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 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 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
  79 {
  80         ENTRY;
  81
  82         ldlm_lock_decref(lockh, mode);
  83
  84         RETURN(0);
  85 }
  86
  87
  88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  89 static int ll_test_inode(struct inode *inode, void *opaque)
  90 {
  91         struct ll_inode_info *lli = ll_i2info(inode);
  92         struct lustre_md     *md = opaque;
  93
  94         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  95                 CERROR("MDS body missing FID\n");
  96                 return 0;
  97         }
  98
  99         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
 100                 return 0;
 101
 102         return 1;
 103 }
 104
 105 static int ll_set_inode(struct inode *inode, void *opaque)
 106 {
 107         struct ll_inode_info *lli = ll_i2info(inode);
 108         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
 109
 110         if (unlikely(!(body->valid & OBD_MD_FLID))) {
 111                 CERROR("MDS body missing FID\n");
 112                 return -EINVAL;
 113         }
 114
 115         lli->lli_fid = body->fid1;
 116         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
 117                 CERROR("Can not initialize inode "DFID" without object type: "
 118                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
 119                 return -EINVAL;
 120         }
 121
 122         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
 123         if (unlikely(inode->i_mode == 0)) {
 124                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 125                 return -EINVAL;
 126         }
 127
 128         ll_lli_init(lli);
 129
 130         return 0;
 131 }
 132
 133
 134 /*
 135  * Get an inode by inode number (already instantiated by the intent lookup).
 136  * Returns inode or NULL
 137  */
 138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 139                       struct lustre_md *md)
 140 {
 141         struct inode         *inode;
 142         ENTRY;
 143
 144         LASSERT(hash != 0);
 145         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 146
 147         if (inode) {
 148                 if (inode->i_state & I_NEW) {
 149                         int rc = 0;
 150
 151                         ll_read_inode2(inode, md);
 152                         if (S_ISREG(inode->i_mode) &&
 153                             ll_i2info(inode)->lli_clob == NULL) {
 154                                 CDEBUG(D_INODE,
 155                                         "%s: apply lsm %p to inode "DFID".\n",
 156                                         ll_get_fsname(sb, NULL, 0), md->lsm,
 157                                         PFID(ll_inode2fid(inode)));
 158                                 rc = cl_file_inode_init(inode, md);
 159                         }
 160                         if (rc != 0) {
 161                                 make_bad_inode(inode);
 162                                 unlock_new_inode(inode);
 163                                 iput(inode);
 164                                 inode = ERR_PTR(rc);
 165                         } else
 166                                 unlock_new_inode(inode);
 167                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 168                         ll_update_inode(inode, md);
 169                         CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
 170                                PFID(&md->body->fid1), inode);
 171         }
 172         RETURN(inode);
 173 }
 174
 175 static void ll_invalidate_negative_children(struct inode *dir)
 176 {
 177         struct dentry *dentry, *tmp_subdir;
 178         DECLARE_LL_D_HLIST_NODE_PTR(p);
 179
 180         ll_lock_dcache(dir);
 181         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
 182                 spin_lock(&dentry->d_lock);
 183                 if (!list_empty(&dentry->d_subdirs)) {
 184                         struct dentry *child;
 185
 186                         list_for_each_entry_safe(child, tmp_subdir,
 187                                                  &dentry->d_subdirs,
 188                                                  d_u.d_child) {
 189                                 if (child->d_inode == NULL)
 190                                         d_lustre_invalidate(child, 1);
 191                         }
 192                 }
 193                 spin_unlock(&dentry->d_lock);
 194         }
 195         ll_unlock_dcache(dir);
 196 }
 197
 198 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 199 {
 200         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 201 }
 202
 203 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 204                        void *data, int flag)
 205 {
 206         struct lustre_handle lockh;
 207         int rc;
 208         ENTRY;
 209
 210         switch (flag) {
 211         case LDLM_CB_BLOCKING:
 212                 ldlm_lock2handle(lock, &lockh);
 213                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 214                 if (rc < 0) {
 215                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 216                         RETURN(rc);
 217                 }
 218                 break;
 219         case LDLM_CB_CANCELING: {
 220                 struct inode *inode = ll_inode_from_resource_lock(lock);
 221                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 222
 223                 /* Inode is set to lock->l_resource->lr_lvb_inode
 224                  * for mdc - bug 24555 */
 225                 LASSERT(lock->l_ast_data == NULL);
 226
 227                 if (inode == NULL)
 228                         break;
 229
 230                 /* Invalidate all dentries associated with this inode */
 231                 LASSERT(ldlm_is_canceling(lock));
 232
 233                 if (!fid_res_name_eq(ll_inode2fid(inode),
 234                                      &lock->l_resource->lr_name)) {
 235                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 236                                    PFID(ll_inode2fid(inode)), inode);
 237                         LBUG();
 238                 }
 239
 240                 if (bits & MDS_INODELOCK_XATTR) {
