                         if (rc) {
2462                                 up(&parent_inode->i_sem);
2463                                 GOTO(cleanup, rc);
2464                         }
2465                 }
2466                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2467                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2468                         new->d_inode->i_generation = body->oa.o_generation;
2469                         mark_inode_dirty(new->d_inode);
2470
2471                         /*
2472                          * avoiding asserts in cache flush case, as
2473                          * @body->oa.o_id should be zero.
2474                          */
2475                         if (body->oa.o_id) {
2476                                 LASSERTF(body->oa.o_id == new->d_inode->i_ino,
2477                                          "BUG 3550: failed to recreate obj "
2478                                          LPU64" -> %lu\n", body->oa.o_id,
2479                                          new->d_inode->i_ino);
2480
2481                                 LASSERTF(body->oa.o_generation ==
2482                                          new->d_inode->i_generation,
2483                                          "BUG 3550: failed to recreate obj/gen "
2484                                          LPU64"/%u -> %lu/%u\n", body->oa.o_id,
2485                                          body->oa.o_generation,
2486                                          new->d_inode->i_ino,
2487                                          new->d_inode->i_generation);
2488                         }
2489                 }
2490
2491                 obdo_from_inode(&repbody->oa, new->d_inode, FILTER_VALID_FLAGS);
2492                 repbody->oa.o_id = new->d_inode->i_ino;
2493                 repbody->oa.o_generation = new->d_inode->i_generation;
2494                 repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FID;
2495
2496                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2497                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2498                         LASSERT(body->oa.o_id != 0);
2499                         LASSERT(body->oa.o_fid != 0);
2500                 }
2501
2502                 mds_inode2id(obd, &id, new->d_inode, fid);
2503                 mds_update_inode_ids(obd, new->d_inode, handle, &id,
2504                                      body->oa.o_valid & OBD_MD_FLID ?
2505                                      NULL : obdo_id(&body->oa));
2506
2507                 /* initializing o_fid after it is allocated. */
2508                 repbody->oa.o_fid = id_fid(&id);
2509                 repbody->oa.o_mds = id_group(&id);
2510
2511                 rc = fsfilt_del_dir_entry(obd, new);
2512                 up(&parent_inode->i_sem);
2513                 if (rc) {
2514                         CERROR("can't remove name for object: %d\n", rc);
2515                         GOTO(cleanup, rc);
2516                 }
2517
2518                 cleanup_phase = 2; /* created directory object */
2519
2520                 CDEBUG(D_OTHER, "created dirobj: %lu/%lu mode %o\n",
2521                        (unsigned long)new->d_inode->i_ino,
2522                        (unsigned long)new->d_inode->i_generation,
2523                        (unsigned)new->d_inode->i_mode);
2524                 cr_inum = new->d_inode->i_ino;
2525         } else {
2526                 up(&parent_inode->i_sem);
2527                 CERROR("%s: can't create dirobj: %d\n", obd->obd_name, rc);
2528                 GOTO(cleanup, rc);
2529         }
2530
2531         if (body->oa.o_valid & OBD_MD_FLID) {
2532                 /* this is new object for splitted dir. We have to prevent
2533                  * recursive splitting on it -bzzz */
2534                 mealen = obd_size_diskmd(mds->mds_md_exp, NULL);
2535
2536                 OBD_ALLOC(mea, mealen);
2537                 if (mea == NULL)
2538                         GOTO(cleanup, rc = -ENOMEM);
2539
2540                 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
2541                 mea->mea_master = body->oa.o_mds;      /* master mds num */
2542                 mea->mea_count = 0;
2543
2544                 obdo2id(&mea->mea_ids[body->oa.o_mds], &body->oa);
2545
2546                 down(&new->d_inode->i_sem);
2547                 rc = fsfilt_set_md(obd, new->d_inode, handle,
2548                                    mea, mealen, EA_MEA);
2549                 up(&new->d_inode->i_sem);
2550                 if (rc)
2551                         CERROR("fsfilt_set_md() failed, "
2552                                "rc = %d\n", rc);
2553
2554                 OBD_FREE(mea, mealen);
2555
2556                 CDEBUG(D_OTHER, "%s: mark non-splittable %lu/%u - %d\n",
2557                        obd->obd_name, new->d_inode->i_ino,
2558                        new->d_inode->i_generation, flags);
2559         } else if (body->oa.o_easize) {
2560                 /* we pass LCK_EX to split routine to signal that we have
2561                  * exclusive access to the directory. simple because nobody
2562                  * knows it already exists -bzzz */
2563                 rc = mds_try_to_split_dir(obd, new, NULL,
2564                                           body->oa.o_easize, LCK_EX);
2565                 if (rc < 0) {
2566                         CERROR("Can't split directory %lu, error = %d.\n",
2567                                new->d_inode->i_ino, rc);
2568                 } else {
2569                         rc = 0;
2570                 }
2571         }
2572
2573         EXIT;
2574 cleanup:
2575         if (rc == 0 && (body->oa.o_flags & OBD_FL_REINT)) {
2576                 rc = mds_fidmap_add(obd, &id);
2577                 if (rc < 0) {
2578                         CERROR("can't create fid->ino mapping, "
2579                                "err %d\n", rc);
2580                 } else {
2581                         rc = 0;
2582                 }
2583         }
2584
2585         switch (cleanup_phase) {
2586         case 2: /* object has been created, but we'll may want to replay it later */
2587                 if (rc == 0)
2588                         ptlrpc_require_repack(req);
2589         case 1: /* transaction */
2590                 rc = mds_finish_transno(mds, parent_inode, handle,
2591                                         req, rc, cr_inum);
2592         }
2593
2594         l_dput(new);
2595         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2596         return rc;
2597 }
2598
2599 static int mdt_get_info(struct ptlrpc_request *req)
2600 {
2601         struct obd_export *exp = req->rq_export;
2602         int keylen, rc = 0;
2603         char *key;
2604         ENTRY;
2605
2606         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2607         if (key == NULL) {
2608                 DEBUG_REQ(D_HA, req, "no get_info key");
2609                 RETURN(-EFAULT);
2610         }
2611         keylen = req->rq_reqmsg->buflens[0];
2612
2613         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
2614             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
2615             (keylen < strlen("lovdesc") || strcmp(key, "lovdesc") != 0) &&
2616             (keylen < strlen("getext") || strcmp(key, "getext") != 0) &&
2617             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
2618                 RETURN(-EPROTO);
2619
2620         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
2621                 struct lustre_id *reply;
2622                 int size = sizeof(*reply);
2623
2624                 rc = lustre_pack_reply(req, 1, &size, NULL);
2625                 if (rc)
2626                         RETURN(rc);
2627
2628                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2629                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2630         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2631                 struct lov_desc *reply;
2632                 int size = sizeof(*reply);
2633
2634                 rc = lustre_pack_reply(req, 1, &size, NULL);
2635                 if (rc)
2636                         RETURN(rc);
2637
2638                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2639                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2640         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2641                 struct fid_extent *reply;
2642                 int size = sizeof(*reply);
2643
2644                 rc = lustre_pack_reply(req, 1, &size, NULL);
2645                 if (rc)
2646                         RETURN(rc);
2647
2648                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2649                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2650         } else {
2651                 obd_id *reply;
2652                 int size = sizeof(*reply);
2653
2654                 rc = lustre_pack_reply(req, 1, &size, NULL);
2655                 if (rc)
2656                         RETURN(rc);
2657
2658                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2659                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2660         }
2661
2662         req->rq_repmsg->status = 0;
2663         RETURN(rc);
2664 }
2665
2666 static int mds_set_info(struct obd_export *exp, __u32 keylen,
2667                         void *key, __u32 vallen, void *val)
2668 {
2669         struct obd_device *obd;
2670         struct mds_obd *mds;
2671         int rc = 0;
2672         ENTRY;
2673
2674         obd = class_exp2obd(exp);
2675         if (obd == NULL) {
2676                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2677                        exp->exp_handle.h_cookie);
2678                 RETURN(-EINVAL);
2679         }
2680
2681         mds = &obd->u.mds;
2682         if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2683                 rc = mds_set_audit(obd, val);
2684                 RETURN(rc);
2685         } else if (keylen >= strlen("ids") && memcmp(key, "ids", keylen) == 0) {
2686                 struct lustre_id *ids = (struct lustre_id *)val;
2687                 struct dentry *de;
2688                 struct inode *inode;
2689                 void *handle;
2690                 int err;
2691
2692                 de = mds_id2dentry(obd, ids, NULL);
2693                 if (IS_ERR(de)) {
2694                         rc = PTR_ERR(de);
2695                         CERROR("lookup by an id error rc=%d\n ", rc);
2696                         RETURN(rc);
2697                 }
2698                 inode = de->d_inode;
2699                 if (inode == NULL)
2700                         GOTO(out_put, rc = -ENOENT);
2701
2702                 down(&inode->i_sem);
2703                 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
2704                 if (IS_ERR(handle)) {
2705                         up(&inode->i_sem);
2706                         GOTO(out_put, rc = PTR_ERR(handle));
2707                 }
2708
2709                 rc = mds_update_inode_ids(obd, inode, handle, NULL, ids + 1);
2710
2711                 err = fsfilt_commit(obd, mds->mds_sb, inode, handle,
2712                                     exp->exp_sync);
2713                 if (err) {
2714                         CERROR("error committing transaction: %d\n", err);
2715                         if (!rc) rc = err;
2716                 }
2717                 up(&inode->i_sem);
2718 out_put:
2719                 l_dput(de);
2720                 RETURN(rc);
2721         }
2722
2723         if (keylen >= strlen("crypto_type") &&
2724              memcmp(key, "crypto_type", keylen) == 0) {
2725                 rc = mds_set_crypto_type(obd, val, vallen);
2726                 RETURN(rc);
2727         }
2728
2729         if (keylen >= strlen("setext") && !memcmp(key, "setext", keylen)) {
2730                 struct fid_extent *ext = val;
2731
2732                 CDEBUG(D_IOCTL, "set last fid to extent ["LPD64"-"LPD64"]\n",
2733                        ext->fe_start, ext->fe_width);
2734
2735                 /* set lastfid into fid extent start. All next object creates
2736                  * will use that fid. */
2737                 mds_set_last_fid(obd, ext->fe_start);
2738
2739                 /* setting the same extent to OSC to avoid ids intersecting in
2740                  * object ids, as all cache MDSs have the same group 0. */
2741                 rc = obd_set_info(mds->mds_dt_exp, strlen("setext"),
2742                                   "setext", sizeof(*ext), ext);
2743                 if (rc) {
2744                         CERROR("can't set extent ["LPD64"-"LPD64"] to %s, "
2745                                "err %d\n", ext->fe_start, ext->fe_width,
2746                                mds->mds_dt_exp->exp_obd->obd_name, rc);
2747                 }
