lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_ver.h>
  52 #include "llite_internal.h"
  53
  54 static int ll_create_it(struct inode *, struct dentry *,
  55                         int, struct lookup_intent *);
  56
  57 /*
  58  * Check if we have something mounted at the named dchild.
  59  * In such a case there would always be dentry present.
  60  */
  61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
  62                            struct qstr *name)
  63 {
  64         int mounted = 0;
  65
  66         if (unlikely(dchild)) {
  67                 mounted = d_mountpoint(dchild);
  68         } else if (dparent) {
  69                 dchild = d_lookup(dparent, name);
  70                 if (dchild) {
  71                         mounted = d_mountpoint(dchild);
  72                         dput(dchild);
  73                 }
  74         }
  75         return mounted;
  76 }
  77
  78 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  79 static int ll_test_inode(struct inode *inode, void *opaque)
  80 {
  81         struct ll_inode_info *lli = ll_i2info(inode);
  82         struct lustre_md     *md = opaque;
  83
  84         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  85                 CERROR("MDS body missing FID\n");
  86                 return 0;
  87         }
  88
  89         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
  90                 return 0;
  91
  92         return 1;
  93 }
  94
  95 static int ll_set_inode(struct inode *inode, void *opaque)
  96 {
  97         struct ll_inode_info *lli = ll_i2info(inode);
  98         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  99
 100         if (unlikely(!(body->valid & OBD_MD_FLID))) {
 101                 CERROR("MDS body missing FID\n");
 102                 return -EINVAL;
 103         }
 104
 105         lli->lli_fid = body->fid1;
 106         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
 107                 CERROR("Can not initialize inode "DFID" without object type: "
 108                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
 109                 return -EINVAL;
 110         }
 111
 112         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
 113         if (unlikely(inode->i_mode == 0)) {
 114                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 115                 return -EINVAL;
 116         }
 117
 118         ll_lli_init(lli);
 119
 120         return 0;
 121 }
 122
 123
 124 /*
 125  * Get an inode by inode number (already instantiated by the intent lookup).
 126  * Returns inode or NULL
 127  */
 128 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 129                       struct lustre_md *md)
 130 {
 131         struct inode         *inode;
 132         ENTRY;
 133
 134         LASSERT(hash != 0);
 135         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 136
 137         if (inode) {
 138                 if (inode->i_state & I_NEW) {
 139                         int rc = 0;
 140
 141                         ll_read_inode2(inode, md);
 142                         if (S_ISREG(inode->i_mode) &&
 143                             ll_i2info(inode)->lli_clob == NULL) {
 144                                 CDEBUG(D_INODE,
 145                                         "%s: apply lsm %p to inode "DFID".\n",
 146                                         ll_get_fsname(sb, NULL, 0), md->lsm,
 147                                         PFID(ll_inode2fid(inode)));
 148                                 rc = cl_file_inode_init(inode, md);
 149                         }
 150                         if (rc != 0) {
 151                                 make_bad_inode(inode);
 152                                 unlock_new_inode(inode);
 153                                 iput(inode);
 154                                 inode = ERR_PTR(rc);
 155                         } else
 156                                 unlock_new_inode(inode);
 157                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 158                         ll_update_inode(inode, md);
 159                         CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
 160                                PFID(&md->body->fid1), inode);
 161         }
 162         RETURN(inode);
 163 }
 164
 165 static void ll_invalidate_negative_children(struct inode *dir)
 166 {
 167         struct dentry *dentry, *tmp_subdir;
 168         DECLARE_LL_D_HLIST_NODE_PTR(p);
 169
 170         ll_lock_dcache(dir);
 171         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
 172                 spin_lock(&dentry->d_lock);
 173                 if (!list_empty(&dentry->d_subdirs)) {
 174                         struct dentry *child;
 175
 176                         list_for_each_entry_safe(child, tmp_subdir,
 177                                                  &dentry->d_subdirs,
 178                                                  d_u.d_child) {
 179                                 if (child->d_inode == NULL)
 180                                         d_lustre_invalidate(child, 1);
 181                         }
 182                 }
 183                 spin_unlock(&dentry->d_lock);
 184         }
 185         ll_unlock_dcache(dir);
 186 }
 187
 188 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 189 {
 190         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 191 }
 192
 193 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 194                        void *data, int flag)
 195 {
 196         struct lustre_handle lockh;
 197         int rc;
 198         ENTRY;
 199
 200         switch (flag) {
 201         case LDLM_CB_BLOCKING:
 202                 ldlm_lock2handle(lock, &lockh);
 203                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 204                 if (rc < 0) {
 205                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 206                         RETURN(rc);
 207                 }
 208                 break;
 209         case LDLM_CB_CANCELING: {
 210                 struct inode *inode = ll_inode_from_resource_lock(lock);
 211                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 212
 213                 /* Inode is set to lock->l_resource->lr_lvb_inode
 214                  * for mdc - bug 24555 */
 215                 LASSERT(lock->l_ast_data == NULL);
 216
 217                 if (inode == NULL)
 218                         break;
 219
 220                 /* Invalidate all dentries associated with this inode */
 221                 LASSERT(ldlm_is_canceling(lock));
 222
 223                 if (!fid_res_name_eq(ll_inode2fid(inode),
