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         if (obd->obd_recovering) {
1466                 CDEBUG(D_INODE, "size for "DLID4" is unknown yet (recovering)\n",
1467                        OLID4(&body->id1));
1468                 RETURN(0);
1469         }
1470
1471         if (atomic_read(&inode->i_writecount)) {
1472                 /* some one has opened the file for write.
1473                  * mds doesn't know actual size */
1474                 CDEBUG(D_INODE, "MDS doesn't know actual size for "DLID4"\n",
1475                        OLID4(&body->id1));
1476                 RETURN(0);
1477         }
1478         CDEBUG(D_INODE, "MDS returns "LPD64"/"LPD64" for"DLID4"\n",
1479                body->size, body->blocks, OLID4(&body->id1));
1480         body->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
1481         RETURN(0);
1482 }
1483
1484 static int mds_getattr_lock(struct ptlrpc_request *req, int offset,
1485                             struct lustre_handle *child_lockh, int child_part)
1486 {
1487         struct obd_device *obd = req->rq_export->exp_obd;
1488         struct mds_obd *mds = &obd->u.mds;
1489         struct ldlm_reply *rep = NULL;
1490         struct lvfs_run_ctxt saved;
1491         struct mds_req_sec_desc *rsd;
1492         struct mds_body *body;
1493         struct dentry *dparent = NULL, *dchild = NULL;
1494         struct lvfs_ucred uc = {NULL, NULL,};
1495         struct lustre_handle parent_lockh[2] = {{0}, {0}};
1496         unsigned int namesize = 0;
1497         int rc = 0, cleanup_phase = 0, resent_req = 0, update_mode, reply_offset;
1498         char *name = NULL;
1499         ENTRY;
1500
1501         LASSERT(!strcmp(obd->obd_type->typ_name, OBD_MDS_DEVICENAME));
1502         MD_COUNTER_INCREMENT(obd, getattr_lock);
1503
1504         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1505         if (!rsd) {
1506                 CERROR("Can't unpack security desc\n");
1507                 RETURN(-EFAULT);
1508         }
1509
1510         /* swab now, before anyone looks inside the request. */
1511         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1512                                   lustre_swab_mds_body);
1513         if (body == NULL) {
1514                 CERROR("Can't swab mds_body\n");
1515                 GOTO(cleanup, rc = -EFAULT);
1516         }
1517
1518         LASSERT_REQSWAB(req, offset + 1);
1519         name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
1520         if (name == NULL) {
1521                 CERROR("Can't unpack name\n");
1522                 GOTO(cleanup, rc = -EFAULT);
1523         }
1524         namesize = req->rq_reqmsg->buflens[offset + 1];
1525
1526         /* namesize less than 2 means we have empty name, probably came from
1527            revalidate by cfid, so no point in having name to be set */
1528         if (namesize <= 1)
1529                 name = NULL;
1530
1531         LASSERT (offset == 1 || offset == 3);
1532         if (offset == 3) {
1533                 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
1534                 reply_offset = 1;
1535         } else {
1536                 reply_offset = 0;
1537         }
1538
1539         rc = mds_init_ucred(&uc, req, rsd);
1540         if (rc) {
1541                 if (child_lockh->cookie == 0)
1542                         mds_audit_auth(req, &uc, AUDIT_STAT, &body->id1,
1543                                        name, namesize);
1544                 GOTO(cleanup, rc);
1545         }
1546
1547         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1548         cleanup_phase = 1; /* kernel context */
1549         intent_set_disposition(rep, DISP_LOOKUP_EXECD);
1550
1551         LASSERT(namesize > 0);
1552         if (child_lockh->cookie != 0) {
1553                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1554                 resent_req = 1;
1555         }
1556 #if HAVE_LOOKUP_RAW
1557         if (body->valid == OBD_MD_FLID) {
1558                 struct mds_body *mds_reply;
1559                 int size = sizeof(*mds_reply);
1560                 struct inode *dir;
1561                 ino_t inum;
1562
1563                 dparent = mds_id2dentry(obd, &body->id1, NULL);
1564                 if (IS_ERR(dparent)) {
1565                         rc = PTR_ERR(dparent);
1566                         GOTO(cleanup, rc);
1567                 }
1568                 /*
1569                  * the user requested ONLY the inode number, so do a raw lookup.
1570                  */
1571                 rc = lustre_pack_reply(req, 1, &size, NULL);
1572                 if (rc) {
1573                         CERROR("out of memory\n");
1574                         l_dput(dparent);
1575                         GOTO(cleanup, rc);
1576                 }
1577                 dir  = dparent->d_inode;
1578                 LASSERT(dir->i_op->lookup_raw != NULL);
1579                 rc = dir->i_op->lookup_raw(dir, name, namesize - 1, &inum);
1580                 l_dput(dparent);
1581                 mds_reply = lustre_msg_buf(req->rq_repmsg, 0,
1582                                            sizeof(*mds_reply));
1583
1584                 id_ino(&mds_reply->id1) = inum;
1585                 mds_reply->valid = OBD_MD_FLID;
1586                 GOTO(cleanup, rc);
1587         }
1588 #endif
1589         if (resent_req == 0) {
1590                 LASSERT(id_fid(&body->id1) != 0);
1591                 if (name) {
1592                         rc = mds_get_parent_child_locked(obd, mds, &body->id1,
1593                                                          parent_lockh, &dparent,
1594                                                          LCK_PR,
1595                                                          MDS_INODELOCK_UPDATE,
1596                                                          &update_mode,
1597                                                          name, namesize,
1598                                                          child_lockh, &dchild,
1599                                                          LCK_PR, child_part);
1600                         if (rc)
1601                                 GOTO(cleanup, rc);
1602
1603                         cleanup_phase = 2; /* dchild, dparent, locks */
1604
1605                         /*
1606                          * let's make sure this name should leave on this mds
1607                          * node.
1608                          */
1609                         rc = mds_check_mds_num(obd, dparent->d_inode, name, namesize);
1610                         if (rc)
1611                                 GOTO(cleanup, rc);
1612                 } else {
1613                         /* we have no dentry here, drop LOOKUP bit */
1614                         /* FIXME: we need MDS_INODELOCK_LOOKUP or not. */
1615                         child_part &= ~MDS_INODELOCK_LOOKUP;
1616                         CDEBUG(D_OTHER, "%s: retrieve attrs for "DLID4"\n",
1617                                obd->obd_name, OLID4(&body->id1));
1618
1619                         dchild = mds_id2locked_dentry(obd, &body->id1, NULL,
1620                                                       LCK_PR, parent_lockh,
1621                                                       &update_mode, NULL, 0,
1622                                                       MDS_INODELOCK_UPDATE);
1623                         if (IS_ERR(dchild)) {
1624                                 CERROR("can't find inode with id "DLID4", err = %d\n",
1625                                        OLID4(&body->id1), (int)PTR_ERR(dchild));
1626                                 rc = PTR_ERR(dchild);
1627                                 dchild = NULL;
1628                                 GOTO(cleanup, rc);
1629                         }
1630                         memcpy(child_lockh, parent_lockh, sizeof(parent_lockh[0]));
1631                 }
1632         } else {
1633                 struct ldlm_lock *granted_lock;
1634
1635                 DEBUG_REQ(D_DLMTRACE, req, "resent, not enqueuing new locks");
1636                 granted_lock = ldlm_handle2lock(child_lockh);
1637
1638                 LASSERTF(granted_lock != NULL, LPU64"/%lu lockh "LPX64"\n",
1639                          id_fid(&body->id1), (unsigned long)id_group(&body->id1),
1640                          child_lockh->cookie);
1641
1642                 if (name) {
1643                         /* usual named request */
1644                         dparent = mds_id2dentry(obd, &body->id1, NULL);
1645                         LASSERT(!IS_ERR(dparent));
1646                         dchild = ll_lookup_one_len(name, dparent, namesize - 1);
1647                         LASSERT(!IS_ERR(dchild));
1648                 } else {
1649                         /* client wants to get attr. by id */
1650                         dchild = mds_id2dentry(obd, &body->id1, NULL);
1651                         LASSERT(!IS_ERR(dchild));
1652                 }
1653                 LDLM_LOCK_PUT(granted_lock);
1654         }
1655
1656         cleanup_phase = 2; /* dchild, dparent, locks */
1657
1658         if (!DENTRY_VALID(dchild)) {
1659                 intent_set_disposition(rep, DISP_LOOKUP_NEG);
1660                 /*
1661                  * in the intent case, the policy clears this error: the
1662                  * disposition is enough.
1663                  */
1664                 rc = -ENOENT;
1665                 GOTO(cleanup, rc);
1666         } else {
1667                 intent_set_disposition(rep, DISP_LOOKUP_POS);
1668         }
1669
1670         if (req->rq_repmsg == NULL) {
1671                 if (dchild->d_flags & DCACHE_CROSS_REF)
1672                         rc = mds_getattr_pack_msg_cf(req, dchild, offset);
1673                 else
1674                         rc = mds_getattr_pack_msg(req, dchild, offset);
1675                 if (rc != 0) {
1676                         CERROR ("mds_getattr_pack_msg: %d\n", rc);
1677                         GOTO (cleanup, rc);
1678                 }
1679         }
1680
1681         rc = mds_getattr_internal(obd, dchild, req, offset, body,
1682                                   reply_offset, rsd);
1683         if (rc)
1684                 GOTO(cleanup, rc); /* returns the lock to the client */
1685
1686         /* probably MDS knows actual size? */
1687         body = lustre_msg_buf(req->rq_repmsg, reply_offset, sizeof(*body));
1688         LASSERT(body != NULL);
1689         mds_getattr_size(obd, dchild, req, body);
1690
1691         GOTO(cleanup, rc);
1692
1693  cleanup:
1694         /* audit stuff for getattr */
1695         if (resent_req == 0 && (dparent || dchild)) {
1696                 struct inode * au_inode = NULL;
1697
1698                 if (dchild && dchild->d_inode)
1699                         au_inode = dchild->d_inode;
1700                 else
1701                         au_inode = dparent->d_inode;
1702
1703                 mds_audit_stat(req, &body->id1, au_inode, name, namesize, rc);
1704         }
1705         switch (cleanup_phase) {
1706         case 2:
1707                 if (resent_req == 0) {
1708                         if (rc && DENTRY_VALID(dchild))
1709                                 ldlm_lock_decref(child_lockh, LCK_PR);
1710                         if (name)
1711                                 ldlm_lock_decref(parent_lockh, LCK_PR);
1712 #ifdef S_PDIROPS
1713                         if (parent_lockh[1].cookie != 0)
1714                                 ldlm_lock_decref(parent_lockh + 1, update_mode);
1715 #endif
1716                         if (dparent)
1717                                 l_dput(dparent);
1718                 }
1719                 l_dput(dchild);
1720         case 1:
1721                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1722         default:
1723                 mds_exit_ucred(&uc);
1724         }
1725         return rc;
1726 }
1727
1728 static int mds_getattr(struct ptlrpc_request *req, int offset)
1729 {
1730         struct obd_device *obd = req->rq_export->exp_obd;
1731         struct lvfs_run_ctxt saved;
1732         struct dentry *de;
1733         struct mds_req_sec_desc *rsd;
1734         struct mds_body *body;
1735         struct lvfs_ucred uc = {NULL, NULL,};
1736         int rc = 0;
1737         ENTRY;
1738
1739         MD_COUNTER_INCREMENT(obd, getattr);
1740
1741         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1742         if (!rsd) {
1743                 CERROR("Can't unpack security desc\n");
1744                 RETURN(-EFAULT);
1745         }
1746
1747         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1748                                   lustre_swab_mds_body);
1749         if (body == NULL) {
1750                 CERROR ("Can't unpack body\n");
1751                 RETURN (-EFAULT);
1752         }
1753
1754         rc = mds_init_ucred(&uc, req, rsd);
1755         if (rc) {
1756                 mds_exit_ucred(&uc);
1757                 RETURN(rc);
1758         }
1759
1760         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1761         de = mds_id2dentry(obd, &body->id1, NULL);
1762         if (IS_ERR(de)) {
1763                 rc = req->rq_status = PTR_ERR(de);
1764                 GOTO(out_pop, rc);
1765         }
1766
1767         rc = mds_getattr_pack_msg(req, de, offset);
1768         if (rc != 0) {
1769                 CERROR("mds_getattr_pack_msg: %d\n", rc);
1770                 GOTO(out_pop, rc);
1771         }
1772
1773         req->rq_status = mds_getattr_internal(obd, de, req, offset, body,
1774                                               0, rsd);
1775         l_dput(de);
1776
1777         EXIT;
1778 out_pop:
1779         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1780         mds_exit_ucred(&uc);
1781         return rc;
1782 }
1783
1784 static int mds_access_check(struct ptlrpc_request *req, int offset)
1785 {
1786         struct obd_device *obd = req->rq_export->exp_obd;
1787         struct lvfs_run_ctxt saved;
1788         struct dentry *de;
1789         struct mds_req_sec_desc *rsd;
1790         struct mds_body *body;
1791         struct lvfs_ucred uc;
1792         int rep_size[2] = {sizeof(*body),
1793                            sizeof(struct mds_remote_perm)};
1794         int rc = 0;
1795         ENTRY;
1796
1797         if (!req->rq_export->exp_mds_data.med_remote) {
1798                 CERROR("from local client "LPU64"\n", req->rq_peer.peer_id.nid);
1799                 RETURN(-EINVAL);
1800         }
1801
1802         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1803         if (!rsd) {
1804                 CERROR("Can't unpack security desc\n");
1805                 RETURN(-EFAULT);
1806         }
1807
1808         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1809                                   lustre_swab_mds_body);
1810         if (body == NULL) {
1811                 CERROR ("Can't unpack body\n");
1812                 RETURN (-EFAULT);
1813         }
1814
1815         MD_COUNTER_INCREMENT(obd, access_check);
1816
1817         rc = mds_init_ucred(&uc, req, rsd);
1818         if (rc) {
1819                 CERROR("init ucred error: %d\n", rc);
1820                 RETURN(rc);
1821         }
1822         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1823
1824         de = mds_id2dentry(obd, &body->id1, NULL);
1825         if (IS_ERR(de)) {
1826                 CERROR("grab ino "LPU64": err %ld\n",
1827                        body->id1.li_stc.u.e3s.l3s_ino, PTR_ERR(de));
1828                 GOTO(out_pop, rc = PTR_ERR(de));
1829         }
1830
1831         rc = lustre_pack_reply(req, 2, rep_size, NULL);
1832         if (rc) {
1833                 CERROR("pack reply error: %d\n", rc);
1834                 GOTO(out_dput, rc = -EINVAL);
1835         }
1836
1837         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1838         LASSERT(body);
1839
1840         rc = mds_pack_remote_perm(req, 1, body, de->d_inode);
1841
1842         EXIT;
1843
1844 out_dput:
1845         l_dput(de);
1846 out_pop:
1847         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1848         mds_exit_ucred(&uc);
1849         return rc;
1850 }
1851
1852 static int mds_obd_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1853                           unsigned long max_age)
1854 {
1855         int rc;
1856         ENTRY;
1857
1858         spin_lock(&obd->obd_osfs_lock);
1859         rc = fsfilt_statfs(obd, obd->u.mds.mds_sb, max_age);
1860         if (rc == 0)
1861                 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
1862         spin_unlock(&obd->obd_osfs_lock);
1863
1864         RETURN(rc);
1865 }
1866
1867 static int mds_statfs(struct ptlrpc_request *req)
1868 {
1869         struct obd_device *obd = req->rq_export->exp_obd;
1870         int rc, size = sizeof(struct obd_statfs);
1871         ENTRY;
1872
1873         /* This will trigger a watchdog timeout */
1874         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
1875                          (MDS_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
1876
1877         rc = lustre_pack_reply(req, 1, &size, NULL);
1878         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
1879                 CERROR("mds: statfs lustre_pack_reply failed: rc = %d\n", rc);
1880                 GOTO(out, rc);
1881         }
1882
1883         OBD_COUNTER_INCREMENT(obd, statfs);
1884
1885         /* We call this so that we can cache a bit - 1 jiffie worth */
1886         rc = mds_obd_statfs(obd, lustre_msg_buf(req->rq_repmsg, 0, size),
1887                             jiffies - HZ);
1888         if (rc) {
1889                 CERROR("mds_obd_statfs failed: rc %d\n", rc);
1890                 GOTO(out, rc);
1891         }
1892
1893         EXIT;
1894 out:
1895         req->rq_status = rc;
1896         return rc;
1897 }
1898
1899 static int mds_sync(struct ptlrpc_request *req, int offset)
1900 {
1901         struct obd_device *obd = req->rq_export->exp_obd;
1902         struct mds_obd *mds = &obd->u.mds;
1903         struct mds_body *body;
1904         int rc, size = sizeof(*body);
1905         ENTRY;
1906
1907         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1908                                   lustre_swab_mds_body);
1909         if (body == NULL)
1910                 GOTO(out, rc = -EPROTO);
1911
1912         rc = lustre_pack_reply(req, 1, &size, NULL);
1913         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) {
1914                 CERROR("fsync lustre_pack_reply failed: rc = %d\n", rc);
1915                 GOTO(out, rc);
1916         }
1917
1918         if (id_ino(&body->id1) == 0) {
1919                 /* an id of zero is taken to mean "sync whole filesystem" */
1920                 rc = fsfilt_sync(obd, mds->mds_sb);
1921                 if (rc)
1922                         GOTO(out, rc);
1923         } else {
1924                 /* just any file to grab fsync method - "file" arg unused */
1925                 struct file *file = mds->mds_rcvd_filp;
1926                 struct mds_body *rep_body;
1927                 struct dentry *de;
1928
1929                 de = mds_id2dentry(obd, &body->id1, NULL);
1930                 if (IS_ERR(de))
1931                         GOTO(out, rc = PTR_ERR(de));
1932
1933                 rc = file->f_op->fsync(NULL, de, 1);
1934                 if (rc)
1935                         GOTO(out, rc);
1936
1937                 rep_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep_body));
1938                 mds_pack_inode2body(obd, rep_body, de->d_inode,
1939                                     (body->valid & OBD_MD_FID) ? 1 : 0);
1940                 l_dput(de);
1941         }
1942
1943         EXIT;
1944 out:
1945         req->rq_status = rc;
1946         return rc;
1947 }
1948
1949 /* mds_readpage does not take a DLM lock on the inode, because the client must
1950  * already have a PR lock.