 241                         ll_xattr_cache_destroy(inode);
 242                         bits &= ~MDS_INODELOCK_XATTR;
 243                 }
 244
 245                 /* For OPEN locks we differentiate between lock modes
 246                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 247                 if (bits & MDS_INODELOCK_OPEN)
 248                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 249
 250                 if (bits & MDS_INODELOCK_OPEN) {
 251                         fmode_t fmode;
 252
 253                         switch (lock->l_req_mode) {
 254                         case LCK_CW:
 255                                 fmode = FMODE_WRITE;
 256                                 break;
 257                         case LCK_PR:
 258                                 fmode = FMODE_EXEC;
 259                                 break;
 260                         case LCK_CR:
 261                                 fmode = FMODE_READ;
 262                                 break;
 263                         default:
 264                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 265                                 LBUG();
 266                         }
 267
 268                         ll_md_real_close(inode, fmode);
 269
 270                         bits &= ~MDS_INODELOCK_OPEN;
 271                 }
 272
 273                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 274                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 275                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 276
 277                 if (bits & MDS_INODELOCK_LAYOUT) {
 278                         struct cl_object_conf conf = {
 279                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 280                                 .coc_inode = inode,
 281                         };
 282
 283                         rc = ll_layout_conf(inode, &conf);
 284                         if (rc < 0)
 285                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 286                                        DFID": rc = %d\n",
 287                                        PFID(ll_inode2fid(inode)), rc);
 288                 }
 289
 290                 if (bits & MDS_INODELOCK_UPDATE) {
 291                         struct ll_inode_info *lli = ll_i2info(inode);
 292
 293                         spin_lock(&lli->lli_lock);
 294                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 295                         spin_unlock(&lli->lli_lock);
 296                 }
 297
 298                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 299                         struct ll_inode_info *lli = ll_i2info(inode);
 300
 301                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 302                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 303                                lli, PFID(&lli->lli_pfid));
 304                         truncate_inode_pages(inode->i_mapping, 0);
 305
 306                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 307                                 struct inode *master_inode = NULL;
 308                                 unsigned long hash;
 309
 310                                 /* This is slave inode, since all of the child
 311                                  * dentry is connected on the master inode, so
 312                                  * we have to invalidate the negative children
 313                                  * on master inode */
 314                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 315                                        PFID(ll_inode2fid(inode)),
 316                                        PFID(&lli->lli_pfid));
 317
 318                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 319                                         ll_need_32bit_api(ll_i2sbi(inode)));
 320
 321                                 master_inode = ilookup5(inode->i_sb, hash,
 322                                                         ll_test_inode_by_fid,
 323                                                         (void *)&lli->lli_pfid);
 324                                 if (master_inode != NULL &&
 325                                         !IS_ERR(master_inode)) {
 326                                         ll_invalidate_negative_children(
 327                                                                 master_inode);
 328                                         iput(master_inode);
 329                                 }
 330                         } else {
 331                                 ll_invalidate_negative_children(inode);
 332                         }
 333                 }
 334
 335                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 336                     inode->i_sb->s_root != NULL &&
 337                     inode != inode->i_sb->s_root->d_inode)
 338                         ll_invalidate_aliases(inode);
 339
 340                 iput(inode);
 341                 break;
 342         }
 343         default:
 344                 LBUG();
 345         }
 346
 347         RETURN(0);
 348 }
 349
 350 __u32 ll_i2suppgid(struct inode *i)
 351 {
 352         if (in_group_p(i->i_gid))
 353                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 354         else
 355                 return (__u32) __kgid_val(INVALID_GID);
 356 }
 357
 358 /* Pack the required supplementary groups into the supplied groups array.
 359  * If we don't need to use the groups from the target inode(s) then we
 360  * instead pack one or more groups from the user's supplementary group
 361  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 362 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 363 {
 364         LASSERT(i1 != NULL);
 365         LASSERT(suppgids != NULL);
 366
 367         suppgids[0] = ll_i2suppgid(i1);
 368
 369         if (i2)
 370                 suppgids[1] = ll_i2suppgid(i2);
 371         else
 372                 suppgids[1] = -1;
 373 }
 374
 375 /*
 376  * try to reuse three types of dentry:
 377  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 378  *    by concurrent .revalidate).
 379  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 380  *    be cleared by others calling d_lustre_revalidate).
 381  * 3. DISCONNECTED alias.
