drivers/staging/lustre/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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #include <linux/fs.h>
  34 #include <linux/sched.h>
  35 #include <linux/mm.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/security.h>
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include "../include/obd_support.h"
  44 #include "../include/lustre_fid.h"
  45 #include "../include/lustre_dlm.h"
  46 #include "../include/lustre_ver.h"
  47 #include "llite_internal.h"
  48
  49 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  50                         struct lookup_intent *it);
  51
  52 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
  53 static int ll_test_inode(struct inode *inode, void *opaque)
  54 {
  55         struct ll_inode_info *lli = ll_i2info(inode);
  56         struct lustre_md     *md = opaque;
  57
  58         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  59                 CERROR("MDS body missing FID\n");
  60                 return 0;
  61         }
  62
  63         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  64                 return 0;
  65
  66         return 1;
  67 }
  68
  69 static int ll_set_inode(struct inode *inode, void *opaque)
  70 {
  71         struct ll_inode_info *lli = ll_i2info(inode);
  72         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  73
  74         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  75                 CERROR("MDS body missing FID\n");
  76                 return -EINVAL;
  77         }
  78
  79         lli->lli_fid = body->mbo_fid1;
  80         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  81                 CERROR("Can not initialize inode " DFID
  82                        " without object type: valid = %#llx\n",
  83                        PFID(&lli->lli_fid), body->mbo_valid);
  84                 return -EINVAL;
  85         }
  86
  87         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  88         if (unlikely(inode->i_mode == 0)) {
  89                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  90                 return -EINVAL;
  91         }
  92
  93         ll_lli_init(lli);
  94
  95         return 0;
  96 }
  97
  98 /**
  99  * Get an inode by inode number(@hash), which is already instantiated by
 100  * the intent lookup).
 101  */
 102 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 103                       struct lustre_md *md)
 104 {
 105         struct inode     *inode;
 106         int rc = 0;
 107
 108         LASSERT(hash != 0);
 109         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 110         if (!inode)
 111                 return ERR_PTR(-ENOMEM);
 112
 113         if (inode->i_state & I_NEW) {
 114                 rc = ll_read_inode2(inode, md);
 115                 if (!rc && S_ISREG(inode->i_mode) &&
 116                     !ll_i2info(inode)->lli_clob) {
 117                         CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
 118                                ll_get_fsname(sb, NULL, 0), md->lsm,
 119                                PFID(ll_inode2fid(inode)));
 120                         rc = cl_file_inode_init(inode, md);
 121                 }
 122                 if (rc) {
 123                         make_bad_inode(inode);
 124                         unlock_new_inode(inode);
 125                         iput(inode);
 126                         inode = ERR_PTR(rc);
 127                 } else {
 128                         unlock_new_inode(inode);
 129                 }
 130         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 131                 rc = ll_update_inode(inode, md);
 132                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 133                        PFID(&md->body->mbo_fid1), inode, rc);
 134                 if (rc) {
 135                         make_bad_inode(inode);
 136                         iput(inode);
 137                         inode = ERR_PTR(rc);
 138                 }
 139         }
 140         return inode;
 141 }
 142
 143 static void ll_invalidate_negative_children(struct inode *dir)
 144 {
 145         struct dentry *dentry, *tmp_subdir;
 146
 147         spin_lock(&dir->i_lock);
 148         hlist_for_each_entry(dentry, &dir->i_dentry, d_u.d_alias) {
 149                 spin_lock(&dentry->d_lock);
 150                 if (!list_empty(&dentry->d_subdirs)) {
 151                         struct dentry *child;
 152
 153                         list_for_each_entry_safe(child, tmp_subdir,
 154                                                  &dentry->d_subdirs,
 155                                                  d_child) {
 156                                 if (d_really_is_negative(child))
 157                                         d_lustre_invalidate(child, 1);
 158                         }
 159                 }
 160                 spin_unlock(&dentry->d_lock);
 161         }
 162         spin_unlock(&dir->i_lock);
 163 }
 164
 165 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 166 {
 167         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 168 }
 169
 170 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 171                        void *data, int flag)
 172 {
 173         struct lustre_handle lockh;
 174         int rc;
 175
 176         switch (flag) {
 177         case LDLM_CB_BLOCKING:
 178                 ldlm_lock2handle(lock, &lockh);
 179                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 180                 if (rc < 0) {
 181                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 182                         return rc;
 183                 }
 184                 break;
 185         case LDLM_CB_CANCELING: {
 186                 struct inode *inode = ll_inode_from_resource_lock(lock);
 187                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 188
 189                 /* Inode is set to lock->l_resource->lr_lvb_inode
 190                  * for mdc - bug 24555