1951  *
1952  * If we were to take another one here, a deadlock will result, if another
1953  * thread is already waiting for a PW lock. */
1954 static int mds_readpage(struct ptlrpc_request *req, int offset)
1955 {
1956         struct obd_device *obd = req->rq_export->exp_obd;
1957         struct vfsmount *mnt;
1958         struct dentry *de;
1959         struct file *file;
1960         struct mds_req_sec_desc *rsd;
1961         struct mds_body *body, *repbody;
1962         struct lvfs_run_ctxt saved;
1963         int rc, size = sizeof(*repbody);
1964         struct lvfs_ucred uc = {NULL, NULL,};
1965         ENTRY;
1966
1967         rc = lustre_pack_reply(req, 1, &size, NULL);
1968         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
1969                 CERROR("mds: out of memory\n");
1970                 GOTO(out, rc = -ENOMEM);
1971         }
1972
1973         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
1974         if (!rsd) {
1975                 CERROR("Can't unpack security desc\n");
1976                 GOTO (out, rc = -EFAULT);
1977         }
1978
1979         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1980                                   lustre_swab_mds_body);
1981         if (body == NULL) {
1982                 CERROR("Can't unpack body\n");
1983                 GOTO (out, rc = -EFAULT);
1984         }
1985
1986         rc = mds_init_ucred(&uc, req, rsd);
1987         if (rc) {
1988                 mds_audit_auth(req, &uc, AUDIT_READDIR, &body->id1,
1989                                NULL, 0);
1990                 GOTO(out, rc);
1991         }
1992
1993         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1994         de = mds_id2dentry(obd, &body->id1, &mnt);
1995         if (IS_ERR(de))
1996                 GOTO(out_pop, rc = PTR_ERR(de));
1997
1998         CDEBUG(D_INODE, "ino %lu\n", de->d_inode->i_ino);
1999
2000         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
2001         /* note: in case of an error, dentry_open puts dentry */
2002         if (IS_ERR(file))
2003                 GOTO(out_pop, rc = PTR_ERR(file));
2004
2005         /* body->size is actually the offset -eeb */
2006         if ((body->size & (de->d_inode->i_blksize - 1)) != 0) {
2007                 CERROR("offset "LPU64" not on a block boundary of %lu\n",
2008                        body->size, de->d_inode->i_blksize);
2009                 GOTO(out_file, rc = -EFAULT);
2010         }
2011
2012         /* body->nlink is actually the #bytes to read -eeb */
2013         if (body->nlink & (de->d_inode->i_blksize - 1)) {
2014                 CERROR("size %u is not multiple of blocksize %lu\n",
2015                        body->nlink, de->d_inode->i_blksize);
2016                 GOTO(out_file, rc = -EFAULT);
2017         }
2018
2019         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*repbody));
2020         repbody->size = file->f_dentry->d_inode->i_size;
2021         repbody->valid = OBD_MD_FLSIZE;
2022
2023         /* to make this asynchronous make sure that the handling function
2024            doesn't send a reply when this function completes. Instead a
2025            callback function would send the reply */
2026         /* body->size is actually the offset -eeb */
2027         rc = mds_sendpage(req, file, body->size, body->nlink);
2028
2029         EXIT;
2030 out_file:
2031         filp_close(file, 0);
2032 out_pop:
2033         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2034 out:
2035         mds_exit_ucred(&uc);
2036         req->rq_status = rc;
2037         return 0;
2038 }
2039
2040 int mds_read_md(struct obd_device *obd, struct lustre_id *id,
2041                 char **data, int *datalen)
2042 {
2043         struct dentry *dentry;
2044         struct mds_obd *mds = &obd->u.mds;
2045         int rc = 0, mea = 0;
2046         char *ea;
2047         ENTRY;
2048
2049         LASSERT(id);
2050         LASSERT(obd);
2051
2052         dentry = mds_id2dentry(obd, id, NULL);
2053         if (IS_ERR(dentry))
2054                 GOTO(out, rc = PTR_ERR(dentry));
2055
2056         if (!dentry->d_inode) {
2057                 CERROR("Can't find object "DLID4".\n",
2058                        OLID4(id));
2059                 GOTO(out_dentry, rc = -EINVAL);
2060         }
2061         if (S_ISDIR(dentry->d_inode->i_mode)) {
2062                 *datalen = obd_packmd(mds->mds_md_exp, NULL, NULL);
2063                 mea = 1;
2064         } else {
2065                 *datalen = obd_packmd(mds->mds_dt_exp, NULL, NULL);
2066                 mea = 0;
2067         }
2068         OBD_ALLOC(ea, *datalen);
2069         if (!ea) {
2070                 *datalen = 0;
2071                 GOTO(out_dentry, rc = PTR_ERR(dentry));
2072         }
2073         *data = ea;
2074         down(&dentry->d_inode->i_sem);
2075         rc = fsfilt_get_md(obd, dentry->d_inode, *data, *datalen,
2076                            (mea ? EA_MEA : EA_LOV));
2077         up(&dentry->d_inode->i_sem);
2078
2079         if (rc < 0)
2080                 CERROR("Error %d reading eadata for ino %lu\n",
2081                         rc, dentry->d_inode->i_ino);
2082 out_dentry:
2083         l_dput(dentry);
2084 out:
2085         RETURN(rc);
2086 }
2087 EXPORT_SYMBOL(mds_read_md);
2088
2089 int mds_reint(struct ptlrpc_request *req, int offset,
2090               struct lustre_handle *lockh)
2091 {
2092         struct mds_update_record *rec;
2093         struct mds_req_sec_desc *rsd;
2094         int rc;
2095         ENTRY;
2096
2097         OBD_ALLOC(rec, sizeof(*rec));
2098         if (rec == NULL)
2099                 RETURN(-ENOMEM);
2100
2101         rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
2102         if (!rsd) {
2103                 CERROR("Can't unpack security desc\n");
2104                 GOTO(out, rc = -EFAULT);
2105         }
2106
2107         rc = mds_update_unpack(req, offset, rec);
2108         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
2109                 CERROR("invalid record\n");
2110                 GOTO(out, req->rq_status = -EINVAL);
2111         }
2112
2113         rc = mds_init_ucred(&rec->ur_uc, req, rsd);
2114         if (rc) {
2115                 audit_op code = AUDIT_NONE;
2116                 char * au_name = NULL;
2117                 int au_len = 0;
2118                 switch (rec->ur_opcode) {
2119                         case REINT_SETATTR:
2120                                 code = AUDIT_SETATTR;
2121                                 break;
2122                         case REINT_CREATE:
2123                                 code = AUDIT_CREATE;
2124                                 au_name = rec->ur_name;
2125                                 au_len = rec->ur_namelen;
2126                                 break;
2127                         case REINT_LINK:
2128                                 code = AUDIT_LINK;
2129                                 break;
2130                         case REINT_UNLINK:
2131                                 code = AUDIT_UNLINK;
2132                                 break;
2133                         case REINT_RENAME:
2134                                 code = AUDIT_RENAME;
2135                                 break;
2136                         case REINT_OPEN:
2137                                 au_name = rec->ur_name;
2138                                 au_len = rec->ur_namelen;
2139                                 code = AUDIT_OPEN;
2140                                 break;
2141                         default:
2142                                 CERROR("Wrong opcode in reint\n");
2143                                 LBUG();
2144                 }
2145
2146                 mds_audit_auth(req, &rec->ur_uc, code, rec->ur_id1,
2147                                au_name, au_len);
2148                 GOTO(out, rc);
2149         }
2150
2151         /* rc will be used to interrupt a for loop over multiple records */
2152         rc = mds_reint_rec(rec, offset, req, lockh);
2153
2154 out:
2155         mds_exit_ucred(&rec->ur_uc);
2156         OBD_FREE(rec, sizeof(*rec));
2157         RETURN(rc);
2158 }
2159
2160 static int mds_filter_recovery_request(struct ptlrpc_request *req,
2161                                        struct obd_device *obd, int *process)
2162 {
2163         switch (req->rq_reqmsg->opc) {
2164         case MDS_CONNECT: /* This will never get here, but for completeness. */
2165         case OST_CONNECT: /* This will never get here, but for completeness. */
2166         case MDS_DISCONNECT:
2167         case OST_DISCONNECT:
2168                *process = 1;
2169                RETURN(0);
2170
2171         case MDS_CLOSE:
2172         case MDS_SYNC: /* used in unmounting */
2173         case OBD_PING:
2174         case MDS_REINT:
2175         case LDLM_ENQUEUE:
2176         case OST_CREATE:
2177                 *process = target_queue_recovery_request(req, obd);
2178                 RETURN(0);
2179
2180         default:
2181                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2182                 *process = 0;
2183                 /* XXX what should we set rq_status to here? */
2184                 req->rq_status = -EAGAIN;
2185                 RETURN(ptlrpc_error(req));
2186         }
2187 }
2188
2189 static char *reint_names[] = {
2190         [REINT_SETATTR] "setattr",
2191         [REINT_CREATE]  "create",
2192         [REINT_LINK]    "link",
2193         [REINT_UNLINK]  "unlink",
2194         [REINT_RENAME]  "rename",
2195         [REINT_OPEN]    "open",
2196 };
2197
2198 #define FILTER_VALID_FLAGS (OBD_MD_FLTYPE | OBD_MD_FLMODE | OBD_MD_FLGENER  | \
2199                             OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ| \
2200                             OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME| \
2201                             OBD_MD_FLID)
2202
2203 static void reconstruct_create(struct ptlrpc_request *req)
2204 {
2205         struct mds_export_data *med = &req->rq_export->exp_mds_data;
2206         struct mds_client_data *mcd = med->med_mcd;
2207         struct dentry *dentry;
2208         struct ost_body *body;
2209         struct lustre_id id;
2210         int rc;
2211         ENTRY;
2212
2213         /* copy rc, transno and disp; steal locks */
2214         mds_req_from_mcd(req, mcd);
2215         if (req->rq_status) {
2216                 EXIT;
2217                 return;
2218         }
2219
2220         id_gen(&id) = 0;
2221         id_group(&id) = 0;
2222
2223         id_ino(&id) = mcd->mcd_last_data;
2224         LASSERT(id_ino(&id) != 0);
2225
2226         dentry = mds_id2dentry(req2obd(req), &id, NULL);
2227         if (IS_ERR(dentry)) {
2228                 CERROR("can't find inode "LPU64"\n", id_ino(&id));
2229                 req->rq_status = PTR_ERR(dentry);
2230                 EXIT;
2231                 return;
2232         }
2233
2234         CWARN("reconstruct reply for x"LPU64" (remote ino) "LPU64" -> %lu/%u\n",
2235               req->rq_xid, id_ino(&id), dentry->d_inode->i_ino,
2236               dentry->d_inode->i_generation);
2237
2238         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
2239         obdo_from_inode(&body->oa, dentry->d_inode, FILTER_VALID_FLAGS);
2240         body->oa.o_id = dentry->d_inode->i_ino;
2241         body->oa.o_generation = dentry->d_inode->i_generation;
2242         body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2243
2244         down(&dentry->d_inode->i_sem);
2245         rc = mds_read_inode_sid(req2obd(req), dentry->d_inode, &id);
2246         up(&dentry->d_inode->i_sem);
2247         if (rc) {
2248                 CERROR("Can't read inode self id, inode %lu, "
2249                        "rc %d\n", dentry->d_inode->i_ino, rc);
2250                 id_fid(&id) = 0;
2251         }
2252
2253         body->oa.o_fid = id_fid(&id);
2254         body->oa.o_mds = id_group(&id);
2255         l_dput(dentry);
2256
2257         EXIT;
2258 }
2259
2260 static int mds_inode_init_acl(struct obd_device *obd, void *handle,
2261                               struct dentry *de, void *xattr, int xattr_size)
2262 {
2263         struct inode *inode = de->d_inode;
2264         struct posix_acl *acl;
2265         mode_t mode;
2266         int rc = 0;
2267
2268         LASSERT(handle);
2269         LASSERT(inode);
2270         LASSERT(xattr);
2271         LASSERT(xattr_size > 0);
2272
2273         if (!inode->i_op->getxattr || !inode->i_op->setxattr) {
2274                 CERROR("backend fs dosen't support xattr\n");
2275                 return -EOPNOTSUPP;
2276         }
2277
2278         /* set default acl */
2279         if (S_ISDIR(inode->i_mode)) {
2280                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_DEFAULT,
2281                                            xattr, xattr_size, 0);
2282                 if (rc) {
2283                         CERROR("set default acl err: %d\n", rc);
2284                         return rc;
2285                 }
2286         }
2287
2288         /* set access acl */
2289         acl = posix_acl_from_xattr(xattr, xattr_size);
2290         if (acl == NULL || IS_ERR(acl)) {
2291                 CERROR("insane attr data\n");
2292                 return PTR_ERR(acl);
2293         }
2294
2295         if (posix_acl_valid(acl)) {
2296                 CERROR("default acl not valid: %d\n", rc);
2297                 rc = -EFAULT;
2298                 goto out;
2299         }
2300
2301         mode = inode->i_mode;
2302         rc = posix_acl_create_masq(acl, &mode);
2303         if (rc < 0) {
2304                 CERROR("create masq err %d\n", rc);
2305                 goto out;
2306         }
2307
2308         if (inode->i_mode != mode) {
2309                 struct iattr iattr = { .ia_valid = ATTR_MODE,
2310                                        .ia_mode = mode };
2311                 int rc2;
2312
2313                 rc2 = fsfilt_setattr(obd, de, handle, &iattr, 0);
2314                 if (rc2) {
2315                         CERROR("setattr mode err: %d\n", rc2);
2316                         rc = rc2;
2317                         goto out;
2318                 }
2319         }
2320
2321         if (rc > 0) {
2322                 /* we didn't change acl except mode bits of some
2323                  * entries, so should be fit into original size.