 382  */
 383 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 384 {
 385         struct dentry *alias, *discon_alias, *invalid_alias;
 386         DECLARE_LL_D_HLIST_NODE_PTR(p);
 387
 388         if (ll_d_hlist_empty(&inode->i_dentry))
 389                 return NULL;
 390
 391         discon_alias = invalid_alias = NULL;
 392
 393         ll_lock_dcache(inode);
 394         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 395                 LASSERT(alias != dentry);
 396
 397                 spin_lock(&alias->d_lock);
 398                 if (alias->d_flags & DCACHE_DISCONNECTED)
 399                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 400                         discon_alias = alias;
 401                 else if (alias->d_parent == dentry->d_parent             &&
 402                          alias->d_name.hash == dentry->d_name.hash       &&
 403                          alias->d_name.len == dentry->d_name.len         &&
 404                          memcmp(alias->d_name.name, dentry->d_name.name,
 405                                 dentry->d_name.len) == 0)
 406                         invalid_alias = alias;
 407                 spin_unlock(&alias->d_lock);
 408
 409                 if (invalid_alias)
 410                         break;
 411         }
 412         alias = invalid_alias ?: discon_alias ?: NULL;
 413         if (alias) {
 414                 spin_lock(&alias->d_lock);
 415                 dget_dlock(alias);
 416                 spin_unlock(&alias->d_lock);
 417         }
 418         ll_unlock_dcache(inode);
 419
 420         return alias;
 421 }
 422
 423 /*
 424  * Similar to d_splice_alias(), but lustre treats invalid alias
 425  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 426  */
 427 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 428 {
 429         struct dentry *new;
 430         int rc;
 431
 432         if (inode) {
 433                 new = ll_find_alias(inode, de);
 434                 if (new) {
 435                         rc = ll_d_init(new);
 436                         if (rc < 0) {
 437                                 dput(new);
 438                                 return ERR_PTR(rc);
 439                         }
 440                         d_move(new, de);
 441                         iput(inode);
 442                         CDEBUG(D_DENTRY,
 443                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 444                               new, new->d_inode, d_count(new), new->d_flags);
 445                         return new;
 446                 }
 447         }
 448         rc = ll_d_init(de);
 449         if (rc < 0)
 450                 return ERR_PTR(rc);
 451         d_add(de, inode);
 452         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 453                de, de->d_inode, d_count(de), de->d_flags);
 454         return de;
 455 }
 456
 457 int ll_lookup_it_finish(struct ptlrpc_request *request,
 458                         struct lookup_intent *it, void *data)
 459 {
 460         struct it_cb_data        *icbd = data;
 461         struct dentry           **de = icbd->icbd_childp;
 462         struct inode             *parent = icbd->icbd_parent;
 463         struct inode             *inode = NULL;
 464         __u64                     bits = 0;
 465         int                       rc;
 466         ENTRY;
 467
 468         /* NB 1 request reference will be taken away by ll_intent_lock()
 469          * when I return */
 470         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 471                it->d.lustre.it_disposition);
 472         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 473                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 474                 if (rc)
 475                         RETURN(rc);
 476
 477                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 478
 479                 /* We used to query real size from OSTs here, but actually
 480                    this is not needed. For stat() calls size would be updated
 481                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 482                    2.4 and
 483                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 484                    Everybody else who needs correct file size would call
 485                    ll_glimpse_size or some equivalent themselves anyway.
 486                    Also see bug 7198. */
 487         }
 488
 489         /* Only hash *de if it is unhashed (new dentry).
 490          * Atoimc_open may passin hashed dentries for open.
 491          */
 492         if (d_unhashed(*de)) {
 493                 struct dentry *alias;
 494
 495                 alias = ll_splice_alias(inode, *de);
 496                 if (IS_ERR(alias))
 497                         RETURN(PTR_ERR(alias));
 498                 *de = alias;
 499         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 500                    !it_disposition(it, DISP_OPEN_CREATE)) {
 501                 /* With DISP_OPEN_CREATE dentry will
 502                    instantiated in ll_create_it. */
 503                 LASSERT((*de)->d_inode == NULL);
 504                 d_instantiate(*de, inode);
 505         }
 506
 507         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 508                 /* we have lookup look - unhide dentry */
 509                 if (bits & MDS_INODELOCK_LOOKUP)
 510                         d_lustre_revalidate(*de);
 511         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 512                 /* If file created on server, don't depend on parent UPDATE
 513                  * lock to unhide it. It is left hidden and next lookup can
 514                  * find it in ll_splice_alias.
 515                  */
 516                 /* Check that parent has UPDATE lock. */
 517                 struct lookup_intent parent_it = {
 518                                         .it_op = IT_GETATTR,
 519                                         .d.lustre.it_lock_handle = 0 };
 520
 521                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 522                                        &ll_i2info(parent)->lli_fid, NULL)) {
 523                         d_lustre_revalidate(*de);
 524                         ll_intent_release(&parent_it);
 525                 }
 526         }
 527
 528         RETURN(0);
 529 }
 530
 531 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 532                                    struct lookup_intent *it, int lookup_flags)
 533 {
 534         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 535         struct dentry *save = dentry, *retval;
 536         struct ptlrpc_request *req = NULL;
 537         struct md_op_data *op_data;
 538         struct it_cb_data icbd;
 539         __u32 opc;
 540         int rc;
 541         ENTRY;
