lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_ver.h>
  52 #include "llite_internal.h"
  53
  54 static int ll_create_it(struct inode *, struct dentry *,
  55                         int, struct lookup_intent *);
  56
  57 /*
  58  * Check if we have something mounted at the named dchild.
  59  * In such a case there would always be dentry present.
  60  */
  61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
  62                            struct qstr *name)
  63 {
  64         int mounted = 0;
  65
  66         if (unlikely(dchild)) {
  67                 mounted = d_mountpoint(dchild);
  68         } else if (dparent) {
  69                 dchild = d_lookup(dparent, name);
  70                 if (dchild) {
  71                         mounted = d_mountpoint(dchild);
  72                         dput(dchild);
  73                 }
  74         }
  75         return mounted;
  76 }
  77
  78 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  79 static int ll_test_inode(struct inode *inode, void *opaque)
  80 {
  81         struct ll_inode_info    *lli = ll_i2info(inode);
  82         struct lustre_md        *md = opaque;
  83
  84         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  85                 CERROR("MDS body missing FID\n");
  86                 return 0;
  87         }
  88
  89         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  90                 return 0;
  91
  92         return 1;
  93 }
  94
  95 static int ll_set_inode(struct inode *inode, void *opaque)
  96 {
  97         struct ll_inode_info *lli = ll_i2info(inode);
  98         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  99
 100         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
 101                 CERROR("MDS body missing FID\n");
 102                 return -EINVAL;
 103         }
 104
 105         lli->lli_fid = body->mbo_fid1;
 106         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
 107                 CERROR("Can not initialize inode "DFID" without object type: "
 108                        "valid = "LPX64"\n",
 109                        PFID(&lli->lli_fid), body->mbo_valid);
 110                 return -EINVAL;
 111         }
 112
 113         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
 114         if (unlikely(inode->i_mode == 0)) {
 115                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 116                 return -EINVAL;
 117         }
 118
 119         ll_lli_init(lli);
 120
 121         return 0;
 122 }
 123
 124
 125 /**
 126  * Get an inode by inode number(@hash), which is already instantiated by
 127  * the intent lookup).
 128  */
 129 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 130                       struct lustre_md *md)
 131 {
 132         struct inode    *inode;
 133         int             rc = 0;
 134
 135         ENTRY;
 136
 137         LASSERT(hash != 0);
 138         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 139         if (inode == NULL)
 140                 RETURN(ERR_PTR(-ENOMEM));
 141
 142         if (inode->i_state & I_NEW) {
 143                 rc = ll_read_inode2(inode, md);
 144                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 145                     ll_i2info(inode)->lli_clob == NULL) {
 146                         CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
 147                                 ll_get_fsname(sb, NULL, 0), md->lsm,
 148                                 PFID(ll_inode2fid(inode)));
 149                         rc = cl_file_inode_init(inode, md);
 150                 }
 151                 if (rc != 0) {
 152                         make_bad_inode(inode);
 153                         unlock_new_inode(inode);
 154                         iput(inode);
 155                         inode = ERR_PTR(rc);
 156                 } else {
 157                         unlock_new_inode(inode);
 158                 }
 159         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 160                 rc = ll_update_inode(inode, md);
 161                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 162                        PFID(&md->body->mbo_fid1), inode, rc);
 163                 if (rc != 0) {
 164                         make_bad_inode(inode);
 165                         iput(inode);
 166                         inode = ERR_PTR(rc);
 167                 }
 168         }
 169
 170         RETURN(inode);
 171 }
 172
 173 static void ll_invalidate_negative_children(struct inode *dir)
 174 {
 175         struct dentry *dentry, *tmp_subdir;
 176         DECLARE_LL_D_HLIST_NODE_PTR(p);
 177
 178         ll_lock_dcache(dir);
 179         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
 180                 spin_lock(&dentry->d_lock);
 181                 if (!list_empty(&dentry->d_subdirs)) {
 182                         struct dentry *child;
 183
 184                         list_for_each_entry_safe(child, tmp_subdir,
 185                                                  &dentry->d_subdirs,
 186                                                  d_u.d_child) {
 187                                 if (child->d_inode == NULL)
 188                                         d_lustre_invalidate(child, 1);
 189                         }
 190                 }
 191                 spin_unlock(&dentry->d_lock);
 192         }
 193         ll_unlock_dcache(dir);
 194 }
 195
 196 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 197 {
 198         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 199 }
 200
 201 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 202                        void *data, int flag)
 203 {
 204         struct lustre_handle lockh;
 205         int rc;
 206         ENTRY;
 207
 208         switch (flag) {
 209         case LDLM_CB_BLOCKING:
 210                 ldlm_lock2handle(lock, &lockh);
 211                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 212                 if (rc < 0) {
 213                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 214                         RETURN(rc);
 215                 }
 216                 break;
 217         case LDLM_CB_CANCELING: {
 218                 struct inode *inode = ll_inode_from_resource_lock(lock);
 219                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 220
 221                 /* Inode is set to lock->l_resource->lr_lvb_inode
 222                  * for mdc - bug 24555 */
 223                 LASSERT(lock->l_ast_data == NULL);
 224
 225                 if (inode == NULL)
 226                         break;
 227
 228                 /* Invalidate all dentries associated with this inode */
 229                 LASSERT(ldlm_is_canceling(lock));