2324                  */
2325                 rc = posix_acl_to_xattr(acl, xattr, xattr_size);
2326                 LASSERT(rc > 0);
2327
2328                 rc = inode->i_op->setxattr(de, XATTR_NAME_ACL_ACCESS,
2329                                            xattr, xattr_size, 0);
2330                 if (rc)
2331                         CERROR("set access acl err: %d\n", rc);
2332         }
2333 out:
2334         posix_acl_release(acl);
2335         return rc;
2336 }
2337
2338 static int mdt_obj_create(struct ptlrpc_request *req)
2339 {
2340         struct obd_device *obd = req->rq_export->exp_obd;
2341         struct mds_obd *mds = &obd->u.mds;
2342         struct ost_body *body, *repbody;
2343         void *acl = NULL;
2344         int acl_size;
2345         char idname[LL_ID_NAMELEN];
2346         int size = sizeof(*repbody);
2347         struct inode *parent_inode;
2348         struct lvfs_run_ctxt saved;
2349         int rc, cleanup_phase = 0;
2350         struct dentry *new = NULL;
2351         struct dentry_params dp;
2352         int mealen, flags = 0;
2353         struct lvfs_ucred uc;
2354         struct lustre_id id;
2355         struct mea *mea;
2356         void *handle = NULL;
2357         unsigned long cr_inum = 0;
2358         __u64 fid = 0;
2359         ENTRY;
2360
2361         DEBUG_REQ(D_HA, req, "create remote object");
2362         parent_inode = mds->mds_unnamed_dir->d_inode;
2363
2364         body = lustre_swab_reqbuf(req, 0, sizeof(*body),
2365                                   lustre_swab_ost_body);
2366         if (body == NULL)
2367                 RETURN(-EFAULT);
2368
2369         /* acl data is packed transparently, no swab here */
2370         LASSERT(req->rq_reqmsg->bufcount >= 2);
2371         acl_size = req->rq_reqmsg->buflens[1];
2372         if (acl_size) {
2373                 acl = lustre_msg_buf(req->rq_reqmsg, 1, acl_size);
2374                 if (!acl) {
2375                         CERROR("No default acl buf?\n");
2376                         RETURN(-EFAULT);
2377                 }
2378         }
2379
2380         rc = lustre_pack_reply(req, 1, &size, NULL);
2381         if (rc)
2382                 RETURN(rc);
2383
2384         MDS_CHECK_RESENT(req, reconstruct_create(req));
2385
2386         uc.luc_lsd = NULL;
2387         uc.luc_ginfo = NULL;
2388         uc.luc_uid = body->oa.o_uid;
2389         uc.luc_gid = body->oa.o_gid;
2390         uc.luc_fsuid = body->oa.o_uid;
2391         uc.luc_fsgid = body->oa.o_gid;
2392
2393         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2394         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
2395
2396         /* in REPLAY case inum should be given (client or other MDS fills it) */
2397         if (body->oa.o_id && ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2398             (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))) {
2399                 /*
2400                  * this is re-create request from MDS holding directory name.
2401                  * we have to lookup given ino/gen first. if it exists (good
2402                  * case) then there is nothing to do. if it does not then we
2403                  * have to recreate it.
2404                  */
2405                 id_ino(&id) = body->oa.o_id;
2406                 id_gen(&id) = body->oa.o_generation;
2407
2408                 new = mds_id2dentry(obd, &id, NULL);
2409                 if (!IS_ERR(new) && new->d_inode) {
2410                         struct lustre_id sid;
2411
2412                         CDEBUG(D_OTHER, "mkdir repairing %lu/%lu\n",
2413                                (unsigned long)id_ino(&id),
2414                                (unsigned long)id_gen(&id));
2415
2416                         obdo_from_inode(&repbody->oa, new->d_inode,
2417                                         FILTER_VALID_FLAGS);
2418
2419                         repbody->oa.o_id = new->d_inode->i_ino;
2420                         repbody->oa.o_generation = new->d_inode->i_generation;
2421                         repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
2422                         cleanup_phase = 1;
2423
2424                         down(&new->d_inode->i_sem);
2425                         rc = mds_read_inode_sid(obd, new->d_inode, &sid);
2426                         up(&new->d_inode->i_sem);
2427                         if (rc) {
2428                                 CERROR("Can't read inode self id "
2429                                        "inode %lu, rc %d.\n",
2430                                        new->d_inode->i_ino, rc);
2431                                 GOTO(cleanup, rc);
2432                         }
2433
2434                         repbody->oa.o_fid = id_fid(&sid);
2435                         repbody->oa.o_mds = id_group(&sid);
2436                         LASSERT(id_fid(&sid) != 0);
2437
2438                         /*
2439                          * here we could use fid passed in body->oa.o_fid and
2440                          * thus avoid mds_read_inode_sid().
2441                          */
2442                         cr_inum = new->d_inode->i_ino;
2443                         GOTO(cleanup, rc = 0);
2444                 }
2445         }
2446
2447         down(&parent_inode->i_sem);
2448         handle = fsfilt_start(obd, parent_inode, FSFILT_OP_MKDIR, NULL);
2449         if (IS_ERR(handle)) {
2450                 up(&parent_inode->i_sem);
2451                 CERROR("fsfilt_start() failed, rc = %d\n",
2452                        (int)PTR_ERR(handle));
2453                 GOTO(cleanup, rc = PTR_ERR(handle));
2454         }
2455         cleanup_phase = 1; /* transaction */
2456
2457 repeat:
2458         rc = sprintf(idname, "%u.%u", ll_insecure_random_int(), current->pid);
2459         new = lookup_one_len(idname, mds->mds_unnamed_dir, rc);
2460         if (IS_ERR(new)) {
2461                 CERROR("%s: can't lookup new inode (%s) for mkdir: %d\n",
2462                        obd->obd_name, idname, (int) PTR_ERR(new));
2463                 fsfilt_commit(obd, mds->mds_sb, new->d_inode, handle, 0);
2464                 up(&parent_inode->i_sem);
2465                 RETURN(PTR_ERR(new));
2466         } else if (new->d_inode) {
2467                 CERROR("%s: name exists. repeat\n", obd->obd_name);
2468                 goto repeat;
2469         }
2470         if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2471              lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2472                 fid = body->oa.o_fid;
2473         } else {
2474                 fid = mds_alloc_fid(obd);
2475         }
2476         new->d_fsdata = (void *)&dp;
2477         dp.p_inum = 0;
2478         dp.p_ptr = req;
2479
2480         dp.p_fid = fid;
2481         dp.p_group = mds->mds_num;
2482
2483         if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) ||
2484             (body->oa.o_flags & OBD_FL_RECREATE_OBJS)) {
2485                 LASSERT(body->oa.o_id != 0);
2486                 dp.p_inum = body->oa.o_id;
2487                 DEBUG_REQ(D_HA, req, "replay create obj %lu/%lu",
2488                           (unsigned long)body->oa.o_id,
2489                           (unsigned long)body->oa.o_generation);
2490         }
2491
2492         rc = vfs_mkdir(parent_inode, new, body->oa.o_mode);
2493         if (rc == 0) {
2494                 if (acl) {
2495                         rc = mds_inode_init_acl(obd, handle, new,
2496                                                 acl, acl_size);
2497                         if (rc) {
2498                                 up(&parent_inode->i_sem);
2499                                 GOTO(cleanup, rc);
2500                         }
2501                 }
2502                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2503                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2504                         new->d_inode->i_generation = body->oa.o_generation;
2505                         mark_inode_dirty(new->d_inode);
2506
2507                         /*
2508                          * avoiding asserts in cache flush case, as
2509                          * @body->oa.o_id should be zero.
2510                          */
2511                         if (body->oa.o_id) {
2512                                 LASSERTF(body->oa.o_id == new->d_inode->i_ino,
2513                                          "BUG 3550: failed to recreate obj "
2514                                          LPU64" -> %lu\n", body->oa.o_id,
2515                                          new->d_inode->i_ino);
2516
2517                                 LASSERTF(body->oa.o_generation ==
2518                                          new->d_inode->i_generation,
2519                                          "BUG 3550: failed to recreate obj/gen "
2520                                          LPU64"/%u -> %lu/%u\n", body->oa.o_id,
2521                                          body->oa.o_generation,
2522                                          new->d_inode->i_ino,
2523                                          new->d_inode->i_generation);
2524                         }
2525                 }
2526
2527                 obdo_from_inode(&repbody->oa, new->d_inode, FILTER_VALID_FLAGS);
2528                 repbody->oa.o_id = new->d_inode->i_ino;
2529                 repbody->oa.o_generation = new->d_inode->i_generation;
2530                 repbody->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FID;
2531
2532                 if ((body->oa.o_flags & OBD_FL_RECREATE_OBJS) ||
2533                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
2534                         LASSERT(body->oa.o_id != 0);
2535                         LASSERT(body->oa.o_fid != 0);
2536                 }
2537
2538                 mds_inode2id(obd, &id, new->d_inode, fid);
2539                 mds_update_inode_ids(obd, new->d_inode, handle, &id,
2540                                      body->oa.o_valid & OBD_MD_FLID ?
2541                                      NULL : obdo_id(&body->oa));
2542
2543                 /* initializing o_fid after it is allocated. */
2544                 repbody->oa.o_fid = id_fid(&id);
2545                 repbody->oa.o_mds = id_group(&id);
2546
2547                 rc = fsfilt_del_dir_entry(obd, new);
2548                 up(&parent_inode->i_sem);
2549                 if (rc) {
2550                         CERROR("can't remove name for object: %d\n", rc);
2551                         GOTO(cleanup, rc);
2552                 }
2553
2554                 cleanup_phase = 2; /* created directory object */
2555
2556                 CDEBUG(D_OTHER, "created dirobj: %lu/%lu mode %o\n",
2557                        (unsigned long)new->d_inode->i_ino,
2558                        (unsigned long)new->d_inode->i_generation,
2559                        (unsigned)new->d_inode->i_mode);
2560                 cr_inum = new->d_inode->i_ino;
2561         } else {
2562                 up(&parent_inode->i_sem);
2563                 CERROR("%s: can't create dirobj: %d\n", obd->obd_name, rc);
2564                 GOTO(cleanup, rc);
2565         }
2566
2567         if (body->oa.o_valid & OBD_MD_FLID) {
2568                 /* this is new object for splitted dir. We have to prevent
2569                  * recursive splitting on it -bzzz */
2570                 mealen = obd_size_diskmd(mds->mds_md_exp, NULL);
2571
2572                 OBD_ALLOC(mea, mealen);
2573                 if (mea == NULL)
2574                         GOTO(cleanup, rc = -ENOMEM);
2575
2576                 mea->mea_magic = MEA_MAGIC_ALL_CHARS;
2577                 mea->mea_master = body->oa.o_mds;      /* master mds num */
2578                 mea->mea_count = 0;
2579
2580                 obdo2id(&mea->mea_ids[body->oa.o_mds], &body->oa);
2581
2582                 down(&new->d_inode->i_sem);
2583                 rc = fsfilt_set_md(obd, new->d_inode, handle,
2584                                    mea, mealen, EA_MEA);
2585                 up(&new->d_inode->i_sem);
2586                 if (rc)
2587                         CERROR("fsfilt_set_md() failed, "
2588                                "rc = %d\n", rc);
2589
2590                 OBD_FREE(mea, mealen);
2591
2592                 CDEBUG(D_OTHER, "%s: mark non-splittable %lu/%u - %d\n",
2593                        obd->obd_name, new->d_inode->i_ino,
2594                        new->d_inode->i_generation, flags);
2595         } else if (body->oa.o_easize) {
2596                 /* we pass LCK_EX to split routine to signal that we have
2597                  * exclusive access to the directory. simple because nobody
2598                  * knows it already exists -bzzz */
2599                 rc = mds_try_to_split_dir(obd, new, NULL,
2600                                           body->oa.o_easize, LCK_EX);
2601                 if (rc < 0) {
2602                         CERROR("Can't split directory %lu, error = %d.\n",
2603                                new->d_inode->i_ino, rc);
2604                 } else {
2605                         rc = 0;
2606                 }
2607         }
2608
2609         EXIT;
2610 cleanup:
2611         if (rc == 0 && (body->oa.o_flags & OBD_FL_REINT)) {
2612                 rc = mds_fidmap_add(obd, &id);
2613                 if (rc < 0) {
2614                         CERROR("can't create fid->ino mapping, "
2615                                "err %d\n", rc);
2616                 } else {
2617                         rc = 0;
2618                 }
2619         }
2620
2621         switch (cleanup_phase) {
2622         case 2: /* object has been created, but we'll may want to replay it later */
2623                 if (rc == 0)
2624                         ptlrpc_require_repack(req);
2625         case 1: /* transaction */
2626                 rc = mds_finish_transno(mds, parent_inode, handle,
2627                                         req, rc, cr_inum);
2628         }
2629
2630         l_dput(new);
2631         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2632         return rc;
2633 }
2634
2635 static int mdt_get_info(struct ptlrpc_request *req)
2636 {
2637         struct obd_export *exp = req->rq_export;
2638         int keylen, rc = 0;
2639         char *key;
2640         ENTRY;
2641
2642         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2643         if (key == NULL) {
2644                 DEBUG_REQ(D_HA, req, "no get_info key");
2645                 RETURN(-EFAULT);
2646         }
2647         keylen = req->rq_reqmsg->buflens[0];
2648
2649         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
2650             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
2651             (keylen < strlen("lovdesc") || strcmp(key, "lovdesc") != 0) &&
2652             (keylen < strlen("getext") || strcmp(key, "getext") != 0) &&
2653             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
2654                 RETURN(-EPROTO);
2655
2656         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
2657                 struct lustre_id *reply;
2658                 int size = sizeof(*reply);
2659
2660                 rc = lustre_pack_reply(req, 1, &size, NULL);
2661                 if (rc)
2662                         RETURN(rc);
2663
2664                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2665                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2666         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2667                 struct lov_desc *reply;
2668                 int size = sizeof(*reply);
2669
2670                 rc = lustre_pack_reply(req, 1, &size, NULL);
2671                 if (rc)
2672                         RETURN(rc);
2673
2674                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2675                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2676         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2677                 struct fid_extent *reply;
2678                 int size = sizeof(*reply);
2679
2680                 rc = lustre_pack_reply(req, 1, &size, NULL);
2681                 if (rc)
2682                         RETURN(rc);
2683
2684                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2685                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2686         } else {
2687                 obd_id *reply;
2688                 int size = sizeof(*reply);
2689
2690                 rc = lustre_pack_reply(req, 1, &size, NULL);
2691                 if (rc)
2692                         RETURN(rc);
2693
2694                 reply = lustre_msg_buf(req->rq_repmsg, 0, size);
2695                 rc = obd_get_info(exp, keylen, key, (__u32 *)&size, reply);
2696         }
2697
2698         req->rq_repmsg->status = 0;
2699         RETURN(rc);
2700 }
2701
2702 static int mds_set_info(struct obd_export *exp, __u32 keylen,
2703                         void *key, __u32 vallen, void *val)
2704 {
2705         struct obd_device *obd;
2706         struct mds_obd *mds;
2707         int rc = 0;
2708         ENTRY;
2709
2710         obd = class_exp2obd(exp);
2711         if (obd == NULL) {
2712                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2713                        exp->exp_handle.h_cookie);
2714                 RETURN(-EINVAL);
2715         }
2716
2717         mds = &obd->u.mds;
2718         if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2719                 rc = mds_set_audit(obd, val);
2720                 RETURN(rc);
2721         } else if (keylen >= strlen("ids") && memcmp(key, "ids", keylen) == 0) {
2722                 struct lustre_id *ids = (struct lustre_id *)val;
2723                 struct dentry *de;
2724                 struct inode *inode;
2725                 void *handle;
2726                 int err;
2727
