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, 2012, 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         struct ll_d_hlist_node *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);
 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);
 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                 /* For OPEN locks we differentiate between lock modes
 231                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 232                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 233                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 234                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 235
 236                 if (bits & MDS_INODELOCK_OPEN)
 237                         ll_have_md_lock(inode, &bits, mode);
 238
 239                 fid = ll_inode2fid(inode);
 240                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
 241                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
 242                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
 243                         LDLM_ERROR(lock, "data mismatch with object "
 244                                    DFID" (%p)", PFID(fid), inode);
 245                 }
 246
 247                 if (bits & MDS_INODELOCK_OPEN) {
 248                         int flags = 0;
 249                         switch (lock->l_req_mode) {
 250                         case LCK_CW:
 251                                 flags = FMODE_WRITE;
 252                                 break;
 253                         case LCK_PR:
 254                                 flags = FMODE_EXEC;
 255                                 break;
 256                         case LCK_CR:
 257                                 flags = FMODE_READ;
 258                                 break;
 259                         default:
 260                                 CERROR("Unexpected lock mode for OPEN lock "
 261                                        "%d, inode %ld\n", lock->l_req_mode,
 262                                        inode->i_ino);
 263                         }
 264                         ll_md_real_close(inode, flags);
 265                 }
 266
 267                 lli = ll_i2info(inode);
 268                 if (bits & MDS_INODELOCK_LAYOUT) {
 269                         struct cl_object_conf conf = { { 0 } };
 270
 271                         conf.coc_opc = OBJECT_CONF_INVALIDATE;
 272                         conf.coc_inode = inode;
 273                         rc = ll_layout_conf(inode, &conf);
 274                         if (rc)
 275                                 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
 276                 }
 277
 278                 if (bits & MDS_INODELOCK_UPDATE)
 279                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 280
 281                 if (S_ISDIR(inode->i_mode) &&
 282                      (bits & MDS_INODELOCK_UPDATE)) {
 283                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 284                                inode->i_ino);
 285                         truncate_inode_pages(inode->i_mapping, 0);
 286                         ll_invalidate_negative_children(inode);
 287                 }
 288
 289                 if (inode->i_sb->s_root &&
 290                     inode != inode->i_sb->s_root->d_inode &&
 291                     (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
 292                         ll_invalidate_aliases(inode);
 293                 iput(inode);
 294                 break;
 295         }
 296         default:
 297                 LBUG();
 298         }
 299
 300         RETURN(0);
 301 }
 302
 303 __u32 ll_i2suppgid(struct inode *i)
 304 {
 305         if (cfs_curproc_is_in_groups(i->i_gid))
 306                 return (__u32)i->i_gid;
 307         else
 308                 return (__u32)(-1);
 309 }
 310
 311 /* Pack the required supplementary groups into the supplied groups array.
 312  * If we don't need to use the groups from the target inode(s) then we
 313  * instead pack one or more groups from the user's supplementary group
 314  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 315 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 316 {
 317 #if 0
 318         int i;
 319 #endif
 320
 321         LASSERT(i1 != NULL);
 322         LASSERT(suppgids != NULL);
 323
 324         suppgids[0] = ll_i2suppgid(i1);
 325
 326         if (i2)
 327                 suppgids[1] = ll_i2suppgid(i2);
 328                 else
 329                         suppgids[1] = -1;
 330
 331 #if 0
 332         for (i = 0; i < current_ngroups; i++) {
 333                 if (suppgids[0] == -1) {
 334                         if (current_groups[i] != suppgids[1])
 335                                 suppgids[0] = current_groups[i];
 336                         continue;
 337                 }
 338                 if (suppgids[1] == -1) {
 339                         if (current_groups[i] != suppgids[0])
 340                                 suppgids[1] = current_groups[i];
 341                         continue;
 342                 }
 343                 break;
 344         }
 345 #endif
 346 }
 347
 348 /*
 349  * try to reuse three types of dentry:
 350  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 351  *    by concurrent .revalidate).
 352  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 353  *    be cleared by others calling d_lustre_revalidate).
 354  * 3. DISCONNECTED alias.