 230
 231                 if (!fid_res_name_eq(ll_inode2fid(inode),
 232                                      &lock->l_resource->lr_name)) {
 233                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 234                                    PFID(ll_inode2fid(inode)), inode);
 235                         LBUG();
 236                 }
 237
 238                 if (bits & MDS_INODELOCK_XATTR) {
 239                         ll_xattr_cache_destroy(inode);
 240                         bits &= ~MDS_INODELOCK_XATTR;
 241                 }
 242
 243                 /* For OPEN locks we differentiate between lock modes
 244                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 245                 if (bits & MDS_INODELOCK_OPEN)
 246                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 247
 248                 if (bits & MDS_INODELOCK_OPEN) {
 249                         fmode_t fmode;
 250
 251                         switch (lock->l_req_mode) {
 252                         case LCK_CW:
 253                                 fmode = FMODE_WRITE;
 254                                 break;
 255                         case LCK_PR:
 256                                 fmode = FMODE_EXEC;
 257                                 break;
 258                         case LCK_CR:
 259                                 fmode = FMODE_READ;
 260                                 break;
 261                         default:
 262                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 263                                 LBUG();
 264                         }
 265
 266                         ll_md_real_close(inode, fmode);
 267
 268                         bits &= ~MDS_INODELOCK_OPEN;
 269                 }
 270
 271                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 272                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 273                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 274
 275                 if (bits & MDS_INODELOCK_LAYOUT) {
 276                         struct cl_object_conf conf = {
 277                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 278                                 .coc_inode = inode,
 279                         };
 280
 281                         rc = ll_layout_conf(inode, &conf);
 282                         if (rc < 0)
 283                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 284                                        DFID": rc = %d\n",
 285                                        PFID(ll_inode2fid(inode)), rc);
 286                 }
 287
 288                 if (bits & MDS_INODELOCK_UPDATE) {
 289                         struct ll_inode_info *lli = ll_i2info(inode);
 290
 291                         spin_lock(&lli->lli_lock);
 292                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 293                         spin_unlock(&lli->lli_lock);
 294                 }
 295
 296                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 297                         struct ll_inode_info *lli = ll_i2info(inode);
 298
 299                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 300                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 301                                lli, PFID(&lli->lli_pfid));
 302                         truncate_inode_pages(inode->i_mapping, 0);
 303
 304                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 305                                 struct inode *master_inode = NULL;
 306                                 unsigned long hash;
 307
 308                                 /* This is slave inode, since all of the child
 309                                  * dentry is connected on the master inode, so
 310                                  * we have to invalidate the negative children
 311                                  * on master inode */
 312                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 313                                        PFID(ll_inode2fid(inode)),
 314                                        PFID(&lli->lli_pfid));
 315
 316                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 317                                         ll_need_32bit_api(ll_i2sbi(inode)));
 318
 319                                 master_inode = ilookup5(inode->i_sb, hash,
 320                                                         ll_test_inode_by_fid,
 321                                                         (void *)&lli->lli_pfid);
 322                                 if (master_inode != NULL &&
 323                                         !IS_ERR(master_inode)) {
 324                                         ll_invalidate_negative_children(
 325                                                                 master_inode);
 326                                         iput(master_inode);
 327                                 }
 328                         } else {
 329                                 ll_invalidate_negative_children(inode);
 330                         }
 331                 }
 332
 333                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 334                     inode->i_sb->s_root != NULL &&
 335                     inode != inode->i_sb->s_root->d_inode)
 336                         ll_invalidate_aliases(inode);
 337
 338                 iput(inode);
 339                 break;
 340         }
 341         default:
 342                 LBUG();
 343         }
 344
 345         RETURN(0);
 346 }
 347
 348 __u32 ll_i2suppgid(struct inode *i)
 349 {
 350         if (in_group_p(i->i_gid))
 351                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 352         else
 353                 return (__u32) __kgid_val(INVALID_GID);
 354 }
 355
 356 /* Pack the required supplementary groups into the supplied groups array.
 357  * If we don't need to use the groups from the target inode(s) then we
 358  * instead pack one or more groups from the user's supplementary group
 359  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 360 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 361 {
 362         LASSERT(i1 != NULL);
 363         LASSERT(suppgids != NULL);
 364
 365         suppgids[0] = ll_i2suppgid(i1);
 366
 367         if (i2)
 368                 suppgids[1] = ll_i2suppgid(i2);
 369         else
 370                 suppgids[1] = -1;
 371 }
 372
 373 /*
 374  * try to reuse three types of dentry:
 375  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 376  *    by concurrent .revalidate).
 377  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 378  *    be cleared by others calling d_lustre_revalidate).
 379  * 3. DISCONNECTED alias.