 191                  */
 192                 LASSERT(!lock->l_ast_data);
 193
 194                 if (!inode)
 195                         break;
 196
 197                 /* Invalidate all dentries associated with this inode */
 198                 LASSERT(ldlm_is_canceling(lock));
 199
 200                 if (!fid_res_name_eq(ll_inode2fid(inode),
 201                                      &lock->l_resource->lr_name)) {
 202                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 203                                    PFID(ll_inode2fid(inode)), inode);
 204                         LBUG();
 205                 }
 206
 207                 if (bits & MDS_INODELOCK_XATTR) {
 208                         ll_xattr_cache_destroy(inode);
 209                         bits &= ~MDS_INODELOCK_XATTR;
 210                 }
 211
 212                 /* For OPEN locks we differentiate between lock modes
 213                  * LCK_CR, LCK_CW, LCK_PR - bug 22891
 214                  */
 215                 if (bits & MDS_INODELOCK_OPEN)
 216                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 217
 218                 if (bits & MDS_INODELOCK_OPEN) {
 219                         fmode_t fmode;
 220
 221                         switch (lock->l_req_mode) {
 222                         case LCK_CW:
 223                                 fmode = FMODE_WRITE;
 224                                 break;
 225                         case LCK_PR:
 226                                 fmode = FMODE_EXEC;
 227                                 break;
 228                         case LCK_CR:
 229                                 fmode = FMODE_READ;
 230                                 break;
 231                         default:
 232                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 233                                 LBUG();
 234                         }
 235
 236                         ll_md_real_close(inode, fmode);
 237                 }
 238
 239                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 240                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 241                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 242
 243                 if (bits & MDS_INODELOCK_LAYOUT) {
 244                         struct cl_object_conf conf = {
 245                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 246                                 .coc_inode = inode,
 247                         };
 248
 249                         rc = ll_layout_conf(inode, &conf);
 250                         if (rc < 0)
 251                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 252                                        DFID": rc = %d\n",
 253                                        PFID(ll_inode2fid(inode)), rc);
 254                 }
 255
 256                 if (bits & MDS_INODELOCK_UPDATE) {
 257                         struct ll_inode_info *lli = ll_i2info(inode);
 258
 259                         spin_lock(&lli->lli_lock);
 260                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 261                         spin_unlock(&lli->lli_lock);
 262                 }
 263
 264                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 265                         struct ll_inode_info *lli = ll_i2info(inode);
 266
 267                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, pfid  = "DFID"\n",
 268                                PFID(ll_inode2fid(inode)), lli,
 269                                PFID(&lli->lli_pfid));
 270
 271                         truncate_inode_pages(inode->i_mapping, 0);
 272
 273                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 274                                 struct inode *master_inode = NULL;
 275                                 unsigned long hash;
 276
 277                                 /*
 278                                  * This is slave inode, since all of the child
 279                                  * dentry is connected on the master inode, so
 280                                  * we have to invalidate the negative children
 281                                  * on master inode
 282                                  */
 283                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 284                                        PFID(ll_inode2fid(inode)),
 285                                        PFID(&lli->lli_pfid));
 286
 287                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 288                                                         ll_need_32bit_api(ll_i2sbi(inode)));
 289
 290                                 master_inode = ilookup5(inode->i_sb, hash,
 291                                                         ll_test_inode_by_fid,
 292                                                         (void *)&lli->lli_pfid);
 293                                 if (master_inode && !IS_ERR(master_inode)) {
 294                                         ll_invalidate_negative_children(master_inode);
 295                                         iput(master_inode);
 296                                 }
 297                         } else {
 298                                 ll_invalidate_negative_children(inode);
 299                         }
 300                 }
 301
 302                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 303                     inode->i_sb->s_root &&
 304                     !is_root_inode(inode))
 305                         ll_invalidate_aliases(inode);
 306
 307                 iput(inode);
 308                 break;
 309         }
 310         default:
 311                 LBUG();
 312         }
 313
 314         return 0;
 315 }
 316
 317 __u32 ll_i2suppgid(struct inode *i)
 318 {
 319         if (in_group_p(i->i_gid))
 320                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 321         else
 322                 return (__u32)(-1);
 323 }
 324
 325 /* Pack the required supplementary groups into the supplied groups array.