2728                 de = mds_id2dentry(obd, ids, NULL);
2729                 if (IS_ERR(de)) {
2730                         rc = PTR_ERR(de);
2731                         CERROR("lookup by an id error rc=%d\n ", rc);
2732                         RETURN(rc);
2733                 }
2734                 inode = de->d_inode;
2735                 if (inode == NULL)
2736                         GOTO(out_put, rc = -ENOENT);
2737
2738                 down(&inode->i_sem);
2739                 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
2740                 if (IS_ERR(handle)) {
2741                         up(&inode->i_sem);
2742                         GOTO(out_put, rc = PTR_ERR(handle));
2743                 }
2744
2745                 rc = mds_update_inode_ids(obd, inode, handle, NULL, ids + 1);
2746
2747                 err = fsfilt_commit(obd, mds->mds_sb, inode, handle,
2748                                     exp->exp_sync);
2749                 if (err) {
2750                         CERROR("error committing transaction: %d\n", err);
2751                         if (!rc) rc = err;
2752                 }
2753                 up(&inode->i_sem);
2754 out_put:
2755                 l_dput(de);
2756                 RETURN(rc);
2757         }
2758
2759         if (keylen >= strlen("crypto_type") &&
2760              memcmp(key, "crypto_type", keylen) == 0) {
2761                 rc = mds_set_crypto_type(obd, val, vallen);
2762                 RETURN(rc);
2763         }
2764
2765         if (keylen >= strlen("setext") && !memcmp(key, "setext", keylen)) {
2766                 struct fid_extent *ext = val;
2767
2768                 CDEBUG(D_IOCTL, "set last fid to extent ["LPD64"-"LPD64"]\n",
2769                        ext->fe_start, ext->fe_width);
2770
2771                 /* set lastfid into fid extent start. All next object creates
2772                  * will use that fid. */
2773                 mds_set_last_fid(obd, ext->fe_start);
2774
2775                 /* setting the same extent to OSC to avoid ids intersecting in
2776                  * object ids, as all cache MDSs have the same group 0. */
2777                 rc = obd_set_info(mds->mds_dt_exp, strlen("setext"),
2778                                   "setext", sizeof(*ext), ext);
2779                 if (rc) {
2780                         CERROR("can't set extent ["LPD64"-"LPD64"] to %s, "
2781                                "err %d\n", ext->fe_start, ext->fe_width,
2782                                mds->mds_dt_exp->exp_obd->obd_name, rc);
2783                 }
2784                 RETURN(rc);
2785         }
2786         CDEBUG(D_IOCTL, "invalid key\n");
2787         RETURN(-EINVAL);
2788 }
2789
2790 static int mdt_set_info(struct ptlrpc_request *req)
2791 {
2792         char *key, *val;
2793         struct obd_export *exp = req->rq_export;
2794         int keylen, rc = 0, vallen = 0;
2795         ENTRY;
2796
2797         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
2798         if (key == NULL) {
2799                 DEBUG_REQ(D_HA, req, "no set_info key");
2800                 RETURN(-EFAULT);
2801         }
2802         keylen = req->rq_reqmsg->buflens[0];
2803
2804         if ((keylen == strlen("crypto_type") &&
2805             memcmp(key, "crypto_type", keylen) == 0)) {
2806                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2807                 if (rc)
2808                         RETURN(rc);
2809
2810                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2811                 vallen = req->rq_reqmsg->buflens[1];
2812
2813                 rc = obd_set_info(exp, keylen, key, vallen, val);
2814                 req->rq_repmsg->status = 0;
2815                 RETURN(rc);
2816         } else if (keylen == 5 && memcmp(key, "audit", 5) == 0) {
2817                 struct audit_attr_msg msg, *p;
2818                 int rc = 0;
2819
2820                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2821                 if (rc)
2822                         RETURN(rc);
2823
2824                 p = lustre_swab_reqbuf(req, 1, sizeof(msg),
2825                                        lustre_swab_audit_attr);
2826
2827                 msg = *p;
2828                 CDEBUG(D_INFO, "Get new audit setting 0x%x\n", (__u32)msg.attr);
2829                 rc = obd_set_info(exp, keylen, key, sizeof(msg), &msg);
2830
2831                 req->rq_repmsg->status = rc;
2832                 RETURN(rc);
2833         } else if (keylen == strlen("ids") &&
2834                    memcmp(key, "ids", keylen) == 0) {
2835                 struct lustre_id *id, ids[2];
2836
2837                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2838                 if (rc)
2839                         RETURN(rc);
2840                 id = lustre_swab_reqbuf(req, 1, sizeof(struct lustre_id),
2841                                         lustre_swab_lustre_id);
2842                 ids[0] = *id;
2843                 id = lustre_swab_reqbuf(req, 2, sizeof(struct lustre_id),
2844                                         lustre_swab_lustre_id);
2845                 ids[1] = *id;
2846
2847                 rc = obd_set_info(exp, keylen, key, vallen, ids);
2848                 req->rq_repmsg->status = rc;
2849                 RETURN(rc);
2850         } else if (keylen == strlen("setext") &&
2851                    memcmp(key, "setext", keylen) == 0) {
2852                 rc = lustre_pack_reply(req, 0, NULL, NULL);
2853                 if (rc)
2854                         RETURN(rc);
2855
2856                 val = lustre_msg_buf(req->rq_reqmsg, 1, 0);
2857                 vallen = req->rq_reqmsg->buflens[1];
2858
2859                 rc = obd_set_info(exp, keylen, key, vallen, val);
2860                 req->rq_repmsg->status = 0;
2861                 RETURN(rc);
2862         }
2863
2864         CDEBUG(D_IOCTL, "invalid key\n");
2865         RETURN(-EINVAL);
2866 }
2867
2868 static void mds_revoke_export_locks(struct obd_export *exp)
2869 {
2870         struct list_head *locklist = &exp->exp_ldlm_data.led_held_locks;
2871         struct list_head  rpc_list;
2872         struct ldlm_lock *lock, *next;
2873         struct ldlm_lock_desc desc;
2874
2875         /* don't do this for local client */
2876         if (!exp->u.eu_mds_data.med_remote)
2877                 return;
2878
2879         /* don't revoke locks during recovery */
2880         if (exp->exp_obd->obd_recovering)
2881                 return;
2882
2883         ENTRY;
2884         INIT_LIST_HEAD(&rpc_list);
2885
2886         spin_lock(&exp->exp_ldlm_data.led_lock);
2887         list_for_each_entry_safe(lock, next, locklist, l_export_chain) {
2888
2889                 lock_res_and_lock(lock);
2890                 if (lock->l_req_mode != lock->l_granted_mode) {
2891                         unlock_res_and_lock(lock);
2892                         continue;
2893                 }
2894
2895                 LASSERT(lock->l_resource);
2896                 if (lock->l_resource->lr_type != LDLM_IBITS &&
2897                     lock->l_resource->lr_type != LDLM_PLAIN) {
2898                         unlock_res_and_lock(lock);
2899                         continue;
2900                 }
2901
2902                 if (lock->l_flags & LDLM_FL_AST_SENT) {
2903                         unlock_res_and_lock(lock);
2904                         continue;
2905                 }
2906
2907                 LASSERT(lock->l_blocking_ast);
2908                 LASSERT(!lock->l_blocking_lock);
2909
2910                 lock->l_flags |= LDLM_FL_AST_SENT;
2911                 unlock_res_and_lock(lock);
2912
2913                 list_move(&lock->l_export_chain, &rpc_list);
2914         }
2915         spin_unlock(&exp->exp_ldlm_data.led_lock);
2916
2917         while (!list_empty(&rpc_list)) {
2918                 lock = list_entry(rpc_list.next, struct ldlm_lock,
2919                                   l_export_chain);
2920                 list_del_init(&lock->l_export_chain);
2921
2922                 /* the desc just pretend to exclusive */
2923                 ldlm_lock2desc(lock, &desc);
2924                 desc.l_req_mode = LCK_EX;
2925                 desc.l_granted_mode = 0;
2926
2927                 lock->l_blocking_ast(lock, &desc, NULL, LDLM_CB_BLOCKING);
2928         }
2929         EXIT;
2930 }
2931
2932 static int mds_msg_check_version(struct lustre_msg *msg)
2933 {
2934         int rc;
2935
2936         switch (msg->opc) {
2937         case MDS_CONNECT:
2938         case MDS_DISCONNECT:
2939         case OBD_PING:
2940                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2941                 if (rc)
2942                         CERROR("bad opc %u version %08x, expecting %08x\n",
2943                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2944                 break;
2945         case MDS_STATFS:
2946         case MDS_GETSTATUS:
2947         case MDS_GETATTR:
2948         case MDS_GETATTR_LOCK:
2949         case MDS_ACCESS_CHECK:
2950         case MDS_READPAGE:
2951         case MDS_REINT:
2952         case MDS_CLOSE:
2953         case MDS_DONE_WRITING:
2954         case MDS_PIN:
2955         case MDS_SYNC:
2956         case MDS_PARSE_ID:
2957                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2958                 if (rc)
2959                         CERROR("bad opc %u version %08x, expecting %08x\n",
2960                                msg->opc, msg->version, LUSTRE_MDS_VERSION);
2961                 break;
2962         case LDLM_ENQUEUE:
2963         case LDLM_CONVERT:
2964         case LDLM_BL_CALLBACK:
2965         case LDLM_CP_CALLBACK:
2966                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2967                 if (rc)
2968                         CERROR("bad opc %u version %08x, expecting %08x\n",
2969                                msg->opc, msg->version, LUSTRE_DLM_VERSION);
2970                 break;
2971         case OBD_LOG_CANCEL:
2972         case LLOG_ORIGIN_HANDLE_OPEN:
2973         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2974         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2975         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2976         case LLOG_ORIGIN_HANDLE_CLOSE:
2977         case LLOG_CATINFO:
2978                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2979                 if (rc)
2980                         CERROR("bad opc %u version %08x, expecting %08x\n",
2981                                msg->opc, msg->version, LUSTRE_LOG_VERSION);
2982                 break;
2983         case OST_CREATE:
2984         case OST_WRITE:
2985         case OST_GET_INFO:
2986         case OST_SET_INFO:
2987                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2988                 if (rc)
2989                         CERROR("bad opc %u version %08x, expecting %08x\n",
2990                                msg->opc, msg->version, LUSTRE_OBD_VERSION);
2991                 break;
2992         case SEC_INIT:
2993         case SEC_INIT_CONTINUE:
2994         case SEC_FINI:
2995                 rc = 0;
2996                 break;
2997         default:
2998                 CERROR("MDS unknown opcode %d\n", msg->opc);
2999                 rc = -ENOTSUPP;
3000                 break;
3001         }
3002
3003         return rc;
3004 }
3005
3006 int mds_handle(struct ptlrpc_request *req)
3007 {
3008         int should_process, fail = OBD_FAIL_MDS_ALL_REPLY_NET;
3009         struct obd_device *obd = NULL;
3010         struct mds_obd *mds = NULL; /* quell gcc overwarning */
3011         int rc = 0;
3012         ENTRY;
3013
3014         OBD_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0);
3015
3016         rc = mds_msg_check_version(req->rq_reqmsg);
3017         if (rc) {
3018                 CERROR("MDS drop mal-formed request\n");
3019                 RETURN(rc);
3020         }
3021
3022         /* Security opc should NOT trigger any recovery events */
3023         if (req->rq_reqmsg->opc == SEC_INIT ||
3024             req->rq_reqmsg->opc == SEC_INIT_CONTINUE) {
3025                 if (req->rq_export) {
3026                         mds_req_add_idmapping(req,
3027                                               &req->rq_export->exp_mds_data);
3028                         mds_revoke_export_locks(req->rq_export);
3029                 }
3030                 GOTO(out, rc = 0);
3031         } else if (req->rq_reqmsg->opc == SEC_FINI) {
3032                 if (req->rq_export) {
3033                         mds_req_del_idmapping(req,
3034                                               &req->rq_export->exp_mds_data);
3035                         mds_revoke_export_locks(req->rq_export);
3036                 }
3037                 GOTO(out, rc = 0);
3038         }
3039
3040         LASSERT(current->journal_info == NULL);
3041         /* XXX identical to OST */
3042         if (req->rq_reqmsg->opc != MDS_CONNECT) {
3043                 struct mds_export_data *med;
3044                 int recovering;
3045
3046                 if (req->rq_export == NULL) {
3047                         CERROR("operation %d on unconnected MDS from %s\n",
3048                                req->rq_reqmsg->opc,
3049                                req->rq_peerstr);
3050                         req->rq_status = -ENOTCONN;
3051                         GOTO(out, rc = -ENOTCONN);
3052                 }
3053
3054                 med = &req->rq_export->exp_mds_data;
3055                 obd = req->rq_export->exp_obd;
3056                 mds = &obd->u.mds;
3057
3058                 /* sanity check: if the xid matches, the request must
3059                  * be marked as a resent or replayed */
3060                 if (req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_xid) ||
3061                    req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_close_xid)) {
3062                         LASSERTF(lustre_msg_get_flags(req->rq_reqmsg) &
3063                                  (MSG_RESENT | MSG_REPLAY),
3064                                  "rq_xid "LPU64" matches last_xid, "
3065                                  "expected RESENT flag\n",
3066                                  req->rq_xid);
3067                 }
3068                 /* else: note the opposite is not always true; a
3069                  * RESENT req after a failover will usually not match
3070                  * the last_xid, since it was likely never
3071                  * committed. A REPLAYed request will almost never
3072                  * match the last xid, however it could for a
3073                  * committed, but still retained, open. */
3074
3075                 spin_lock_bh(&obd->obd_processing_task_lock);
3076                 recovering = obd->obd_recovering;
3077                 spin_unlock_bh(&obd->obd_processing_task_lock);
3078                 if (recovering) {
3079                         rc = mds_filter_recovery_request(req, obd,
3080                                                          &should_process);
3081                         if (rc || should_process == 0) {
3082                                 RETURN(rc);
3083                         } else if (should_process < 0) {
3084                                 req->rq_status = should_process;
3085                                 rc = ptlrpc_error(req);
3086                                 RETURN(rc);
3087                         }
3088                 }
3089         }
3090
3091         switch (req->rq_reqmsg->opc) {
3092         case MDS_CONNECT:
3093                 DEBUG_REQ(D_INODE, req, "connect");
3094                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
3095                 rc = target_handle_connect(req);
3096                 if (!rc) {
3097                         struct mds_export_data *med;
3098
3099                         LASSERT(req->rq_export);
3100                         med = &req->rq_export->u.eu_mds_data;
3101                         mds_init_export_data(req, med);
3102                         mds_req_add_idmapping(req, med);
3103
3104                         /* Now that we have an export, set mds. */
3105                         obd = req->rq_export->exp_obd;
3106                         mds = mds_req2mds(req);
3107                 }
3108                 break;
3109
3110         case MDS_DISCONNECT:
3111                 DEBUG_REQ(D_INODE, req, "disconnect");
3112                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
3113                 rc = target_handle_disconnect(req);
3114                 req->rq_status = rc;            /* superfluous? */
3115                 break;
3116
3117         case MDS_GETSTATUS:
3118                 DEBUG_REQ(D_INODE, req, "getstatus");
3119                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
3120                 rc = mds_getstatus(req);
3121                 break;
3122
3123         case MDS_GETATTR:
3124                 DEBUG_REQ(D_INODE, req, "getattr");
3125                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
3126                 rc = mds_getattr(req, MDS_REQ_REC_OFF);
3127                 break;
3128
3129         case MDS_ACCESS_CHECK:
3130                 DEBUG_REQ(D_INODE, req, "access_check");
3131                 OBD_FAIL_RETURN(OBD_FAIL_MDS_ACCESS_CHECK_NET, 0);
3132                 rc = mds_access_check(req, MDS_REQ_REC_OFF);
3133                 break;
3134
3135         case MDS_GETATTR_LOCK: {
3136                 struct lustre_handle lockh;
3137                 DEBUG_REQ(D_INODE, req, "getattr_lock");
3138                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_LOCK_NET, 0);
3139
3140                 /* If this request gets a reconstructed reply, we won't be
3141                  * acquiring any new locks in mds_getattr_lock, so we don't
3142                  * want to cancel.