 355  */
 356 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 357 {
 358         struct dentry *alias, *discon_alias, *invalid_alias;
 359         struct ll_d_hlist_node *p;
 360
 361         if (ll_d_hlist_empty(&inode->i_dentry))
 362                 return NULL;
 363
 364         discon_alias = invalid_alias = NULL;
 365
 366         ll_lock_dcache(inode);
 367         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 368                 LASSERT(alias != dentry);
 369
 370                 spin_lock(&alias->d_lock);
 371                 if (alias->d_flags & DCACHE_DISCONNECTED)
 372                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 373                         discon_alias = alias;
 374                 else if (alias->d_parent == dentry->d_parent             &&
 375                          alias->d_name.hash == dentry->d_name.hash       &&
 376                          alias->d_name.len == dentry->d_name.len         &&
 377                          memcmp(alias->d_name.name, dentry->d_name.name,
 378                                 dentry->d_name.len) == 0)
 379                         invalid_alias = alias;
 380                 spin_unlock(&alias->d_lock);
 381
 382                 if (invalid_alias)
 383                         break;
 384         }
 385         alias = invalid_alias ?: discon_alias ?: NULL;
 386         if (alias) {
 387                 spin_lock(&alias->d_lock);
 388                 dget_dlock(alias);
 389                 spin_unlock(&alias->d_lock);
 390         }
 391         ll_unlock_dcache(inode);
 392
 393         return alias;
 394 }
 395
 396 /*
 397  * Similar to d_splice_alias(), but lustre treats invalid alias
 398  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 399  */
 400 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 401 {
 402         struct dentry *new;
 403
 404         if (inode) {
 405                 new = ll_find_alias(inode, de);
 406                 if (new) {
 407                         ll_dops_init(new, 1, 1);
 408                         d_move(new, de);
 409                         iput(inode);
 410                         CDEBUG(D_DENTRY,
 411                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 412                               new, new->d_inode, d_refcount(new), new->d_flags);
 413                         return new;
 414                 }
 415         }
 416         ll_dops_init(de, 1, 1);
 417         __d_lustre_invalidate(de);
 418         d_add(de, inode);
 419         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 420                de, de->d_inode, d_refcount(de), de->d_flags);
 421         return de;
 422 }
 423
 424 int ll_lookup_it_finish(struct ptlrpc_request *request,
 425                         struct lookup_intent *it, void *data)
 426 {
 427         struct it_cb_data *icbd = data;
 428         struct dentry **de = icbd->icbd_childp;
 429         struct inode *parent = icbd->icbd_parent;
 430         struct inode *inode = NULL;
 431         __u64 bits = 0;
 432         int rc;
 433         ENTRY;
 434
 435         /* NB 1 request reference will be taken away by ll_intent_lock()
 436          * when I return */
 437         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 438                it->d.lustre.it_disposition);
 439         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 440                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 441                 if (rc)
 442                         RETURN(rc);
 443
 444                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 445
 446                 /* We used to query real size from OSTs here, but actually
 447                    this is not needed. For stat() calls size would be updated
 448                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 449                    2.4 and
 450                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 451                    Everybody else who needs correct file size would call
 452                    ll_glimpse_size or some equivalent themselves anyway.
 453                    Also see bug 7198. */
 454         }
 455
 456         /* Only hash *de if it is unhashed (new dentry).
 457          * Atoimc_open may passin hashed dentries for open.
 458          */
 459         if (d_unhashed(*de))
 460                 *de = ll_splice_alias(inode, *de);
 461
 462         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 463                 /* we have lookup look - unhide dentry */
 464                 if (bits & MDS_INODELOCK_LOOKUP)
 465                         d_lustre_revalidate(*de);
 466         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 467                 /* If file created on server, don't depend on parent UPDATE
 468                  * lock to unhide it. It is left hidden and next lookup can
 469                  * find it in ll_splice_alias.