 380  */
 381 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 382 {
 383         struct dentry *alias, *discon_alias, *invalid_alias;
 384         DECLARE_LL_D_HLIST_NODE_PTR(p);
 385
 386         if (ll_d_hlist_empty(&inode->i_dentry))
 387                 return NULL;
 388
 389         discon_alias = invalid_alias = NULL;
 390
 391         ll_lock_dcache(inode);
 392         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 393                 LASSERT(alias != dentry);
 394
 395                 spin_lock(&alias->d_lock);
 396                 if (alias->d_flags & DCACHE_DISCONNECTED)
 397                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 398                         discon_alias = alias;
 399                 else if (alias->d_parent == dentry->d_parent             &&
 400                          alias->d_name.hash == dentry->d_name.hash       &&
 401                          alias->d_name.len == dentry->d_name.len         &&
 402                          memcmp(alias->d_name.name, dentry->d_name.name,
 403                                 dentry->d_name.len) == 0)
 404                         invalid_alias = alias;
 405                 spin_unlock(&alias->d_lock);
 406
 407                 if (invalid_alias)
 408                         break;
 409         }
 410         alias = invalid_alias ?: discon_alias ?: NULL;
 411         if (alias) {
 412                 spin_lock(&alias->d_lock);
 413                 dget_dlock(alias);
 414                 spin_unlock(&alias->d_lock);
 415         }
 416         ll_unlock_dcache(inode);
 417
 418         return alias;
 419 }
 420
 421 /*
 422  * Similar to d_splice_alias(), but lustre treats invalid alias
 423  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 424  */
 425 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 426 {
 427         struct dentry *new;
 428         int rc;
 429
 430         if (inode) {
 431                 new = ll_find_alias(inode, de);
 432                 if (new) {
 433                         rc = ll_d_init(new);
 434                         if (rc < 0) {
 435                                 dput(new);
 436                                 return ERR_PTR(rc);
 437                         }
 438                         d_move(new, de);
 439                         iput(inode);
 440                         CDEBUG(D_DENTRY,
 441                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 442                               new, new->d_inode, d_count(new), new->d_flags);
 443                         return new;
 444                 }
 445         }
 446         rc = ll_d_init(de);
 447         if (rc < 0)
 448                 return ERR_PTR(rc);
 449         d_add(de, inode);
 450         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 451                de, de->d_inode, d_count(de), de->d_flags);
 452         return de;
 453 }
 454
 455 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 456                                struct lookup_intent *it,
 457                                struct inode *parent, struct dentry **de)
 458 {
 459         struct inode             *inode = NULL;
 460         __u64                     bits = 0;
 461         int                       rc;
 462         ENTRY;
 463
 464         /* NB 1 request reference will be taken away by ll_intent_lock()
 465          * when I return */
 466         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 467                it->d.lustre.it_disposition);
 468         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 469                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 470                 if (rc)
 471                         RETURN(rc);
 472
 473                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 474
 475                 /* We used to query real size from OSTs here, but actually
 476                    this is not needed. For stat() calls size would be updated
 477                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 478                    2.4 and
 479                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 480                    Everybody else who needs correct file size would call
 481                    ll_glimpse_size or some equivalent themselves anyway.
 482                    Also see bug 7198. */
 483         }
 484
 485         /* Only hash *de if it is unhashed (new dentry).
 486          * Atoimc_open may passin hashed dentries for open.
 487          */
 488         if (d_unhashed(*de)) {
 489                 struct dentry *alias;
 490
 491                 alias = ll_splice_alias(inode, *de);
 492                 if (IS_ERR(alias))
 493                         RETURN(PTR_ERR(alias));
 494                 *de = alias;
 495         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 496                    !it_disposition(it, DISP_OPEN_CREATE)) {
 497                 /* With DISP_OPEN_CREATE dentry will
 498                    instantiated in ll_create_it. */
 499                 LASSERT((*de)->d_inode == NULL);
 500                 d_instantiate(*de, inode);
 501         }
 502
 503         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 504                 /* we have lookup look - unhide dentry */
 505                 if (bits & MDS_INODELOCK_LOOKUP)
 506                         d_lustre_revalidate(*de);
 507         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 508                 /* If file created on server, don't depend on parent UPDATE
 509                  * lock to unhide it. It is left hidden and next lookup can
 510                  * find it in ll_splice_alias.