3143                  */
3144                 lockh.cookie = 0;
3145                 rc = mds_getattr_lock(req, MDS_REQ_REC_OFF, &lockh,
3146                                       MDS_INODELOCK_UPDATE);
3147                 /* this non-intent call (from an ioctl) is special */
3148                 req->rq_status = rc;
3149                 if (rc == 0 && lockh.cookie)
3150                         ldlm_lock_decref(&lockh, LCK_PR);
3151                 break;
3152         }
3153         case MDS_STATFS:
3154                 DEBUG_REQ(D_INODE, req, "statfs");
3155                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
3156                 rc = mds_statfs(req);
3157                 break;
3158
3159         case MDS_READPAGE:
3160                 DEBUG_REQ(D_INODE, req, "readpage");
3161                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
3162                 rc = mds_readpage(req, MDS_REQ_REC_OFF);
3163
3164                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_SENDPAGE)) {
3165                         if (req->rq_reply_state) {
3166                                 lustre_free_reply_state (req->rq_reply_state);
3167                                 req->rq_reply_state = NULL;
3168                         }
3169                         RETURN(0);
3170                 }
3171
3172                 break;
3173         case MDS_REINT: {
3174                 __u32 *opcp = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
3175                                              sizeof (*opcp));
3176                 __u32  opc;
3177                 int size[3] = {sizeof(struct mds_body), mds->mds_max_mdsize,
3178                                mds->mds_max_cookiesize};
3179                 int bufcount;
3180
3181                 /* NB only peek inside req now; mds_reint() will swab it */
3182                 if (opcp == NULL) {
3183                         CERROR ("Can't inspect opcode\n");
3184                         rc = -EINVAL;
3185                         break;
3186                 }
3187                 opc = *opcp;
3188                 if (lustre_msg_swabbed (req->rq_reqmsg))
3189                         __swab32s(&opc);
3190
3191                 DEBUG_REQ(D_INODE, req, "reint %d (%s)", opc,
3192                           (opc < sizeof(reint_names) / sizeof(reint_names[0]) ||
3193                            reint_names[opc] == NULL) ? reint_names[opc] :
3194                                                        "unknown opcode");
3195
3196                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
3197
3198                 if (opc == REINT_UNLINK || opc == REINT_RENAME)
3199                         bufcount = 3;
3200                 else if (opc == REINT_OPEN)
3201                         bufcount = 2;
3202                 else
3203                         bufcount = 1;
3204
3205                 /* for SETATTR: I have different reply setting for
3206                  * remote setfacl, so delay the reply buffer allocation.
3207                  */
3208                 if (opc != REINT_SETATTR) {
3209                         rc = lustre_pack_reply(req, bufcount, size, NULL);
3210                         if (rc)
3211                                 break;
3212                 }
3213
3214                 rc = mds_reint(req, MDS_REQ_REC_OFF, NULL);
3215                 fail = OBD_FAIL_MDS_REINT_NET_REP;
3216                 break;
3217         }
3218
3219         case MDS_CLOSE:
3220                 DEBUG_REQ(D_INODE, req, "close");
3221                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
3222                 rc = mds_close(req, MDS_REQ_REC_OFF);
3223                 break;
3224
3225         case MDS_DONE_WRITING:
3226                 DEBUG_REQ(D_INODE, req, "done_writing");
3227                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DONE_WRITING_NET, 0);
3228                 rc = mds_done_writing(req, MDS_REQ_REC_OFF);
3229                 break;
3230
3231         case MDS_PIN:
3232                 DEBUG_REQ(D_INODE, req, "pin");
3233                 OBD_FAIL_RETURN(OBD_FAIL_MDS_PIN_NET, 0);
3234                 rc = mds_pin(req, MDS_REQ_REC_OFF);
3235                 break;
3236
3237         case MDS_SYNC:
3238                 DEBUG_REQ(D_INODE, req, "sync");
3239                 OBD_FAIL_RETURN(OBD_FAIL_MDS_SYNC_NET, 0);
3240                 rc = mds_sync(req, MDS_REQ_REC_OFF);
3241                 break;
3242         case MDS_PARSE_ID:
3243                 DEBUG_REQ(D_INODE, req, "parseid");
3244                 rc = mds_parse_id(req);
3245                 break;
3246         case OBD_PING:
3247                 DEBUG_REQ(D_INODE, req, "ping");
3248                 rc = target_handle_ping(req);
3249                 break;
3250
3251         case OBD_LOG_CANCEL:
3252                 CDEBUG(D_INODE, "log cancel\n");
3253                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
3254                 rc = -ENOTSUPP; /* la la la */
3255                 break;
3256
3257         case LDLM_ENQUEUE:
3258                 DEBUG_REQ(D_INODE, req, "enqueue");
3259                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
3260                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
3261                                          ldlm_server_blocking_ast, NULL);
3262                 fail = OBD_FAIL_LDLM_REPLY;
3263                 break;
3264         case LDLM_CONVERT:
3265                 DEBUG_REQ(D_INODE, req, "convert");
3266                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
3267                 rc = ldlm_handle_convert(req);
3268                 break;
3269         case LDLM_BL_CALLBACK:
3270         case LDLM_CP_CALLBACK:
3271                 DEBUG_REQ(D_INODE, req, "callback");
3272                 CERROR("callbacks should not happen on MDS\n");
3273                 LBUG();
3274                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
3275                 break;
3276         case LLOG_ORIGIN_HANDLE_OPEN:
3277                 DEBUG_REQ(D_INODE, req, "llog_init");
3278                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3279                 rc = llog_origin_handle_open(req);
3280                 break;
3281         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3282                 DEBUG_REQ(D_INODE, req, "llog next block");
3283                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3284                 rc = llog_origin_handle_next_block(req);
3285                 break;
3286         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3287                 DEBUG_REQ(D_INODE, req, "llog prev block");
3288                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3289                 rc = llog_origin_handle_prev_block(req);
3290                 break;
3291         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3292                 DEBUG_REQ(D_INODE, req, "llog read header");
3293                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3294                 rc = llog_origin_handle_read_header(req);
3295                 break;
3296         case LLOG_ORIGIN_HANDLE_CLOSE:
3297                 DEBUG_REQ(D_INODE, req, "llog close");
3298                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3299                 rc = llog_origin_handle_close(req);
3300                 break;
3301         case OST_CREATE:
3302                 DEBUG_REQ(D_INODE, req, "ost_create");
3303                 rc = mdt_obj_create(req);
3304                 break;
3305         case OST_GET_INFO:
3306                 DEBUG_REQ(D_INODE, req, "get_info");
3307                 rc = mdt_get_info(req);
3308                 break;
3309         case OST_SET_INFO:
3310                 DEBUG_REQ(D_INODE, req, "set_info");
3311                 rc = mdt_set_info(req);
3312                 break;
3313         case OST_WRITE:
3314                 CDEBUG(D_INODE, "write\n");
3315                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
3316                 rc = ost_brw_write(req, NULL);
3317                 LASSERT(current->journal_info == NULL);
3318                 /* mdt_brw sends its own replies */
3319                 RETURN(rc);
3320                 break;
3321         case LLOG_CATINFO:
3322                 DEBUG_REQ(D_INODE, req, "llog catinfo");
3323                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
3324                 rc = llog_catinfo(req);
3325                 break;
3326         default:
3327                 req->rq_status = -ENOTSUPP;
3328                 rc = ptlrpc_error(req);
3329                 RETURN(rc);
3330         }
3331
3332         LASSERT(current->journal_info == NULL);
3333
3334         /* If we're DISCONNECTing, the mds_export_data is already freed */
3335         if (!rc && req->rq_reqmsg->opc != MDS_DISCONNECT) {
3336                 struct mds_export_data *med = &req->rq_export->exp_mds_data;
3337                 struct obd_device *obd = list_entry(mds, struct obd_device,
3338                                                     u.mds);
3339                 req->rq_repmsg->last_xid =
3340                         le64_to_cpu(med->med_mcd->mcd_last_xid);
3341
3342                 if (!obd->obd_no_transno) {
3343                         req->rq_repmsg->last_committed =
3344                                 obd->obd_last_committed;
3345                 } else {
3346                         DEBUG_REQ(D_IOCTL, req,
3347                                   "not sending last_committed update");
3348                 }
3349                 CDEBUG(D_INFO, "last_transno "LPU64", last_committed "LPU64
3350                        ", xid "LPU64"\n",
3351                        mds->mds_last_transno, obd->obd_last_committed,
3352                        req->rq_xid);
3353         }
3354  out:
3355
3356
3357         target_send_reply(req, rc, fail);
3358         RETURN(0);
3359 }
3360
3361 /* Update the server data on disk.  This stores the new mount_count and also the
3362  * last_rcvd value to disk.  If we don't have a clean shutdown, then the server
3363  * last_rcvd value may be less than that of the clients.  This will alert us
3364  * that we may need to do client recovery.
3365  *
3366  * Also assumes for mds_last_transno that we are not modifying it (no locking).
3367  */
3368 int mds_update_server_data(struct obd_device *obd, int force_sync)
3369 {
3370         struct mds_obd *mds = &obd->u.mds;
3371         struct mds_server_data *msd = mds->mds_server_data;
3372         struct file *filp = mds->mds_rcvd_filp;
3373         struct lvfs_run_ctxt saved;
3374         loff_t off = 0;
3375         int rc;
3376         ENTRY;
3377
3378         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3379         msd->msd_last_transno = cpu_to_le64(mds->mds_last_transno);
3380
3381         CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
3382                mds->mds_mount_count, mds->mds_last_transno);
3383         rc = fsfilt_write_record(obd, filp, msd, sizeof(*msd), &off, force_sync);
3384         if (rc)
3385                 CERROR("error writing MDS server data: rc = %d\n", rc);
3386         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3387
3388         RETURN(rc);
3389 }
3390
3391 /* saves last allocated fid counter to file. */
3392 int mds_update_last_fid(struct obd_device *obd, void *handle,
3393                         int force_sync)
3394 {
3395         struct mds_obd *mds = &obd->u.mds;
3396         struct file *filp = mds->mds_fid_filp;
3397         struct lvfs_run_ctxt saved;
3398         loff_t off = 0;
3399         __u64 last_fid;
3400         int rc = 0;
3401         ENTRY;
3402
3403         spin_lock(&mds->mds_last_fid_lock);
3404         last_fid = mds->mds_last_fid;
3405         spin_unlock(&mds->mds_last_fid_lock);
3406
3407         CDEBUG(D_SUPER, "MDS last_fid is #"LPU64"\n",
3408                last_fid);
3409
3410         if (handle) {
3411                 fsfilt_add_journal_cb(obd, mds->mds_sb, last_fid,
3412                                       handle, mds_commit_last_fid_cb,
3413                                       NULL);
3414         }
3415
3416         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3417         rc = fsfilt_write_record(obd, filp, &last_fid, sizeof(last_fid),
3418                                  &off, force_sync);
3419         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3420
3421         if (rc) {
3422                 CERROR("error writing MDS last_fid #"LPU64
3423                        ", err = %d\n", last_fid, rc);
3424                 RETURN(rc);
3425         }
3426
3427         CDEBUG(D_SUPER, "wrote fid #"LPU64" at idx "
3428                "%llu: err = %d\n", last_fid, off, rc);
3429
3430         RETURN(rc);
3431 }
3432
3433 void mds_set_last_fid(struct obd_device *obd, __u64 fid)
3434 {
3435         struct mds_obd *mds = &obd->u.mds;
3436
3437         spin_lock(&mds->mds_last_fid_lock);
3438         if (fid > mds->mds_last_fid)
3439                 mds->mds_last_fid = fid;
3440         spin_unlock(&mds->mds_last_fid_lock);
3441 }
3442
3443 void mds_commit_last_transno_cb(struct obd_device *obd,
3444                                 __u64 transno, void *data,
3445                                 int error)
3446 {
3447         obd_transno_commit_cb(obd, transno, error);
3448 }
3449
3450 void mds_commit_last_fid_cb(struct obd_device *obd,
3451                             __u64 fid, void *data,
3452                             int error)
3453 {
3454         if (error) {
3455                 CERROR("%s: fid "LPD64" commit error: %d\n",
3456                        obd->obd_name, fid, error);
3457                 return;
3458         }
3459
3460         CDEBUG(D_HA, "%s: fid "LPD64" committed\n",
3461                obd->obd_name, fid);
3462 }
3463
3464 __u64 mds_alloc_fid(struct obd_device *obd)
3465 {
3466         struct mds_obd *mds = &obd->u.mds;
3467         __u64 fid;
3468
3469         spin_lock(&mds->mds_last_fid_lock);
3470         fid = ++mds->mds_last_fid;
3471         spin_unlock(&mds->mds_last_fid_lock);
3472
3473         return fid;
3474 }
3475
3476 /*
3477  * reads inode self id from inode EA. Probably later this should be replaced by
3478  * caching inode self id to avoid raeding it every time it is needed.