 470                  */
 471                 /* Check that parent has UPDATE lock. */
 472                 struct lookup_intent parent_it = {
 473                                         .it_op = IT_GETATTR,
 474                                         .d.lustre.it_lock_handle = 0 };
 475
 476                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 477                                        &ll_i2info(parent)->lli_fid, NULL)) {
 478                         d_lustre_revalidate(*de);
 479                         ll_intent_release(&parent_it);
 480                 }
 481         }
 482
 483         RETURN(0);
 484 }
 485
 486 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 487                                    struct lookup_intent *it, int lookup_flags)
 488 {
 489         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 490         struct dentry *save = dentry, *retval;
 491         struct ptlrpc_request *req = NULL;
 492         struct md_op_data *op_data;
 493         struct it_cb_data icbd;
 494         __u32 opc;
 495         int rc;
 496         ENTRY;
 497
 498         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 499                 RETURN(ERR_PTR(-ENAMETOOLONG));
 500
 501         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 502                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 503                parent->i_generation, parent, LL_IT2STR(it));
 504
 505         if (d_mountpoint(dentry))
 506                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 507
 508         ll_frob_intent(&it, &lookup_it);
 509
 510         /* As do_lookup is called before follow_mount, root dentry may be left
 511          * not valid, revalidate it here. */
 512         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 513             (it->it_op & (IT_OPEN | IT_CREAT))) {
 514                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
 515                                             MDS_INODELOCK_LOOKUP);
 516                 if (rc)
 517                         RETURN(ERR_PTR(rc));
 518         }
 519
 520         if (it->it_op == IT_GETATTR) {
 521                 rc = ll_statahead_enter(parent, &dentry, 0);
 522                 if (rc == 1) {
 523                         if (dentry == save)
 524                                 GOTO(out, retval = NULL);
 525                         GOTO(out, retval = dentry);
 526                 }
 527         }
 528
 529         icbd.icbd_childp = &dentry;
 530         icbd.icbd_parent = parent;
 531
 532         if (it->it_op & IT_CREAT ||
 533             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
 534                 opc = LUSTRE_OPC_CREATE;
 535         else
 536                 opc = LUSTRE_OPC_ANY;
 537
 538         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 539                                      dentry->d_name.len, lookup_flags, opc,
 540                                      NULL);
 541         if (IS_ERR(op_data))
 542                 RETURN((void *)op_data);
 543
 544         /* enforce umask if acl disabled or MDS doesn't support umask */
 545         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 546                 it->it_create_mode &= ~cfs_curproc_umask();
 547
 548         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 549                             lookup_flags, &req, ll_md_blocking_ast, 0);
 550         ll_finish_md_op_data(op_data);
 551         if (rc < 0)
 552                 GOTO(out, retval = ERR_PTR(rc));
 553
 554         rc = ll_lookup_it_finish(req, it, &icbd);
 555         if (rc != 0) {
 556                 ll_intent_release(it);
 557                 GOTO(out, retval = ERR_PTR(rc));
 558         }
 559
 560         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 561             !S_ISREG(dentry->d_inode->i_mode) &&
 562             !S_ISDIR(dentry->d_inode->i_mode)) {
 563                 ll_release_openhandle(dentry, it);
 564         }
 565         ll_lookup_finish_locks(it, dentry);
 566
 567         if (dentry == save)
 568                 GOTO(out, retval = NULL);
 569         else
 570                 GOTO(out, retval = dentry);
 571  out:
 572         if (req)
 573                 ptlrpc_req_finished(req);
 574         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 575                 ll_statahead_mark(parent, dentry);
 576         return retval;
 577 }
 578
 579 #ifdef HAVE_IOP_ATOMIC_OPEN
 580 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 581                                    unsigned int flags)
 582 {
 583         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 584         struct dentry *de;
 585
 586         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
 587                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 588                parent->i_generation, parent, flags);
 589
 590         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 591         if ((flags & LOOKUP_CREATE ) && !(flags & LOOKUP_OPEN)) {
 592                 ll_dops_init(dentry, 1, 1);
 593                 __d_lustre_invalidate(dentry);
 594                 d_add(dentry, NULL);
 595                 return NULL;
 596         }
 597
 598         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 599                 itp = NULL;
 600         else
 601                 itp = &it;
 602         de = ll_lookup_it(parent, dentry, itp, 0);
 603
 604         if (itp != NULL)
 605                 ll_intent_release(itp);
 606
 607         return de;
 608 }
 609
 610 /*
 611  * For cached negative dentry and new dentry, handle lookup/create/open
 612  * together.
 613  */
 614 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 615                           struct file *file, unsigned open_flags,
 616                           umode_t mode, int *opened)
 617 {
 618         struct lookup_intent *it;
 619         struct dentry *de;
 620         long long lookup_flags = LOOKUP_OPEN;
 621         int rc = 0;
 622         ENTRY;
 623
 624         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
 625                            "open_flags %x,mode %x opened %d\n",
 626                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 627                dir->i_generation, dir, file, open_flags, mode, *opened);
 628
 629         OBD_ALLOC(it, sizeof(*it));
 630         if (!it)
 631                 RETURN(-ENOMEM);
 632
 633         it->it_op = IT_OPEN;
 634         if (mode) {
 635                 it->it_op |= IT_CREAT;
 636                 lookup_flags |= LOOKUP_CREATE;
 637         }
 638         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 639         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 640
 641         /* Dentry added to dcache tree in ll_lookup_it */
 642         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 643         if (IS_ERR(de))
 644                 rc = PTR_ERR(de);
 645         else if (de != NULL)
 646                 dentry = de;
 647
 648         if (!rc) {
 649                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 650                         /* Dentry instantiated in ll_create_it. */
 651                         rc = ll_create_it(dir, dentry, mode, it);
 652                         if (rc) {
 653                                 /* We dget in ll_splice_alias. */
 654                                 if (de != NULL)
 655                                         dput(de);
 656                                 goto out_release;
 657                         }
 658
 659                         *opened |= FILE_CREATED;
 660                 }
 661                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 662                         /* Open dentry. */
 663                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 664                                 /* We cannot call open here as it would
 665                                  * deadlock.