 224                                      &lock->l_resource->lr_name)) {
 225                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 226                                    PFID(ll_inode2fid(inode)), inode);
 227                         LBUG();
 228                 }
 229
 230                 if (bits & MDS_INODELOCK_XATTR) {
 231                         ll_xattr_cache_destroy(inode);
 232                         bits &= ~MDS_INODELOCK_XATTR;
 233                 }
 234
 235                 /* For OPEN locks we differentiate between lock modes
 236                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 237                 if (bits & MDS_INODELOCK_OPEN)
 238                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 239
 240                 if (bits & MDS_INODELOCK_OPEN) {
 241                         fmode_t fmode;
 242
 243                         switch (lock->l_req_mode) {
 244                         case LCK_CW:
 245                                 fmode = FMODE_WRITE;
 246                                 break;
 247                         case LCK_PR:
 248                                 fmode = FMODE_EXEC;
 249                                 break;
 250                         case LCK_CR:
 251                                 fmode = FMODE_READ;
 252                                 break;
 253                         default:
 254                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 255                                 LBUG();
 256                         }
 257
 258                         ll_md_real_close(inode, fmode);
 259
 260                         bits &= ~MDS_INODELOCK_OPEN;
 261                 }
 262
 263                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 264                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 265                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 266
 267                 if (bits & MDS_INODELOCK_LAYOUT) {
 268                         struct cl_object_conf conf = {
 269                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 270                                 .coc_inode = inode,
 271                         };
 272
 273                         rc = ll_layout_conf(inode, &conf);
 274                         if (rc < 0)
 275                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 276                                        DFID": rc = %d\n",
 277                                        PFID(ll_inode2fid(inode)), rc);
 278                 }
 279
 280                 if (bits & MDS_INODELOCK_UPDATE) {
 281                         struct ll_inode_info *lli = ll_i2info(inode);
 282
 283                         spin_lock(&lli->lli_lock);
 284                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 285                         spin_unlock(&lli->lli_lock);
 286                 }
 287
 288                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 289                         struct ll_inode_info *lli = ll_i2info(inode);
 290
 291                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 292                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 293                                lli, PFID(&lli->lli_pfid));
 294                         truncate_inode_pages(inode->i_mapping, 0);
 295
 296                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 297                                 struct inode *master_inode = NULL;
 298                                 unsigned long hash;
 299
 300                                 /* This is slave inode, since all of the child
 301                                  * dentry is connected on the master inode, so
 302                                  * we have to invalidate the negative children
 303                                  * on master inode */
 304                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 305                                        PFID(ll_inode2fid(inode)),
 306                                        PFID(&lli->lli_pfid));
 307
 308                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 309                                         ll_need_32bit_api(ll_i2sbi(inode)));
 310
 311                                 master_inode = ilookup5(inode->i_sb, hash,
 312                                                         ll_test_inode_by_fid,
 313                                                         (void *)&lli->lli_pfid);
 314                                 if (master_inode != NULL &&
 315                                         !IS_ERR(master_inode)) {
 316                                         ll_invalidate_negative_children(
 317                                                                 master_inode);
 318                                         iput(master_inode);
 319                                 }
 320                         } else {
 321                                 ll_invalidate_negative_children(inode);
 322                         }
 323                 }
 324
 325                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 326                     inode->i_sb->s_root != NULL &&
 327                     inode != inode->i_sb->s_root->d_inode)
 328                         ll_invalidate_aliases(inode);
 329
 330                 iput(inode);
 331                 break;
 332         }
 333         default:
 334                 LBUG();
 335         }
 336
 337         RETURN(0);
 338 }
 339
 340 __u32 ll_i2suppgid(struct inode *i)
 341 {
 342         if (in_group_p(i->i_gid))
 343                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 344         else
 345                 return (__u32) __kgid_val(INVALID_GID);
 346 }
 347
 348 /* Pack the required supplementary groups into the supplied groups array.
 349  * If we don't need to use the groups from the target inode(s) then we
 350  * instead pack one or more groups from the user's supplementary group
 351  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 352 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 353 {
 354         LASSERT(i1 != NULL);
 355         LASSERT(suppgids != NULL);
 356
 357         suppgids[0] = ll_i2suppgid(i1);
 358
 359         if (i2)
 360                 suppgids[1] = ll_i2suppgid(i2);
 361         else
 362                 suppgids[1] = -1;
 363 }
 364
 365 /*
 366  * try to reuse three types of dentry:
 367  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 368  *    by concurrent .revalidate).
 369  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 370  *    be cleared by others calling d_lustre_revalidate).