 542
 543         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 544                 RETURN(ERR_PTR(-ENAMETOOLONG));
 545
 546         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 547                dentry->d_name.len, dentry->d_name.name,
 548                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 549
 550         if (d_mountpoint(dentry))
 551                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 552
 553         ll_frob_intent(&it, &lookup_it);
 554
 555         if (it->it_op == IT_GETATTR) {
 556                 rc = ll_statahead_enter(parent, &dentry, 0);
 557                 if (rc == 1) {
 558                         if (dentry == save)
 559                                 GOTO(out, retval = NULL);
 560                         GOTO(out, retval = dentry);
 561                 }
 562         }
 563
 564         icbd.icbd_childp = &dentry;
 565         icbd.icbd_parent = parent;
 566
 567         if (it->it_op & IT_CREAT)
 568                 opc = LUSTRE_OPC_CREATE;
 569         else
 570                 opc = LUSTRE_OPC_ANY;
 571
 572         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 573                                      dentry->d_name.len, lookup_flags, opc,
 574                                      NULL);
 575         if (IS_ERR(op_data))
 576                 RETURN((void *)op_data);
 577
 578         /* enforce umask if acl disabled or MDS doesn't support umask */
 579         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 580                 it->it_create_mode &= ~current_umask();
 581
 582         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 583                             lookup_flags, &req, ll_md_blocking_ast, 0);
 584         ll_finish_md_op_data(op_data);
 585         if (rc < 0)
 586                 GOTO(out, retval = ERR_PTR(rc));
 587
 588         rc = ll_lookup_it_finish(req, it, &icbd);
 589         if (rc != 0) {
 590                 ll_intent_release(it);
 591                 GOTO(out, retval = ERR_PTR(rc));
 592         }
 593
 594         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 595             !S_ISREG(dentry->d_inode->i_mode) &&
 596             !S_ISDIR(dentry->d_inode->i_mode)) {
 597                 ll_release_openhandle(dentry, it);
 598         }
 599         ll_lookup_finish_locks(it, dentry);
 600
 601         if (dentry == save)
 602                 GOTO(out, retval = NULL);
 603         else
 604                 GOTO(out, retval = dentry);
 605  out:
 606         if (req)
 607                 ptlrpc_req_finished(req);
 608         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 609                 ll_statahead_mark(parent, dentry);
 610         return retval;
 611 }
 612
 613 #ifdef HAVE_IOP_ATOMIC_OPEN
 614 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 615                                    unsigned int flags)
 616 {
 617         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 618         struct dentry *de;
 619
 620         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 621                dentry->d_name.len, dentry->d_name.name,
 622                PFID(ll_inode2fid(parent)), parent, flags);
 623
 624         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 625         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 626                 return NULL;
 627
 628         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 629                 itp = NULL;
 630         else
 631                 itp = &it;
 632         de = ll_lookup_it(parent, dentry, itp, 0);
 633
 634         if (itp != NULL)
 635                 ll_intent_release(itp);
 636
 637         return de;
 638 }
 639
 640 /*
 641  * For cached negative dentry and new dentry, handle lookup/create/open
 642  * together.
 643  */
 644 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 645                           struct file *file, unsigned open_flags,
 646                           umode_t mode, int *opened)
 647 {
 648         struct lookup_intent *it;
 649         struct dentry *de;
 650         long long lookup_flags = LOOKUP_OPEN;
 651         int rc = 0;
 652         ENTRY;
 653
 654         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 655                            "open_flags %x, mode %x opened %d\n",
 656                dentry->d_name.len, dentry->d_name.name,
 657                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 658
 659         OBD_ALLOC(it, sizeof(*it));
 660         if (!it)
 661                 RETURN(-ENOMEM);
 662
 663         it->it_op = IT_OPEN;
 664         if (open_flags & O_CREAT) {
 665                 it->it_op |= IT_CREAT;
 666                 lookup_flags |= LOOKUP_CREATE;
 667         }
 668         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 669         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 670
 671         /* Dentry added to dcache tree in ll_lookup_it */
 672         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 673         if (IS_ERR(de))
 674                 rc = PTR_ERR(de);
 675         else if (de != NULL)
 676                 dentry = de;
 677
 678         if (!rc) {
 679                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 680                         /* Dentry instantiated in ll_create_it. */
 681                         rc = ll_create_it(dir, dentry, mode, it);
 682                         if (rc) {
 683                                 /* We dget in ll_splice_alias. */
 684                                 if (de != NULL)
 685                                         dput(de);
 686                                 goto out_release;
 687                         }
 688
 689                         *opened |= FILE_CREATED;
 690                 }
 691                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 692                         /* Open dentry. */
 693                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 694                                 /* We cannot call open here as it would
 695                                  * deadlock.
 696                                  */
 697                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 698                                         ptlrpc_req_finished(
 699                                                        (struct ptlrpc_request *)
 700                                                           it->d.lustre.it_data);
 701                                 rc = finish_no_open(file, de);
 702                         } else {
 703                                 file->private_data = it;
 704                                 rc = finish_open(file, dentry, NULL, opened);
 705                                 /* We dget in ll_splice_alias. finish_open takes
 706                                  * care of dget for fd open.