 511                  */
 512                 /* Check that parent has UPDATE lock. */
 513                 struct lookup_intent parent_it = {
 514                                         .it_op = IT_GETATTR,
 515                                         .d.lustre.it_lock_handle = 0 };
 516                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 517
 518                 /* If it is striped directory, get the real stripe parent */
 519                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 520                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 521                                                  ll_i2info(parent)->lli_lsm_md,
 522                                                  (*de)->d_name.name,
 523                                                  (*de)->d_name.len, &fid);
 524                         if (rc != 0)
 525                                 RETURN(rc);
 526                 }
 527
 528                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 529                                        NULL)) {
 530                         d_lustre_revalidate(*de);
 531                         ll_intent_release(&parent_it);
 532                 }
 533         }
 534
 535         RETURN(0);
 536 }
 537
 538 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 539                                    struct lookup_intent *it, int lookup_flags)
 540 {
 541         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 542         struct dentry *save = dentry, *retval;
 543         struct ptlrpc_request *req = NULL;
 544         struct md_op_data *op_data;
 545         __u32 opc;
 546         int rc;
 547         ENTRY;
 548
 549         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 550                 RETURN(ERR_PTR(-ENAMETOOLONG));
 551
 552         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 553                dentry->d_name.len, dentry->d_name.name,
 554                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 555
 556         if (d_mountpoint(dentry))
 557                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 558
 559         if (it == NULL || it->it_op == IT_GETXATTR)
 560                 it = &lookup_it;
 561
 562         if (it->it_op == IT_GETATTR) {
 563                 rc = ll_statahead_enter(parent, &dentry, 0);
 564                 if (rc == 1) {
 565                         if (dentry == save)
 566                                 GOTO(out, retval = NULL);
 567                         GOTO(out, retval = dentry);
 568                 }
 569         }
 570
 571         if (it->it_op & IT_CREAT)
 572                 opc = LUSTRE_OPC_CREATE;
 573         else
 574                 opc = LUSTRE_OPC_ANY;
 575
 576         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 577                                      dentry->d_name.len, lookup_flags, opc,
 578                                      NULL);
 579         if (IS_ERR(op_data))
 580                 RETURN((void *)op_data);
 581
 582         /* enforce umask if acl disabled or MDS doesn't support umask */
 583         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 584                 it->it_create_mode &= ~current_umask();
 585
 586         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 587                             &ll_md_blocking_ast, 0);
 588         ll_finish_md_op_data(op_data);
 589         if (rc < 0)
 590                 GOTO(out, retval = ERR_PTR(rc));
 591
 592         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 593         if (rc != 0) {
 594                 ll_intent_release(it);
 595                 GOTO(out, retval = ERR_PTR(rc));
 596         }
 597
 598         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 599             !S_ISREG(dentry->d_inode->i_mode) &&
 600             !S_ISDIR(dentry->d_inode->i_mode)) {
 601                 ll_release_openhandle(dentry, it);
 602         }
 603         ll_lookup_finish_locks(it, dentry);
 604
 605         if (dentry == save)
 606                 GOTO(out, retval = NULL);
 607         else
 608                 GOTO(out, retval = dentry);
 609  out:
 610         if (req)
 611                 ptlrpc_req_finished(req);
 612         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 613                 ll_statahead_mark(parent, dentry);
 614         return retval;
 615 }
 616
 617 #ifdef HAVE_IOP_ATOMIC_OPEN
 618 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 619                                    unsigned int flags)
 620 {
 621         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 622         struct dentry *de;
 623
 624         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 625                dentry->d_name.len, dentry->d_name.name,
 626                PFID(ll_inode2fid(parent)), parent, flags);
 627
 628         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 629         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 630                 return NULL;
 631
 632         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 633                 itp = NULL;
 634         else
 635                 itp = &it;
 636         de = ll_lookup_it(parent, dentry, itp, 0);
 637
 638         if (itp != NULL)
 639                 ll_intent_release(itp);
 640
 641         return de;
 642 }
 643
 644 /*
 645  * For cached negative dentry and new dentry, handle lookup/create/open
 646  * together.
 647  */
 648 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 649                           struct file *file, unsigned open_flags,
 650                           umode_t mode, int *opened)
 651 {
 652         struct lookup_intent *it;
 653         struct dentry *de;
 654         long long lookup_flags = LOOKUP_OPEN;
 655         int rc = 0;
 656         ENTRY;
 657
 658         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 659                            "open_flags %x, mode %x opened %d\n",
 660                dentry->d_name.len, dentry->d_name.name,
 661                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 662
 663         OBD_ALLOC(it, sizeof(*it));
 664         if (!it)
 665                 RETURN(-ENOMEM);
 666
 667         it->it_op = IT_OPEN;
 668         if (open_flags & O_CREAT) {
 669                 it->it_op |= IT_CREAT;
 670                 lookup_flags |= LOOKUP_CREATE;
 671         }
 672         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 673         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 674
 675         /* Dentry added to dcache tree in ll_lookup_it */
 676         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 677         if (IS_ERR(de))
 678                 rc = PTR_ERR(de);
 679         else if (de != NULL)
 680                 dentry = de;
 681
 682         if (!rc) {
 683                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 684                         /* Dentry instantiated in ll_create_it. */
 685                         rc = ll_create_it(dir, dentry, mode, it);
 686                         if (rc) {
 687                                 /* We dget in ll_splice_alias. */
 688                                 if (de != NULL)
 689                                         dput(de);
 690                                 goto out_release;
 691                         }
 692
 693                         *opened |= FILE_CREATED;
 694                 }
 695                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 696                         /* Open dentry. */
 697                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 698                                 /* We cannot call open here as it would
 699                                  * deadlock.
 700                                  */
 701                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 702                                         ptlrpc_req_finished(
 703                                                        (struct ptlrpc_request *)
 704                                                           it->d.lustre.it_data);
 705                                 rc = finish_no_open(file, de);
 706                         } else {
 707                                 file->private_data = it;
 708                                 rc = finish_open(file, dentry, NULL, opened);
 709                                 /* We dget in ll_splice_alias. finish_open takes
 710                                  * care of dget for fd open.