 371  * 3. DISCONNECTED alias.
 372  */
 373 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 374 {
 375         struct dentry *alias, *discon_alias, *invalid_alias;
 376         DECLARE_LL_D_HLIST_NODE_PTR(p);
 377
 378         if (ll_d_hlist_empty(&inode->i_dentry))
 379                 return NULL;
 380
 381         discon_alias = invalid_alias = NULL;
 382
 383         ll_lock_dcache(inode);
 384         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 385                 LASSERT(alias != dentry);
 386
 387                 spin_lock(&alias->d_lock);
 388                 if (alias->d_flags & DCACHE_DISCONNECTED)
 389                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 390                         discon_alias = alias;
 391                 else if (alias->d_parent == dentry->d_parent             &&
 392                          alias->d_name.hash == dentry->d_name.hash       &&
 393                          alias->d_name.len == dentry->d_name.len         &&
 394                          memcmp(alias->d_name.name, dentry->d_name.name,
 395                                 dentry->d_name.len) == 0)
 396                         invalid_alias = alias;
 397                 spin_unlock(&alias->d_lock);
 398
 399                 if (invalid_alias)
 400                         break;
 401         }
 402         alias = invalid_alias ?: discon_alias ?: NULL;
 403         if (alias) {
 404                 spin_lock(&alias->d_lock);
 405                 dget_dlock(alias);
 406                 spin_unlock(&alias->d_lock);
 407         }
 408         ll_unlock_dcache(inode);
 409
 410         return alias;
 411 }
 412
 413 /*
 414  * Similar to d_splice_alias(), but lustre treats invalid alias
 415  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 416  */
 417 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 418 {
 419         struct dentry *new;
 420         int rc;
 421
 422         if (inode) {
 423                 new = ll_find_alias(inode, de);
 424                 if (new) {
 425                         rc = ll_d_init(new);
 426                         if (rc < 0) {
 427                                 dput(new);
 428                                 return ERR_PTR(rc);
 429                         }
 430                         d_move(new, de);
 431                         iput(inode);
 432                         CDEBUG(D_DENTRY,
 433                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 434                               new, new->d_inode, d_count(new), new->d_flags);
 435                         return new;
 436                 }
 437         }
 438         rc = ll_d_init(de);
 439         if (rc < 0)
 440                 return ERR_PTR(rc);
 441         d_add(de, inode);
 442         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 443                de, de->d_inode, d_count(de), de->d_flags);
 444         return de;
 445 }
 446
 447 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 448                                struct lookup_intent *it,
 449                                struct inode *parent, struct dentry **de)
 450 {
 451         struct inode             *inode = NULL;
 452         __u64                     bits = 0;
 453         int                       rc;
 454         ENTRY;
 455
 456         /* NB 1 request reference will be taken away by ll_intent_lock()
 457          * when I return */
 458         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 459                it->d.lustre.it_disposition);
 460         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 461                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 462                 if (rc)
 463                         RETURN(rc);
 464
 465                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 466
 467                 /* We used to query real size from OSTs here, but actually
 468                    this is not needed. For stat() calls size would be updated
 469                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 470                    2.4 and
 471                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 472                    Everybody else who needs correct file size would call
 473                    ll_glimpse_size or some equivalent themselves anyway.
 474                    Also see bug 7198. */
 475         }
 476
 477         /* Only hash *de if it is unhashed (new dentry).
 478          * Atoimc_open may passin hashed dentries for open.
 479          */
 480         if (d_unhashed(*de)) {
 481                 struct dentry *alias;
 482
 483                 alias = ll_splice_alias(inode, *de);
 484                 if (IS_ERR(alias))
 485                         RETURN(PTR_ERR(alias));
 486                 *de = alias;
 487         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 488                    !it_disposition(it, DISP_OPEN_CREATE)) {
 489                 /* With DISP_OPEN_CREATE dentry will
 490                    instantiated in ll_create_it. */
 491                 LASSERT((*de)->d_inode == NULL);
 492                 d_instantiate(*de, inode);
 493         }
 494
 495         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 496                 /* we have lookup look - unhide dentry */
 497                 if (bits & MDS_INODELOCK_LOOKUP)
 498                         d_lustre_revalidate(*de);
 499         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 500                 /* If file created on server, don't depend on parent UPDATE
 501                  * lock to unhide it. It is left hidden and next lookup can
 502                  * find it in ll_splice_alias.