3479  */
3480 int mds_read_inode_sid(struct obd_device *obd, struct inode *inode,
3481                        struct lustre_id *id)
3482 {
3483         int rc;
3484         ENTRY;
3485
3486         LASSERT(id != NULL);
3487         LASSERT(obd != NULL);
3488         LASSERT(inode != NULL);
3489
3490         rc = fsfilt_get_md(obd, inode, &id->li_fid,
3491                            sizeof(id->li_fid), EA_SID);
3492         if (rc < 0) {
3493                 CERROR("fsfilt_get_md() failed, "
3494                        "rc = %d\n", rc);
3495                 RETURN(rc);
3496         } else if (!rc) {
3497                 rc = -ENODATA;
3498                 RETURN(rc);
3499         } else {
3500                 rc = 0;
3501         }
3502         id_ino(id) = inode->i_ino;
3503         id_gen(id) = inode->i_generation;
3504         id_type(id) = S_IFMT & inode->i_mode;
3505
3506         RETURN(rc);
3507 }
3508
3509 int mds_read_inode_pid(struct obd_device *obd, struct inode *inode,
3510                        struct lustre_id *id)
3511 {
3512         int rc;
3513         ENTRY;
3514
3515         LASSERT(inode && id);
3516
3517         rc = fsfilt_get_md(obd, inode, id, sizeof(*id), EA_PID);
3518         if (rc < 0)
3519                CERROR("get parent id from EA failed, rc=%d\n", rc);
3520         else if (!rc)
3521                 rc = -ENODATA;
3522         else
3523                 rc = 0;
3524
3525         RETURN(rc);
3526 }
3527
3528 /* updates inode self id in EA. */
3529 int mds_update_inode_ids(struct obd_device *obd, struct inode *inode,
3530                          void *handle, struct lustre_id *id,
3531                          struct lustre_id *pid)
3532 {
3533         int rc = 0;
3534         ENTRY;
3535
3536         LASSERT(id || pid);
3537         LASSERT(id == NULL || id_fid(id) != 0);
3538         LASSERT(pid == NULL || id_fid(pid) != 0);
3539         LASSERT(obd != NULL);
3540         LASSERT(inode != NULL);
3541
3542         if (id) {
3543                 mds_set_last_fid(obd, id_fid(id));
3544                 rc = fsfilt_set_md(obd, inode, handle, &id->li_fid,
3545                                    sizeof(id->li_fid), EA_SID);
3546                 LASSERTF(rc == 0, "failed to update fid:  %d\n", rc);
3547         }
3548         if (pid) {
3549                 rc = fsfilt_set_md(obd, inode, handle, pid,
3550                                    sizeof(*pid), EA_PID);
3551                 LASSERTF(rc == 0, "failed to update parent fid:  %d\n", rc);
3552         }
3553
3554         RETURN(rc);
3555 }
3556
3557 /* mount the file system (secretly) */
3558 static int mds_setup(struct obd_device *obd, obd_count len, void *buf)
3559 {
3560         struct lustre_cfg* lcfg = buf;
3561         struct mds_obd *mds = &obd->u.mds;
3562         struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
3563         char *options = NULL;
3564         struct vfsmount *mnt;
3565         char ns_name[48];
3566         unsigned long page;
3567         struct crypto_tfm *tfm = NULL;
3568         int rc = 0;
3569         ENTRY;
3570
3571         if (lcfg->lcfg_bufcount < 3)
3572                 RETURN(rc = -EINVAL);
3573
3574         if (LUSTRE_CFG_BUFLEN(lcfg, 1) == 0 || LUSTRE_CFG_BUFLEN(lcfg, 2) == 0)
3575                 RETURN(rc = -EINVAL);
3576
3577         obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
3578         if (IS_ERR(obd->obd_fsops))
3579                 RETURN(rc = PTR_ERR(obd->obd_fsops));
3580
3581         mds->mds_max_mdsize = sizeof(struct lov_mds_md);
3582
3583         page = get_zeroed_page(GFP_KERNEL);
3584         if (!page)
3585                 RETURN(-ENOMEM);
3586
3587         options = (char *)page;
3588
3589         /*
3590          * here we use "iopen_nopriv" hardcoded, because it affects MDS utility
3591          * and the rest of options are passed by mount options. Probably this
3592          * should be moved to somewhere else like startup scripts or lconf. */
3593         sprintf(options, "iopen_nopriv");
3594
3595         if (LUSTRE_CFG_BUFLEN(lcfg, 4) > 0 && lustre_cfg_buf(lcfg, 4))
3596                 sprintf(options + strlen(options), ",%s",
3597                         lustre_cfg_string(lcfg, 4));
3598
3599         /* we have to know mdsnum before touching underlying fs -bzzz */
3600         atomic_set(&mds->mds_open_count, 0);
3601         sema_init(&mds->mds_md_sem, 1);
3602         mds->mds_md_connected = 0;
3603         mds->mds_md_name = NULL;
3604
3605         if (LUSTRE_CFG_BUFLEN(lcfg, 5) > 0 && lustre_cfg_buf(lcfg, 5) &&
3606             strncmp(lustre_cfg_string(lcfg, 5), "dumb", LUSTRE_CFG_BUFLEN(lcfg, 5))) {
3607                 class_uuid_t uuid;
3608
3609                 generate_random_uuid(uuid);
3610                 class_uuid_unparse(uuid, &mds->mds_md_uuid);
3611
3612                 OBD_ALLOC(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3613                 if (mds->mds_md_name == NULL)
3614                         RETURN(rc = -ENOMEM);
3615
3616                 memcpy(mds->mds_md_name, lustre_cfg_buf(lcfg, 5),
3617                        LUSTRE_CFG_BUFLEN(lcfg, 5));
3618
3619                 CDEBUG(D_OTHER, "MDS: %s is master for %s\n",
3620                        obd->obd_name, mds->mds_md_name);
3621
3622                 rc = mds_md_connect(obd, mds->mds_md_name);
3623                 if (rc) {
3624                         OBD_FREE(mds->mds_md_name, LUSTRE_CFG_BUFLEN(lcfg, 5));
3625                         GOTO(err_ops, rc);
3626                 }
3627         }
3628
3629         mds->mds_obd_type = MDS_MASTER_OBD;
3630
3631         if (LUSTRE_CFG_BUFLEN(lcfg, 6) > 0 && lustre_cfg_buf(lcfg, 6) &&
3632             strncmp(lustre_cfg_string(lcfg, 6), "dumb",
3633                     LUSTRE_CFG_BUFLEN(lcfg, 6))) {
3634                 if (!memcmp(lustre_cfg_string(lcfg, 6), "master",
3635                             strlen("master"))) {
3636                         mds->mds_obd_type = MDS_MASTER_OBD;
3637                 } else if (!memcmp(lustre_cfg_string(lcfg, 6), "cache",
3638                                    strlen("cache"))) {
3639                         mds->mds_obd_type = MDS_CACHE_OBD;
3640                 }
3641         }
3642
3643         rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
3644                            lustre_cfg_string(lcfg, 2),
3645                            options, 0, &lvfs_ctxt);
3646
3647         free_page(page);
3648
3649         if (rc || !lvfs_ctxt) {
3650                 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
3651                 GOTO(err_ops, rc);
3652         }
3653
3654         mnt = lvfs_ctxt->loc_mnt;
3655         mds->mds_lvfs_ctxt = lvfs_ctxt;
3656         ll_clear_rdonly(ll_sbdev(mnt->mnt_sb));
3657
3658         CDEBUG(D_SUPER, "%s: mnt = %p\n", lustre_cfg_string(lcfg, 1), mnt);
3659
3660         mds->mds_fidext_thumb = 0;
3661         sema_init(&mds->mds_epoch_sem, 1);
3662         atomic_set(&mds->mds_real_clients, 0);
3663         spin_lock_init(&mds->mds_fidext_lock);
3664         spin_lock_init(&mds->mds_fidmap_lock);
3665         spin_lock_init(&mds->mds_transno_lock);
3666         spin_lock_init(&mds->mds_last_fid_lock);
3667         sema_init(&mds->mds_orphan_recovery_sem, 1);
3668         mds->mds_max_cookiesize = sizeof(struct llog_cookie);
3669
3670         sprintf(ns_name, "mds-%s", obd->obd_uuid.uuid);
3671         obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
3672
3673         if (obd->obd_namespace == NULL) {
3674                 mds_cleanup(obd, 0);
3675                 GOTO(err_put, rc = -ENOMEM);
3676         }
3677         ldlm_register_intent(obd->obd_namespace, mds_intent_policy);
3678
3679         tfm = crypto_alloc_tfm(CAPA_HMAC_ALG, 0);
3680         if (!tfm)
3681                 GOTO(err_ns, rc = -ENOSYS);
3682
3683         mds->mds_capa_hmac = tfm;
3684         mds->mds_capa_timeout = CAPA_TIMEOUT;
3685         mds->mds_capa_key_timeout = CAPA_KEY_TIMEOUT;
3686
3687         rc = mds_fs_setup(obd, mnt);
3688         if (rc) {
3689                 CERROR("%s: MDS filesystem method init failed: rc = %d\n",
3690                        obd->obd_name, rc);
3691                 GOTO(err_capa, rc);
3692         }
3693
3694         rc = llog_start_commit_thread();
3695         if (rc < 0)
3696
3697                 GOTO(err_fs, rc);
3698
3699         if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3) &&
3700             strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3701                     LUSTRE_CFG_BUFLEN(lcfg, 3))) {
3702                 class_uuid_t uuid;
3703
3704                 generate_random_uuid(uuid);
3705                 class_uuid_unparse(uuid, &mds->mds_dt_uuid);
3706
3707                 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
3708                 if (mds->mds_profile == NULL)
3709                         GOTO(err_fs, rc = -ENOMEM);
3710
3711                 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
3712                         LUSTRE_CFG_BUFLEN(lcfg, 3));
3713         }
3714
3715         /*
3716          * setup root dir and files ID dir if lmv already connected, or there is
3717          * not lmv at all.
3718          */
3719         if (mds->mds_md_exp || (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 &&
3720                                 lustre_cfg_buf(lcfg, 3) &&
3721                                 strncmp(lustre_cfg_string(lcfg, 3), "dumb",
3722                                         LUSTRE_CFG_BUFLEN(lcfg, 3)))) {
3723                 rc = mds_fs_setup_rootid(obd);
3724                 if (rc)
3725                         GOTO(err_fs, rc);
3726
3727                 rc = mds_fs_setup_virtid(obd);
3728                 if (rc)
3729                         GOTO(err_fs, rc);
3730
3731         }
3732
3733         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3734                            "mds_ldlm_client", &obd->obd_ldlm_client);
3735         obd->obd_replayable = 1;
3736
3737         mds->mds_crypto_type = NO_CRYPTO;
3738
3739         rc = mds_postsetup(obd);
3740         if (rc)
3741                 GOTO(err_fs, rc);
3742
3743         RETURN(0);
3744
3745 err_fs:
3746         /* No extra cleanup needed for llog_init_commit_thread() */
3747         mds_fs_cleanup(obd, 0);
3748 err_capa:
3749         crypto_free_tfm(mds->mds_capa_hmac);
3750 err_ns:
3751         ldlm_namespace_free(obd->obd_namespace, 0);
3752         obd->obd_namespace = NULL;
3753 err_put:
3754         unlock_kernel();
3755         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3756         mds->mds_sb = 0;
3757         lock_kernel();
3758 err_ops:
3759         fsfilt_put_ops(obd->obd_fsops);
3760         return rc;
3761 }
3762
3763 static int mds_fs_post_setup(struct obd_device *obd)
3764 {
3765         struct mds_obd *mds = &obd->u.mds;
3766         struct dentry *dentry;
3767         int rc = 0;
3768         ENTRY;
3769
3770         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
3771         if (IS_ERR(dentry)) {
3772                 CERROR("Can't find ROOT, err = %d\n",
3773                        (int)PTR_ERR(dentry));
3774                 RETURN(PTR_ERR(dentry));
3775         }
3776         rc = fsfilt_post_setup(obd, dentry);
3777         //set id2name function handler
3778         fsfilt_set_info(obd, mds->mds_sb, NULL, 7, "id2name",
3779                         sizeof(mds_audit_id2name), mds_audit_id2name);
3780
3781         l_dput(dentry);
3782         RETURN(rc);
3783 }
3784
3785 static int mds_postsetup(struct obd_device *obd)
3786 {
3787         struct mds_obd *mds = &obd->u.mds;
3788         int rc = 0;
3789         ENTRY;
3790
3791         rc = obd_llog_setup(obd, &obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT,
3792                             obd, 0, NULL, &llog_lvfs_ops);
3793         if (rc)
3794                 RETURN(rc);
3795
3796         if (mds->mds_profile) {
3797                 struct llog_ctxt *lgctxt;
3798                 struct lvfs_run_ctxt saved;
3799                 struct lustre_profile *lprof;
3800                 struct config_llog_instance cfg;
3801
3802                 cfg.cfg_instance = NULL;
3803                 cfg.cfg_uuid = mds->mds_dt_uuid;
3804                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3805
3806                 lgctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
3807                 if (!lgctxt) {
3808                         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3809                         GOTO(err_llog, rc = -EINVAL);
3810                 }
3811
3812                 rc = class_config_process_llog(lgctxt, mds->mds_profile, &cfg);
3813                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3814
3815                 if (rc)
3816                         GOTO(err_llog, rc);
3817
3818                 lprof = class_get_profile(mds->mds_profile);
3819                 if (lprof == NULL) {
3820                         CERROR("No profile found: %s\n", mds->mds_profile);
3821                         GOTO(err_cleanup, rc = -ENOENT);
3822                 }
3823                 rc = mds_dt_connect(obd, lprof->lp_lov);
3824                 if (rc)
3825                         GOTO(err_cleanup, rc);
3826
3827                 rc = mds_md_postsetup(obd);
3828                 if (rc)
3829                         GOTO(err_cleanup, rc);
3830         }
3831         rc = mds_fs_post_setup(obd);
3832         if (rc)
3833                 CERROR("can not post setup fsfilt\n");
3834         RETURN(rc);
3835 err_cleanup:
3836         mds_dt_clean(obd);
3837 err_llog:
3838         obd_llog_cleanup(llog_get_context(&obd->obd_llogs,
3839                                           LLOG_CONFIG_ORIG_CTXT));
3840         return rc;
3841 }
3842
3843 int mds_postrecov_common(struct obd_device *obd)
3844 {
3845         struct mds_obd *mds = &obd->u.mds;
3846         struct llog_ctxt *ctxt;
3847         int rc, item = 0, valsize;
3848          __u32 group;
3849         ENTRY;
3850
3851         LASSERT(!obd->obd_recovering);
3852         ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT);
3853         LASSERT(ctxt != NULL);
3854
3855         /* clean PENDING dir */
3856         rc = mds_cleanup_orphans(obd);
3857         if (rc < 0)
3858                 GOTO(out, rc);
3859         item = rc;
3860
3861         group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
3862         valsize = sizeof(group);
3863         rc = obd_set_info(mds->mds_dt_exp, strlen("mds_conn"),
3864                           "mds_conn", valsize, &group);
3865         if (rc)
3866                 GOTO(out, rc);
3867
3868         rc = llog_connect(ctxt, obd->u.mds.mds_dt_desc.ld_tgt_count,
3869                           NULL, NULL, NULL);
3870         if (rc) {
3871                 CERROR("%s: failed at llog_origin_connect: %d\n",
3872                        obd->obd_name, rc);
3873                 GOTO(out, rc);
3874         }
3875
3876         /* remove the orphaned precreated objects */
3877         rc = mds_dt_clear_orphans(mds, NULL /* all OSTs */);
3878         if (rc)
3879                 GOTO(err_llog, rc);
3880
3881 out:
3882         RETURN(rc < 0 ? rc : item);
3883
3884 err_llog:
3885         /* cleanup all llogging subsystems */
3886         rc = obd_llog_finish(obd, &obd->obd_llogs,
3887                              mds->mds_dt_desc.ld_tgt_count);
3888         if (rc)
3889                 CERROR("%s: failed to cleanup llogging subsystems\n",
3890                         obd->obd_name);
3891         goto out;
3892 }
3893