 711                                  */
 712                                 if (de != NULL)
 713                                         dput(de);
 714                         }
 715                 } else {
 716                         rc = finish_no_open(file, de);
 717                 }
 718         }
 719
 720 out_release:
 721         ll_intent_release(it);
 722         OBD_FREE(it, sizeof(*it));
 723
 724         RETURN(rc);
 725 }
 726
 727 #else /* !HAVE_IOP_ATOMIC_OPEN */
 728 static struct lookup_intent *
 729 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 730 {
 731         struct lookup_intent *it;
 732
 733         OBD_ALLOC(it, sizeof(*it));
 734         if (!it)
 735                 return ERR_PTR(-ENOMEM);
 736
 737         if (lookup_flags & LOOKUP_OPEN) {
 738                 it->it_op = IT_OPEN;
 739                 if (lookup_flags & LOOKUP_CREATE)
 740                         it->it_op |= IT_CREAT;
 741                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 742                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 743         } else {
 744                 it->it_op = IT_GETATTR;
 745         }
 746
 747         return it;
 748 }
 749
 750 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 751                                    struct nameidata *nd)
 752 {
 753         struct dentry *de;
 754         ENTRY;
 755
 756         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 757                 struct lookup_intent *it;
 758
 759                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 760                         it = ll_d2d(dentry)->lld_it;
 761                         ll_d2d(dentry)->lld_it = NULL;
 762                 } else {
 763                         if ((nd->flags & LOOKUP_CREATE) &&
 764                             !(nd->flags & LOOKUP_OPEN))
 765                                 RETURN(NULL);
 766
 767                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 768                         if (IS_ERR(it))
 769                                 RETURN((struct dentry *)it);
 770                 }
 771
 772                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 773                 if (de)
 774                         dentry = de;
 775                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 776                         if (dentry->d_inode &&
 777                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 778                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 779                                         // We cannot call open here as it would
 780                                         // deadlock.
 781                                         ptlrpc_req_finished(
 782                                                        (struct ptlrpc_request *)
 783                                                           it->d.lustre.it_data);
 784                                 } else {
 785                                         struct file *filp;
 786
 787                                         nd->intent.open.file->private_data = it;
 788                                         filp = lookup_instantiate_filp(nd,
 789                                                                        dentry,
 790                                                                        NULL);
 791                                         if (IS_ERR(filp)) {
 792                                                 if (de)
 793                                                         dput(de);
 794                                                 de = (struct dentry *)filp;
 795                                         }
 796                                 }
 797                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 798                                 // XXX This can only reliably work on assumption
 799                                 // that there are NO hashed negative dentries.
 800                                 ll_d2d(dentry)->lld_it = it;
 801                                 it = NULL; /* Will be freed in ll_create_nd */
 802                                 /* We absolutely depend on ll_create_nd to be
 803                                  * called to not leak this intent and possible
 804                                  * data attached to it */
 805                         }
 806                 }
 807
 808                 if (it) {
 809                         ll_intent_release(it);
 810                         OBD_FREE(it, sizeof(*it));
 811                 }
 812         } else {
 813                 de = ll_lookup_it(parent, dentry, NULL, 0);
 814         }
 815
 816         RETURN(de);
 817 }
 818 #endif /* HAVE_IOP_ATOMIC_OPEN */
 819
 820 /* We depend on "mode" being set with the proper file type/umask by now */
 821 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 822 {
 823         struct inode *inode = NULL;
 824         struct ptlrpc_request *request = NULL;
 825         struct ll_sb_info *sbi = ll_i2sbi(dir);
 826         int rc;
 827         ENTRY;
 828
 829         LASSERT(it && it->d.lustre.it_disposition);
 830
 831         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 832         request = it->d.lustre.it_data;
 833         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 834         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 835         if (rc)
 836                 GOTO(out, inode = ERR_PTR(rc));
 837
 838         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 839
 840         /* We asked for a lock on the directory, but were granted a
 841          * lock on the inode.  Since we finally have an inode pointer,
 842          * stuff it in the lock. */
 843         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 844                PFID(ll_inode2fid(inode)), inode);
 845         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 846         EXIT;
 847  out:
 848         ptlrpc_req_finished(request);
 849         return inode;
 850 }
 851
 852 /*
 853  * By the time this is called, we already have created the directory cache
 854  * entry for the new file, but it is so far negative - it has no inode.
 855  *
 856  * We defer creating the OBD object(s) until open, to keep the intent and
 857  * non-intent code paths similar, and also because we do not have the MDS
 858  * inode number before calling ll_create_node() (which is needed for LOV),
 859  * so we would need to do yet another RPC to the MDS to store the LOV EA
 860  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 861  * lmm_size in datalen (the MDS still has code which will handle that).
 862  *
 863  * If the create succeeds, we fill in the inode information
 864  * with d_instantiate().