 503                  */
 504                 /* Check that parent has UPDATE lock. */
 505                 struct lookup_intent parent_it = {
 506                                         .it_op = IT_GETATTR,
 507                                         .d.lustre.it_lock_handle = 0 };
 508
 509                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 510                                        &ll_i2info(parent)->lli_fid, NULL)) {
 511                         d_lustre_revalidate(*de);
 512                         ll_intent_release(&parent_it);
 513                 }
 514         }
 515
 516         RETURN(0);
 517 }
 518
 519 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 520                                    struct lookup_intent *it, int lookup_flags)
 521 {
 522         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 523         struct dentry *save = dentry, *retval;
 524         struct ptlrpc_request *req = NULL;
 525         struct md_op_data *op_data;
 526         __u32 opc;
 527         int rc;
 528         ENTRY;
 529
 530         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 531                 RETURN(ERR_PTR(-ENAMETOOLONG));
 532
 533         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 534                dentry->d_name.len, dentry->d_name.name,
 535                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 536
 537         if (d_mountpoint(dentry))
 538                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 539
 540         if (it == NULL || it->it_op == IT_GETXATTR)
 541                 it = &lookup_it;
 542
 543         if (it->it_op == IT_GETATTR) {
 544                 rc = ll_statahead_enter(parent, &dentry, 0);
 545                 if (rc == 1) {
 546                         if (dentry == save)
 547                                 GOTO(out, retval = NULL);
 548                         GOTO(out, retval = dentry);
 549                 }
 550         }
 551
 552         if (it->it_op & IT_CREAT)
 553                 opc = LUSTRE_OPC_CREATE;
 554         else
 555                 opc = LUSTRE_OPC_ANY;
 556
 557         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 558                                      dentry->d_name.len, lookup_flags, opc,
 559                                      NULL);
 560         if (IS_ERR(op_data))
 561                 RETURN((void *)op_data);
 562
 563         /* enforce umask if acl disabled or MDS doesn't support umask */
 564         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 565                 it->it_create_mode &= ~current_umask();
 566
 567         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 568                             lookup_flags, &req, ll_md_blocking_ast, 0);
 569         ll_finish_md_op_data(op_data);
 570         if (rc < 0)
 571                 GOTO(out, retval = ERR_PTR(rc));
 572
 573         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 574         if (rc != 0) {
 575                 ll_intent_release(it);
 576                 GOTO(out, retval = ERR_PTR(rc));
 577         }
 578
 579         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 580             !S_ISREG(dentry->d_inode->i_mode) &&
 581             !S_ISDIR(dentry->d_inode->i_mode)) {
 582                 ll_release_openhandle(dentry, it);
 583         }
 584         ll_lookup_finish_locks(it, dentry);
 585
 586         if (dentry == save)
 587                 GOTO(out, retval = NULL);
 588         else
 589                 GOTO(out, retval = dentry);
 590  out:
 591         if (req)
 592                 ptlrpc_req_finished(req);
 593         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 594                 ll_statahead_mark(parent, dentry);
 595         return retval;
 596 }
 597
 598 #ifdef HAVE_IOP_ATOMIC_OPEN
 599 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 600                                    unsigned int flags)
 601 {
 602         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 603         struct dentry *de;
 604
 605         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 606                dentry->d_name.len, dentry->d_name.name,
 607                PFID(ll_inode2fid(parent)), parent, flags);
 608
 609         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 610         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 611                 return NULL;
 612
 613         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 614                 itp = NULL;
 615         else
 616                 itp = &it;
 617         de = ll_lookup_it(parent, dentry, itp, 0);
 618
 619         if (itp != NULL)
 620                 ll_intent_release(itp);
 621
 622         return de;
 623 }
 624
 625 /*
 626  * For cached negative dentry and new dentry, handle lookup/create/open
 627  * together.
 628  */
 629 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 630                           struct file *file, unsigned open_flags,
 631                           umode_t mode, int *opened)
 632 {
 633         struct lookup_intent *it;
 634         struct dentry *de;
 635         long long lookup_flags = LOOKUP_OPEN;
 636         int rc = 0;
 637         ENTRY;
 638
 639         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 640                            "open_flags %x, mode %x opened %d\n",
 641                dentry->d_name.len, dentry->d_name.name,
 642                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 643
 644         OBD_ALLOC(it, sizeof(*it));
 645         if (!it)
 646                 RETURN(-ENOMEM);
 647
 648         it->it_op = IT_OPEN;
 649         if (open_flags & O_CREAT) {
 650                 it->it_op |= IT_CREAT;
 651                 lookup_flags |= LOOKUP_CREATE;
 652         }
 653         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 654         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 655
 656         /* Dentry added to dcache tree in ll_lookup_it */
 657         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 658         if (IS_ERR(de))
 659                 rc = PTR_ERR(de);
 660         else if (de != NULL)
 661                 dentry = de;
 662
 663         if (!rc) {
 664                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 665                         /* Dentry instantiated in ll_create_it. */
 666                         rc = ll_create_it(dir, dentry, mode, it);
 667                         if (rc) {
 668                                 /* We dget in ll_splice_alias. */
 669                                 if (de != NULL)
 670                                         dput(de);
 671                                 goto out_release;
 672                         }
 673
 674                         *opened |= FILE_CREATED;
 675                 }
 676                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 677                         /* Open dentry. */
 678                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 679                                 /* We cannot call open here as it would
 680                                  * deadlock.