2748                 RETURN(rc);
2749         }
2750         CDEBUG(D_IOCTL, "invalid key\n");
2751         RETURN(-EINVAL);
2752 }
2753
2754 static int mdt_set_info(struct ptlrpc_request *req)
2755 {
2756         char *key, *val;
2757         struct obd_export *exp = req->rq_export;
2758         int keylen, rc = 0, vallen = 0;
2759         ENTRY;
2760
2761         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2762         if (key == NULL) {
2763                 DEBUG_REQ(D_HA, req, "no set_info key");
2764                 RETURN(-EFAULT);
2765         }
2766         keylen = req->rq_reqmsg->buflens[0];
2767
2768         if ((keylen == strlen("crypto_type") &&
2769             memcmp(key, "crypto_type", keylen) == 0)) {
2770                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2771                 if (rc)
2772                         RETURN(rc);
2773
2774                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2775                 vallen = req->rq_reqmsg->buflens[1];
2776
2777                 rc = obd_set_info(exp, keylen, key, vallen, val);
2778                 req->rq_repmsg->status = 0;
2779                 RETURN(rc);
2780         } else if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2781                 struct audit_attr_msg msg, *p;
2782                 int rc = 0;
2783
2784                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2785                 if (rc)
2786                         RETURN(rc);
2787
2788                 p = lustre_swab_reqbuf(req, 1, sizeof(msg),
2789                                        lustre_swab_audit_attr);
2790
2791                 msg = *p;
2792                 CDEBUG(D_INFO, "Get new audit setting 0x%x\n", (__u32)msg.attr);
2793                 rc = obd_set_info(exp, keylen, key, sizeof(msg), &msg);
2794
2795                 req->rq_repmsg->status = rc;
2796                 RETURN(rc);
2797         } else if (keylen == strlen("ids") &&
2798                    memcmp(key, "ids", keylen) == 0) {
2799                 struct lustre_id *id, ids[2];
2800
2801                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2802                 if (rc)
2803                         RETURN(rc);
2804                 id = lustre_swab_reqbuf(req, 1, sizeof(struct lustre_id),
2805                                         lustre_swab_lustre_id);
2806                 ids[0] = *id;
2807                 id = lustre_swab_reqbuf(req, 2, sizeof(struct lustre_id),
2808                                         lustre_swab_lustre_id);
2809                 ids[1] = *id;
2810
2811                 rc = obd_set_info(exp, keylen, key, vallen, ids);
2812                 req->rq_repmsg->status = rc;
2813                 RETURN(rc);
2814         } else if (keylen == strlen("setext") &&
2815                    memcmp(key, "setext", keylen) == 0) {
2816                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2817                 if (rc)
2818                         RETURN(rc);
2819
2820                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2821                 vallen = req->rq_reqmsg->buflens[1];
2822
2823                 rc = obd_set_info(exp, keylen, key, vallen, val);
2824                 req->rq_repmsg->status = 0;
2825                 RETURN(rc);
2826         }
2827
2828         CDEBUG(D_IOCTL, "invalid key\n");
2829         RETURN(-EINVAL);
2830 }
2831
2832 static void mds_revoke_export_locks(struct obd_export *exp)
2833 {
2834         struct list_head *locklist = &exp->exp_ldlm_data.led_held_locks;
2835         struct list_head  rpc_list;
2836         struct ldlm_lock *lock, *next;
2837         struct ldlm_lock_desc desc;
2838
2839         /* don't do this for local client */
2840         if (!exp->u.eu_mds_data.med_remote)
2841                 return;
2842
2843         /* don't revoke locks during recovery */
2844         if (exp->exp_obd->obd_recovering)
2845                 return;
2846
2847         ENTRY;
2848         INIT_LIST_HEAD(&rpc_list);
2849
2850         spin_lock(&exp->exp_ldlm_data.led_lock);
2851         list_for_each_entry_safe(lock, next, locklist, l_export_chain) {
2852
2853                 lock_res_and_lock(lock);
2854                 if (lock->l_req_mode != lock->l_granted_mode) {
2855                         unlock_res_and_lock(lock);
2856                         continue;
2857                 }
2858
2859                 LASSERT(lock->l_resource);
2860                 if (lock->l_resource->lr_type != LDLM_IBITS &&
2861                     lock->l_resource->lr_type != LDLM_PLAIN) {
2862                         unlock_res_and_lock(lock);
2863                         continue;
2864                 }
2865
2866                 if (lock->l_flags & LDLM_FL_AST_SENT) {
2867                         unlock_res_and_lock(lock);
2868                         continue;
2869                 }
2870
2871                 LASSERT(lock->l_blocking_ast);
2872                 LASSERT(!lock->l_blocking_lock);
2873
2874                 lock->l_flags |= LDLM_FL_AST_SENT;
2875                 unlock_res_and_lock(lock);
2876
2877                 list_move(&lock->l_export_chain, &rpc_list);
2878         }
2879         spin_unlock(&exp->exp_ldlm_data.led_lock);
2880
2881         while (!list_empty(&rpc_list)) {
2882                 lock = list_entry(rpc_list.next, struct ldlm_lock,
2883                                   l_export_chain);
2884                 list_del_init(&lock->l_export_chain);
2885
2886                 /* the desc just pretend to exclusive */
2887                 ldlm_lock2desc(lock, &desc);
2888                 desc.l_req_mode = LCK_EX;
2889                 desc.l_granted_mode = 0;
2890
2891                 lock->l_blocking_ast(lock, &desc, NULL, LDLM_CB_BLOCKING);
2892         }
2893         EXIT;
2894 }
2895
2896 static int mds_msg_check_version(struct lustre_msg *msg)
2897 {
2898         int rc;
2899
2900         switch (msg->opc) {
2901         case MDS_CONNECT:
2902         case MDS_DISCONNECT:
2903         case OBD_PING:
2904                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2905                 if (rc)
2906                         CERROR("bad opc %u version %08x, expecting %08x\n",
2907                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2908                 break;
2909         case MDS_STATFS:
2910         case MDS_GETSTATUS:
2911         case MDS_GETATTR:
2912         case MDS_GETATTR_LOCK:
2913         case MDS_ACCESS_CHECK:
2914         case MDS_READPAGE:
2915         case MDS_REINT:
2916         case MDS_CLOSE:
2917         case MDS_DONE_WRITING:
2918         case MDS_PIN:
2919         case MDS_SYNC:
2920         case MDS_PARSE_ID:
2921                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2922                 if (rc)
2923                         CERROR("bad opc %u version %08x, expecting %08x\n",
2924                                msg->opc, msg->version, LUSTRE_MDS_VERSION);
2925                 break;
2926         case LDLM_ENQUEUE:
2927         case LDLM_CONVERT:
2928         case LDLM_BL_CALLBACK:
2929         case LDLM_CP_CALLBACK:
2930         case LDLM_FLK_DEADLOCK_CHK:
2931                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2932                 if (rc)
2933                         CERROR("bad opc %u version %08x, expecting %08x\n",
2934                                msg->opc, msg->version, LUSTRE_DLM_VERSION);
2935                 break;
2936         case OBD_LOG_CANCEL:
2937         case LLOG_ORIGIN_HANDLE_OPEN:
2938         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2939         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2940         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2941         case LLOG_ORIGIN_HANDLE_CLOSE:
2942         case LLOG_CATINFO:
2943                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2944                 if (rc)
2945                         CERROR("bad opc %u version %08x, expecting %08x\n",
2946                                msg->opc, msg->version, LUSTRE_LOG_VERSION);
2947                 break;
2948         case OST_CREATE:
2949         case OST_WRITE:
2950         case OST_GET_INFO:
2951         case OST_SET_INFO:
2952                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2953                 if (rc)
2954                         CERROR("bad opc %u version %08x, expecting %08x\n",
2955                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2956                 break;
2957         case SEC_INIT:
2958         case SEC_INIT_CONTINUE:
2959         case SEC_FINI:
2960                 rc = 0;
2961                 break;
2962         default:
2963                 CERROR("MDS unknown opcode %d\n", msg->opc);
2964                 rc = -ENOTSUPP;
2965                 break;
2966         }
2967
2968         return rc;
2969 }
2970
2971 int mds_handle(struct ptlrpc_request *req)
2972 {
2973         int should_process, fail = OBD_FAIL_MDS_ALL_REPLY_NET;
2974         struct obd_device *obd = NULL;
2975         struct mds_obd *mds = NULL; /* quell gcc overwarning */
2976         int rc = 0;
2977         ENTRY;
2978
2979         OBD_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0);
2980
2981         rc = mds_msg_check_version(req->rq_reqmsg);
2982         if (rc) {
2983                 CERROR("MDS drop mal-formed request\n");
2984                 RETURN(rc);
2985         }
2986
2987         /* Security opc should NOT trigger any recovery events */
2988         if (req->rq_reqmsg->opc == SEC_INIT ||
2989             req->rq_reqmsg->opc == SEC_INIT_CONTINUE) {
2990                 if (req->rq_export) {
2991                         mds_req_add_idmapping(req,
2992                                               &req->rq_export->exp_mds_data);
2993                         mds_revoke_export_locks(req->rq_export);
2994                 }
2995                 GOTO(out, rc = 0);
2996         } else if (req->rq_reqmsg->opc == SEC_FINI) {
2997                 if (req->rq_export) {
2998                         mds_req_del_idmapping(req,
2999                                               &req->rq_export->exp_mds_data);
3000                         mds_revoke_export_locks(req->rq_export);
3001                 }
3002                 GOTO(out, rc = 0);
3003         }
3004
3005         LASSERT(current->journal_info == NULL);
3006         /* XXX identical to OST */
3007         if (req->rq_reqmsg->opc != MDS_CONNECT) {
3008                 struct mds_export_data *med;
3009                 int recovering;
3010
3011                 if (req->rq_export == NULL) {
3012                         CERROR("operation %d on unconnected MDS from %s\n",
3013                                req->rq_reqmsg->opc,
3014                                req->rq_peerstr);
3015                         req->rq_status = -ENOTCONN;
3016                         GOTO(out, rc = -ENOTCONN);
3017                 }
3018
3019                 med = &req->rq_export->exp_mds_data;
3020                 obd = req->rq_export->exp_obd;
3021                 mds = &obd->u.mds;
3022
3023                 /* sanity check: if the xid matches, the request must
3024                  * be marked as a resent or replayed */
3025                 if (req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_xid) ||
3026                    req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_close_xid)) {
3027                         LASSERTF(lustre_msg_get_flags(req->rq_reqmsg) &
3028                                  (MSG_RESENT | MSG_REPLAY),
3029                                  "rq_xid "LPU64" matches last_xid, "
3030                                  "expected RESENT flag\n",
3031                                  req->rq_xid);
3032                 }
3033                 /* else: note the opposite is not always true; a
3034                  * RESENT req after a failover will usually not match
3035                  * the last_xid, since it was likely never
3036                  * committed. A REPLAYed request will almost never
3037                  * match the last xid, however it could for a
3038                  * committed, but still retained, open. */
3039
3040                 spin_lock_bh(&obd->obd_processing_task_lock);
3041                 recovering = obd->obd_recovering;
3042                 spin_unlock_bh(&obd->obd_processing_task_lock);
3043                 if (recovering) {
3044                         rc = mds_filter_recovery_request(req, obd,
3045                                                          &should_process);
3046                         if (rc || should_process == 0) {
3047                                 RETURN(rc);
3048                         } else if (should_process < 0) {
3049                                 req->rq_status = should_process;
3050                                 rc = ptlrpc_error(req);
3051                                 RETURN(rc);