 865  */
 866 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 867                         struct lookup_intent *it)
 868 {
 869         struct inode *inode;
 870         int rc = 0;
 871         ENTRY;
 872
 873         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 874                dentry->d_name.len, dentry->d_name.name,
 875                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 876
 877         rc = it_open_error(DISP_OPEN_CREATE, it);
 878         if (rc)
 879                 RETURN(rc);
 880
 881         inode = ll_create_node(dir, it);
 882         if (IS_ERR(inode))
 883                 RETURN(PTR_ERR(inode));
 884
 885         d_instantiate(dentry, inode);
 886         RETURN(0);
 887 }
 888
 889 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 890 {
 891         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 892                                                        &RMF_MDT_BODY);
 893
 894         LASSERT(body);
 895         if (body->mbo_valid & OBD_MD_FLMTIME &&
 896             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 897                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 898                        "\n", PFID(ll_inode2fid(inode)),
 899                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 900                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 901         }
 902
 903         if (body->mbo_valid & OBD_MD_FLCTIME &&
 904             body->mbo_ctime > LTIME_S(inode->i_ctime))
 905                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 906 }
 907
 908 static int ll_new_node(struct inode *dir, struct qstr *name,
 909                        const char *tgt, int mode, int rdev,
 910                        struct dentry *dchild, __u32 opc)
 911 {
 912         struct ptlrpc_request *request = NULL;
 913         struct md_op_data *op_data;
 914         struct inode *inode = NULL;
 915         struct ll_sb_info *sbi = ll_i2sbi(dir);
 916         int tgt_len = 0;
 917         int err;
 918
 919         ENTRY;
 920         if (unlikely(tgt != NULL))
 921                 tgt_len = strlen(tgt) + 1;
 922
 923         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 924                                      name->len, 0, opc, NULL);
 925         if (IS_ERR(op_data))
 926                 GOTO(err_exit, err = PTR_ERR(op_data));
 927
 928         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 929                         from_kuid(&init_user_ns, current_fsuid()),
 930                         from_kgid(&init_user_ns, current_fsgid()),
 931                         cfs_curproc_cap_pack(), rdev, &request);
 932         ll_finish_md_op_data(op_data);
 933         if (err)
 934                 GOTO(err_exit, err);
 935
 936         ll_update_times(request, dir);
 937
 938         if (dchild) {
 939                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 940                 if (err)
 941                      GOTO(err_exit, err);
 942
 943                 d_instantiate(dchild, inode);
 944         }
 945         EXIT;
 946 err_exit:
 947         ptlrpc_req_finished(request);
 948
 949         return err;
 950 }
 951
 952 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 953                             unsigned rdev, struct dentry *dchild)
 954 {
 955         int err;
 956         ENTRY;
 957
 958         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
 959                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
 960                mode, rdev);
 961
 962         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 963                 mode &= ~current_umask();
 964
 965         switch (mode & S_IFMT) {
 966         case 0:
 967                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 968         case S_IFREG:
 969         case S_IFCHR:
 970         case S_IFBLK:
 971         case S_IFIFO:
 972         case S_IFSOCK:
 973                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 974                                   LUSTRE_OPC_MKNOD);
 975                 break;
 976         case S_IFDIR:
 977                 err = -EPERM;
 978                 break;
 979         default:
 980                 err = -EINVAL;
 981         }
 982
 983         if (!err)
 984                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 985
 986         RETURN(err);
 987 }
 988
 989 #ifdef HAVE_IOP_ATOMIC_OPEN
 990 /*
 991  * Plain create. Intent create is handled in atomic_open.
 992  */
 993 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 994                         umode_t mode, bool want_excl)
 995 {
 996         int rc;
 997
 998         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
 999                            "flags=%u, excl=%d\n", dentry->d_name.len,
1000                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1001                dir, mode, want_excl);
1002
1003         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1004
1005         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1006
1007         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1008                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1009
1010         return rc;
1011 }
1012 #else /* !HAVE_IOP_ATOMIC_OPEN */
1013 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1014                         ll_umode_t mode, struct nameidata *nd)
1015 {
1016         struct ll_dentry_data *lld = ll_d2d(dentry);
1017         struct lookup_intent *it = NULL;
1018         int rc;
1019
1020         if (lld != NULL)
1021                 it = lld->lld_it;
1022
1023         if (!it)
1024                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1025
1026         lld->lld_it = NULL;
1027
1028         /* Was there an error? Propagate it! */
1029         if (it->d.lustre.it_status) {
1030                 rc = it->d.lustre.it_status;
1031                 goto out;
1032         }
1033
1034         rc = ll_create_it(dir, dentry, mode, it);
1035         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1036                 struct file *filp;
1037
1038                 nd->intent.open.file->private_data = it;
1039                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1040                 if (IS_ERR(filp))
1041                         rc = PTR_ERR(filp);
1042         }
1043
1044 out:
1045         ll_intent_release(it);
1046         OBD_FREE(it, sizeof(*it));
1047
1048         if (!rc)
1049                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1050
1051         return rc;
1052 }
1053 #endif /* HAVE_IOP_ATOMIC_OPEN */
1054
1055 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1056                               const char *tgt, struct dentry *dchild)
1057 {
1058         int err;
1059         ENTRY;
1060
1061         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1062                name->len, name->name, PFID(ll_inode2fid(dir)),
1063                dir, 3000, tgt);
1064
1065         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1066                           0, dchild, LUSTRE_OPC_SYMLINK);
1067
1068         if (!err)
1069                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1070
1071         RETURN(err);
1072 }
1073
1074 static int ll_link_generic(struct inode *src,  struct inode *dir,