 666                                  */
 667                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 668                                         ptlrpc_req_finished(
 669                                                        (struct ptlrpc_request *)
 670                                                           it->d.lustre.it_data);
 671                                 rc = finish_no_open(file, de);
 672                         } else {
 673                                 file->private_data = it;
 674                                 rc = finish_open(file, dentry, NULL, opened);
 675                                 /* We dget in ll_splice_alias. finish_open takes
 676                                  * care of dget for fd open.
 677                                  */
 678                                 if (de != NULL)
 679                                         dput(de);
 680                         }
 681                 } else {
 682                         rc = finish_no_open(file, de);
 683                 }
 684         }
 685
 686 out_release:
 687         ll_intent_release(it);
 688         OBD_FREE(it, sizeof(*it));
 689
 690         RETURN(rc);
 691 }
 692
 693 #else /* !HAVE_IOP_ATOMIC_OPEN */
 694 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 695                                         int lookup_flags)
 696 {
 697         struct lookup_intent *it;
 698
 699         OBD_ALLOC(it, sizeof(*it));
 700         if (!it)
 701                 return ERR_PTR(-ENOMEM);
 702
 703         if (lookup_flags & LOOKUP_OPEN) {
 704                 it->it_op = IT_OPEN;
 705                 if (lookup_flags & LOOKUP_CREATE)
 706                         it->it_op |= IT_CREAT;
 707                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 708                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 709         } else {
 710                 it->it_op = IT_GETATTR;
 711         }
 712
 713         return it;
 714 }
 715
 716 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 717                                    struct nameidata *nd)
 718 {
 719         struct dentry *de;
 720         ENTRY;
 721
 722         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 723                 struct lookup_intent *it;
 724
 725                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 726                         it = ll_d2d(dentry)->lld_it;
 727                         ll_d2d(dentry)->lld_it = NULL;
 728                 } else {
 729                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
 730                                 /* We are sure this is new dentry, so we need to create
 731                                    our private data and set the dentry ops */
 732                                 ll_dops_init(dentry, 1, 1);
 733                                 __d_lustre_invalidate(dentry);
 734                                 d_add(dentry, NULL);
 735                                 RETURN(NULL);
 736                         }
 737                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 738                         if (IS_ERR(it))
 739                                 RETURN((struct dentry *)it);
 740                 }
 741
 742                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 743                 if (de)
 744                         dentry = de;
 745                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 746                         if (dentry->d_inode &&
 747                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 748                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 749                                         // We cannot call open here as it would
 750                                         // deadlock.
 751                                         ptlrpc_req_finished(
 752                                                        (struct ptlrpc_request *)
 753                                                           it->d.lustre.it_data);
 754                                 } else {
 755                                         struct file *filp;
 756
 757                                         nd->intent.open.file->private_data = it;
 758                                         filp = lookup_instantiate_filp(nd,
 759                                                                        dentry,
 760                                                                        NULL);
 761                                         if (IS_ERR(filp)) {
 762                                                 if (de)
 763                                                         dput(de);
 764                                                 de = (struct dentry *)filp;
 765                                         }
 766                                 }
 767                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 768                                 // XXX This can only reliably work on assumption
 769                                 // that there are NO hashed negative dentries.
 770                                 ll_d2d(dentry)->lld_it = it;
 771                                 it = NULL; /* Will be freed in ll_create_nd */
 772                                 /* We absolutely depend on ll_create_nd to be
 773                                  * called to not leak this intent and possible
 774                                  * data attached to it */
 775                         }
 776                 }
 777
 778                 if (it) {
 779                         ll_intent_release(it);
 780                         OBD_FREE(it, sizeof(*it));
 781                 }
 782         } else {
 783                 de = ll_lookup_it(parent, dentry, NULL, 0);
 784         }
 785
 786         RETURN(de);
 787 }
 788 #endif /* HAVE_IOP_ATOMIC_OPEN */
 789
 790 /* We depend on "mode" being set with the proper file type/umask by now */
 791 static struct inode *ll_create_node(struct inode *dir, const char *name,
 792                                     int namelen, const void *data, int datalen,
 793                                     int mode, __u64 extra,
 794                                     struct lookup_intent *it)
 795 {
 796         struct inode *inode = NULL;
 797         struct ptlrpc_request *request = NULL;
 798         struct ll_sb_info *sbi = ll_i2sbi(dir);
 799         int rc;
 800         ENTRY;
 801
 802         LASSERT(it && it->d.lustre.it_disposition);
 803
 804         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 805         request = it->d.lustre.it_data;
 806         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 807         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 808         if (rc)
 809                 GOTO(out, inode = ERR_PTR(rc));
 810
 811         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 812
 813         /* We asked for a lock on the directory, but were granted a
 814          * lock on the inode.  Since we finally have an inode pointer,
 815          * stuff it in the lock. */
 816         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 817                inode, inode->i_ino, inode->i_generation);
 818         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 819         EXIT;
 820  out:
 821         ptlrpc_req_finished(request);
 822         return inode;
 823 }
 824
 825 /*
 826  * By the time this is called, we already have created the directory cache
 827  * entry for the new file, but it is so far negative - it has no inode.
