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