1075                            struct qstr *name, struct dentry *dchild)
1076 {
1077         struct ll_sb_info *sbi = ll_i2sbi(dir);
1078         struct ptlrpc_request *request = NULL;
1079         struct md_op_data *op_data;
1080         int err;
1081
1082         ENTRY;
1083         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1084                "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1085                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1086
1087         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1088                                      0, LUSTRE_OPC_ANY, NULL);
1089         if (IS_ERR(op_data))
1090                 RETURN(PTR_ERR(op_data));
1091
1092         err = md_link(sbi->ll_md_exp, op_data, &request);
1093         ll_finish_md_op_data(op_data);
1094         if (err)
1095                 GOTO(out, err);
1096
1097         ll_update_times(request, dir);
1098         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1099         EXIT;
1100 out:
1101         ptlrpc_req_finished(request);
1102         RETURN(err);
1103 }
1104
1105 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1106                             int mode, struct dentry *dchild)
1107
1108 {
1109         int err;
1110         ENTRY;
1111
1112         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1113                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1114
1115         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1116                 mode &= ~current_umask();
1117         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1118         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1119
1120         if (!err)
1121                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1122
1123         RETURN(err);
1124 }
1125
1126 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1127                             struct dentry *dchild, struct qstr *name)
1128 {
1129         struct ptlrpc_request *request = NULL;
1130         struct md_op_data *op_data;
1131         int rc;
1132         ENTRY;
1133
1134         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1135                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1136
1137         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1138                 RETURN(-EBUSY);
1139
1140         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1141                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1142         if (IS_ERR(op_data))
1143                 RETURN(PTR_ERR(op_data));
1144
1145         if (dchild != NULL && dchild->d_inode != NULL)
1146                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1147         op_data->op_fid2 = op_data->op_fid3;
1148         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1149         ll_finish_md_op_data(op_data);
1150         if (rc == 0) {
1151                 ll_update_times(request, dir);
1152                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1153         }
1154
1155         ptlrpc_req_finished(request);
1156         RETURN(rc);
1157 }
1158
1159 /**
1160  * Remove dir entry
1161  **/
1162 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1163 {
1164         struct ptlrpc_request *request = NULL;
1165         struct md_op_data *op_data;
1166         int rc;
1167         ENTRY;
1168
1169         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1170                namelen, name, PFID(ll_inode2fid(dir)), dir);
1171
1172         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1173                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1174         if (IS_ERR(op_data))
1175                 RETURN(PTR_ERR(op_data));
1176         op_data->op_cli_flags |= CLI_RM_ENTRY;
1177         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1178         ll_finish_md_op_data(op_data);
1179         if (rc == 0) {
1180                 ll_update_times(request, dir);
1181                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1182         }
1183
1184         ptlrpc_req_finished(request);
1185         RETURN(rc);
1186 }
1187
1188 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1189 {
1190         struct mdt_body *body;
1191         struct lov_mds_md *eadata;
1192         struct lov_stripe_md *lsm = NULL;
1193         struct obd_trans_info oti = { 0 };
1194         struct obdo *oa;
1195         struct obd_capa *oc = NULL;
1196         int rc;
1197         ENTRY;
1198
1199         /* req is swabbed so this is safe */
1200         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1201         if (!(body->mbo_valid & OBD_MD_FLEASIZE))
1202                 RETURN(0);
1203
1204         if (body->mbo_eadatasize == 0) {
1205                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1206                 GOTO(out, rc = -EPROTO);
1207         }
1208
1209         /* The MDS sent back the EA because we unlinked the last reference
1210          * to this file. Use this EA to unlink the objects on the OST.
1211          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1212          * check it is complete and sensible. */
1213         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1214                                               body->mbo_eadatasize);
1215         LASSERT(eadata != NULL);
1216
1217         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->mbo_eadatasize);
1218         if (rc < 0) {
1219                 CERROR("obd_unpackmd: %d\n", rc);
1220                 GOTO(out, rc);
1221         }
1222         LASSERT(rc >= sizeof(*lsm));
1223
1224         OBDO_ALLOC(oa);
1225         if (oa == NULL)
1226                 GOTO(out_free_memmd, rc = -ENOMEM);
1227
1228         oa->o_oi = lsm->lsm_oi;
1229         oa->o_mode = body->mbo_mode & S_IFMT;
1230         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1231
1232         if (body->mbo_valid & OBD_MD_FLCOOKIE) {
1233                 oa->o_valid |= OBD_MD_FLCOOKIE;
1234                 oti.oti_logcookies =
1235                         req_capsule_server_sized_get(&request->rq_pill,
1236                                                      &RMF_LOGCOOKIES,
1237                                                    sizeof(struct llog_cookie) *
1238                                                      lsm->lsm_stripe_count);
1239                 if (oti.oti_logcookies == NULL) {
1240                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1241                         body->mbo_valid &= ~OBD_MD_FLCOOKIE;
1242                 }
1243         }
1244
1245         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1246                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1247                 if (rc)
1248                         GOTO(out_free_memmd, rc);
1249         }
1250
1251         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1252                          ll_i2mdexp(dir), oc);
1253         capa_put(oc);
1254         if (rc)
1255                 CERROR("obd destroy objid "DOSTID" error %d\n",
1256                        POSTID(&lsm->lsm_oi), rc);
1257 out_free_memmd:
1258         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1259         OBDO_FREE(oa);
1260 out:
1261         return rc;
1262 }
1263
1264 /* ll_unlink_generic() doesn't update the inode with the new link count.