 828  *
 829  * We defer creating the OBD object(s) until open, to keep the intent and
 830  * non-intent code paths similar, and also because we do not have the MDS
 831  * inode number before calling ll_create_node() (which is needed for LOV),
 832  * so we would need to do yet another RPC to the MDS to store the LOV EA
 833  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 834  * lmm_size in datalen (the MDS still has code which will handle that).
 835  *
 836  * If the create succeeds, we fill in the inode information
 837  * with d_instantiate().
 838  */
 839 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 840                         struct lookup_intent *it)
 841 {
 842         struct inode *inode;
 843         int rc = 0;
 844         ENTRY;
 845
 846         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 847                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 848                dir->i_generation, dir, LL_IT2STR(it));
 849
 850         rc = it_open_error(DISP_OPEN_CREATE, it);
 851         if (rc)
 852                 RETURN(rc);
 853
 854         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 855                                NULL, 0, mode, 0, it);
 856         if (IS_ERR(inode))
 857                 RETURN(PTR_ERR(inode));
 858
 859         if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
 860                 ll_i2info(inode)->lli_volatile = true;
 861
 862         d_instantiate(dentry, inode);
 863         RETURN(0);
 864 }
 865
 866 static void ll_update_times(struct ptlrpc_request *request,
 867                             struct inode *inode)
 868 {
 869         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 870                                                        &RMF_MDT_BODY);
 871
 872         LASSERT(body);
 873         if (body->valid & OBD_MD_FLMTIME &&
 874             body->mtime > LTIME_S(inode->i_mtime)) {
 875                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
 876                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 877                 LTIME_S(inode->i_mtime) = body->mtime;
 878         }
 879         if (body->valid & OBD_MD_FLCTIME &&
 880             body->ctime > LTIME_S(inode->i_ctime))
 881                 LTIME_S(inode->i_ctime) = body->ctime;
 882 }
 883
 884 static int ll_new_node(struct inode *dir, struct qstr *name,
 885                        const char *tgt, int mode, int rdev,
 886                        struct dentry *dchild, __u32 opc)
 887 {
 888         struct ptlrpc_request *request = NULL;
 889         struct md_op_data *op_data;
 890         struct inode *inode = NULL;
 891         struct ll_sb_info *sbi = ll_i2sbi(dir);
 892         int tgt_len = 0;
 893         int err;
 894
 895         ENTRY;
 896         if (unlikely(tgt != NULL))
 897                 tgt_len = strlen(tgt) + 1;
 898
 899         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 900                                      name->len, 0, opc, NULL);
 901         if (IS_ERR(op_data))
 902                 GOTO(err_exit, err = PTR_ERR(op_data));
 903
 904         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 905                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 906                         cfs_curproc_cap_pack(), rdev, &request);
 907         ll_finish_md_op_data(op_data);
 908         if (err)
 909                 GOTO(err_exit, err);
 910
 911         ll_update_times(request, dir);
 912
 913         if (dchild) {
 914                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 915                 if (err)
 916                      GOTO(err_exit, err);
 917
 918                 d_instantiate(dchild, inode);
 919         }
 920         EXIT;
 921 err_exit:
 922         ptlrpc_req_finished(request);
 923
 924         return err;
 925 }
 926
 927 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 928                             unsigned rdev, struct dentry *dchild)
 929 {
 930         int err;
 931         ENTRY;
 932
 933         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 934                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 935                mode, rdev);
 936
 937         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 938                 mode &= ~cfs_curproc_umask();
 939
 940         switch (mode & S_IFMT) {
 941         case 0:
 942                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 943         case S_IFREG:
 944         case S_IFCHR:
 945         case S_IFBLK:
 946         case S_IFIFO:
 947         case S_IFSOCK:
 948                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 949                                   LUSTRE_OPC_MKNOD);
 950                 break;
 951         case S_IFDIR:
 952                 err = -EPERM;
 953                 break;
 954         default:
 955                 err = -EINVAL;
 956         }
 957
 958         if (!err)
 959                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 960
 961         RETURN(err);
 962 }
 963
 964 #ifdef HAVE_IOP_ATOMIC_OPEN
 965 /*
 966  * Plain create. Intent create is handled in atomic_open.