3052                         }
3053                 }
3054         }
3055
3056         switch (req->rq_reqmsg->opc) {
3057         case MDS_CONNECT:
3058                 DEBUG_REQ(D_INODE, req, "connect");
3059                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
3060                 rc = target_handle_connect(req);
3061                 if (!rc) {
3062                         struct mds_export_data *med;
3063
3064                         LASSERT(req->rq_export);
3065                         med = &req->rq_export->u.eu_mds_data;
3066                         mds_init_export_data(req, med);
3067                         mds_req_add_idmapping(req, med);
3068
3069                         /* Now that we have an export, set mds. */
3070                         obd = req->rq_export->exp_obd;
3071                         mds = mds_req2mds(req);
3072                 }
3073                 break;
3074
3075         case MDS_DISCONNECT:
3076                 DEBUG_REQ(D_INODE, req, "disconnect");
3077                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
3078                 rc = target_handle_disconnect(req);
3079                 req->rq_status = rc;            /* superfluous? */
3080                 break;
3081
3082         case MDS_GETSTATUS:
3083                 DEBUG_REQ(D_INODE, req, "getstatus");
3084                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
3085                 rc = mds_getstatus(req);
3086                 break;
3087
3088         case MDS_GETATTR:
3089                 DEBUG_REQ(D_INODE, req, "getattr");
3090                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
3091                 rc = mds_getattr(req, MDS_REQ_REC_OFF);
3092                 break;
3093
3094         case MDS_ACCESS_CHECK:
3095                 DEBUG_REQ(D_INODE, req, "access_check");
3096                 OBD_FAIL_RETURN(OBD_FAIL_MDS_ACCESS_CHECK_NET, 0);
3097                 rc = mds_access_check(req, MDS_REQ_REC_OFF);
3098                 break;
3099
3100         case MDS_GETATTR_LOCK: {
3101                 struct lustre_handle lockh;
3102                 DEBUG_REQ(D_INODE, req, "getattr_lock");
3103                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_LOCK_NET, 0);
3104
3105                 /* If this request gets a reconstructed reply, we won't be
3106                  * acquiring any new locks in mds_getattr_lock, so we don't
3107                  * want to cancel.
3108                  */
3109                 lockh.cookie = 0;
3110                 rc = mds_getattr_lock(req, MDS_REQ_REC_OFF, &lockh,
3111                                       MDS_INODELOCK_UPDATE);
3112                 /* this non-intent call (from an ioctl) is special */
3113                 req->rq_status = rc;
3114                 if (rc == 0 && lockh.cookie)
3115                         ldlm_lock_decref(&lockh, LCK_PR);
3116                 break;
3117         }
3118         case MDS_STATFS:
3119                 DEBUG_REQ(D_INODE, req, "statfs");
3120                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
3121                 rc = mds_statfs(req);
3122                 break;
3123
3124         case MDS_READPAGE:
3125                 DEBUG_REQ(D_INODE, req, "readpage");
3126                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
3127                 rc = mds_readpage(req, MDS_REQ_REC_OFF);
3128
3129                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_SENDPAGE)) {
3130                         if (req->rq_reply_state) {
3131                                 lustre_free_reply_state (req->rq_reply_state);
3132                                 req->rq_reply_state = NULL;
3133                         }
3134                         RETURN(0);
3135                 }
3136
3137                 break;
3138         case MDS_REINT: {
3139                 __u32 *opcp = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
3140                                              sizeof (*opcp));
3141                 __u32  opc;
3142                 int size[3] = {sizeof(struct mds_body), mds->mds_max_mdsize,
3143                                mds->mds_max_cookiesize};
3144                 int bufcount;
3145
3146                 /* NB only peek inside req now; mds_reint() will swab it */
3147                 if (opcp == NULL) {
3148                         CERROR ("Can't inspect opcode\n");
3149                         rc = -EINVAL;
3150                         break;
3151                 }
3152                 opc = *opcp;
3153                 if (lustre_msg_swabbed (req->rq_reqmsg))
3154                         __swab32s(&opc);
3155
3156                 DEBUG_REQ(D_INODE, req, "reint %d (%s)", opc,
3157                           (opc < sizeof(reint_names) / sizeof(reint_names[0]) ||
3158                            reint_names[opc] == NULL) ? reint_names[opc] :
3159                                                        "unknown opcode");
3160
3161                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
3162
3163                 if (opc == REINT_UNLINK || opc == REINT_RENAME)
3164                         bufcount = 3;
3165                 else if (opc == REINT_OPEN)
3166                         bufcount = 2;
3167                 else
3168                         bufcount = 1;
3169
3170                 /* for SETATTR: I have different reply setting for
3171                  * remote setfacl, so delay the reply buffer allocation.
3172                  */
3173                 if (opc != REINT_SETATTR) {
3174                         rc = lustre_pack_reply(req, bufcount, size, NULL);
3175                         if (rc)
3176                                 break;
3177                 }
3178
3179                 rc = mds_reint(req, MDS_REQ_REC_OFF, NULL);
3180                 fail = OBD_FAIL_MDS_REINT_NET_REP;
3181                 break;
3182         }
3183
3184         case MDS_CLOSE:
3185                 DEBUG_REQ(D_INODE, req, "close");
3186                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
3187                 rc = mds_close(req, MDS_REQ_REC_OFF);
3188                 break;
3189
3190         case MDS_DONE_WRITING:
3191                 DEBUG_REQ(D_INODE, req, "done_writing");
3192                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DONE_WRITING_NET, 0);
3193                 rc = mds_done_writing(req, MDS_REQ_REC_OFF);
3194                 break;
3195
3196         case MDS_PIN:
3197                 DEBUG_REQ(D_INODE, req, "pin");
3198                 OBD_FAIL_RETURN(OBD_FAIL_MDS_PIN_NET, 0);
3199                 rc = mds_pin(req, MDS_REQ_REC_OFF);
3200                 break;
3201
3202         case MDS_SYNC:
3203                 DEBUG_REQ(D_INODE, req, "sync");
3204                 OBD_FAIL_RETURN(OBD_FAIL_MDS_SYNC_NET, 0);
3205                 rc = mds_sync(req, MDS_REQ_REC_OFF);
3206                 break;
3207         case MDS_PARSE_ID:
3208                 DEBUG_REQ(D_INODE, req, "parseid");
3209                 rc = mds_parse_id(req);
3210                 break;
3211         case OBD_PING:
3212                 DEBUG_REQ(D_INODE, req, "ping");
3213                 rc = target_handle_ping(req);
3214                 break;
3215
3216         case OBD_LOG_CANCEL:
3217                 CDEBUG(D_INODE, "log cancel\n");
3218                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
3219                 rc = -ENOTSUPP; /* la la la */
3220                 break;
3221
3222         case LDLM_ENQUEUE:
3223                 DEBUG_REQ(D_INODE, req, "enqueue");
3224                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
3225                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
3226                                          ldlm_server_blocking_ast, NULL);
3227                 fail = OBD_FAIL_LDLM_REPLY;
3228                 break;
3229         case LDLM_CONVERT:
3230                 DEBUG_REQ(D_INODE, req, "convert");
3231                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
3232                 rc = ldlm_handle_convert(req);
3233                 break;
3234         case LDLM_BL_CALLBACK:
3235         case LDLM_CP_CALLBACK:
3236                 DEBUG_REQ(D_INODE, req, "callback");
3237                 CERROR("callbacks should not happen on MDS\n");
3238                 LBUG();
3239                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
3240                 break;
3241         case LDLM_FLK_DEADLOCK_CHK:
3242                 DEBUG_REQ(D_INODE, req, "flock deadlock check");
3243                 rc = ldlm_handle_flock_deadlock_check(req);
3244                 break;
3245         case LLOG_ORIGIN_HANDLE_OPEN:
3246                 DEBUG_REQ(D_INODE, req, "llog_init");
3247                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3248                 rc = llog_origin_handle_open(req);
3249                 break;
3250         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3251                 DEBUG_REQ(D_INODE, req, "llog next block");
3252                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3253                 rc = llog_origin_handle_next_block(req);
3254                 break;
3255         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3256                 DEBUG_REQ(D_INODE, req, "llog prev block");
3257                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3258                 rc = llog_origin_handle_prev_block(req);
3259                 break;
3260         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3261                 DEBUG_REQ(D_INODE, req, "llog read header");
3262                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3263                 rc = llog_origin_handle_read_header(req);
3264                 break;
3265         case LLOG_ORIGIN_HANDLE_CLOSE:
3266                 DEBUG_REQ(D_INODE, req, "llog close");
3267                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3268                 rc = llog_origin_handle_close(req);
3269                 break;
3270         case OST_CREATE:
3271                 DEBUG_REQ(D_INODE, req, "ost_create");
3272                 rc = mdt_obj_create(req);
3273                 break;
3274         case OST_GET_INFO:
3275                 DEBUG_REQ(D_INODE, req, "get_info");
3276                 rc = mdt_get_info(req);
3277                 break;
3278         case OST_SET_INFO:
3279                 DEBUG_REQ(D_INODE, req, "set_info");
3280                 rc = mdt_set_info(req);
3281                 break;
3282         case OST_WRITE:
3283                 CDEBUG(D_INODE, "write\n");
3284                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
3285                 rc = ost_brw_write(req, NULL);
3286                 LASSERT(current->journal_info == NULL);
3287                 /* mdt_brw sends its own replies */
3288                 RETURN(rc);
3289                 break;
3290         case LLOG_CATINFO:
3291                 DEBUG_REQ(D_INODE, req, "llog catinfo");
3292                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3293                 rc = llog_catinfo(req);
3294                 break;
3295         default:
3296                 req->rq_status = -ENOTSUPP;
3297                 rc = ptlrpc_error(req);
3298                 RETURN(rc);
3299         }
3300
3301         LASSERT(current->journal_info == NULL);
3302
3303         /* If we're DISCONNECTing, the mds_export_data is already freed */
3304         if (!rc && req->rq_reqmsg->opc != MDS_DISCONNECT) {
3305                 struct mds_export_data *med = &req->rq_export->exp_mds_data;
3306                 struct obd_device *obd = list_entry(mds, struct obd_device,
3307                                                     u.mds);
3308                 req->rq_repmsg->last_xid =
3309                         le64_to_cpu(med->med_mcd->mcd_last_xid);
3310
3311                 if (!obd->obd_no_transno) {
3312                         req->rq_repmsg->last_committed =
3313                                 obd->obd_last_committed;
3314                 } else {
3315                         DEBUG_REQ(D_IOCTL, req,
3316                                   "not sending last_committed update");
3317                 }
3318                 CDEBUG(D_INFO, "last_transno "LPU64", last_committed "LPU64
3319                        ", xid "LPU64"\n",
3320                        mds->mds_last_transno, obd->obd_last_committed,
3321                        req->rq_xid);
3322         }
3323  out:
3324
3325
3326         target_send_reply(req, rc, fail);
3327         RETURN(0);
3328 }
3329
3330 /* Update the server data on disk.  This stores the new mount_count and also the
3331  * last_rcvd value to disk.  If we don't have a clean shutdown, then the server
3332  * last_rcvd value may be less than that of the clients.  This will alert us
3333  * that we may need to do client recovery.