 707                                  */
 708                                 if (de != NULL)
 709                                         dput(de);
 710                         }
 711                 } else {
 712                         rc = finish_no_open(file, de);
 713                 }
 714         }
 715
 716 out_release:
 717         ll_intent_release(it);
 718         OBD_FREE(it, sizeof(*it));
 719
 720         RETURN(rc);
 721 }
 722
 723 #else /* !HAVE_IOP_ATOMIC_OPEN */
 724 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 725                                         int lookup_flags)
 726 {
 727         struct lookup_intent *it;
 728
 729         OBD_ALLOC(it, sizeof(*it));
 730         if (!it)
 731                 return ERR_PTR(-ENOMEM);
 732
 733         if (lookup_flags & LOOKUP_OPEN) {
 734                 it->it_op = IT_OPEN;
 735                 if (lookup_flags & LOOKUP_CREATE)
 736                         it->it_op |= IT_CREAT;
 737                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 738                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 739         } else {
 740                 it->it_op = IT_GETATTR;
 741         }
 742
 743         return it;
 744 }
 745
 746 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 747                                    struct nameidata *nd)
 748 {
 749         struct dentry *de;
 750         ENTRY;
 751
 752         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 753                 struct lookup_intent *it;
 754
 755                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 756                         it = ll_d2d(dentry)->lld_it;
 757                         ll_d2d(dentry)->lld_it = NULL;
 758                 } else {
 759                         if ((nd->flags & LOOKUP_CREATE) &&
 760                             !(nd->flags & LOOKUP_OPEN))
 761                                 RETURN(NULL);
 762
 763                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 764                         if (IS_ERR(it))
 765                                 RETURN((struct dentry *)it);
 766                 }
 767
 768                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 769                 if (de)
 770                         dentry = de;
 771                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 772                         if (dentry->d_inode &&
 773                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 774                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 775                                         // We cannot call open here as it would
 776                                         // deadlock.
 777                                         ptlrpc_req_finished(
 778                                                        (struct ptlrpc_request *)
 779                                                           it->d.lustre.it_data);
 780                                 } else {
 781                                         struct file *filp;
 782
 783                                         nd->intent.open.file->private_data = it;
 784                                         filp = lookup_instantiate_filp(nd,
 785                                                                        dentry,
 786                                                                        NULL);
 787                                         if (IS_ERR(filp)) {
 788                                                 if (de)
 789                                                         dput(de);
 790                                                 de = (struct dentry *)filp;
 791                                         }
 792                                 }
 793                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 794                                 // XXX This can only reliably work on assumption
 795                                 // that there are NO hashed negative dentries.
 796                                 ll_d2d(dentry)->lld_it = it;
 797                                 it = NULL; /* Will be freed in ll_create_nd */
 798                                 /* We absolutely depend on ll_create_nd to be
 799                                  * called to not leak this intent and possible
 800                                  * data attached to it */
 801                         }
 802                 }
 803
 804                 if (it) {
 805                         ll_intent_release(it);
 806                         OBD_FREE(it, sizeof(*it));
 807                 }
 808         } else {
 809                 de = ll_lookup_it(parent, dentry, NULL, 0);
 810         }
 811
 812         RETURN(de);
 813 }
 814 #endif /* HAVE_IOP_ATOMIC_OPEN */
 815
 816 /* We depend on "mode" being set with the proper file type/umask by now */
 817 static struct inode *ll_create_node(struct inode *dir, const char *name,
 818                                     int namelen, const void *data, int datalen,
 819                                     int mode, __u64 extra,
 820                                     struct lookup_intent *it)
 821 {
 822         struct inode *inode = NULL;
 823         struct ptlrpc_request *request = NULL;
 824         struct ll_sb_info *sbi = ll_i2sbi(dir);
 825         int rc;
 826         ENTRY;
 827
 828         LASSERT(it && it->d.lustre.it_disposition);
 829
 830         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 831         request = it->d.lustre.it_data;
 832         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 833         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 834         if (rc)
 835                 GOTO(out, inode = ERR_PTR(rc));
 836
 837         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 838
 839         /* We asked for a lock on the directory, but were granted a
 840          * lock on the inode.  Since we finally have an inode pointer,
 841          * stuff it in the lock. */
 842         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 843                PFID(ll_inode2fid(inode)), inode);
 844         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 845         EXIT;
 846  out:
 847         ptlrpc_req_finished(request);
 848         return inode;
 849 }
 850
 851 /*
 852  * By the time this is called, we already have created the directory cache
 853  * entry for the new file, but it is so far negative - it has no inode.
 854  *
 855  * We defer creating the OBD object(s) until open, to keep the intent and
 856  * non-intent code paths similar, and also because we do not have the MDS
 857  * inode number before calling ll_create_node() (which is needed for LOV),
 858  * so we would need to do yet another RPC to the MDS to store the LOV EA
 859  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 860  * lmm_size in datalen (the MDS still has code which will handle that).
 861  *
 862  * If the create succeeds, we fill in the inode information
 863  * with d_instantiate().