 326  * If we don't need to use the groups from the target inode(s) then we
 327  * instead pack one or more groups from the user's supplementary group
 328  * array in case it might be useful.  Not needed if doing an MDS-side upcall.
 329  */
 330 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 331 {
 332         LASSERT(i1);
 333
 334         suppgids[0] = ll_i2suppgid(i1);
 335
 336         if (i2)
 337                 suppgids[1] = ll_i2suppgid(i2);
 338                 else
 339                         suppgids[1] = -1;
 340 }
 341
 342 /*
 343  * try to reuse three types of dentry:
 344  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 345  *    by concurrent .revalidate).
 346  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 347  *    be cleared by others calling d_lustre_revalidate).
 348  * 3. DISCONNECTED alias.
 349  */
 350 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 351 {
 352         struct dentry *alias, *discon_alias, *invalid_alias;
 353
 354         if (hlist_empty(&inode->i_dentry))
 355                 return NULL;
 356
 357         discon_alias = NULL;
 358         invalid_alias = NULL;
 359
 360         spin_lock(&inode->i_lock);
 361         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 362                 LASSERT(alias != dentry);
 363
 364                 spin_lock(&alias->d_lock);
 365                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 366                     S_ISDIR(inode->i_mode))
 367                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 368                         discon_alias = alias;
 369                 else if (alias->d_parent == dentry->d_parent         &&
 370                          alias->d_name.hash == dentry->d_name.hash       &&
 371                          alias->d_name.len == dentry->d_name.len         &&
 372                          memcmp(alias->d_name.name, dentry->d_name.name,
 373                                 dentry->d_name.len) == 0)
 374                         invalid_alias = alias;
 375                 spin_unlock(&alias->d_lock);
 376
 377                 if (invalid_alias)
 378                         break;
 379         }
 380         alias = invalid_alias ?: discon_alias ?: NULL;
 381         if (alias) {
 382                 spin_lock(&alias->d_lock);
 383                 dget_dlock(alias);
 384                 spin_unlock(&alias->d_lock);
 385         }
 386         spin_unlock(&inode->i_lock);
 387
 388         return alias;
 389 }
 390
 391 /*
 392  * Similar to d_splice_alias(), but lustre treats invalid alias
 393  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 394  */
 395 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 396 {
 397         struct dentry *new;
 398         int rc;
 399
 400         if (inode) {
 401                 new = ll_find_alias(inode, de);
 402                 if (new) {
 403                         rc = ll_d_init(new);
 404                         if (rc < 0) {
 405                                 dput(new);
 406                                 return ERR_PTR(rc);
 407                         }
 408                         d_move(new, de);
 409                         iput(inode);
 410                         CDEBUG(D_DENTRY,
 411                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 412                               new, d_inode(new), d_count(new), new->d_flags);
 413                         return new;
 414                 }
 415         }
 416         rc = ll_d_init(de);
 417         if (rc < 0)
 418                 return ERR_PTR(rc);
 419         d_add(de, inode);
 420         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 421                de, d_inode(de), d_count(de), de->d_flags);
 422         return de;
 423 }
 424
 425 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 426                                struct lookup_intent *it,
 427                                struct inode *parent, struct dentry **de)
 428 {
 429         struct inode *inode = NULL;
 430         __u64 bits = 0;
 431         int rc = 0;
 432         struct dentry *alias;
 433
 434         /* NB 1 request reference will be taken away by ll_intent_lock()
 435          * when I return
 436          */
 437         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 438                it->it_disposition);
 439         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 440                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 441                 if (rc)
 442                         return rc;
 443
 444                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 445
 446                 /* We used to query real size from OSTs here, but actually
 447                  * this is not needed. For stat() calls size would be updated
 448                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 449                  * 2.4 and
 450                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 451                  * Everybody else who needs correct file size would call
 452                  * ll_glimpse_size or some equivalent themselves anyway.