3334  *
3335  * Also assumes for mds_last_transno that we are not modifying it (no locking).
3336  */
3337 int mds_update_server_data(struct obd_device *obd, int force_sync)
3338 {
3339         struct mds_obd *mds = &obd->u.mds;
3340         struct mds_server_data *msd = mds->mds_server_data;
3341         struct file *filp = mds->mds_rcvd_filp;
3342         struct lvfs_run_ctxt saved;
3343         loff_t off = 0;
3344         int rc;
3345         ENTRY;
3346
3347         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3348         msd->msd_last_transno = cpu_to_le64(mds->mds_last_transno);
3349
3350         CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
3351                mds->mds_mount_count, mds->mds_last_transno);
3352         rc = fsfilt_write_record(obd, filp, msd, sizeof(*msd), &off, force_sync);
3353         if (rc)
3354                 CERROR("error writing MDS server data: rc = %d\n", rc);
3355         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3356
3357         RETURN(rc);
3358 }
3359
3360 /* saves last allocated fid counter to file. */
3361 int mds_update_last_fid(struct obd_device *obd, void *handle,
3362                         int force_sync)
3363 {
3364         struct mds_obd *mds = &obd->u.mds;
3365         struct file *filp = mds->mds_fid_filp;
3366         struct lvfs_run_ctxt saved;
3367         loff_t off = 0;
3368         __u64 last_fid;
3369         int rc = 0;
3370         ENTRY;
3371
3372         spin_lock(&mds->mds_last_fid_lock);
3373         last_fid = mds->mds_last_fid;
3374         spin_unlock(&mds->mds_last_fid_lock);
3375
3376         CDEBUG(D_SUPER, "MDS last_fid is #"LPU64"\n",
3377                last_fid);
3378
3379         if (handle) {
3380                 fsfilt_add_journal_cb(obd, mds->mds_sb, last_fid,
3381                                       handle, mds_commit_last_fid_cb,
3382                                       NULL);
3383         }
3384
3385         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3386         rc = fsfilt_write_record(obd, filp, &last_fid, sizeof(last_fid),
3387                                  &off, force_sync);
3388         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3389
3390         if (rc) {
3391                 CERROR("error writing MDS last_fid #"LPU64
3392                        ", err = %d\n", last_fid, rc);
3393                 RETURN(rc);
3394         }
3395
3396         CDEBUG(D_SUPER, "wrote fid #"LPU64" at idx "
3397                "%llu: err = %d\n", last_fid, off, rc);
3398
3399         RETURN(rc);
3400 }
3401
3402 void mds_set_last_fid(struct obd_device *obd, __u64 fid)
3403 {
3404         struct mds_obd *mds = &obd->u.mds;
3405
3406         spin_lock(&mds->mds_last_fid_lock);
3407         if (fid > mds->mds_last_fid)
3408                 mds->mds_last_fid = fid;
3409         spin_unlock(&mds->mds_last_fid_lock);
3410 }
3411
3412 void mds_commit_last_transno_cb(struct obd_device *obd,
3413                                 __u64 transno, void *data,
3414                                 int error)
3415 {
3416         obd_transno_commit_cb(obd, transno, error);
3417 }
3418
3419 void mds_commit_last_fid_cb(struct obd_device *obd,
3420                             __u64 fid, void *data,
3421                             int error)
3422 {
3423         if (error) {
3424                 CERROR("%s: fid "LPD64" commit error: %d\n",
3425                        obd->obd_name, fid, error);
3426                 return;
3427         }
3428
3429         CDEBUG(D_HA, "%s: fid "LPD64" committed\n",
3430                obd->obd_name, fid);
3431 }
3432
3433 __u64 mds_alloc_fid(struct obd_device *obd)
3434 {
3435         struct mds_obd *mds = &obd->u.mds;
3436         __u64 fid;
3437
3438         spin_lock(&mds->mds_last_fid_lock);
3439         fid = ++mds->mds_last_fid;
3440         spin_unlock(&mds->mds_last_fid_lock);
3441
3442         return fid;
3443 }
3444
3445 /*
3446  * reads inode self id from inode EA. Probably later this should be replaced by
3447  * caching inode self id to avoid raeding it every time it is needed.
3448  */
3449 int mds_read_inode_sid(struct obd_device *obd, struct inode *inode,
3450                        struct lustre_id *id)
3451 {
3452         int rc;
3453         ENTRY;
3454
3455         LASSERT(id != NULL);
3456         LASSERT(obd != NULL);
3457         LASSERT(inode != NULL);
3458
3459         rc = fsfilt_get_md(obd, inode, &id->li_fid,
3460                            sizeof(id->li_fid), EA_SID);
3461         if (rc < 0) {
3462                 CERROR("fsfilt_get_md() failed, "
3463                        "rc = %d\n", rc);
3464                 RETURN(rc);
3465         } else if (!rc) {
3466                 rc = -ENODATA;
3467                 RETURN(rc);
3468         } else {
3469                 rc = 0;
3470         }
3471         id_ino(id) = inode->i_ino;
3472         id_gen(id) = inode->i_generation;
3473         id_type(id) = S_IFMT & inode->i_mode;
3474
3475         RETURN(rc);
3476 }
3477
3478 int mds_read_inode_pid(struct obd_device *obd, struct inode *inode,
3479                        struct lustre_id *id)
3480 {
3481         int rc;
3482         ENTRY;
3483
3484         LASSERT(inode && id);
3485
3486         rc = fsfilt_get_md(obd, inode, id, sizeof(*id), EA_PID);
3487         if (rc < 0)
3488                CERROR("get parent id from EA failed, rc=%d\n", rc);
3489         else if (!rc)
3490                 rc = -ENODATA;
3491         else
3492                 rc = 0;
3493
3494         RETURN(rc);
3495 }
3496
3497 /* updates inode self id in EA. */
3498 int mds_update_inode_ids(struct obd_device *obd, struct inode *inode,
3499                          void *handle, struct lustre_id *id,
3500                          struct lustre_id *pid)
3501 {
3502         int rc = 0;
3503         ENTRY;
3504
3505         LASSERT(id || pid);
3506         LASSERT(id == NULL || id_fid(id) != 0);
3507         LASSERT(pid == NULL || id_fid(pid) != 0);
3508         LASSERT(obd != NULL);
3509         LASSERT(inode != NULL);
3510
3511         if (id) {
3512                 mds_set_last_fid(obd, id_fid(id));
3513                 rc = fsfilt_set_md(obd, inode, handle, &id->li_fid,
3514                                    sizeof(id->li_fid), EA_SID);
3515                 LASSERTF(rc == 0, "failed to update fid:  %d\n", rc);
3516         }
3517         if (pid) {
3518                 rc = fsfilt_set_md(obd, inode, handle, pid,
3519                                    sizeof(*pid), EA_PID);
3520                 LASSERTF(rc == 0, "failed to update parent fid:  %d\n", rc);
3521         }
3522
3523         RETURN(rc);
3524 }
3525
3526 /* mount the file system (secretly) */
3527 static int mds_setup(struct obd_device *obd, obd_count len, void *buf)
3528 {
3529         struct lustre_cfg* lcfg = buf;
3530         struct mds_obd *mds = &obd->u.mds;
3531         struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
3532         char *options = NULL;
3533         struct vfsmount *mnt;
3534         char ns_name[48];
3535         unsigned long page;
3536         struct crypto_tfm *tfm = NULL;
3537         int rc = 0;
3538         ENTRY;
3539
3540         if (lcfg->lcfg_bufcount < 3)
3541                 RETURN(rc = -EINVAL);
3542
3543         if (LUSTRE_CFG_BUFLEN(lcfg, 1) == 0 || LUSTRE_CFG_BUFLEN(lcfg, 2) == 0)
3544                 RETURN(rc = -EINVAL);
3545
3546         obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
3547         if (IS_ERR(obd->obd_fsops))
3548                 RETURN(rc = PTR_ERR(obd->obd_fsops));
3549
3550         mds->mds_max_mdsize = sizeof(struct lov_mds_md);
3551
3552         page = get_zeroed_page(GFP_KERNEL);
3553         if (!page)
3554                 RETURN(-ENOMEM);
3555
3556         options = (char *)page;
3557
3558         /*
3559          * here we use "iopen_nopriv" hardcoded, because it affects MDS utility
3560          * and the rest of options are passed by mount options. Probably this
3561          * should be moved to somewhere else like startup scripts or lconf. */
3562         sprintf(options, "iopen_nopriv");
3563
3564         if (LUSTRE_CFG_BUFLEN(lcfg, 4) > 0 && lustre_cfg_buf(lcfg, 4))
3565                 sprintf(options + strlen(options), ",%s",
3566                         lustre_cfg_string(lcfg, 4));
3567
3568         /* we have to know mdsnum before touching underlying fs -bzzz */
3569         atomic_set(&mds->mds_open_count, 0);
3570         sema_init(&mds->mds_md_sem, 1);
3571         mds->mds_md_connected = 0;
3572         mds->mds_md_name = NULL;
3573
3574         if (LUSTRE_CFG_BUFLEN(lcfg, 5) > 0 && lustre_cfg_buf(lcfg, 5) &&
3575             strncmp(lustre_cfg_string(lcfg, 5), "dumb", LUSTRE_CFG_BUFLEN(lcfg, 5))) {
3576                 class_uuid_t uuid;
3577
3578                 generate_random_uuid(uuid);
3579                 class_uuid_unparse(uuid, &mds->mds_md_uuid);
3580
3581                 OBD_ALLOC(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3582                 if (mds->mds_md_name == NULL)
3583                         RETURN(rc = -ENOMEM);
3584
3585                 memcpy(mds->mds_md_name, lustre_cfg_buf(lcfg, 5),
3586                        LUSTRE_CFG_BUFLEN(lcfg, 5));
3587
3588                 CDEBUG(D_OTHER, "MDS: %s is master for %s\n",
3589                        obd->obd_name, mds->mds_md_name);
3590
3591                 rc = mds_md_connect(obd, mds->mds_md_name);
3592                 if (rc) {
3593                         OBD_FREE(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3594                         GOTO(err_ops, rc);
3595                 }
3596         }
3597
3598         mds->mds_obd_type = MDS_MASTER_OBD;
3599
3600         if (LUSTRE_CFG_BUFLEN(lcfg, 6) > 0 && lustre_cfg_buf(lcfg, 6) &&
3601             strncmp(lustre_cfg_string(lcfg, 6), "dumb",
3602                     LUSTRE_CFG_BUFLEN(lcfg, 6))) {
3603                 if (!memcmp(lustre_cfg_string(lcfg, 6), "master",
3604                             strlen("master"))) {
3605                         mds->mds_obd_type = MDS_MASTER_OBD;
3606                 } else if (!memcmp(lustre_cfg_string(lcfg, 6), "cache",
3607                                    strlen("cache"))) {
3608                         mds->mds_obd_type = MDS_CACHE_OBD;
3609                 }
3610         }
3611
3612         rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
3613                            lustre_cfg_string(lcfg, 2),
3614                            options, 0, &lvfs_ctxt);
3615
3616         free_page(page);
3617
3618         if (rc || !lvfs_ctxt) {