 681                                  */
 682                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 683                                         ptlrpc_req_finished(
 684                                                        (struct ptlrpc_request *)
 685                                                           it->d.lustre.it_data);
 686                                 rc = finish_no_open(file, de);
 687                         } else {
 688                                 file->private_data = it;
 689                                 rc = finish_open(file, dentry, NULL, opened);
 690                                 /* We dget in ll_splice_alias. finish_open takes
 691                                  * care of dget for fd open.
 692                                  */
 693                                 if (de != NULL)
 694                                         dput(de);
 695                         }
 696                 } else {
 697                         rc = finish_no_open(file, de);
 698                 }
 699         }
 700
 701 out_release:
 702         ll_intent_release(it);
 703         OBD_FREE(it, sizeof(*it));
 704
 705         RETURN(rc);
 706 }
 707
 708 #else /* !HAVE_IOP_ATOMIC_OPEN */
 709 static struct lookup_intent *
 710 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 711 {
 712         struct lookup_intent *it;
 713
 714         OBD_ALLOC(it, sizeof(*it));
 715         if (!it)
 716                 return ERR_PTR(-ENOMEM);
 717
 718         if (lookup_flags & LOOKUP_OPEN) {
 719                 it->it_op = IT_OPEN;
 720                 if (lookup_flags & LOOKUP_CREATE)
 721                         it->it_op |= IT_CREAT;
 722                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 723                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 724         } else {
 725                 it->it_op = IT_GETATTR;
 726         }
 727
 728         return it;
 729 }
 730
 731 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 732                                    struct nameidata *nd)
 733 {
 734         struct dentry *de;
 735         ENTRY;
 736
 737         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 738                 struct lookup_intent *it;
 739
 740                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 741                         it = ll_d2d(dentry)->lld_it;
 742                         ll_d2d(dentry)->lld_it = NULL;
 743                 } else {
 744                         if ((nd->flags & LOOKUP_CREATE) &&
 745                             !(nd->flags & LOOKUP_OPEN))
 746                                 RETURN(NULL);
 747
 748                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 749                         if (IS_ERR(it))
 750                                 RETURN((struct dentry *)it);
 751                 }
 752
 753                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 754                 if (de)
 755                         dentry = de;
 756                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 757                         if (dentry->d_inode &&
 758                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 759                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 760                                         // We cannot call open here as it would
 761                                         // deadlock.
 762                                         ptlrpc_req_finished(
 763                                                        (struct ptlrpc_request *)
 764                                                           it->d.lustre.it_data);
 765                                 } else {
 766                                         struct file *filp;
 767
 768                                         nd->intent.open.file->private_data = it;
 769                                         filp = lookup_instantiate_filp(nd,
 770                                                                        dentry,
 771                                                                        NULL);
 772                                         if (IS_ERR(filp)) {
 773                                                 if (de)
 774                                                         dput(de);
 775                                                 de = (struct dentry *)filp;
 776                                         }
 777                                 }
 778                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 779                                 // XXX This can only reliably work on assumption
 780                                 // that there are NO hashed negative dentries.
 781                                 ll_d2d(dentry)->lld_it = it;
 782                                 it = NULL; /* Will be freed in ll_create_nd */
 783                                 /* We absolutely depend on ll_create_nd to be
 784                                  * called to not leak this intent and possible
 785                                  * data attached to it */
 786                         }
 787                 }
 788
 789                 if (it) {
 790                         ll_intent_release(it);
 791                         OBD_FREE(it, sizeof(*it));
 792                 }
 793         } else {
 794                 de = ll_lookup_it(parent, dentry, NULL, 0);
 795         }
 796
 797         RETURN(de);
 798 }
 799 #endif /* HAVE_IOP_ATOMIC_OPEN */
 800
 801 /* We depend on "mode" being set with the proper file type/umask by now */
 802 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 803 {
 804         struct inode *inode = NULL;
 805         struct ptlrpc_request *request = NULL;
 806         struct ll_sb_info *sbi = ll_i2sbi(dir);
 807         int rc;
 808         ENTRY;
 809
 810         LASSERT(it && it->d.lustre.it_disposition);
 811
 812         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 813         request = it->d.lustre.it_data;
 814         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 815         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 816         if (rc)
 817                 GOTO(out, inode = ERR_PTR(rc));
 818
 819         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 820
 821         /* We asked for a lock on the directory, but were granted a
 822          * lock on the inode.  Since we finally have an inode pointer,
 823          * stuff it in the lock. */
 824         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 825                PFID(ll_inode2fid(inode)), inode);
 826         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 827         EXIT;
 828  out:
 829         ptlrpc_req_finished(request);
 830         return inode;
 831 }
 832
 833 /*
 834  * By the time this is called, we already have created the directory cache
 835  * entry for the new file, but it is so far negative - it has no inode.