3894 int mds_postrecov(struct obd_device *obd)
3895 {
3896         int rc;
3897         ENTRY;
3898         rc = mds_postrecov_common(obd);
3899         if (rc == 0)
3900                 rc = mds_md_reconnect(obd);
3901         RETURN(rc);
3902 }
3903
3904 int mds_dt_clean(struct obd_device *obd)
3905 {
3906         struct mds_obd *mds = &obd->u.mds;
3907         ENTRY;
3908
3909         if (mds->mds_profile) {
3910                 char * cln_prof;
3911                 struct llog_ctxt *llctx;
3912                 struct lvfs_run_ctxt saved;
3913                 struct config_llog_instance cfg;
3914                 int len = strlen(mds->mds_profile) + sizeof("-clean") + 1;
3915
3916                 OBD_ALLOC(cln_prof, len);
3917                 if (!cln_prof) {
3918                         CERROR("can't allocate memory, processing cleanup "
3919                                "profile is skipped\n");
3920                         goto out;
3921                 }
3922
3923                 sprintf(cln_prof, "%s-clean", mds->mds_profile);
3924
3925                 cfg.cfg_instance = NULL;
3926                 cfg.cfg_uuid = mds->mds_dt_uuid;
3927
3928                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3929                 llctx = llog_get_context(&obd->obd_llogs,
3930                                          LLOG_CONFIG_ORIG_CTXT);
3931                 class_config_process_llog(llctx, cln_prof, &cfg);
3932                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
3933
3934                 OBD_FREE(cln_prof, len);
3935         out:
3936                 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
3937                 mds->mds_profile = NULL;
3938         }
3939         RETURN(0);
3940 }
3941
3942 int mds_md_clean(struct obd_device *obd)
3943 {
3944         struct mds_obd *mds = &obd->u.mds;
3945         ENTRY;
3946
3947         if (mds->mds_md_name) {
3948                 OBD_FREE(mds->mds_md_name, strlen(mds->mds_md_name) + 1);
3949                 mds->mds_md_name = NULL;
3950         }
3951         RETURN(0);
3952 }
3953
3954 static int mds_precleanup(struct obd_device *obd, int flags)
3955 {
3956         int rc = 0;
3957         ENTRY;
3958
3959         mds_md_clean(obd);
3960         mds_dt_disconnect(obd, flags);
3961         mds_dt_clean(obd);
3962         obd_llog_cleanup(llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT));
3963         RETURN(rc);
3964 }
3965
3966 extern void lgss_svc_cache_purge_all(void);
3967 static int mds_cleanup(struct obd_device *obd, int flags)
3968 {
3969         struct mds_obd *mds = &obd->u.mds;
3970         ENTRY;
3971
3972         if (mds->mds_sb == NULL)
3973                 RETURN(0);
3974
3975         mds_update_server_data(obd, 1);
3976         mds_update_last_fid(obd, NULL, 1);
3977
3978         if (mds->mds_dt_objids != NULL) {
3979                 int size = mds->mds_dt_desc.ld_tgt_count *
3980                         sizeof(obd_id);
3981                 OBD_FREE(mds->mds_dt_objids, size);
3982         }
3983         mds_fs_cleanup(obd, flags);
3984
3985         unlock_kernel();
3986
3987         /* 2 seems normal on mds, (may_umount() also expects 2
3988           fwiw), but we only see 1 at this point in obdfilter. */
3989         lvfs_umount_fs(mds->mds_lvfs_ctxt);
3990
3991         mds->mds_sb = 0;
3992
3993         ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
3994
3995         spin_lock_bh(&obd->obd_processing_task_lock);
3996         if (obd->obd_recovering) {
3997                 target_cancel_recovery_timer(obd);
3998                 obd->obd_recovering = 0;
3999         }
4000         spin_unlock_bh(&obd->obd_processing_task_lock);
4001
4002         lock_kernel();
4003         fsfilt_put_ops(obd->obd_fsops);
4004
4005 #ifdef ENABLE_GSS
4006         /* XXX */
4007         lgss_svc_cache_purge_all();
4008 #endif
4009
4010         spin_lock(&mds->mds_denylist_lock);
4011         while (!list_empty( &mds->mds_denylist ) ) {
4012                 deny_sec_t *p_deny_sec = list_entry(mds->mds_denylist.next,
4013                                                     deny_sec_t, list);
4014                 list_del(&p_deny_sec->list);
4015                 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
4016         }
4017         spin_unlock(&mds->mds_denylist_lock);
4018
4019         mds_capa_keys_cleanup(obd);
4020
4021         if (mds->mds_capa_hmac)
4022                 crypto_free_tfm(mds->mds_capa_hmac);
4023         RETURN(0);
4024 }
4025
4026 static int set_security(const char *value, char **sec)
4027 {
4028         if (!strcmp(value, "null"))
4029                 *sec = "null";
4030         else if (!strcmp(value, "krb5i"))
4031                 *sec = "krb5i";
4032         else if (!strcmp(value, "krb5p"))
4033                 *sec = "krb5p";
4034         else {
4035                 CERROR("Unrecognized security flavor %s\n", value);
4036                 return -EINVAL;
4037         }
4038
4039         return 0;
4040 }
4041
4042 static int mds_process_config(struct obd_device *obd, obd_count len, void *buf)
4043 {
4044         struct lustre_cfg *lcfg = buf;
4045         struct mds_obd *mds = &obd->u.mds;
4046         int rc = 0;
4047         ENTRY;
4048
4049         switch(lcfg->lcfg_command) {
4050         case LCFG_SET_SECURITY: {
4051                 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
4052                     (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
4053                         GOTO(out, rc = -EINVAL);
4054
4055                 if (!strcmp(lustre_cfg_string(lcfg, 1), "mds_sec"))
4056                         rc = set_security(lustre_cfg_string(lcfg, 2),
4057                                           &mds->mds_mds_sec);
4058                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "oss_sec"))
4059                         rc = set_security(lustre_cfg_string(lcfg, 2),
4060                                           &mds->mds_ost_sec);
4061                 else if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
4062                         spin_lock(&mds->mds_denylist_lock);
4063                         rc = add_deny_security(lustre_cfg_string(lcfg, 2),
4064                                                &mds->mds_denylist);
4065                         spin_unlock(&mds->mds_denylist_lock);
4066                 } else {
4067                         CERROR("Unrecognized key\n");
4068                         rc = -EINVAL;
4069                 }
4070                 break;
4071         }
4072         default:
4073                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
4074                 GOTO(out, rc = -EINVAL);
4075         }
4076 out:
4077         RETURN(rc);
4078 }
4079
4080 static void fixup_handle_for_resent_req(struct ptlrpc_request *req,
4081                                         int offset,
4082                                         struct ldlm_lock *new_lock,
4083                                         struct ldlm_lock **old_lock,
4084                                         struct lustre_handle *lockh)
4085 {
4086         struct obd_export *exp = req->rq_export;
4087         struct obd_device *obd = exp->exp_obd;
4088         struct ldlm_request *dlmreq =
4089                 lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*dlmreq));
4090         struct lustre_handle remote_hdl = dlmreq->lock_handle1;
4091         struct list_head *iter;
4092
4093         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4094                 return;
4095
4096         spin_lock(&obd->obd_namespace->ns_hash_lock);
4097         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
4098                 struct ldlm_lock *lock;
4099                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
4100                 if (lock == new_lock)
4101                         continue;
4102                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
4103                         lockh->cookie = lock->l_handle.h_cookie;
4104                         LDLM_DEBUG(lock, "restoring lock cookie");
4105                         DEBUG_REQ(D_HA, req, "restoring lock cookie "LPX64,
4106                                   lockh->cookie);
4107                         if (old_lock)
4108                                 *old_lock = LDLM_LOCK_GET(lock);
4109                         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4110                         return;
4111                 }
4112         }
4113         spin_unlock(&obd->obd_namespace->ns_hash_lock);
4114
4115         /* If the xid matches, then we know this is a resent request,
4116          * and allow it. (It's probably an OPEN, for which we don't
4117          * send a lock */
4118         if (req->rq_xid ==
4119             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_xid))
4120                 return;
4121
4122         if (req->rq_xid ==
4123             le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_close_xid))
4124                 return;
4125
4126         /* This remote handle isn't enqueued, so we never received or
4127          * processed this request.  Clear MSG_RESENT, because it can
4128          * be handled like any normal request now. */
4129
4130         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4131
4132         DEBUG_REQ(D_HA, req, "no existing lock with rhandle "LPX64,
4133                   remote_hdl.cookie);
4134 }
4135
4136 int intent_disposition(struct ldlm_reply *rep, int flag)
4137 {
4138         if (!rep)
4139                 return 0;
4140         return (rep->lock_policy_res1 & flag);
4141 }
4142
4143 void intent_set_disposition(struct ldlm_reply *rep, int flag)
4144 {
4145         if (!rep)
4146                 return;
4147         rep->lock_policy_res1 |= flag;
4148 }
4149
4150 static int mds_intent_prepare_reply_buffers(struct ptlrpc_request *req,
4151                                             struct ldlm_intent *it)
4152 {
4153         struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
4154         struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
4155         int rc, reply_buffers;
4156         int repsize[8] = {sizeof(struct ldlm_reply),
4157                           sizeof(struct mds_body),
4158                           mds->mds_max_mdsize};
4159         ENTRY;
4160
4161         reply_buffers = 3;
4162         if (it->opc & ( IT_OPEN | IT_GETATTR | IT_LOOKUP | IT_CHDIR )) {
4163                 /*Ugly here, Actually, we should prepare the reply buffer
4164                  *after we know whether these stuff exist or not, which should
4165                  * be fixed in future, Now each item is in the fix position,
4166                  * the sequence is lsm, acl, crypto ea, capa.*/
4167                 repsize[reply_buffers++] = sizeof(int);
4168                 if (med->med_remote)
4169                         repsize[reply_buffers++] =
4170                                 sizeof(struct mds_remote_perm);
4171                 else
4172                         repsize[reply_buffers++] =
4173                                 xattr_acl_size(LL_ACL_MAX_ENTRIES);
4174
4175                 repsize[reply_buffers++] = sizeof(int);
4176                 repsize[reply_buffers++] = sizeof(struct crypto_key);
4177
4178                 /* XXX: if new buffer is to be added, capability reply
4179                  *      buffer should always been reserved. */
4180                 if (it->opc & IT_OPEN)
4181                         repsize[reply_buffers++] = sizeof(struct lustre_capa);
4182         }
4183
4184         rc = lustre_pack_reply(req, reply_buffers, repsize, NULL);
4185
4186         RETURN(rc);
4187 }
4188
4189 static int mds_intent_policy(struct ldlm_namespace *ns,
4190                              struct ldlm_lock **lockp, void *req_cookie,
4191                              ldlm_mode_t mode, int flags, void *data)
4192 {
4193         struct ptlrpc_request *req = req_cookie;
4194         struct ldlm_lock *lock = *lockp;
4195         struct ldlm_intent *it;
4196         struct ldlm_reply *rep;
4197         struct lustre_handle lockh[2] = {{0}, {0}};
4198         struct ldlm_lock *new_lock = NULL;
4199         int getattr_part = MDS_INODELOCK_UPDATE;
4200         int rc;
4201
4202         int offset = MDS_REQ_INTENT_REC_OFF;
4203         ENTRY;
4204
4205         LASSERT(req != NULL);
4206         MD_COUNTER_INCREMENT(req->rq_export->exp_obd, intent_lock);
4207
4208         if (req->rq_reqmsg->bufcount <= MDS_REQ_INTENT_IT_OFF) {
4209                 /* No intent was provided */
4210                 int size = sizeof(struct ldlm_reply);
4211                 rc = lustre_pack_reply(req, 1, &size, NULL);
4212                 LASSERT(rc == 0);
4213                 RETURN(0);
4214         }
4215
4216         it = lustre_swab_reqbuf(req, MDS_REQ_INTENT_IT_OFF, sizeof(*it),
4217                                 lustre_swab_ldlm_intent);
4218         if (it == NULL) {
4219                 CERROR("Intent missing\n");
4220                 RETURN(req->rq_status = -EFAULT);
4221         }
4222
4223         LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
4224
4225         rc = mds_intent_prepare_reply_buffers(req, it);
4226
4227         if (rc)
4228                 RETURN(req->rq_status = rc);
4229
4230         rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
4231         LASSERT(rep != NULL);
4232
4233         intent_set_disposition(rep, DISP_IT_EXECD);
4234
4235         /* execute policy */
4236         switch ((long)it->opc) {
4237         case IT_OPEN:
4238         case IT_CREAT|IT_OPEN:
4239                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4240                                             lock, NULL, lockh);
4241                 /* XXX swab here to assert that an mds_open reint
4242                  * packet is following */
4243                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4244                                             lock, NULL, lockh);
4245                 rep->lock_policy_res2 = mds_reint(req, offset, lockh);
4246
4247                 if (rep->lock_policy_res2) {
4248                         /*
4249                          * mds_open() returns ENOLCK where it should return
4250                          * zero, but it has no lock to return.
4251                          */
4252                         if (rep->lock_policy_res2 == ENOLCK)
4253                                 rep->lock_policy_res2 = 0;
4254
4255                         RETURN(ELDLM_LOCK_ABORTED);
4256                 }
4257
4258                 /*
4259                  * IT_OPEN may return lock on cross-node dentry that we want to
4260                  * hold during attr retrival -bzzz
4261                  */
4262                 if (lockh[0].cookie == 0)
4263                         RETURN(ELDLM_LOCK_ABORTED);
4264
4265                 break;
4266         case IT_LOOKUP:
4267                 getattr_part = MDS_INODELOCK_LOOKUP;
4268         case IT_CHDIR:
4269         case IT_GETATTR:
4270                 getattr_part |= MDS_INODELOCK_LOOKUP;
4271         case IT_READDIR:
4272                 fixup_handle_for_resent_req(req, MDS_REQ_INTENT_LOCKREQ_OFF,
4273                                             lock, &new_lock, lockh);
4274                 rep->lock_policy_res2 = mds_getattr_lock(req, offset, lockh,
4275                                                          getattr_part);
4276                 /* FIXME: LDLM can set req->rq_status. MDS sets
4277                    policy_res{1,2} with disposition and status.