3619                 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
3620                 GOTO(err_ops, rc);
3621         }
3622
3623         mnt = lvfs_ctxt->loc_mnt;
3624         mds->mds_lvfs_ctxt = lvfs_ctxt;
3625         ll_clear_rdonly(ll_sbdev(mnt->mnt_sb));
3626
3627         CDEBUG(D_SUPER, "%s: mnt = %p\n", lustre_cfg_string(lcfg, 1), mnt);
3628
3629         mds->mds_fidext_thumb = 0;
3630         sema_init(&mds->mds_epoch_sem, 1);
3631         atomic_set(&mds->mds_real_clients, 0);
3632         spin_lock_init(&mds->mds_fidext_lock);
3633         spin_lock_init(&mds->mds_fidmap_lock);
3634         spin_lock_init(&mds->mds_transno_lock);
3635         spin_lock_init(&mds->mds_last_fid_lock);
3636         sema_init(&mds->mds_orphan_recovery_sem, 1);
3637         mds->mds_max_cookiesize = sizeof(struct llog_cookie);
3638
3639         sprintf(ns_name, "mds-%s", obd->obd_uuid.uuid);
3640         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
3641
3642         if (obd->obd_namespace == NULL) {
3643                 mds_cleanup(obd, 0);
3644                 GOTO(err_put, rc = -ENOMEM);
3645         }
3646         ldlm_register_intent(obd->obd_namespace, mds_intent_policy);
3647
3648         tfm = crypto_alloc_tfm(CAPA_HMAC_ALG, 0);
3649         if (!tfm)
3650                 GOTO(err_ns, rc = -ENOSYS);
3651
3652         mds->mds_capa_hmac = tfm;
3653         mds->mds_capa_timeout = CAPA_TIMEOUT;
3654         mds->mds_capa_key_timeout = CAPA_KEY_TIMEOUT;
3655
3656         rc = mds_fs_setup(obd, mnt);
3657         if (rc) {
3658                 CERROR("%s: MDS filesystem method init failed: rc = %d\n",
3659                        obd->obd_name, rc);
3660                 GOTO(err_capa, rc);
3661         }
3662
3663         rc = llog_start_commit_thread();
3664         if (rc < 0)
3665
3666                 GOTO(err_fs, rc);
3667
3668         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3) &&
3669             strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3670                     LUSTRE_CFG_BUFLEN(lcfg, 3))) {
3671                 class_uuid_t uuid;
3672
3673                 generate_random_uuid(uuid);
3674                 class_uuid_unparse(uuid, &mds->mds_dt_uuid);
3675
3676                 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
3677                 if (mds->mds_profile == NULL)
3678                         GOTO(err_fs, rc = -ENOMEM);
3679
3680                 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
3681                         LUSTRE_CFG_BUFLEN(lcfg, 3));
3682         }
3683
3684         /*
3685          * setup root dir and files ID dir if lmv already connected, or there is
3686          * not lmv at all.
3687          */
3688         if (mds->mds_md_exp || (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 &&
3689                                 lustre_cfg_buf(lcfg, 3) &&
3690                                 strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3691                                         LUSTRE_CFG_BUFLEN(lcfg, 3)))) {
3692                 rc = mds_fs_setup_rootid(obd);
3693                 if (rc)
3694                         GOTO(err_fs, rc);
3695
3696                 rc = mds_fs_setup_virtid(obd);
3697                 if (rc)
3698                         GOTO(err_fs, rc);
3699
3700         }
3701
3702         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3703                            "mds_ldlm_client", &obd->obd_ldlm_client);
3704         obd->obd_replayable = 1;
3705
3706         mds->mds_crypto_type = NO_CRYPTO;
3707
3708         rc = mds_postsetup(obd);
3709         if (rc)
3710                 GOTO(err_fs, rc);
3711
3712         RETURN(0);
3713
3714 err_fs:
3715         /* No extra cleanup needed for llog_init_commit_thread() */
3716         mds_fs_cleanup(obd, 0);
3717 err_capa:
3718         crypto_free_tfm(mds->mds_capa_hmac);
3719 err_ns:
3720         ldlm_namespace_free(obd->obd_namespace, 0);
3721         obd->obd_namespace = NULL;
3722 err_put:
3723         unlock_kernel();
3724         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3725         mds->mds_sb = 0;
3726         lock_kernel();
3727 err_ops:
3728         fsfilt_put_ops(obd->obd_fsops);
3729         return rc;
3730 }
3731
3732 static int mds_fs_post_setup(struct obd_device *obd)
3733 {
3734         struct mds_obd *mds = &obd->u.mds;
3735         struct dentry *dentry;
3736         int rc = 0;
3737         ENTRY;
3738
3739         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
3740         if (IS_ERR(dentry)) {
3741                 CERROR("Can't find ROOT, err = %d\n",
3742                        (int)PTR_ERR(dentry));
3743                 RETURN(PTR_ERR(dentry));
3744         }
3745         rc = fsfilt_post_setup(obd, dentry);
3746         //set id2name function handler
3747         fsfilt_set_info(obd, mds->mds_sb, NULL, 7, "id2name",
3748                         sizeof(mds_audit_id2name), mds_audit_id2name);
3749
3750         l_dput(dentry);
3751         RETURN(rc);
3752 }
3753
3754 static int mds_postsetup(struct obd_device *obd)
3755 {
3756         struct mds_obd *mds = &obd->u.mds;
3757         int rc = 0;
3758         ENTRY;
3759
3760         rc = obd_llog_setup(obd, &obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT,
3761                             obd, 0, NULL, &llog_lvfs_ops);
3762         if (rc)
3763                 RETURN(rc);
3764
3765         if (mds->mds_profile) {
3766                 struct llog_ctxt *lgctxt;
3767                 struct lvfs_run_ctxt saved;
3768                 struct lustre_profile *lprof;
3769                 struct config_llog_instance cfg;
3770
3771                 cfg.cfg_instance = NULL;
3772                 cfg.cfg_uuid = mds->mds_dt_uuid;
3773                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3774
3775                 lgctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
3776                 if (!lgctxt) {
3777                         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3778                         GOTO(err_llog, rc = -EINVAL);
3779                 }
3780
3781                 rc = class_config_process_llog(lgctxt, mds->mds_profile, &cfg);
3782                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3783
3784                 if (rc)
3785                         GOTO(err_llog, rc);
3786
3787                 lprof = class_get_profile(mds->mds_profile);
3788                 if (lprof == NULL) {
3789                         CERROR("No profile found: %s\n", mds->mds_profile);
3790                         GOTO(err_cleanup, rc = -ENOENT);
3791                 }
3792                 rc = mds_dt_connect(obd, lprof->lp_lov);
3793                 if (rc)
3794                         GOTO(err_cleanup, rc);
3795
3796                 rc = mds_md_postsetup(obd);
3797                 if (rc)
3798                         GOTO(err_cleanup, rc);
3799         }
3800         rc = mds_fs_post_setup(obd);
3801         if (rc)
3802                 CERROR("can not post setup fsfilt\n");
3803         RETURN(rc);
3804 err_cleanup:
3805         mds_dt_clean(obd);
3806 err_llog:
3807         obd_llog_cleanup(llog_get_context(&obd->obd_llogs,
3808                                           LLOG_CONFIG_ORIG_CTXT));
3809         return rc;
3810 }
3811
3812 int mds_postrecov_common(struct obd_device *obd)
3813 {
3814         struct mds_obd *mds = &obd->u.mds;
3815         struct llog_ctxt *ctxt;
3816         int rc, item = 0, valsize;
3817         struct timeval tstart, now;
3818          __u32 group;
3819         ENTRY;
3820
3821         CDEBUG(D_ERROR, "%s: post-recovery\n", obd->obd_name);
3822         LASSERT(!obd->obd_recovering);
3823         ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT);
3824         LASSERT(ctxt != NULL);
3825
3826         /* clean PENDING dir */
3827         do_gettimeofday(&tstart);
3828         rc = mds_cleanup_orphans(obd);
3829         do_gettimeofday(&now);
3830         CDEBUG(D_ERROR, "%s: cleaned %d MDS orphans in %us\n",
3831                obd->obd_name, rc, (unsigned) (now.tv_sec - tstart.tv_sec));
3832         if (rc < 0)
3833                 GOTO(out, rc);
3834         item = rc;
3835
3836         group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
3837         valsize = sizeof(group);
3838         rc = obd_set_info(mds->mds_dt_exp, strlen("mds_conn"),
3839                           "mds_conn", valsize, &group);
3840         if (rc) {
3841                 CERROR("%s: failed to obd_set_info(): %d\n",
3842                        obd->obd_name, rc);
3843                 GOTO(out, rc);
3844         }
3845
3846         rc = llog_connect(ctxt, obd->u.mds.mds_dt_desc.ld_tgt_count,
3847                           NULL, NULL, NULL);
3848         if (rc) {
3849                 CERROR("%s: failed at llog_origin_connect: %d\n",
3850                        obd->obd_name, rc);
3851                 GOTO(out, rc);
3852         }
3853
3854         /* remove the orphaned precreated objects */
3855         do_gettimeofday(&tstart);
3856         rc = mds_dt_clear_orphans(mds, NULL /* all OSTs */);
3857         do_gettimeofday(&now);
3858         CDEBUG(D_ERROR, "%s: cleaned OSS orphans in %us: %d\n",
3859                obd->obd_name, (unsigned) (now.tv_sec - tstart.tv_sec), rc);
3860         if (rc) {
3861                 CERROR("%s: can't clear OSS orphans: %d\n",
3862                        obd->obd_name, rc);
3863                 GOTO(err_llog, rc);
3864         }
3865
3866 out:
3867         RETURN(rc < 0 ? rc : item);
3868
3869 err_llog:
3870         /* cleanup all llogging subsystems */
3871         rc = obd_llog_finish(obd, &obd->obd_llogs,
3872                              mds->mds_dt_desc.ld_tgt_count);
3873         if (rc)
3874                 CERROR("%s: failed to cleanup llogging subsystems\n",
3875                         obd->obd_name);
3876         goto out;
3877 }
3878
3879 int mds_postrecov(struct obd_device *obd)
3880 {
3881         int rc;
3882         ENTRY;
3883         rc = mds_postrecov_common(obd);
3884         if (rc >= 0)
3885                 rc = mds_md_reconnect(obd);
3886         RETURN(rc);
3887 }
3888
3889 int mds_dt_clean(struct obd_device *obd)
3890 {
3891         struct mds_obd *mds = &obd->u.mds;
3892         ENTRY;
3893
3894         if (mds->mds_profile) {
3895                 char * cln_prof;
3896                 struct llog_ctxt *llctx;
3897                 struct lvfs_run_ctxt saved;
3898                 struct config_llog_instance cfg;
3899                 int len = strlen(mds->mds_profile) + sizeof("-clean") + 1;