 453                  * Also see bug 7198.
 454                  */
 455         }
 456
 457         alias = ll_splice_alias(inode, *de);
 458         if (IS_ERR(alias)) {
 459                 rc = PTR_ERR(alias);
 460                 goto out;
 461         }
 462         *de = alias;
 463
 464         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 465                 /* we have lookup look - unhide dentry */
 466                 if (bits & MDS_INODELOCK_LOOKUP)
 467                         d_lustre_revalidate(*de);
 468         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 469                 /* If file created on server, don't depend on parent UPDATE
 470                  * lock to unhide it. It is left hidden and next lookup can
 471                  * find it in ll_splice_alias.
 472                  */
 473                 /* Check that parent has UPDATE lock. */
 474                 struct lookup_intent parent_it = {
 475                                         .it_op = IT_GETATTR,
 476                                         .it_lock_handle = 0 };
 477                 struct lu_fid fid = ll_i2info(parent)->lli_fid;
 478
 479                 /* If it is striped directory, get the real stripe parent */
 480                 if (unlikely(ll_i2info(parent)->lli_lsm_md)) {
 481                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 482                                                  ll_i2info(parent)->lli_lsm_md,
 483                                                  (*de)->d_name.name,
 484                                                  (*de)->d_name.len, &fid);
 485                         if (rc)
 486                                 return rc;
 487                 }
 488
 489                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 490                                        NULL)) {
 491                         d_lustre_revalidate(*de);
 492                         ll_intent_release(&parent_it);
 493                 }
 494         }
 495
 496 out:
 497         if (rc != 0 && it->it_op & IT_OPEN)
 498                 ll_open_cleanup((*de)->d_sb, request);
 499
 500         return rc;
 501 }
 502
 503 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 504                                    struct lookup_intent *it)
 505 {
 506         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 507         struct dentry *save = dentry, *retval;
 508         struct ptlrpc_request *req = NULL;
 509         struct inode *inode;
 510         struct md_op_data *op_data;
 511         __u32 opc;
 512         int rc;
 513
 514         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 515                 return ERR_PTR(-ENAMETOOLONG);
 516
 517         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),intent=%s\n",
 518                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 519
 520         if (d_mountpoint(dentry))
 521                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 522
 523         if (!it || it->it_op == IT_GETXATTR)
 524                 it = &lookup_it;
 525
 526         if (it->it_op == IT_GETATTR) {
 527                 rc = ll_statahead_enter(parent, &dentry, 0);
 528                 if (rc == 1) {
 529                         if (dentry == save)
 530                                 retval = NULL;
 531                         else
 532                                 retval = dentry;
 533                         goto out;
 534                 }
 535         }
 536
 537         if (it->it_op & IT_CREAT)
 538                 opc = LUSTRE_OPC_CREATE;
 539         else
 540                 opc = LUSTRE_OPC_ANY;
 541
 542         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 543                                      dentry->d_name.len, 0, opc, NULL);
 544         if (IS_ERR(op_data))
 545                 return (void *)op_data;
 546
 547         /* enforce umask if acl disabled or MDS doesn't support umask */
 548         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 549                 it->it_create_mode &= ~current_umask();
 550
 551         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 552                             &ll_md_blocking_ast, 0);
 553         ll_finish_md_op_data(op_data);
 554         if (rc < 0) {
 555                 retval = ERR_PTR(rc);
 556                 goto out;
 557         }
 558
 559         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 560         if (rc != 0) {
 561                 ll_intent_release(it);
 562                 retval = ERR_PTR(rc);
 563                 goto out;
 564         }
 565
 566         inode = d_inode(dentry);
 567         if ((it->it_op & IT_OPEN) && inode &&
 568             !S_ISREG(inode->i_mode) &&
 569             !S_ISDIR(inode->i_mode)) {
 570                 ll_release_openhandle(inode, it);
 571         }
 572         ll_lookup_finish_locks(it, inode);
 573
 574         if (dentry == save)
 575                 retval = NULL;
 576         else
 577                 retval = dentry;
 578  out:
 579         if (req)
 580                 ptlrpc_req_finished(req);
 581         if (it->it_op == IT_GETATTR && (!retval || retval == dentry))
 582                 ll_statahead_mark(parent, dentry);
 583         return retval;
 584 }
 585
 586 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 587                                    unsigned int flags)
 588 {
 589         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 590         struct dentry *de;
 591
 592         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),flags=%u\n",
 593                dentry, PFID(ll_inode2fid(parent)), parent, flags);
 594
 595         /* Optimize away (CREATE && !OPEN). Let .create handle the race.