 864  */
 865 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 866                         struct lookup_intent *it)
 867 {
 868         struct inode *inode;
 869         int rc = 0;
 870         ENTRY;
 871
 872         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 873                dentry->d_name.len, dentry->d_name.name,
 874                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 875
 876         rc = it_open_error(DISP_OPEN_CREATE, it);
 877         if (rc)
 878                 RETURN(rc);
 879
 880         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 881                                NULL, 0, mode, 0, it);
 882         if (IS_ERR(inode))
 883                 RETURN(PTR_ERR(inode));
 884
 885         if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
 886                 ll_i2info(inode)->lli_volatile = true;
 887
 888         d_instantiate(dentry, inode);
 889         RETURN(0);
 890 }
 891
 892 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 893 {
 894         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 895                                                        &RMF_MDT_BODY);
 896
 897         LASSERT(body);
 898         if (body->valid & OBD_MD_FLMTIME &&
 899             body->mtime > LTIME_S(inode->i_mtime)) {
 900                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 901                                 "\n", PFID(ll_inode2fid(inode)),
 902                                 LTIME_S(inode->i_mtime), body->mtime);
 903                 LTIME_S(inode->i_mtime) = body->mtime;
 904         }
 905         if (body->valid & OBD_MD_FLCTIME &&
 906             body->ctime > LTIME_S(inode->i_ctime))
 907                 LTIME_S(inode->i_ctime) = body->ctime;
 908 }
 909
 910 static int ll_new_node(struct inode *dir, struct qstr *name,
 911                        const char *tgt, int mode, int rdev,
 912                        struct dentry *dchild, __u32 opc)
 913 {
 914         struct ptlrpc_request *request = NULL;
 915         struct md_op_data *op_data;
 916         struct inode *inode = NULL;
 917         struct ll_sb_info *sbi = ll_i2sbi(dir);
 918         int tgt_len = 0;
 919         int err;
 920
 921         ENTRY;
 922         if (unlikely(tgt != NULL))
 923                 tgt_len = strlen(tgt) + 1;
 924
 925         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 926                                      name->len, 0, opc, NULL);
 927         if (IS_ERR(op_data))
 928                 GOTO(err_exit, err = PTR_ERR(op_data));
 929
 930         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 931                         from_kuid(&init_user_ns, current_fsuid()),
 932                         from_kgid(&init_user_ns, current_fsgid()),
 933                         cfs_curproc_cap_pack(), rdev, &request);
 934         ll_finish_md_op_data(op_data);
 935         if (err)
 936                 GOTO(err_exit, err);
 937
 938         ll_update_times(request, dir);
 939
 940         if (dchild) {
 941                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 942                 if (err)
 943                      GOTO(err_exit, err);
 944
 945                 d_instantiate(dchild, inode);
 946         }
 947         EXIT;
 948 err_exit:
 949         ptlrpc_req_finished(request);
 950
 951         return err;
 952 }
 953
 954 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 955                             unsigned rdev, struct dentry *dchild)
 956 {
 957         int err;
 958         ENTRY;
 959
 960         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
 961                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
 962                mode, rdev);
 963
 964         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 965                 mode &= ~current_umask();
 966
 967         switch (mode & S_IFMT) {
 968         case 0:
 969                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 970         case S_IFREG:
 971         case S_IFCHR:
 972         case S_IFBLK:
 973         case S_IFIFO:
 974         case S_IFSOCK:
 975                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 976                                   LUSTRE_OPC_MKNOD);
 977                 break;
 978         case S_IFDIR:
 979                 err = -EPERM;
 980                 break;
 981         default:
 982                 err = -EINVAL;
 983         }
 984
 985         if (!err)
 986                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 987
 988         RETURN(err);
 989 }
 990
 991 #ifdef HAVE_IOP_ATOMIC_OPEN
 992 /*
 993  * Plain create. Intent create is handled in atomic_open.
 994  */
 995 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 996                         umode_t mode, bool want_excl)
 997 {
 998         int rc;
 999
1000         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1001                            "flags=%u, excl=%d\n", dentry->d_name.len,
1002                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1003                dir, mode, want_excl);
1004
1005         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1006
1007         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1008
1009         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1010                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1011
1012         return rc;
1013 }
1014 #else /* !HAVE_IOP_ATOMIC_OPEN */
1015 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1016                         ll_umode_t mode, struct nameidata *nd)
1017 {
1018         struct ll_dentry_data *lld = ll_d2d(dentry);
1019         struct lookup_intent *it = NULL;
1020         int rc;
1021
1022         if (lld != NULL)
1023                 it = lld->lld_it;
1024
1025         if (!it)
1026                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1027
1028         lld->lld_it = NULL;
1029
1030         /* Was there an error? Propagate it! */
1031         if (it->d.lustre.it_status) {
1032                 rc = it->d.lustre.it_status;
1033                 goto out;
1034         }
1035
1036         rc = ll_create_it(dir, dentry, mode, it);
1037         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1038                 struct file *filp;
1039
1040                 nd->intent.open.file->private_data = it;
1041                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1042                 if (IS_ERR(filp))
1043                         rc = PTR_ERR(filp);
1044         }
1045
1046 out:
1047         ll_intent_release(it);
1048         OBD_FREE(it, sizeof(*it));
1049
1050         if (!rc)
1051                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1052
1053         return rc;
1054 }
1055 #endif /* HAVE_IOP_ATOMIC_OPEN */
1056
1057 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1058                               const char *tgt, struct dentry *dchild)
1059 {
1060         int err;
1061         ENTRY;
1062
1063         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1064                name->len, name->name, PFID(ll_inode2fid(dir)),
1065                dir, 3000, tgt);
1066
1067         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1068                           0, dchild, LUSTRE_OPC_SYMLINK);
1069
1070         if (!err)
1071                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1072