3900
3901                 OBD_ALLOC(cln_prof, len);
3902                 if (!cln_prof) {
3903                         CERROR("can't allocate memory, processing cleanup "
3904                                "profile is skipped\n");
3905                         goto out;
3906                 }
3907
3908                 sprintf(cln_prof, "%s-clean", mds->mds_profile);
3909
3910                 cfg.cfg_instance = NULL;
3911                 cfg.cfg_uuid = mds->mds_dt_uuid;
3912
3913                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3914                 llctx = llog_get_context(&obd->obd_llogs,
3915                                          LLOG_CONFIG_ORIG_CTXT);
3916                 class_config_process_llog(llctx, cln_prof, &cfg);
3917                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3918
3919                 OBD_FREE(cln_prof, len);
3920         out:
3921                 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
3922                 mds->mds_profile = NULL;
3923         }
3924         RETURN(0);
3925 }
3926
3927 int mds_md_clean(struct obd_device *obd)
3928 {
3929         struct mds_obd *mds = &obd->u.mds;
3930         ENTRY;
3931
3932         if (mds->mds_md_name) {
3933                 OBD_FREE(mds->mds_md_name, strlen(mds->mds_md_name) + 1);
3934                 mds->mds_md_name = NULL;
3935         }
3936         RETURN(0);
3937 }
3938
3939 static int mds_precleanup(struct obd_device *obd, int flags)
3940 {
3941         int rc = 0;
3942         ENTRY;
3943
3944         mds_md_clean(obd);
3945         mds_dt_disconnect(obd, flags);
3946         mds_dt_clean(obd);
3947         obd_llog_cleanup(llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT));
3948         RETURN(rc);
3949 }
3950
3951 extern void lgss_svc_cache_purge_all(void);
3952 static int mds_cleanup(struct obd_device *obd, int flags)
3953 {
3954         struct mds_obd *mds = &obd->u.mds;
3955         ENTRY;
3956
3957         if (mds->mds_sb == NULL)
3958                 RETURN(0);
3959
3960         mds_update_server_data(obd, 1);
3961         mds_update_last_fid(obd, NULL, 1);
3962
3963         if (mds->mds_dt_objids != NULL) {
3964                 int size = mds->mds_dt_desc.ld_tgt_count *
3965                         sizeof(obd_id);
3966                 OBD_FREE(mds->mds_dt_objids, size);
3967         }
3968         mds_fs_cleanup(obd, flags);
3969
3970         unlock_kernel();
3971
3972         /* 2 seems normal on mds, (may_umount() also expects 2
3973           fwiw), but we only see 1 at this point in obdfilter. */
3974         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3975
3976         mds->mds_sb = 0;
3977
3978         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
3979
3980         spin_lock_bh(&obd->obd_processing_task_lock);
3981         if (obd->obd_recovering) {
3982                 target_cancel_recovery_timer(obd);
3983                 obd->obd_recovering = 0;
3984         }
3985         spin_unlock_bh(&obd->obd_processing_task_lock);
3986
3987         lock_kernel();
3988         fsfilt_put_ops(obd->obd_fsops);
3989
3990 #ifdef ENABLE_GSS
3991         /* XXX */
3992         lgss_svc_cache_purge_all();
3993 #endif
3994
3995         spin_lock(&mds->mds_denylist_lock);
3996         while (!list_empty( &mds->mds_denylist ) ) {
3997                 deny_sec_t *p_deny_sec = list_entry(mds->mds_denylist.next,
3998                                                     deny_sec_t, list);
3999                 list_del(&p_deny_sec->list);
4000                 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
4001         }
4002         spin_unlock(&mds->mds_denylist_lock);
4003
4004         mds_capa_keys_cleanup(obd);
4005
4006         if (mds->mds_capa_hmac)
4007                 crypto_free_tfm(mds->mds_capa_hmac);
4008         RETURN(0);
4009 }
4010
4011 static int set_security(const char *value, char **sec)
4012 {
4013         if (!strcmp(value, "null"))
4014                 *sec = "null";
4015         else if (!strcmp(value, "krb5i"))
4016                 *sec = "krb5i";
4017         else if (!strcmp(value, "krb5p"))
4018                 *sec = "krb5p";
4019         else {
4020                 CERROR("Unrecognized security flavor %s\n", value);
4021                 return -EINVAL;
4022         }
4023
4024         return 0;
4025 }
4026
4027 static int mds_process_config(struct obd_device *obd, obd_count len, void *buf)
4028 {
4029         struct lustre_cfg *lcfg = buf;
4030         struct mds_obd *mds = &obd->u.mds;
4031         int rc = 0;
4032         ENTRY;
4033
4034         switch(lcfg->lcfg_command) {
4035         case LCFG_SET_SECURITY: {
4036                 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
4037                     (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
4038                         GOTO(out, rc = -EINVAL);
4039
4040                 if (!strcmp(lustre_cfg_string(lcfg, 1), "mds_sec"))
4041                         rc = set_security(lustre_cfg_string(lcfg, 2),
4042                                           &mds->mds_mds_sec);
4043                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "oss_sec"))
4044                         rc = set_security(lustre_cfg_string(lcfg, 2),
4045                                           &mds->mds_ost_sec);
4046                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
4047                         spin_lock(&mds->mds_denylist_lock);
4048                         rc = add_deny_security(lustre_cfg_string(lcfg, 2),
4049                                                &mds->mds_denylist);
4050                         spin_unlock(&mds->mds_denylist_lock);
4051                 } else {
4052                         CERROR("Unrecognized key\n");
4053                         rc = -EINVAL;
4054                 }
4055                 break;
4056         }
4057         default:
4058                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
4059                 GOTO(out, rc = -EINVAL);
4060         }
4061 out:
4062         RETURN(rc);
4063 }
4064
4065 static void fixup_handle_for_resent_req(struct ptlrpc_request *req,
4066                                         int offset,
4067                                         struct ldlm_lock *new_lock,
4068                                         struct ldlm_lock **old_lock,
4069                                         struct lustre_handle *lockh)
4070 {
4071         struct obd_export *exp = req->rq_export;
4072         struct obd_device *obd = exp->exp_obd;
4073         struct ldlm_request *dlmreq =
4074                 lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*dlmreq));
4075         struct lustre_handle remote_hdl = dlmreq->lock_handle1;
4076         struct list_head *iter;
4077
4078         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4079                 return;
4080
4081         spin_lock(&obd->obd_namespace->ns_hash_lock);
4082         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
4083                 struct ldlm_lock *lock;
4084                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
4085                 if (lock == new_lock)
4086                         continue;
4087                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
4088                         lockh->cookie = lock->l_handle.h_cookie;
4089                         LDLM_DEBUG(lock, "restoring lock cookie");
4090                         DEBUG_REQ(D_HA, req, "restoring lock cookie "LPX64,
4091                                   lockh->cookie);
4092                         if (old_lock)
4093                                 *old_lock = LDLM_LOCK_GET(lock);
4094                         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4095                         return;
4096                 }
4097         }
4098         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4099
4100         /* If the xid matches, then we know this is a resent request,
4101          * and allow it. (It's probably an OPEN, for which we don't
4102          * send a lock */
4103         if (req->rq_xid ==
4104             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_xid))
4105                 return;
4106
4107         if (req->rq_xid ==
4108             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_close_xid))
4109                 return;
4110
4111         /* This remote handle isn't enqueued, so we never received or
4112          * processed this request.  Clear MSG_RESENT, because it can
4113          * be handled like any normal request now. */
4114
4115         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4116
4117         DEBUG_REQ(D_HA, req, "no existing lock with rhandle "LPX64,
4118                   remote_hdl.cookie);
4119 }
4120
4121 int intent_disposition(struct ldlm_reply *rep, int flag)
4122 {
4123         if (!rep)
4124                 return 0;
4125         return (rep->lock_policy_res1 & flag);
4126 }
4127
4128 void intent_set_disposition(struct ldlm_reply *rep, int flag)
4129 {
4130         if (!rep)
4131                 return;
4132         rep->lock_policy_res1 |= flag;
4133 }
4134
4135 static int mds_intent_prepare_reply_buffers(struct ptlrpc_request *req,
4136                                             struct ldlm_intent *it)
4137 {
4138         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
4139         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
4140         int rc, reply_buffers;
4141         int repsize[8] = {sizeof(struct ldlm_reply),
4142                           sizeof(struct mds_body),
4143                           mds->mds_max_mdsize};
4144         ENTRY;
4145
4146         reply_buffers = 3;
4147         if (it->opc & ( IT_OPEN | IT_GETATTR | IT_LOOKUP | IT_CHDIR )) {
4148                 /*Ugly here, Actually, we should prepare the reply buffer
4149                  *after we know whether these stuff exist or not, which should
4150                  * be fixed in future, Now each item is in the fix position,
4151                  * the sequence is lsm, acl, crypto ea, capa.*/
4152                 repsize[reply_buffers++] = sizeof(int);
4153                 if (med->med_remote)
4154                         repsize[reply_buffers++] =
4155                                 sizeof(struct mds_remote_perm);
4156                 else
4157                         repsize[reply_buffers++] =
4158                                 xattr_acl_size(LL_ACL_MAX_ENTRIES);
4159
4160                 repsize[reply_buffers++] = sizeof(int);
4161                 repsize[reply_buffers++] = sizeof(struct crypto_key);