 596          * but only if we have write permissions there, otherwise we need
 597          * to proceed with lookup. LU-4185
 598          */
 599         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 600             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 601                 return NULL;
 602
 603         if (flags & (LOOKUP_PARENT | LOOKUP_OPEN | LOOKUP_CREATE))
 604                 itp = NULL;
 605         else
 606                 itp = &it;
 607         de = ll_lookup_it(parent, dentry, itp);
 608
 609         if (itp)
 610                 ll_intent_release(itp);
 611
 612         return de;
 613 }
 614
 615 /*
 616  * For cached negative dentry and new dentry, handle lookup/create/open
 617  * together.
 618  */
 619 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 620                           struct file *file, unsigned open_flags,
 621                           umode_t mode, int *opened)
 622 {
 623         struct lookup_intent *it;
 624         struct dentry *de;
 625         int rc = 0;
 626
 627         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),file %p,open_flags %x,mode %x opened %d\n",
 628                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 629                *opened);
 630
 631         /* Only negative dentries enter here */
 632         LASSERT(!d_inode(dentry));
 633
 634         if (!d_in_lookup(dentry)) {
 635                 /* A valid negative dentry that just passed revalidation,
 636                  * there's little point to try and open it server-side,
 637                  * even though there's a minuscle chance it might succeed.
 638                  * Either way it's a valid race to just return -ENOENT here.
 639                  */
 640                 if (!(open_flags & O_CREAT))
 641                         return -ENOENT;
 642
 643                 /* Otherwise we just unhash it to be rehashed afresh via
 644                  * lookup if necessary
 645                  */
 646                 d_drop(dentry);
 647         }
 648
 649         it = kzalloc(sizeof(*it), GFP_NOFS);
 650         if (!it)
 651                 return -ENOMEM;
 652
 653         it->it_op = IT_OPEN;
 654         if (open_flags & O_CREAT)
 655                 it->it_op |= IT_CREAT;
 656         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 657         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 658         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 659
 660         /* Dentry added to dcache tree in ll_lookup_it */
 661         de = ll_lookup_it(dir, dentry, it);
 662         if (IS_ERR(de))
 663                 rc = PTR_ERR(de);
 664         else if (de)
 665                 dentry = de;
 666
 667         if (!rc) {
 668                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 669                         /* Dentry instantiated in ll_create_it. */
 670                         rc = ll_create_it(dir, dentry, it);
 671                         if (rc) {
 672                                 /* We dget in ll_splice_alias. */
 673                                 if (de)
 674                                         dput(de);
 675                                 goto out_release;
 676                         }
 677
 678                         *opened |= FILE_CREATED;
 679                 }
 680                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 681                         /* Open dentry. */
 682                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 683                                 /* We cannot call open here as it might
 684                                  * deadlock. This case is unreachable in
 685                                  * practice because of OBD_CONNECT_NODEVOH.
 686                                  */
 687                                 rc = finish_no_open(file, de);
 688                         } else {
 689                                 file->private_data = it;
 690                                 rc = finish_open(file, dentry, NULL, opened);
 691                                 /* We dget in ll_splice_alias. finish_open takes
 692                                  * care of dget for fd open.
 693                                  */
 694                                 if (de)
 695                                         dput(de);
 696                         }
 697                 } else {
 698                         rc = finish_no_open(file, de);
 699                 }
 700         }
 701
 702 out_release:
 703         ll_intent_release(it);
 704         kfree(it);
 705
 706         return rc;
 707 }
 708
 709 /* We depend on "mode" being set with the proper file type/umask by now */
 710 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 711 {
 712         struct inode *inode = NULL;
 713         struct ptlrpc_request *request = NULL;
 714         struct ll_sb_info *sbi = ll_i2sbi(dir);
 715         int rc;
 716
 717         LASSERT(it && it->it_disposition);
 718
 719         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 720         request = it->it_request;
 721         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 722         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 723         if (rc) {
 724                 inode = ERR_PTR(rc);
 725                 goto out;
 726         }
 727
 728         LASSERT(hlist_empty(&inode->i_dentry));
 729
 730         /* We asked for a lock on the directory, but were granted a
 731          * lock on the inode.  Since we finally have an inode pointer,
 732          * stuff it in the lock.
