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