1073         RETURN(err);
1074 }
1075
1076 static int ll_link_generic(struct inode *src,  struct inode *dir,
1077                            struct qstr *name, struct dentry *dchild)
1078 {
1079         struct ll_sb_info *sbi = ll_i2sbi(dir);
1080         struct ptlrpc_request *request = NULL;
1081         struct md_op_data *op_data;
1082         int err;
1083
1084         ENTRY;
1085         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1086                "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1087                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1088
1089         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1090                                      0, LUSTRE_OPC_ANY, NULL);
1091         if (IS_ERR(op_data))
1092                 RETURN(PTR_ERR(op_data));
1093
1094         err = md_link(sbi->ll_md_exp, op_data, &request);
1095         ll_finish_md_op_data(op_data);
1096         if (err)
1097                 GOTO(out, err);
1098
1099         ll_update_times(request, dir);
1100         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1101         EXIT;
1102 out:
1103         ptlrpc_req_finished(request);
1104         RETURN(err);
1105 }
1106
1107 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1108                             int mode, struct dentry *dchild)
1109
1110 {
1111         int err;
1112         ENTRY;
1113
1114         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1115                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1116
1117         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1118                 mode &= ~current_umask();
1119         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1120         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1121
1122         if (!err)
1123                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1124
1125         RETURN(err);
1126 }
1127
1128 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1129                             struct dentry *dchild, struct qstr *name)
1130 {
1131         struct ptlrpc_request *request = NULL;
1132         struct md_op_data *op_data;
1133         int rc;
1134         ENTRY;
1135
1136         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1137                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1138
1139         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1140                 RETURN(-EBUSY);
1141
1142         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1143                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1144         if (IS_ERR(op_data))
1145                 RETURN(PTR_ERR(op_data));
1146
1147         if (dchild != NULL && dchild->d_inode != NULL)
1148                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1149         op_data->op_fid2 = op_data->op_fid3;
1150         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1151         ll_finish_md_op_data(op_data);
1152         if (rc == 0) {
1153                 ll_update_times(request, dir);
1154                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1155         }
1156
1157         ptlrpc_req_finished(request);
1158         RETURN(rc);
1159 }
1160
1161 /**
1162  * Remove dir entry
1163  **/
1164 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1165 {
1166         struct ptlrpc_request *request = NULL;
1167         struct md_op_data *op_data;
1168         int rc;
1169         ENTRY;
1170
1171         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1172                namelen, name, PFID(ll_inode2fid(dir)), dir);
1173
1174         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1175                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1176         if (IS_ERR(op_data))
1177                 RETURN(PTR_ERR(op_data));
1178         op_data->op_cli_flags |= CLI_RM_ENTRY;
1179         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1180         ll_finish_md_op_data(op_data);
1181         if (rc == 0) {
1182                 ll_update_times(request, dir);
1183                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1184         }
1185
1186         ptlrpc_req_finished(request);
1187         RETURN(rc);
1188 }
1189
1190 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1191 {
1192         struct mdt_body *body;
1193         struct lov_mds_md *eadata;
1194         struct lov_stripe_md *lsm = NULL;
1195         struct obd_trans_info oti = { 0 };
1196         struct obdo *oa;
1197         struct obd_capa *oc = NULL;
1198         int rc;
1199         ENTRY;
1200
1201         /* req is swabbed so this is safe */
1202         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1203         if (!(body->valid & OBD_MD_FLEASIZE))
1204                 RETURN(0);
1205
1206         if (body->eadatasize == 0) {
1207                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1208                 GOTO(out, rc = -EPROTO);
1209         }
1210
1211         /* The MDS sent back the EA because we unlinked the last reference
1212          * to this file. Use this EA to unlink the objects on the OST.
1213          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1214          * check it is complete and sensible. */
1215         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1216                                               body->eadatasize);
1217         LASSERT(eadata != NULL);
1218
1219         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1220         if (rc < 0) {
1221                 CERROR("obd_unpackmd: %d\n", rc);
1222                 GOTO(out, rc);
1223         }
1224         LASSERT(rc >= sizeof(*lsm));
1225
1226         OBDO_ALLOC(oa);
1227         if (oa == NULL)
1228                 GOTO(out_free_memmd, rc = -ENOMEM);
1229
1230         oa->o_oi = lsm->lsm_oi;
1231         oa->o_mode = body->mode & S_IFMT;
1232         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1233
1234         if (body->valid & OBD_MD_FLCOOKIE) {
1235                 oa->o_valid |= OBD_MD_FLCOOKIE;
1236                 oti.oti_logcookies =
1237                         req_capsule_server_sized_get(&request->rq_pill,
1238                                                      &RMF_LOGCOOKIES,
1239                                                    sizeof(struct llog_cookie) *
1240                                                      lsm->lsm_stripe_count);
1241                 if (oti.oti_logcookies == NULL) {
1242                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1243                         body->valid &= ~OBD_MD_FLCOOKIE;
1244                 }
1245         }
1246
1247         if (body->valid & OBD_MD_FLOSSCAPA) {
1248                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1249                 if (rc)
1250                         GOTO(out_free_memmd, rc);
1251         }
1252
1253         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1254                          ll_i2mdexp(dir), oc);
1255         capa_put(oc);
1256         if (rc)
1257                 CERROR("obd destroy objid "DOSTID" error %d\n",
1258                        POSTID(&lsm->lsm_oi), rc);
1259 out_free_memmd:
1260         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1261         OBDO_FREE(oa);
1262 out:
1263         return rc;
1264 }
1265
1266 /* ll_unlink_generic() doesn't update the inode with the new link count.