 733          */
 734         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 735                PFID(ll_inode2fid(dir)), inode);
 736         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 737  out:
 738         ptlrpc_req_finished(request);
 739         return inode;
 740 }
 741
 742 /*
 743  * By the time this is called, we already have created the directory cache
 744  * entry for the new file, but it is so far negative - it has no inode.
 745  *
 746  * We defer creating the OBD object(s) until open, to keep the intent and
 747  * non-intent code paths similar, and also because we do not have the MDS
 748  * inode number before calling ll_create_node() (which is needed for LOV),
 749  * so we would need to do yet another RPC to the MDS to store the LOV EA
 750  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 751  * lmm_size in datalen (the MDS still has code which will handle that).
 752  *
 753  * If the create succeeds, we fill in the inode information
 754  * with d_instantiate().
 755  */
 756 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 757                         struct lookup_intent *it)
 758 {
 759         struct inode *inode;
 760         int rc = 0;
 761
 762         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 763                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 764
 765         rc = it_open_error(DISP_OPEN_CREATE, it);
 766         if (rc)
 767                 return rc;
 768
 769         inode = ll_create_node(dir, it);
 770         if (IS_ERR(inode))
 771                 return PTR_ERR(inode);
 772
 773         d_instantiate(dentry, inode);
 774         return 0;
 775 }
 776
 777 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 778 {
 779         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 780                                                        &RMF_MDT_BODY);
 781
 782         LASSERT(body);
 783         if (body->mbo_valid & OBD_MD_FLMTIME &&
 784             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 785                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu\n",
 786                        PFID(ll_inode2fid(inode)), LTIME_S(inode->i_mtime),
 787                        body->mbo_mtime);
 788                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 789         }
 790         if (body->mbo_valid & OBD_MD_FLCTIME &&
 791             body->mbo_ctime > LTIME_S(inode->i_ctime))
 792                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 793 }
 794
 795 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 796                        const char *tgt, umode_t mode, int rdev,
 797                        __u32 opc)
 798 {
 799         struct ptlrpc_request *request = NULL;
 800         struct md_op_data *op_data;
 801         struct inode *inode = NULL;
 802         struct ll_sb_info *sbi = ll_i2sbi(dir);
 803         int tgt_len = 0;
 804         int err;
 805
 806         if (unlikely(tgt))
 807                 tgt_len = strlen(tgt) + 1;
 808
 809         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 810                                      dentry->d_name.name,
 811                                      dentry->d_name.len,
 812                                      0, opc, NULL);
 813         if (IS_ERR(op_data)) {
 814                 err = PTR_ERR(op_data);
 815                 goto err_exit;
 816         }
 817
 818         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 819                         from_kuid(&init_user_ns, current_fsuid()),
 820                         from_kgid(&init_user_ns, current_fsgid()),
 821                         cfs_curproc_cap_pack(), rdev, &request);
 822         ll_finish_md_op_data(op_data);
 823         if (err)
 824                 goto err_exit;
 825
 826         ll_update_times(request, dir);
 827
 828         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 829         if (err)
 830                 goto err_exit;
 831
 832         d_instantiate(dentry, inode);
 833 err_exit:
 834         ptlrpc_req_finished(request);
 835
 836         return err;
 837 }
 838
 839 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 840                     umode_t mode, dev_t rdev)
 841 {
 842         int err;
 843
 844         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
 845                dchild, PFID(ll_inode2fid(dir)), dir, mode,
 846                old_encode_dev(rdev));
 847
 848         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 849                 mode &= ~current_umask();
 850
 851         switch (mode & S_IFMT) {
 852         case 0:
 853                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 854         case S_IFREG:
 855         case S_IFCHR:
 856         case S_IFBLK:
 857         case S_IFIFO:
 858         case S_IFSOCK:
 859                 err = ll_new_node(dir, dchild, NULL, mode,
 860                                   old_encode_dev(rdev),
 861                                   LUSTRE_OPC_MKNOD);
 862                 break;
 863         case S_IFDIR:
 864                 err = -EPERM;
 865                 break;
 866         default:
 867                 err = -EINVAL;
 868         }
 869
 870         if (!err)
 871                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 872
 873         return err;
 874 }
 875
 876 /*
 877  * Plain create. Intent create is handled in atomic_open.