 967  */
 968 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 969                         umode_t mode, bool want_excl)
 970 {
 971         int rc;
 972
 973         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
 974                            "flags=%u, excl=%d\n",
 975                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 976                dir->i_generation, dir, mode, want_excl);
 977
 978         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 979
 980         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 981
 982         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
 983                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
 984
 985         return rc;
 986 }
 987 #else /* !HAVE_IOP_ATOMIC_OPEN */
 988 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 989                         ll_umode_t mode, struct nameidata *nd)
 990 {
 991         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 992         int rc;
 993
 994         if (!it)
 995                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 996
 997         ll_d2d(dentry)->lld_it = NULL;
 998
 999         /* Was there an error? Propagate it! */
1000         if (it->d.lustre.it_status) {
1001                 rc = it->d.lustre.it_status;
1002                 goto out;
1003         }
1004
1005         rc = ll_create_it(dir, dentry, mode, it);
1006         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1007                 struct file *filp;
1008
1009                 nd->intent.open.file->private_data = it;
1010                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1011                 if (IS_ERR(filp))
1012                         rc = PTR_ERR(filp);
1013         }
1014
1015 out:
1016         ll_intent_release(it);
1017         OBD_FREE(it, sizeof(*it));
1018
1019         if (!rc)
1020                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1021
1022         return rc;
1023 }
1024 #endif /* HAVE_IOP_ATOMIC_OPEN */
1025
1026 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1027                               const char *tgt, struct dentry *dchild)
1028 {
1029         int err;
1030         ENTRY;
1031
1032         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1033                name->len, name->name, dir->i_ino, dir->i_generation,
1034                dir, 3000, tgt);
1035
1036         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1037                           0, dchild, LUSTRE_OPC_SYMLINK);
1038
1039         if (!err)
1040                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1041
1042         RETURN(err);
1043 }
1044
1045 static int ll_link_generic(struct inode *src,  struct inode *dir,
1046                            struct qstr *name, struct dentry *dchild)
1047 {
1048         struct ll_sb_info *sbi = ll_i2sbi(dir);
1049         struct ptlrpc_request *request = NULL;
1050         struct md_op_data *op_data;
1051         int err;
1052
1053         ENTRY;
1054         CDEBUG(D_VFSTRACE,
1055                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1056                src->i_ino, src->i_generation, src, dir->i_ino,
1057                dir->i_generation, dir, name->len, name->name);
1058
1059         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1060                                      0, LUSTRE_OPC_ANY, NULL);
1061         if (IS_ERR(op_data))
1062                 RETURN(PTR_ERR(op_data));
1063
1064         err = md_link(sbi->ll_md_exp, op_data, &request);
1065         ll_finish_md_op_data(op_data);
1066         if (err)
1067                 GOTO(out, err);
1068
1069         ll_update_times(request, dir);
1070         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1071         EXIT;
1072 out:
1073         ptlrpc_req_finished(request);
1074         RETURN(err);
1075 }
1076
1077 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1078                             int mode, struct dentry *dchild)
1079
1080 {
1081         int err;
1082         ENTRY;
1083
1084         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1085                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1086
1087         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1088                 mode &= ~cfs_curproc_umask();
1089         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1090         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1091
1092         if (!err)
1093                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1094
1095         RETURN(err);
1096 }
1097
1098 /* Try to find the child dentry by its name.