4162
4163                 /* XXX: if new buffer is to be added, capability reply
4164                  *      buffer should always been reserved. */
4165                 if (it->opc & IT_OPEN)
4166                         repsize[reply_buffers++] = sizeof(struct lustre_capa);
4167         }
4168
4169         rc = lustre_pack_reply(req, reply_buffers, repsize, NULL);
4170
4171         RETURN(rc);
4172 }
4173
4174 static int mds_intent_policy(struct ldlm_namespace *ns,
4175                              struct ldlm_lock **lockp, void *req_cookie,
4176                              ldlm_mode_t mode, int flags, void *data)
4177 {
4178         struct ptlrpc_request *req = req_cookie;
4179         struct ldlm_lock *lock = *lockp;
4180         struct ldlm_intent *it;
4181         struct ldlm_reply *rep;
4182         struct lustre_handle lockh[2] = {{0}, {0}};
4183         struct ldlm_lock *new_lock = NULL;
4184         int getattr_part = MDS_INODELOCK_UPDATE;
4185         int rc;
4186
4187         int offset = MDS_REQ_INTENT_REC_OFF;
4188         ENTRY;
4189
4190         LASSERT(req != NULL);
4191         MD_COUNTER_INCREMENT(req->rq_export->exp_obd, intent_lock);
4192
4193         if (req->rq_reqmsg->bufcount <= MDS_REQ_INTENT_IT_OFF) {
4194                 /* No intent was provided */
4195                 int size = sizeof(struct ldlm_reply);
4196                 rc = lustre_pack_reply(req, 1, &size, NULL);
4197                 LASSERT(rc == 0);
4198                 RETURN(0);
4199         }
4200
4201         it = lustre_swab_reqbuf(req, MDS_REQ_INTENT_IT_OFF, sizeof(*it),
4202                                 lustre_swab_ldlm_intent);
4203         if (it == NULL) {
4204                 CERROR("Intent missing\n");
4205                 RETURN(req->rq_status = -EFAULT);
4206         }
4207
4208         LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
4209
4210         rc = mds_intent_prepare_reply_buffers(req, it);
4211
4212         if (rc)
4213                 RETURN(req->rq_status = rc);
4214
4215         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
4216         LASSERT(rep != NULL);
4217
4218         intent_set_disposition(rep, DISP_IT_EXECD);
4219
4220         /* execute policy */
4221         switch ((long)it->opc) {
4222         case IT_OPEN:
4223         case IT_CREAT|IT_OPEN:
4224                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4225                                             lock, NULL, lockh);
4226                 /* XXX swab here to assert that an mds_open reint
4227                  * packet is following */
4228                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4229                                             lock, NULL, lockh);
4230                 rep->lock_policy_res2 = mds_reint(req, offset, lockh);
4231
4232                 if (rep->lock_policy_res2) {
4233                         /*
4234                          * mds_open() returns ENOLCK where it should return
4235                          * zero, but it has no lock to return.
4236                          */
4237                         if (rep->lock_policy_res2 == ENOLCK)
4238                                 rep->lock_policy_res2 = 0;
4239
4240                         RETURN(ELDLM_LOCK_ABORTED);
4241                 }
4242
4243                 /*
4244                  * IT_OPEN may return lock on cross-node dentry that we want to
4245                  * hold during attr retrival -bzzz
4246                  */
4247                 if (lockh[0].cookie == 0)
4248                         RETURN(ELDLM_LOCK_ABORTED);
4249
4250                 break;
4251         case IT_LOOKUP:
4252                 getattr_part = MDS_INODELOCK_LOOKUP;
4253         case IT_CHDIR:
4254         case IT_GETATTR:
4255                 getattr_part |= MDS_INODELOCK_LOOKUP;
4256         case IT_READDIR:
4257                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4258                                             lock, &new_lock, lockh);
4259                 rep->lock_policy_res2 = mds_getattr_lock(req, offset, lockh,
4260                                                          getattr_part);
4261                 /* FIXME: LDLM can set req->rq_status. MDS sets
4262                    policy_res{1,2} with disposition and status.
4263                    - replay: returns 0 & req->status is old status
4264                    - otherwise: returns req->status */
4265                 if (intent_disposition(rep, DISP_LOOKUP_NEG))
4266                         rep->lock_policy_res2 = 0;
4267                 if (!intent_disposition(rep, DISP_LOOKUP_POS) ||
4268                     rep->lock_policy_res2)
4269                         RETURN(ELDLM_LOCK_ABORTED);
4270                 if (req->rq_status != 0) {
4271                         LBUG();
4272                         rep->lock_policy_res2 = req->rq_status;
4273                         RETURN(ELDLM_LOCK_ABORTED);
4274                 }
4275                 break;
4276         case IT_UNLINK:
4277                 rc = mds_lock_and_check_slave(offset, req, lockh);
4278                 if ((rep->lock_policy_res2 = rc)) {
4279                         if (rc == ENOLCK)
4280                                 rep->lock_policy_res2 = 0;
4281                         RETURN(ELDLM_LOCK_ABORTED);
4282                 }
4283                 break;
4284         default:
4285                 CERROR("Unhandled intent "LPD64"\n", it->opc);
4286                 LBUG();
4287         }
4288
4289         /* By this point, whatever function we called above must have either
4290          * filled in 'lockh', been an intent replay, or returned an error.  We
4291          * want to allow replayed RPCs to not get a lock, since we would just
4292          * drop it below anyways because lock replay is done separately by the
4293          * client afterwards.  For regular RPCs we want to give the new lock to
4294          * the client instead of whatever lock it was about to get. */
4295         if (new_lock == NULL)
4296                 new_lock = ldlm_handle2lock(&lockh[0]);
4297         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY))
4298                 RETURN(0);
4299
4300         LASSERTF(new_lock != NULL, "op "LPX64" lockh "LPX64"\n",
4301                  it->opc, lockh[0].cookie);
4302
4303         /* If we've already given this lock to a client once, then we should
4304          * have no readers or writers.  Otherwise, we should have one reader
4305          * _or_ writer ref (which will be zeroed below) before returning the
4306          * lock to a client. */
4307         if (new_lock->l_export == req->rq_export) {
4308                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4309         } else {
4310                 LASSERT(new_lock->l_export == NULL);
4311                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4312         }
4313
4314         *lockp = new_lock;
4315
4316         if (new_lock->l_export == req->rq_export) {
4317                 /* Already gave this to the client, which means that we
4318                  * reconstructed a reply. */
4319                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4320                         MSG_RESENT);
4321                 RETURN(ELDLM_LOCK_REPLACED);
4322         }
4323
4324         /* Fixup the lock to be given to the client */
4325         lock_res_and_lock(new_lock);
4326         new_lock->l_readers = 0;
4327         new_lock->l_writers = 0;
4328
4329         new_lock->l_export = class_export_get(req->rq_export);
4330
4331         spin_lock(&new_lock->l_export->exp_ldlm_data.led_lock);
4332         list_add(&new_lock->l_export_chain,
4333                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
4334         spin_unlock(&new_lock->l_export->exp_ldlm_data.led_lock);
4335
4336         new_lock->l_blocking_ast = lock->l_blocking_ast;
4337         new_lock->l_completion_ast = lock->l_completion_ast;
4338
4339         memcpy(&new_lock->l_remote_handle, &lock->l_remote_handle,
4340                sizeof(lock->l_remote_handle));
4341
4342         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4343
4344         unlock_res_and_lock(new_lock);
4345         LDLM_LOCK_PUT(new_lock);
4346
4347         RETURN(ELDLM_LOCK_REPLACED);
4348 }
4349
4350 int mds_attach(struct obd_device *dev, obd_count len, void *data)
4351 {
4352         struct lprocfs_static_vars lvars;
4353         int rc = 0;
4354         struct mds_obd *mds = &dev->u.mds;
4355
4356         spin_lock_init(&mds->mds_denylist_lock);
4357         INIT_LIST_HEAD(&mds->mds_denylist);
4358
4359         lprocfs_init_multi_vars(0, &lvars);
4360
4361         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
4362         if (rc)
4363                 return rc;
4364
4365         return lprocfs_alloc_md_stats(dev, 0);
4366 }
4367
4368 int mds_detach(struct obd_device *dev)
4369 {
4370         lprocfs_free_md_stats(dev);
4371         return lprocfs_obd_detach(dev);
4372 }
4373
4374 int mdt_attach(struct obd_device *dev, obd_count len, void *data)
4375 {
4376         struct lprocfs_static_vars lvars;
4377
4378         lprocfs_init_multi_vars(1, &lvars);
4379         return lprocfs_obd_attach(dev, lvars.obd_vars);
4380 }
4381
4382 int mdt_detach(struct obd_device *dev)
4383 {
4384         return lprocfs_obd_detach(dev);
4385 }
4386
4387 static int mdt_setup(struct obd_device *obd, obd_count len, void *buf)
4388 {
4389         struct mds_obd *mds = &obd->u.mds;
4390         int rc = 0;
4391         ENTRY;
4392
4393         mds->mds_service =
4394                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4395                                 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
4396                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4397                                 mds_handle, "mds", obd->obd_proc_entry);
4398
4399         if (!mds->mds_service) {
4400                 CERROR("failed to start service\n");
4401                 RETURN(-ENOMEM);
4402         }
4403
4404         rc = ptlrpc_start_n_threads(obd, mds->mds_service, MDT_NUM_THREADS,
4405                                     "ll_mdt");
4406         if (rc)
4407                 GOTO(err_thread, rc);
4408
4409         mds->mds_setattr_service =
4410                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4411                                 MDS_SETATTR_PORTAL, MDC_REPLY_PORTAL,
4412                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4413                                 mds_handle, "mds_setattr",