 878  */
 879 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 880                         umode_t mode, bool want_excl)
 881 {
 882         int rc;
 883
 884         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
 885                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
 886
 887         rc = ll_mknod(dir, dentry, mode, 0);
 888
 889         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 890
 891         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 892                dentry, d_unhashed(dentry));
 893
 894         return rc;
 895 }
 896
 897 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 898 {
 899         struct mdt_body *body;
 900         struct lov_mds_md *eadata;
 901         struct lov_stripe_md *lsm = NULL;
 902         struct obd_trans_info oti = { 0 };
 903         struct obdo *oa;
 904         int rc;
 905
 906         /* req is swabbed so this is safe */
 907         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 908         if (!(body->mbo_valid & OBD_MD_FLEASIZE))
 909                 return 0;
 910
 911         if (body->mbo_eadatasize == 0) {
 912                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 913                 rc = -EPROTO;
 914                 goto out;
 915         }
 916
 917         /* The MDS sent back the EA because we unlinked the last reference
 918          * to this file. Use this EA to unlink the objects on the OST.
 919          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 920          * check it is complete and sensible.
 921          */
 922         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 923                                               body->mbo_eadatasize);
 924         LASSERT(eadata);
 925
 926         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->mbo_eadatasize);
 927         if (rc < 0) {
 928                 CERROR("obd_unpackmd: %d\n", rc);
 929                 goto out;
 930         }
 931         LASSERT(rc >= sizeof(*lsm));
 932
 933         oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
 934         if (!oa) {
 935                 rc = -ENOMEM;
 936                 goto out_free_memmd;
 937         }
 938
 939         oa->o_oi = lsm->lsm_oi;
 940         oa->o_mode = body->mbo_mode & S_IFMT;
 941         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 942
 943         if (body->mbo_valid & OBD_MD_FLCOOKIE) {
 944                 oa->o_valid |= OBD_MD_FLCOOKIE;
 945                 oti.oti_logcookies =
 946                         req_capsule_server_sized_get(&request->rq_pill,
 947                                                      &RMF_LOGCOOKIES,
 948                                                    sizeof(struct llog_cookie) *
 949                                                      lsm->lsm_stripe_count);
 950                 if (!oti.oti_logcookies) {
 951                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 952                         body->mbo_valid &= ~OBD_MD_FLCOOKIE;
 953                 }
 954         }
 955
 956         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 957                          ll_i2mdexp(dir));
 958         if (rc)
 959                 CERROR("obd destroy objid "DOSTID" error %d\n",
 960                        POSTID(&lsm->lsm_oi), rc);
 961 out_free_memmd:
 962         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 963         kmem_cache_free(obdo_cachep, oa);
 964 out:
 965         return rc;
 966 }
 967
 968 /* ll_unlink() doesn't update the inode with the new link count.