1265  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1266  * is any lock existing. They will recycle dentries and inodes based upon locks
1267  * too. b=20433 */
1268 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1269                              struct dentry *dchild, struct qstr *name)
1270 {
1271         struct ptlrpc_request *request = NULL;
1272         struct md_op_data *op_data;
1273         int rc;
1274         ENTRY;
1275         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1276                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1277
1278         /*
1279          * XXX: unlink bind mountpoint maybe call to here,
1280          * just check it as vfs_unlink does.
1281          */
1282         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1283                 RETURN(-EBUSY);
1284
1285         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1286                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1287         if (IS_ERR(op_data))
1288                 RETURN(PTR_ERR(op_data));
1289
1290         if (dchild != NULL && dchild->d_inode != NULL)
1291                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1292
1293         op_data->op_fid2 = op_data->op_fid3;
1294         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1295         ll_finish_md_op_data(op_data);
1296         if (rc)
1297                 GOTO(out, rc);
1298
1299         ll_update_times(request, dir);
1300         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1301
1302         rc = ll_objects_destroy(request, dir);
1303  out:
1304         ptlrpc_req_finished(request);
1305         RETURN(rc);
1306 }
1307
1308 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1309                              struct dentry *src_dchild, struct qstr *src_name,
1310                              struct inode *tgt, struct dentry *tgt_dparent,
1311                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1312 {
1313         struct ptlrpc_request *request = NULL;
1314         struct ll_sb_info *sbi = ll_i2sbi(src);
1315         struct md_op_data *op_data;
1316         int err;
1317         ENTRY;
1318         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1319                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1320                src_name->len, src_name->name,
1321                PFID(ll_inode2fid(src)), src, tgt_name->len,
1322                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1323
1324         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1325             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1326                 RETURN(-EBUSY);
1327
1328         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1329                                      LUSTRE_OPC_ANY, NULL);
1330         if (IS_ERR(op_data))
1331                 RETURN(PTR_ERR(op_data));
1332
1333         if (src_dchild != NULL && src_dchild->d_inode != NULL)
1334                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1335         if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1336                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1337
1338         err = md_rename(sbi->ll_md_exp, op_data,
1339                         src_name->name, src_name->len,
1340                         tgt_name->name, tgt_name->len, &request);
1341         ll_finish_md_op_data(op_data);
1342         if (!err) {
1343                 ll_update_times(request, src);
1344                 ll_update_times(request, tgt);
1345                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1346                 err = ll_objects_destroy(request, src);
1347         }
1348
1349         ptlrpc_req_finished(request);
1350
1351         RETURN(err);
1352 }
1353
1354 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1355                     dev_t rdev)
1356 {
1357         return ll_mknod_generic(dir, &dchild->d_name, mode,
1358                                 old_encode_dev(rdev), dchild);
1359 }
1360
1361 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1362 {
1363         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1364 }
1365
1366 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1367 {
1368         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1369 }
1370
1371 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1372 {
1373         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1374 }
1375
1376 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1377                       const char *oldname)
1378 {
1379         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1380 }
1381
1382 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1383                    struct dentry *new_dentry)
1384 {
1385         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1386                                new_dentry);
1387 }
1388
1389 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1390                      struct inode *new_dir, struct dentry *new_dentry)
1391 {
1392         int err;
1393         err = ll_rename_generic(old_dir, NULL,
1394                                  old_dentry, &old_dentry->d_name,
1395                                  new_dir, NULL, new_dentry,
1396                                  &new_dentry->d_name);
1397         if (!err)
1398                         d_move(old_dentry, new_dentry);
1399         return err;
1400 }
1401
1402 const struct inode_operations ll_dir_inode_operations = {
1403         .mknod              = ll_mknod,
1404 #ifdef HAVE_IOP_ATOMIC_OPEN
1405         .atomic_open        = ll_atomic_open,
1406 #endif
1407         .lookup             = ll_lookup_nd,
1408         .create             = ll_create_nd,
1409         /* We need all these non-raw things for NFSD, to not patch it. */
1410         .unlink             = ll_unlink,
1411         .mkdir              = ll_mkdir,
1412         .rmdir              = ll_rmdir,
1413         .symlink            = ll_symlink,
1414         .link               = ll_link,
1415         .rename             = ll_rename,
1416         .setattr            = ll_setattr,
1417         .getattr            = ll_getattr,
1418         .permission         = ll_inode_permission,
1419         .setxattr           = ll_setxattr,
1420         .getxattr           = ll_getxattr,
1421         .listxattr          = ll_listxattr,
1422         .removexattr        = ll_removexattr,
1423 #ifdef HAVE_IOP_GET_ACL
1424         .get_acl            = ll_get_acl,
1425 #endif
1426 };
1427
1428 const struct inode_operations ll_special_inode_operations = {
1429         .setattr        = ll_setattr,
1430         .getattr        = ll_getattr,
1431         .permission     = ll_inode_permission,
1432         .setxattr       = ll_setxattr,
1433         .getxattr       = ll_getxattr,
1434         .listxattr      = ll_listxattr,
1435         .removexattr    = ll_removexattr,
1436 #ifdef HAVE_IOP_GET_ACL
1437         .get_acl            = ll_get_acl,
1438 #endif
1439 };