 836  *
 837  * We defer creating the OBD object(s) until open, to keep the intent and
 838  * non-intent code paths similar, and also because we do not have the MDS
 839  * inode number before calling ll_create_node() (which is needed for LOV),
 840  * so we would need to do yet another RPC to the MDS to store the LOV EA
 841  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 842  * lmm_size in datalen (the MDS still has code which will handle that).
 843  *
 844  * If the create succeeds, we fill in the inode information
 845  * with d_instantiate().
 846  */
 847 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 848                         struct lookup_intent *it)
 849 {
 850         struct inode *inode;
 851         int rc = 0;
 852         ENTRY;
 853
 854         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 855                dentry->d_name.len, dentry->d_name.name,
 856                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 857
 858         rc = it_open_error(DISP_OPEN_CREATE, it);
 859         if (rc)
 860                 RETURN(rc);
 861
 862         inode = ll_create_node(dir, it);
 863         if (IS_ERR(inode))
 864                 RETURN(PTR_ERR(inode));
 865
 866         d_instantiate(dentry, inode);
 867         RETURN(0);
 868 }
 869
 870 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 871 {
 872         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 873                                                        &RMF_MDT_BODY);
 874
 875         LASSERT(body);
 876         if (body->valid & OBD_MD_FLMTIME &&
 877             body->mtime > LTIME_S(inode->i_mtime)) {
 878                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 879                                 "\n", PFID(ll_inode2fid(inode)),
 880                                 LTIME_S(inode->i_mtime), body->mtime);
 881                 LTIME_S(inode->i_mtime) = body->mtime;
 882         }
 883         if (body->valid & OBD_MD_FLCTIME &&
 884             body->ctime > LTIME_S(inode->i_ctime))
 885                 LTIME_S(inode->i_ctime) = body->ctime;
 886 }
 887
 888 static int ll_new_node(struct inode *dir, struct qstr *name,
 889                        const char *tgt, int mode, int rdev,
 890                        struct dentry *dchild, __u32 opc)
 891 {
 892         struct ptlrpc_request *request = NULL;
 893         struct md_op_data *op_data;
 894         struct inode *inode = NULL;
 895         struct ll_sb_info *sbi = ll_i2sbi(dir);
 896         int tgt_len = 0;
 897         int err;
 898
 899         ENTRY;
 900         if (unlikely(tgt != NULL))
 901                 tgt_len = strlen(tgt) + 1;
 902
 903         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 904                                      name->len, 0, opc, NULL);
 905         if (IS_ERR(op_data))
 906                 GOTO(err_exit, err = PTR_ERR(op_data));
 907
 908         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 909                         from_kuid(&init_user_ns, current_fsuid()),
 910                         from_kgid(&init_user_ns, current_fsgid()),
 911                         cfs_curproc_cap_pack(), rdev, &request);
 912         ll_finish_md_op_data(op_data);
 913         if (err)
 914                 GOTO(err_exit, err);
 915
 916         ll_update_times(request, dir);
 917
 918         if (dchild) {
 919                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 920                 if (err)
 921                      GOTO(err_exit, err);
 922
 923                 d_instantiate(dchild, inode);
 924         }
 925         EXIT;
 926 err_exit:
 927         ptlrpc_req_finished(request);
 928
 929         return err;
 930 }
 931
 932 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 933                             unsigned rdev, struct dentry *dchild)
 934 {
 935         int err;
 936         ENTRY;
 937
 938         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
 939                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
 940                mode, rdev);
 941
 942         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 943                 mode &= ~current_umask();
 944
 945         switch (mode & S_IFMT) {
 946         case 0:
 947                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 948         case S_IFREG:
 949         case S_IFCHR:
 950         case S_IFBLK:
 951         case S_IFIFO:
 952         case S_IFSOCK:
 953                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 954                                   LUSTRE_OPC_MKNOD);
 955                 break;
 956         case S_IFDIR:
 957                 err = -EPERM;
 958                 break;
 959         default:
 960                 err = -EINVAL;
 961         }
 962
 963         if (!err)
 964                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 965
 966         RETURN(err);
 967 }
 968
 969 #ifdef HAVE_IOP_ATOMIC_OPEN
 970 /*
 971  * Plain create. Intent create is handled in atomic_open.