4414                                 obd->obd_proc_entry);
4415         if (!mds->mds_setattr_service) {
4416                 CERROR("failed to start getattr service\n");
4417                 GOTO(err_thread, rc = -ENOMEM);
4418         }
4419
4420         rc = ptlrpc_start_n_threads(obd, mds->mds_setattr_service,
4421                                     MDT_NUM_THREADS, "ll_mdt_attr");
4422         if (rc)
4423                 GOTO(err_thread2, rc);
4424
4425         mds->mds_readpage_service =
4426                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4427                                 MDS_READPAGE_PORTAL, MDC_REPLY_PORTAL,
4428                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4429                                 mds_handle, "mds_readpage",
4430                                 obd->obd_proc_entry);
4431         if (!mds->mds_readpage_service) {
4432                 CERROR("failed to start readpage service\n");
4433                 GOTO(err_thread2, rc = -ENOMEM);
4434         }
4435
4436         rc = ptlrpc_start_n_threads(obd, mds->mds_readpage_service,
4437                                     MDT_NUM_THREADS, "ll_mdt_rdpg");
4438
4439         if (rc)
4440                 GOTO(err_thread3, rc);
4441
4442         mds->mds_close_service =
4443                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4444                                 MDS_CLOSE_PORTAL, MDC_REPLY_PORTAL,
4445                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4446                                 mds_handle, "mds_close",
4447                                 obd->obd_proc_entry);
4448         if (!mds->mds_close_service) {
4449                 CERROR("failed to start close service\n");
4450                 GOTO(err_thread3, rc = -ENOMEM);
4451         }
4452
4453         rc = ptlrpc_start_n_threads(obd, mds->mds_close_service,
4454                                     MDT_NUM_THREADS, "ll_mdt_clos");
4455
4456         if (rc)
4457                 GOTO(err_thread4, rc);
4458         RETURN(0);
4459
4460 err_thread4:
4461         ptlrpc_unregister_service(mds->mds_close_service);
4462 err_thread3:
4463         ptlrpc_unregister_service(mds->mds_readpage_service);
4464 err_thread2:
4465         ptlrpc_unregister_service(mds->mds_setattr_service);
4466 err_thread:
4467         ptlrpc_unregister_service(mds->mds_service);
4468         return rc;
4469 }
4470
4471 static int mdt_cleanup(struct obd_device *obd, int flags)
4472 {
4473         struct mds_obd *mds = &obd->u.mds;
4474         ENTRY;
4475
4476         ptlrpc_stop_all_threads(mds->mds_close_service);
4477         ptlrpc_unregister_service(mds->mds_close_service);
4478
4479         ptlrpc_stop_all_threads(mds->mds_readpage_service);
4480         ptlrpc_unregister_service(mds->mds_readpage_service);
4481
4482         ptlrpc_stop_all_threads(mds->mds_setattr_service);
4483         ptlrpc_unregister_service(mds->mds_setattr_service);
4484
4485         ptlrpc_stop_all_threads(mds->mds_service);
4486         ptlrpc_unregister_service(mds->mds_service);
4487
4488         RETURN(0);
4489 }
4490
4491 static struct dentry *mds_lvfs_id2dentry(__u64 ino, __u32 gen,
4492                                          __u64 gr, void *data)
4493 {
4494         struct lustre_id id;
4495         struct obd_device *obd = data;
4496
4497         id_ino(&id) = ino;
4498         id_gen(&id) = gen;
4499         return mds_id2dentry(obd, &id, NULL);
4500 }
4501
4502 static int mds_get_info(struct obd_export *exp, __u32 keylen,
4503                         void *key, __u32 *valsize, void *val)
4504 {
4505         struct obd_device *obd;
4506         struct mds_obd *mds;
4507         int rc = 0;
4508         ENTRY;
4509
4510         obd = class_exp2obd(exp);
4511         mds = &obd->u.mds;
4512
4513         if (obd == NULL) {
4514                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
4515                        exp->exp_handle.h_cookie);
4516                 RETURN(-EINVAL);
4517         }
4518
4519         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
4520                 /* get log_context handle. */
4521                 struct llog_ctxt *ctxt;
4522                 unsigned long *llh_handle = val;
4523                 *valsize = sizeof(unsigned long);
4524                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_REINT_ORIG_CTXT);
4525                 if (!ctxt) {
4526                         CERROR("Cannot get REINT llog context\n");
4527                         RETURN(-ENOENT);
4528                 }
4529                 *llh_handle = (unsigned long)ctxt;
4530                 RETURN(0);
4531         }
4532         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
4533                 /* get log_context handle. */
4534                 unsigned long *sb = val;
4535                 *valsize = sizeof(unsigned long);
4536                 *sb = (unsigned long)obd->u.mds.mds_sb;
4537                 RETURN(0);
4538         }
4539
4540         if (keylen >= strlen("mdsize") && memcmp(key, "mdsize", keylen) == 0) {
4541                 __u32 *mdsize = val;
4542                 *valsize = sizeof(*mdsize);
4543                 *mdsize = mds->mds_max_mdsize;
4544                 RETURN(0);
4545         }
4546
4547         if (keylen >= strlen("mdsnum") && strcmp(key, "mdsnum") == 0) {
4548                 __u32 *mdsnum = val;
4549                 *valsize = sizeof(*mdsnum);
4550                 *mdsnum = mds->mds_num;
4551                 RETURN(0);
4552         }
4553
4554         if (keylen >= strlen("rootid") && strcmp(key, "rootid") == 0) {
4555                 struct lustre_id *rootid = val;
4556                 *valsize = sizeof(*rootid);
4557                 *rootid = mds->mds_rootid;
4558                 RETURN(0);
4559         }
4560
4561         if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
4562                 struct lov_desc *desc = val;
4563                 *valsize = sizeof(*desc);
4564                 *desc = mds->mds_dt_desc;
4565                 RETURN(0);
4566         }
4567
4568         if (keylen >= strlen("getext") && strcmp(key, "getext") == 0) {
4569                 struct fid_extent *ext = val;
4570                 *valsize = sizeof(*ext);
4571
4572                 spin_lock(&mds->mds_fidext_lock);
4573                 ext->fe_width = MDS_FIDEXT_SIZE;
4574                 ext->fe_start = mds->mds_fidext_thumb + 1;
4575                 mds->mds_fidext_thumb += MDS_FIDEXT_SIZE;
4576                 spin_unlock(&mds->mds_fidext_lock);
4577
4578                 RETURN(0);
4579         }
4580
4581         rc = fsfilt_get_info(obd, mds->mds_sb, NULL, keylen, key, valsize, val);
4582         if (rc)
4583                 CDEBUG(D_IOCTL, "invalid key\n");
4584
4585         RETURN(rc);
4586 }
4587
4588 struct lvfs_callback_ops mds_lvfs_ops = {
4589         l_id2dentry:     mds_lvfs_id2dentry,
4590 };
4591
4592 int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
4593                 int objcount, struct obd_ioobj *obj,
4594                 int niocount, struct niobuf_remote *nb,
4595                 struct niobuf_local *res,
4596                 struct obd_trans_info *oti, struct lustre_capa *capa);
4597
4598 int mds_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
4599                  int objcount, struct obd_ioobj *obj, int niocount,
4600                  struct niobuf_local *res, struct obd_trans_info *oti,
4601                  int rc);
4602
4603 /* use obd ops to offer management infrastructure */
4604 static struct obd_ops mds_obd_ops = {
4605         .o_owner           = THIS_MODULE,
4606         .o_attach          = mds_attach,
4607         .o_detach          = mds_detach,
4608         .o_connect         = mds_connect,
4609         .o_connect_post    = mds_connect_post,
4610         .o_init_export     = mds_init_export,
4611         .o_destroy_export  = mds_destroy_export,
4612         .o_disconnect      = mds_disconnect,
4613         .o_setup           = mds_setup,
4614         .o_precleanup      = mds_precleanup,
4615         .o_cleanup         = mds_cleanup,
4616         .o_process_config  = mds_process_config,
4617         .o_postrecov       = mds_postrecov,
4618         .o_statfs          = mds_obd_statfs,
4619         .o_iocontrol       = mds_iocontrol,
4620         .o_create          = mds_obd_create,
4621         .o_destroy         = mds_obd_destroy,
4622         .o_llog_init       = mds_llog_init,
4623         .o_llog_finish     = mds_llog_finish,
4624         .o_notify          = mds_notify,
4625         .o_get_info        = mds_get_info,
4626         .o_set_info        = mds_set_info,
4627         .o_preprw          = mds_preprw,
4628         .o_commitrw        = mds_commitrw,
4629 };
4630
4631 static struct obd_ops mdt_obd_ops = {
4632         .o_owner           = THIS_MODULE,
4633         .o_attach          = mdt_attach,
4634         .o_detach          = mdt_detach,
4635         .o_setup           = mdt_setup,
4636         .o_cleanup         = mdt_cleanup,
4637 };
4638
4639 static int __init mds_init(void)
4640 {
4641         struct lprocfs_static_vars lvars;
4642         int rc = 0;
4643         ENTRY;
4644
4645         mds_init_lsd_cache();
4646         mds_init_rmtacl_upcall_cache();
4647
4648         lprocfs_init_multi_vars(0, &lvars);
4649         class_register_type(&mds_obd_ops, NULL, lvars.module_vars,
4650                             OBD_MDS_DEVICENAME);
4651         lprocfs_init_multi_vars(1, &lvars);
4652         class_register_type(&mdt_obd_ops, NULL, lvars.module_vars,
4653                             OBD_MDT_DEVICENAME);
4654
4655         rc = mds_capa_key_start_thread();
4656         if (rc) {
4657                 class_unregister_type(OBD_MDT_DEVICENAME);
4658                 class_unregister_type(OBD_MDS_DEVICENAME);
4659                 mds_cleanup_lsd_cache();
4660         }
4661
4662         mds_eck_timer.function = mds_capa_key_timer_callback;
4663         mds_eck_timer.data = 0;
4664         init_timer(&mds_eck_timer);
4665
4666         RETURN(rc);
4667 }
4668
4669 static void /*__exit*/ mds_exit(void)
4670 {
4671         mds_capa_key_stop_thread();
4672         mds_cleanup_rmtacl_upcall_cache();
4673         mds_cleanup_lsd_cache();
4674
4675         class_unregister_type(OBD_MDS_DEVICENAME);
4676         class_unregister_type(OBD_MDT_DEVICENAME);
4677 }
4678
4679 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4680 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
4681 MODULE_LICENSE("GPL");
4682
4683 module_init(mds_init);
4684 module_exit(mds_exit);