1099    If found, put the result fid into @fid. */
1100 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1101                              struct lu_fid *fid)
1102 {
1103         struct dentry *parent, *child;
1104
1105         parent = ll_d_hlist_entry(dir->i_dentry, struct dentry, d_alias);
1106         child = d_lookup(parent, name);
1107         if (child) {
1108                 if (child->d_inode)
1109                         *fid = *ll_inode2fid(child->d_inode);
1110                 dput(child);
1111         }
1112 }
1113
1114 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1115                             struct dentry *dchild, struct qstr *name)
1116 {
1117         struct ptlrpc_request *request = NULL;
1118         struct md_op_data *op_data;
1119         int rc;
1120         ENTRY;
1121
1122         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1123                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1124
1125         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1126                 RETURN(-EBUSY);
1127
1128         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1129                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1130         if (IS_ERR(op_data))
1131                 RETURN(PTR_ERR(op_data));
1132
1133         ll_get_child_fid(dir, name, &op_data->op_fid3);
1134         op_data->op_fid2 = op_data->op_fid3;
1135         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1136         ll_finish_md_op_data(op_data);
1137         if (rc == 0) {
1138                 ll_update_times(request, dir);
1139                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1140         }
1141
1142         ptlrpc_req_finished(request);
1143         RETURN(rc);
1144 }
1145
1146 /**
1147  * Remove dir entry
1148  **/
1149 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1150 {
1151         struct ptlrpc_request *request = NULL;
1152         struct md_op_data *op_data;
1153         int rc;
1154         ENTRY;
1155
1156         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1157                namelen, name, dir->i_ino, dir->i_generation, dir);
1158
1159         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1160                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1161         if (IS_ERR(op_data))
1162                 RETURN(PTR_ERR(op_data));
1163         op_data->op_cli_flags |= CLI_RM_ENTRY;
1164         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1165         ll_finish_md_op_data(op_data);
1166         if (rc == 0) {
1167                 ll_update_times(request, dir);
1168                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1169         }
1170
1171         ptlrpc_req_finished(request);
1172         RETURN(rc);
1173 }
1174
1175 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1176 {
1177         struct mdt_body *body;
1178         struct lov_mds_md *eadata;
1179         struct lov_stripe_md *lsm = NULL;
1180         struct obd_trans_info oti = { 0 };
1181         struct obdo *oa;
1182         struct obd_capa *oc = NULL;
1183         int rc;
1184         ENTRY;
1185
1186         /* req is swabbed so this is safe */
1187         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1188         if (!(body->valid & OBD_MD_FLEASIZE))
1189                 RETURN(0);
1190
1191         if (body->eadatasize == 0) {
1192                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1193                 GOTO(out, rc = -EPROTO);
1194         }
1195
1196         /* The MDS sent back the EA because we unlinked the last reference
1197          * to this file. Use this EA to unlink the objects on the OST.
1198          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1199          * check it is complete and sensible. */
1200         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1201                                               body->eadatasize);
1202         LASSERT(eadata != NULL);
1203
1204         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1205         if (rc < 0) {
1206                 CERROR("obd_unpackmd: %d\n", rc);
1207                 GOTO(out, rc);
1208         }
1209         LASSERT(rc >= sizeof(*lsm));
1210
1211         OBDO_ALLOC(oa);
1212         if (oa == NULL)
1213                 GOTO(out_free_memmd, rc = -ENOMEM);
1214
1215         oa->o_id = lsm->lsm_object_id;
1216         oa->o_seq = lsm->lsm_object_seq;
1217         oa->o_mode = body->mode & S_IFMT;
1218         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1219
1220         if (body->valid & OBD_MD_FLCOOKIE) {
1221                 oa->o_valid |= OBD_MD_FLCOOKIE;
1222                 oti.oti_logcookies =
1223                         req_capsule_server_sized_get(&request->rq_pill,
1224                                                      &RMF_LOGCOOKIES,
1225                                                    sizeof(struct llog_cookie) *
1226                                                      lsm->lsm_stripe_count);
1227                 if (oti.oti_logcookies == NULL) {
1228                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1229                         body->valid &= ~OBD_MD_FLCOOKIE;
1230                 }
1231         }
1232
1233         if (body->valid & OBD_MD_FLOSSCAPA) {
1234                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1235                 if (rc)
1236                         GOTO(out_free_memmd, rc);
1237         }
1238
1239         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1240                          ll_i2mdexp(dir), oc);
1241         capa_put(oc);
1242         if (rc)
1243                 CERROR("obd destroy objid "LPX64" error %d\n",
1244                        lsm->lsm_object_id, rc);
1245  out_free_memmd:
1246         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1247         OBDO_FREE(oa);
1248  out:
1249         return rc;
1250 }
1251
1252 /* ll_unlink_generic() doesn't update the inode with the new link count.
1253  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1254  * is any lock existing. They will recycle dentries and inodes based upon locks
1255  * too. b=20433 */
1256 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1257                              struct dentry *dchild, struct qstr *name)
1258 {
1259         struct ptlrpc_request *request = NULL;
1260         struct md_op_data *op_data;
1261         int rc;
1262         ENTRY;
1263         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1264                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1265
1266         /*
1267          * XXX: unlink bind mountpoint maybe call to here,
1268          * just check it as vfs_unlink does.