 972  */
 973 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 974                         umode_t mode, bool want_excl)
 975 {
 976         int rc;
 977
 978         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
 979                            "flags=%u, excl=%d\n", dentry->d_name.len,
 980                dentry->d_name.name, PFID(ll_inode2fid(dir)),
 981                dir, mode, want_excl);
 982
 983         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 984
 985         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 986
 987         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
 988                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
 989
 990         return rc;
 991 }
 992 #else /* !HAVE_IOP_ATOMIC_OPEN */
 993 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 994                         ll_umode_t mode, struct nameidata *nd)
 995 {
 996         struct ll_dentry_data *lld = ll_d2d(dentry);
 997         struct lookup_intent *it = NULL;
 998         int rc;
 999
1000         if (lld != NULL)
1001                 it = lld->lld_it;
1002
1003         if (!it)
1004                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1005
1006         lld->lld_it = NULL;
1007
1008         /* Was there an error? Propagate it! */
1009         if (it->d.lustre.it_status) {
1010                 rc = it->d.lustre.it_status;
1011                 goto out;
1012         }
1013
1014         rc = ll_create_it(dir, dentry, mode, it);
1015         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1016                 struct file *filp;
1017
1018                 nd->intent.open.file->private_data = it;
1019                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1020                 if (IS_ERR(filp))
1021                         rc = PTR_ERR(filp);
1022         }
1023
1024 out:
1025         ll_intent_release(it);
1026         OBD_FREE(it, sizeof(*it));
1027
1028         if (!rc)
1029                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1030
1031         return rc;
1032 }
1033 #endif /* HAVE_IOP_ATOMIC_OPEN */
1034
1035 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1036                               const char *tgt, struct dentry *dchild)
1037 {
1038         int err;
1039         ENTRY;
1040
1041         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1042                name->len, name->name, PFID(ll_inode2fid(dir)),
1043                dir, 3000, tgt);
1044
1045         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1046                           0, dchild, LUSTRE_OPC_SYMLINK);
1047
1048         if (!err)
1049                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1050
1051         RETURN(err);
1052 }
1053
1054 static int ll_link_generic(struct inode *src,  struct inode *dir,
1055                            struct qstr *name, struct dentry *dchild)
1056 {
1057         struct ll_sb_info *sbi = ll_i2sbi(dir);
1058         struct ptlrpc_request *request = NULL;
1059         struct md_op_data *op_data;
1060         int err;
1061
1062         ENTRY;
1063         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1064                "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1065                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1066
1067         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1068                                      0, LUSTRE_OPC_ANY, NULL);
1069         if (IS_ERR(op_data))
1070                 RETURN(PTR_ERR(op_data));
1071
1072         err = md_link(sbi->ll_md_exp, op_data, &request);
1073         ll_finish_md_op_data(op_data);
1074         if (err)
1075                 GOTO(out, err);
1076
1077         ll_update_times(request, dir);
1078         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1079         EXIT;
1080 out:
1081         ptlrpc_req_finished(request);
1082         RETURN(err);
1083 }
1084
1085 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1086                             int mode, struct dentry *dchild)
1087
1088 {
1089         int err;
1090         ENTRY;
1091
1092         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1093                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1094
1095         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1096                 mode &= ~current_umask();
1097         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1098         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1099
1100         if (!err)
1101                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1102
1103         RETURN(err);
1104 }
1105
1106 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1107                             struct dentry *dchild, struct qstr *name)
1108 {
1109         struct ptlrpc_request *request = NULL;
1110         struct md_op_data *op_data;
1111         int rc;
1112         ENTRY;
1113
1114         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1115                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1116
1117         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1118                 RETURN(-EBUSY);
1119
1120         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1121                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1122         if (IS_ERR(op_data))
1123                 RETURN(PTR_ERR(op_data));
1124
1125         if (dchild != NULL && dchild->d_inode != NULL)
1126                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1127         op_data->op_fid2 = op_data->op_fid3;
1128         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1129         ll_finish_md_op_data(op_data);
1130         if (rc == 0) {
1131                 ll_update_times(request, dir);
1132                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1133         }
1134
1135         ptlrpc_req_finished(request);
1136         RETURN(rc);
1137 }
1138
1139 /**
1140  * Remove dir entry
1141  **/
1142 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1143 {
1144         struct ptlrpc_request *request = NULL;
1145         struct md_op_data *op_data;
1146         int rc;
1147         ENTRY;
1148
1149         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1150                namelen, name, PFID(ll_inode2fid(dir)), dir);
1151
1152         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1153                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1154         if (IS_ERR(op_data))
1155                 RETURN(PTR_ERR(op_data));
1156         op_data->op_cli_flags |= CLI_RM_ENTRY;
1157         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1158         ll_finish_md_op_data(op_data);
1159         if (rc == 0) {
1160                 ll_update_times(request, dir);
1161                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1162         }
1163
1164         ptlrpc_req_finished(request);
1165         RETURN(rc);
1166 }
1167
1168 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1169 {
1170         struct mdt_body *body;
1171         struct lov_mds_md *eadata;
1172         struct lov_stripe_md *lsm = NULL;
1173         struct obd_trans_info oti = { 0 };
1174         struct obdo *oa;
1175         struct obd_capa *oc = NULL;
1176         int rc;
1177         ENTRY;
1178
1179         /* req is swabbed so this is safe */
1180         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1181         if (!(body->valid & OBD_MD_FLEASIZE))
1182                 RETURN(0);
1183
1184         if (body->eadatasize == 0) {
1185                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1186                 GOTO(out, rc = -EPROTO);
1187         }
1188
1189         /* The MDS sent back the EA because we unlinked the last reference
1190          * to this file. Use this EA to unlink the objects on the OST.