4278                    - replay: returns 0 & req->status is old status
4279                    - otherwise: returns req->status */
4280                 if (intent_disposition(rep, DISP_LOOKUP_NEG))
4281                         rep->lock_policy_res2 = 0;
4282                 if (!intent_disposition(rep, DISP_LOOKUP_POS) ||
4283                     rep->lock_policy_res2)
4284                         RETURN(ELDLM_LOCK_ABORTED);
4285                 if (req->rq_status != 0) {
4286                         LBUG();
4287                         rep->lock_policy_res2 = req->rq_status;
4288                         RETURN(ELDLM_LOCK_ABORTED);
4289                 }
4290                 break;
4291         case IT_UNLINK:
4292                 rc = mds_lock_and_check_slave(offset, req, lockh);
4293                 if ((rep->lock_policy_res2 = rc)) {
4294                         if (rc == ENOLCK)
4295                                 rep->lock_policy_res2 = 0;
4296                         RETURN(ELDLM_LOCK_ABORTED);
4297                 }
4298                 break;
4299         default:
4300                 CERROR("Unhandled intent "LPD64"\n", it->opc);
4301                 LBUG();
4302         }
4303
4304         /* By this point, whatever function we called above must have either
4305          * filled in 'lockh', been an intent replay, or returned an error.  We
4306          * want to allow replayed RPCs to not get a lock, since we would just
4307          * drop it below anyways because lock replay is done separately by the
4308          * client afterwards.  For regular RPCs we want to give the new lock to
4309          * the client instead of whatever lock it was about to get. */
4310         if (new_lock == NULL)
4311                 new_lock = ldlm_handle2lock(&lockh[0]);
4312         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY))
4313                 RETURN(0);
4314
4315         LASSERTF(new_lock != NULL, "op "LPX64" lockh "LPX64"\n",
4316                  it->opc, lockh[0].cookie);
4317
4318         /* If we've already given this lock to a client once, then we should
4319          * have no readers or writers.  Otherwise, we should have one reader
4320          * _or_ writer ref (which will be zeroed below) before returning the
4321          * lock to a client. */
4322         if (new_lock->l_export == req->rq_export) {
4323                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4324         } else {
4325                 LASSERT(new_lock->l_export == NULL);
4326                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4327         }
4328
4329         *lockp = new_lock;
4330
4331         if (new_lock->l_export == req->rq_export) {
4332                 /* Already gave this to the client, which means that we
4333                  * reconstructed a reply. */
4334                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4335                         MSG_RESENT);
4336                 RETURN(ELDLM_LOCK_REPLACED);
4337         }
4338
4339         /* Fixup the lock to be given to the client */
4340         lock_res_and_lock(new_lock);
4341         new_lock->l_readers = 0;
4342         new_lock->l_writers = 0;
4343
4344         new_lock->l_export = class_export_get(req->rq_export);
4345
4346         spin_lock(&new_lock->l_export->exp_ldlm_data.led_lock);
4347         list_add(&new_lock->l_export_chain,
4348                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
4349         spin_unlock(&new_lock->l_export->exp_ldlm_data.led_lock);
4350
4351         new_lock->l_blocking_ast = lock->l_blocking_ast;
4352         new_lock->l_completion_ast = lock->l_completion_ast;
4353
4354         memcpy(&new_lock->l_remote_handle, &lock->l_remote_handle,
4355                sizeof(lock->l_remote_handle));
4356
4357         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4358
4359         unlock_res_and_lock(new_lock);
4360         LDLM_LOCK_PUT(new_lock);
4361
4362         RETURN(ELDLM_LOCK_REPLACED);
4363 }
4364
4365 int mds_attach(struct obd_device *dev, obd_count len, void *data)
4366 {
4367         struct lprocfs_static_vars lvars;
4368         int rc = 0;
4369         struct mds_obd *mds = &dev->u.mds;
4370
4371         spin_lock_init(&mds->mds_denylist_lock);
4372         INIT_LIST_HEAD(&mds->mds_denylist);
4373
4374         lprocfs_init_multi_vars(0, &lvars);
4375
4376         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
4377         if (rc)
4378                 return rc;
4379
4380         return lprocfs_alloc_md_stats(dev, 0);
4381 }
4382
4383 int mds_detach(struct obd_device *dev)
4384 {
4385         lprocfs_free_md_stats(dev);
4386         return lprocfs_obd_detach(dev);
4387 }
4388
4389 int mdt_attach(struct obd_device *dev, obd_count len, void *data)
4390 {
4391         struct lprocfs_static_vars lvars;
4392
4393         lprocfs_init_multi_vars(1, &lvars);
4394         return lprocfs_obd_attach(dev, lvars.obd_vars);
4395 }
4396
4397 int mdt_detach(struct obd_device *dev)
4398 {
4399         return lprocfs_obd_detach(dev);
4400 }
4401
4402 static int mdt_setup(struct obd_device *obd, obd_count len, void *buf)
4403 {
4404         struct mds_obd *mds = &obd->u.mds;
4405         int rc = 0;
4406         ENTRY;
4407
4408         mds->mds_service =
4409                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4410                                 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
4411                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4412                                 mds_handle, "mds", obd->obd_proc_entry);
4413
4414         if (!mds->mds_service) {
4415                 CERROR("failed to start service\n");
4416                 RETURN(-ENOMEM);
4417         }
4418
4419         rc = ptlrpc_start_n_threads(obd, mds->mds_service, MDT_NUM_THREADS,
4420                                     "ll_mdt");
4421         if (rc)
4422                 GOTO(err_thread, rc);
4423
4424         mds->mds_setattr_service =
4425                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4426                                 MDS_SETATTR_PORTAL, MDC_REPLY_PORTAL,
4427                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4428                                 mds_handle, "mds_setattr",
4429                                 obd->obd_proc_entry);
4430         if (!mds->mds_setattr_service) {
4431                 CERROR("failed to start getattr service\n");
4432                 GOTO(err_thread, rc = -ENOMEM);
4433         }
4434
4435         rc = ptlrpc_start_n_threads(obd, mds->mds_setattr_service,
4436                                     MDT_NUM_THREADS, "ll_mdt_attr");
4437         if (rc)
4438                 GOTO(err_thread2, rc);
4439
4440         mds->mds_readpage_service =
4441                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4442                                 MDS_READPAGE_PORTAL, MDC_REPLY_PORTAL,
4443                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4444                                 mds_handle, "mds_readpage",
4445                                 obd->obd_proc_entry);
4446         if (!mds->mds_readpage_service) {
4447                 CERROR("failed to start readpage service\n");
4448                 GOTO(err_thread2, rc = -ENOMEM);
4449         }
4450
4451         rc = ptlrpc_start_n_threads(obd, mds->mds_readpage_service,
4452                                     MDT_NUM_THREADS, "ll_mdt_rdpg");
4453
4454         if (rc)
4455                 GOTO(err_thread3, rc);
4456
4457         mds->mds_close_service =
4458                 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
4459                                 MDS_CLOSE_PORTAL, MDC_REPLY_PORTAL,
4460                                 MDS_SERVICE_WATCHDOG_TIMEOUT,
4461                                 mds_handle, "mds_close",
4462                                 obd->obd_proc_entry);
4463         if (!mds->mds_close_service) {
4464                 CERROR("failed to start close service\n");
4465                 GOTO(err_thread3, rc = -ENOMEM);
4466         }
4467
4468         rc = ptlrpc_start_n_threads(obd, mds->mds_close_service,
4469                                     MDT_NUM_THREADS, "ll_mdt_clos");
4470
4471         if (rc)
4472                 GOTO(err_thread4, rc);
4473         RETURN(0);
4474
4475 err_thread4:
4476         ptlrpc_unregister_service(mds->mds_close_service);
4477 err_thread3:
4478         ptlrpc_unregister_service(mds->mds_readpage_service);
4479 err_thread2:
4480         ptlrpc_unregister_service(mds->mds_setattr_service);
4481 err_thread:
4482         ptlrpc_unregister_service(mds->mds_service);
4483         return rc;
4484 }
4485
4486 static int mdt_cleanup(struct obd_device *obd, int flags)
4487 {
4488         struct mds_obd *mds = &obd->u.mds;
4489         ENTRY;
4490
4491         ptlrpc_stop_all_threads(mds->mds_close_service);
4492         ptlrpc_unregister_service(mds->mds_close_service);
4493
4494         ptlrpc_stop_all_threads(mds->mds_readpage_service);
4495         ptlrpc_unregister_service(mds->mds_readpage_service);
4496
4497         ptlrpc_stop_all_threads(mds->mds_setattr_service);
4498         ptlrpc_unregister_service(mds->mds_setattr_service);
4499
4500         ptlrpc_stop_all_threads(mds->mds_service);
4501         ptlrpc_unregister_service(mds->mds_service);
4502
4503         RETURN(0);
4504 }
4505
4506 static struct dentry *mds_lvfs_id2dentry(__u64 ino, __u32 gen,
4507                                          __u64 gr, void *data)
4508 {
4509         struct lustre_id id;
4510         struct obd_device *obd = data;
4511
4512         id_ino(&id) = ino;
4513         id_gen(&id) = gen;
4514         return mds_id2dentry(obd, &id, NULL);
4515 }
4516
4517 static int mds_get_info(struct obd_export *exp, __u32 keylen,
4518                         void *key, __u32 *valsize, void *val)
4519 {
4520         struct obd_device *obd;
4521         struct mds_obd *mds;
4522         int rc = 0;
4523         ENTRY;
4524
4525         obd = class_exp2obd(exp);
4526         mds = &obd->u.mds;
4527
4528         if (obd == NULL) {
4529                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
4530                        exp->exp_handle.h_cookie);
4531                 RETURN(-EINVAL);
4532         }
4533
4534         if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
4535                 /* get log_context handle. */
4536                 struct llog_ctxt *ctxt;
4537                 unsigned long *llh_handle = val;
4538                 *valsize = sizeof(unsigned long);
4539                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_REINT_ORIG_CTXT);
4540                 if (!ctxt) {
4541                         CERROR("Cannot get REINT llog context\n");
4542                         RETURN(-ENOENT);
4543                 }
4544                 *llh_handle = (unsigned long)ctxt;
4545                 RETURN(0);
4546         }
4547         if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
4548                 /* get log_context handle. */
4549                 unsigned long *sb = val;
4550                 *valsize = sizeof(unsigned long);
4551                 *sb = (unsigned long)obd->u.mds.mds_sb;
4552                 RETURN(0);
4553         }
4554
4555         if (keylen >= strlen("mdsize") && memcmp(key, "mdsize", keylen) == 0) {
4556                 __u32 *mdsize = val;
4557                 *valsize = sizeof(*mdsize);
4558                 *mdsize = mds->mds_max_mdsize;
4559                 RETURN(0);
4560         }
4561
4562         if (keylen >= strlen("mdsnum") && strcmp(key, "mdsnum") == 0) {
4563                 __u32 *mdsnum = val;
4564                 *valsize = sizeof(*mdsnum);
4565                 *mdsnum = mds->mds_num;
4566                 RETURN(0);
4567         }
4568
4569         if (keylen >= strlen("rootid") && strcmp(key, "rootid") == 0) {
4570                 struct lustre_id *rootid = val;
4571                 *valsize = sizeof(*rootid);
4572                 *rootid = mds->mds_rootid;
4573                 RETURN(0);
4574         }
4575
4576         if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
4577                 struct lov_desc *desc = val;
4578                 *valsize = sizeof(*desc);
4579                 *desc = mds->mds_dt_desc;
4580                 RETURN(0);
4581         }
4582
4583         if (keylen >= strlen("getext") && strcmp(key, "getext") == 0) {
4584                 struct fid_extent *ext = val;
4585                 *valsize = sizeof(*ext);
4586
4587                 spin_lock(&mds->mds_fidext_lock);
4588                 ext->fe_width = MDS_FIDEXT_SIZE;
4589                 ext->fe_start = mds->mds_fidext_thumb + 1;
4590                 mds->mds_fidext_thumb += MDS_FIDEXT_SIZE;
4591                 spin_unlock(&mds->mds_fidext_lock);
4592
4593                 RETURN(0);
4594         }
4595
4596         rc = fsfilt_get_info(obd, mds->mds_sb, NULL, keylen, key, valsize, val);
4597         if (rc)
4598                 CDEBUG(D_IOCTL, "invalid key\n");
4599
4600         RETURN(rc);
4601 }
4602
4603 struct lvfs_callback_ops mds_lvfs_ops = {
4604         l_id2dentry:     mds_lvfs_id2dentry,
4605 };
4606
4607 int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
4608                 int objcount, struct obd_ioobj *obj,
4609                 int niocount, struct niobuf_remote *nb,
4610                 struct niobuf_local *res,
4611                 struct obd_trans_info *oti, struct lustre_capa *capa);
4612
4613 int mds_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
4614                  int objcount, struct obd_ioobj *obj, int niocount,
4615                  struct niobuf_local *res, struct obd_trans_info *oti,
4616                  int rc);
4617
4618 /* use obd ops to offer management infrastructure */
4619 static struct obd_ops mds_obd_ops = {
4620         .o_owner           = THIS_MODULE,
4621         .o_attach          = mds_attach,
4622         .o_detach          = mds_detach,
4623         .o_connect         = mds_connect,
4624         .o_connect_post    = mds_connect_post,
4625         .o_init_export     = mds_init_export,
4626         .o_destroy_export  = mds_destroy_export,
4627         .o_disconnect      = mds_disconnect,
4628         .o_setup           = mds_setup,
4629         .o_precleanup      = mds_precleanup,
4630         .o_cleanup         = mds_cleanup,
4631         .o_process_config  = mds_process_config,
4632         .o_postrecov       = mds_postrecov,
4633         .o_statfs          = mds_obd_statfs,
4634         .o_iocontrol       = mds_iocontrol,
4635         .o_create          = mds_obd_create,
4636         .o_destroy         = mds_obd_destroy,
4637         .o_llog_init       = mds_llog_init,
4638         .o_llog_finish     = mds_llog_finish,
4639         .o_notify          = mds_notify,
4640         .o_get_info        = mds_get_info,
4641         .o_set_info        = mds_set_info,
4642         .o_preprw          = mds_preprw,
4643         .o_commitrw        = mds_commitrw,
4644 };
4645
4646 static struct obd_ops mdt_obd_ops = {
4647         .o_owner           = THIS_MODULE,
4648         .o_attach          = mdt_attach,
4649         .o_detach          = mdt_detach,
4650         .o_setup           = mdt_setup,
4651         .o_cleanup         = mdt_cleanup,
4652 };
4653
4654 static int __init mds_init(void)
4655 {
4656         struct lprocfs_static_vars lvars;
4657         int rc = 0;
4658         ENTRY;
4659
4660         mds_init_lsd_cache();
4661         mds_init_rmtacl_upcall_cache();
4662
4663         lprocfs_init_multi_vars(0, &lvars);
4664         class_register_type(&mds_obd_ops, NULL, lvars.module_vars,
4665                             OBD_MDS_DEVICENAME);
4666         lprocfs_init_multi_vars(1, &lvars);
4667         class_register_type(&mdt_obd_ops, NULL, lvars.module_vars,
4668                             OBD_MDT_DEVICENAME);
4669
4670         rc = mds_capa_key_start_thread();
4671         if (rc) {
4672                 class_unregister_type(OBD_MDT_DEVICENAME);
4673                 class_unregister_type(OBD_MDS_DEVICENAME);
4674                 mds_cleanup_lsd_cache();
4675         }
4676
4677         mds_eck_timer.function = mds_capa_key_timer_callback;
4678         mds_eck_timer.data = 0;
4679         init_timer(&mds_eck_timer);
4680
4681         RETURN(rc);
4682 }
4683
4684 static void /*__exit*/ mds_exit(void)
4685 {
4686         mds_capa_key_stop_thread();
4687         mds_cleanup_rmtacl_upcall_cache();
4688         mds_cleanup_lsd_cache();
4689
4690         class_unregister_type(OBD_MDS_DEVICENAME);
4691         class_unregister_type(OBD_MDT_DEVICENAME);
4692 }
4693
4694 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4695 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
4696 MODULE_LICENSE("GPL");
4697
4698 module_init(mds_init);
4699 module_exit(mds_exit);