1269          */
1270         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1271                 RETURN(-EBUSY);
1272
1273         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1274                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1275         if (IS_ERR(op_data))
1276                 RETURN(PTR_ERR(op_data));
1277
1278         ll_get_child_fid(dir, name, &op_data->op_fid3);
1279         op_data->op_fid2 = op_data->op_fid3;
1280         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1281         ll_finish_md_op_data(op_data);
1282         if (rc)
1283                 GOTO(out, rc);
1284
1285         ll_update_times(request, dir);
1286         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1287
1288         rc = ll_objects_destroy(request, dir);
1289  out:
1290         ptlrpc_req_finished(request);
1291         RETURN(rc);
1292 }
1293
1294 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1295                              struct dentry *src_dchild, struct qstr *src_name,
1296                              struct inode *tgt, struct dentry *tgt_dparent,
1297                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1298 {
1299         struct ptlrpc_request *request = NULL;
1300         struct ll_sb_info *sbi = ll_i2sbi(src);
1301         struct md_op_data *op_data;
1302         int err;
1303         ENTRY;
1304         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1305                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1306                src->i_ino, src->i_generation, src, tgt_name->len,
1307                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1308
1309         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1310             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1311                 RETURN(-EBUSY);
1312
1313         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1314                                      LUSTRE_OPC_ANY, NULL);
1315         if (IS_ERR(op_data))
1316                 RETURN(PTR_ERR(op_data));
1317
1318         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1319         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1320         err = md_rename(sbi->ll_md_exp, op_data,
1321                         src_name->name, src_name->len,
1322                         tgt_name->name, tgt_name->len, &request);
1323         ll_finish_md_op_data(op_data);
1324         if (!err) {
1325                 ll_update_times(request, src);
1326                 ll_update_times(request, tgt);
1327                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1328                 err = ll_objects_destroy(request, src);
1329         }
1330
1331         ptlrpc_req_finished(request);
1332
1333         RETURN(err);
1334 }
1335
1336 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1337                     dev_t rdev)
1338 {
1339         return ll_mknod_generic(dir, &dchild->d_name, mode,
1340                                 old_encode_dev(rdev), dchild);
1341 }
1342
1343 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1344 {
1345         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1346 }
1347
1348 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1349 {
1350         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1351 }
1352
1353 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1354 {
1355         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1356 }
1357
1358 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1359                       const char *oldname)
1360 {
1361         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1362 }
1363
1364 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1365                    struct dentry *new_dentry)
1366 {
1367         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1368                                new_dentry);
1369 }
1370
1371 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1372                      struct inode *new_dir, struct dentry *new_dentry)
1373 {
1374         int err;
1375         err = ll_rename_generic(old_dir, NULL,
1376                                  old_dentry, &old_dentry->d_name,
1377                                  new_dir, NULL, new_dentry,
1378                                  &new_dentry->d_name);
1379         if (!err) {
1380 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1381                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1382 #endif
1383                         d_move(old_dentry, new_dentry);
1384         }
1385         return err;
1386 }
1387
1388 struct inode_operations ll_dir_inode_operations = {
1389         .mknod              = ll_mknod,
1390 #ifdef HAVE_IOP_ATOMIC_OPEN
1391         .atomic_open        = ll_atomic_open,
1392 #endif
1393         .lookup             = ll_lookup_nd,
1394         .create             = ll_create_nd,
1395         /* We need all these non-raw things for NFSD, to not patch it. */
1396         .unlink             = ll_unlink,
1397         .mkdir              = ll_mkdir,
1398         .rmdir              = ll_rmdir,
1399         .symlink            = ll_symlink,
1400         .link               = ll_link,
1401         .rename             = ll_rename,
1402         .setattr            = ll_setattr,
1403         .getattr            = ll_getattr,
1404         .permission         = ll_inode_permission,
1405         .setxattr           = ll_setxattr,
1406         .getxattr           = ll_getxattr,
1407         .listxattr          = ll_listxattr,
1408         .removexattr        = ll_removexattr,
1409 #ifdef HAVE_IOP_GET_ACL
1410         .get_acl            = ll_get_acl,
1411 #endif
1412 };
1413
1414 struct inode_operations ll_special_inode_operations = {
1415         .setattr        = ll_setattr,
1416         .getattr        = ll_getattr,
1417         .permission     = ll_inode_permission,
1418         .setxattr       = ll_setxattr,
1419         .getxattr       = ll_getxattr,
1420         .listxattr      = ll_listxattr,
1421         .removexattr    = ll_removexattr,
1422 #ifdef HAVE_IOP_GET_ACL
1423         .get_acl            = ll_get_acl,
1424 #endif
1425 };