1191          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1192          * check it is complete and sensible. */
1193         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1194                                               body->eadatasize);
1195         LASSERT(eadata != NULL);
1196
1197         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1198         if (rc < 0) {
1199                 CERROR("obd_unpackmd: %d\n", rc);
1200                 GOTO(out, rc);
1201         }
1202         LASSERT(rc >= sizeof(*lsm));
1203
1204         OBDO_ALLOC(oa);
1205         if (oa == NULL)
1206                 GOTO(out_free_memmd, rc = -ENOMEM);
1207
1208         oa->o_oi = lsm->lsm_oi;
1209         oa->o_mode = body->mode & S_IFMT;
1210         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1211
1212         if (body->valid & OBD_MD_FLCOOKIE) {
1213                 oa->o_valid |= OBD_MD_FLCOOKIE;
1214                 oti.oti_logcookies =
1215                         req_capsule_server_sized_get(&request->rq_pill,
1216                                                      &RMF_LOGCOOKIES,
1217                                                    sizeof(struct llog_cookie) *
1218                                                      lsm->lsm_stripe_count);
1219                 if (oti.oti_logcookies == NULL) {
1220                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1221                         body->valid &= ~OBD_MD_FLCOOKIE;
1222                 }
1223         }
1224
1225         if (body->valid & OBD_MD_FLOSSCAPA) {
1226                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1227                 if (rc)
1228                         GOTO(out_free_memmd, rc);
1229         }
1230
1231         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1232                          ll_i2mdexp(dir), oc);
1233         capa_put(oc);
1234         if (rc)
1235                 CERROR("obd destroy objid "DOSTID" error %d\n",
1236                        POSTID(&lsm->lsm_oi), rc);
1237 out_free_memmd:
1238         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1239         OBDO_FREE(oa);
1240 out:
1241         return rc;
1242 }
1243
1244 /* ll_unlink_generic() doesn't update the inode with the new link count.
1245  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1246  * is any lock existing. They will recycle dentries and inodes based upon locks
1247  * too. b=20433 */
1248 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1249                              struct dentry *dchild, struct qstr *name)
1250 {
1251         struct ptlrpc_request *request = NULL;
1252         struct md_op_data *op_data;
1253         int rc;
1254         ENTRY;
1255         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1256                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1257
1258         /*
1259          * XXX: unlink bind mountpoint maybe call to here,
1260          * just check it as vfs_unlink does.
1261          */
1262         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1263                 RETURN(-EBUSY);
1264
1265         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1266                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1267         if (IS_ERR(op_data))
1268                 RETURN(PTR_ERR(op_data));
1269
1270         if (dchild != NULL && dchild->d_inode != NULL)
1271                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1272
1273         op_data->op_fid2 = 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="DFID
1299                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1300                src_name->len, src_name->name,
1301                PFID(ll_inode2fid(src)), src, tgt_name->len,
1302                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1303
1304         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1305             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1306                 RETURN(-EBUSY);
1307
1308         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1309                                      LUSTRE_OPC_ANY, NULL);
1310         if (IS_ERR(op_data))
1311                 RETURN(PTR_ERR(op_data));
1312
1313         if (src_dchild != NULL && src_dchild->d_inode != NULL)
1314                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1315         if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1316                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1317
1318         err = md_rename(sbi->ll_md_exp, op_data,
1319                         src_name->name, src_name->len,
1320                         tgt_name->name, tgt_name->len, &request);
1321         ll_finish_md_op_data(op_data);
1322         if (!err) {
1323                 ll_update_times(request, src);
1324                 ll_update_times(request, tgt);
1325                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1326                 err = ll_objects_destroy(request, src);
1327         }
1328
1329         ptlrpc_req_finished(request);
1330
1331         RETURN(err);
1332 }
1333
1334 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1335                     dev_t rdev)
1336 {
1337         return ll_mknod_generic(dir, &dchild->d_name, mode,
1338                                 old_encode_dev(rdev), dchild);
1339 }
1340
1341 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1342 {
1343         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1344 }
1345
1346 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1347 {
1348         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1349 }
1350
1351 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1352 {
1353         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1354 }
1355
1356 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1357                       const char *oldname)
1358 {
1359         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1360 }
1361
1362 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1363                    struct dentry *new_dentry)
1364 {
1365         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1366                                new_dentry);
1367 }
1368
1369 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1370                      struct inode *new_dir, struct dentry *new_dentry)
1371 {
1372         int err;
1373         err = ll_rename_generic(old_dir, NULL,
1374                                  old_dentry, &old_dentry->d_name,
1375                                  new_dir, NULL, new_dentry,
1376                                  &new_dentry->d_name);
1377         if (!err)
1378                         d_move(old_dentry, new_dentry);
1379         return err;
1380 }
1381
1382 const 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 const 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 };