1267  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1268  * is any lock existing. They will recycle dentries and inodes based upon locks
1269  * too. b=20433 */
1270 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1271                              struct dentry *dchild, struct qstr *name)
1272 {
1273         struct ptlrpc_request *request = NULL;
1274         struct md_op_data *op_data;
1275         int rc;
1276         ENTRY;
1277         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1278                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1279
1280         /*
1281          * XXX: unlink bind mountpoint maybe call to here,
1282          * just check it as vfs_unlink does.
1283          */
1284         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1285                 RETURN(-EBUSY);
1286
1287         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1288                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1289         if (IS_ERR(op_data))
1290                 RETURN(PTR_ERR(op_data));
1291
1292         if (dchild != NULL && dchild->d_inode != NULL)
1293                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1294
1295         op_data->op_fid2 = op_data->op_fid3;
1296         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1297         ll_finish_md_op_data(op_data);
1298         if (rc)
1299                 GOTO(out, rc);
1300
1301         ll_update_times(request, dir);
1302         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1303
1304         rc = ll_objects_destroy(request, dir);
1305  out:
1306         ptlrpc_req_finished(request);
1307         RETURN(rc);
1308 }
1309
1310 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1311                              struct dentry *src_dchild, struct qstr *src_name,
1312                              struct inode *tgt, struct dentry *tgt_dparent,
1313                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1314 {
1315         struct ptlrpc_request *request = NULL;
1316         struct ll_sb_info *sbi = ll_i2sbi(src);
1317         struct md_op_data *op_data;
1318         int err;
1319         ENTRY;
1320         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1321                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1322                src_name->len, src_name->name,
1323                PFID(ll_inode2fid(src)), src, tgt_name->len,
1324                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1325
1326         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1327             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1328                 RETURN(-EBUSY);
1329
1330         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1331                                      LUSTRE_OPC_ANY, NULL);
1332         if (IS_ERR(op_data))
1333                 RETURN(PTR_ERR(op_data));
1334
1335         if (src_dchild != NULL && src_dchild->d_inode != NULL)
1336                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1337         if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1338                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1339
1340         err = md_rename(sbi->ll_md_exp, op_data,
1341                         src_name->name, src_name->len,
1342                         tgt_name->name, tgt_name->len, &request);
1343         ll_finish_md_op_data(op_data);
1344         if (!err) {
1345                 ll_update_times(request, src);
1346                 ll_update_times(request, tgt);
1347                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1348                 err = ll_objects_destroy(request, src);
1349         }
1350
1351         ptlrpc_req_finished(request);
1352
1353         RETURN(err);
1354 }
1355
1356 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1357                     dev_t rdev)
1358 {
1359         return ll_mknod_generic(dir, &dchild->d_name, mode,
1360                                 old_encode_dev(rdev), dchild);
1361 }
1362
1363 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1364 {
1365         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1366 }
1367
1368 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1369 {
1370         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1371 }
1372
1373 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1374 {
1375         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1376 }
1377
1378 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1379                       const char *oldname)
1380 {
1381         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1382 }
1383
1384 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1385                    struct dentry *new_dentry)
1386 {
1387         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1388                                new_dentry);
1389 }
1390
1391 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1392                      struct inode *new_dir, struct dentry *new_dentry)
1393 {
1394         int err;
1395         err = ll_rename_generic(old_dir, NULL,
1396                                  old_dentry, &old_dentry->d_name,
1397                                  new_dir, NULL, new_dentry,
1398                                  &new_dentry->d_name);
1399         if (!err)
1400                         d_move(old_dentry, new_dentry);
1401         return err;
1402 }
1403
1404 struct inode_operations ll_dir_inode_operations = {
1405         .mknod              = ll_mknod,
1406 #ifdef HAVE_IOP_ATOMIC_OPEN
1407         .atomic_open        = ll_atomic_open,
1408 #endif
1409         .lookup             = ll_lookup_nd,
1410         .create             = ll_create_nd,
1411         /* We need all these non-raw things for NFSD, to not patch it. */
1412         .unlink             = ll_unlink,
1413         .mkdir              = ll_mkdir,
1414         .rmdir              = ll_rmdir,
1415         .symlink            = ll_symlink,
1416         .link               = ll_link,
1417         .rename             = ll_rename,
1418         .setattr            = ll_setattr,
1419         .getattr            = ll_getattr,
1420         .permission         = ll_inode_permission,
1421         .setxattr           = ll_setxattr,
1422         .getxattr           = ll_getxattr,
1423         .listxattr          = ll_listxattr,
1424         .removexattr        = ll_removexattr,
1425 #ifdef HAVE_IOP_GET_ACL
1426         .get_acl            = ll_get_acl,
1427 #endif
1428 };
1429
1430 struct inode_operations ll_special_inode_operations = {
1431         .setattr        = ll_setattr,
1432         .getattr        = ll_getattr,
1433         .permission     = ll_inode_permission,
1434         .setxattr       = ll_setxattr,
1435         .getxattr       = ll_getxattr,
1436         .listxattr      = ll_listxattr,
1437         .removexattr    = ll_removexattr,
1438 #ifdef HAVE_IOP_GET_ACL
1439         .get_acl            = ll_get_acl,
1440 #endif
1441 };