 969  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 970  * is any lock existing. They will recycle dentries and inodes based upon locks
 971  * too. b=20433
 972  */
 973 static int ll_unlink(struct inode *dir, struct dentry *dchild)
 974 {
 975         struct ptlrpc_request *request = NULL;
 976         struct md_op_data *op_data;
 977         int rc;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 980                dchild, dir->i_ino, dir->i_generation, dir);
 981
 982         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 983                                      dchild->d_name.name,
 984                                      dchild->d_name.len,
 985                                      0, LUSTRE_OPC_ANY, NULL);
 986         if (IS_ERR(op_data))
 987                 return PTR_ERR(op_data);
 988
 989         if (dchild->d_inode)
 990                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
 991
 992         op_data->op_fid2 = op_data->op_fid3;
 993         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 994         ll_finish_md_op_data(op_data);
 995         if (rc)
 996                 goto out;
 997
 998         ll_update_times(request, dir);
 999         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1000
1001         rc = ll_objects_destroy(request, dir);
1002  out:
1003         ptlrpc_req_finished(request);
1004         return rc;
1005 }
1006
1007 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1008 {
1009         int err;
1010
1011         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir"DFID"(%p)\n",
1012                dentry, PFID(ll_inode2fid(dir)), dir);
1013
1014         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1015                 mode &= ~current_umask();
1016         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
1017
1018         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1019         if (!err)
1020                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1021
1022         return err;
1023 }
1024
1025 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1026 {
1027         struct ptlrpc_request *request = NULL;
1028         struct md_op_data *op_data;
1029         int rc;
1030
1031         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1032                dchild, PFID(ll_inode2fid(dir)), dir);
1033
1034         op_data = ll_prep_md_op_data(NULL, dir, NULL,
1035                                      dchild->d_name.name,
1036                                      dchild->d_name.len,
1037                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1038         if (IS_ERR(op_data))
1039                 return PTR_ERR(op_data);
1040
1041         if (dchild->d_inode)
1042                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1043
1044         op_data->op_fid2 = op_data->op_fid3;
1045         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1046         ll_finish_md_op_data(op_data);
1047         if (rc == 0) {
1048                 ll_update_times(request, dir);
1049                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1050         }
1051
1052         ptlrpc_req_finished(request);
1053         return rc;
1054 }
1055
1056 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1057                       const char *oldname)
1058 {
1059         int err;
1060
1061         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),target=%.*s\n",
1062                dentry, PFID(ll_inode2fid(dir)), dir, 3000, oldname);
1063
1064         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1065                           0, LUSTRE_OPC_SYMLINK);
1066
1067         if (!err)
1068                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1069
1070         return err;
1071 }
1072
1073 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1074                    struct dentry *new_dentry)
1075 {
1076         struct inode *src = d_inode(old_dentry);
1077         struct ll_sb_info *sbi = ll_i2sbi(dir);
1078         struct ptlrpc_request *request = NULL;
1079         struct md_op_data *op_data;
1080         int err;
1081
1082         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1083                PFID(ll_inode2fid(src)), src, PFID(ll_inode2fid(dir)), dir,
1084                new_dentry);
1085
1086         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1087                                      new_dentry->d_name.len,
1088                                      0, LUSTRE_OPC_ANY, NULL);
1089         if (IS_ERR(op_data))
1090                 return PTR_ERR(op_data);
1091
1092         err = md_link(sbi->ll_md_exp, op_data, &request);
1093         ll_finish_md_op_data(op_data);
1094         if (err)
1095                 goto out;
1096
1097         ll_update_times(request, dir);
1098         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1099 out:
1100         ptlrpc_req_finished(request);
1101         return err;
1102 }
1103
1104 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1105                      struct inode *tgt, struct dentry *tgt_dchild)
1106 {
1107         struct ptlrpc_request *request = NULL;
1108         struct ll_sb_info *sbi = ll_i2sbi(src);
1109         struct md_op_data *op_data;
1110         int err;
1111
1112         CDEBUG(D_VFSTRACE,
1113                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1114                src_dchild, PFID(ll_inode2fid(src)), src,
1115                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1116
1117         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1118                                      LUSTRE_OPC_ANY, NULL);
1119         if (IS_ERR(op_data))
1120                 return PTR_ERR(op_data);
1121
1122         if (src_dchild->d_inode)
1123                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1124         if (tgt_dchild->d_inode)
1125                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1126
1127         err = md_rename(sbi->ll_md_exp, op_data,
1128                         src_dchild->d_name.name,
1129                         src_dchild->d_name.len,
1130                         tgt_dchild->d_name.name,
1131                         tgt_dchild->d_name.len, &request);
1132         ll_finish_md_op_data(op_data);
1133         if (!err) {
1134                 ll_update_times(request, src);
1135                 ll_update_times(request, tgt);
1136                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1137                 err = ll_objects_destroy(request, src);
1138         }
1139
1140         ptlrpc_req_finished(request);
1141         if (!err)
1142                 d_move(src_dchild, tgt_dchild);
1143         return err;
1144 }
1145
1146 const struct inode_operations ll_dir_inode_operations = {
1147         .mknod        = ll_mknod,
1148         .atomic_open        = ll_atomic_open,
1149         .lookup      = ll_lookup_nd,
1150         .create      = ll_create_nd,
1151         /* We need all these non-raw things for NFSD, to not patch it. */
1152         .unlink      = ll_unlink,
1153         .mkdir        = ll_mkdir,
1154         .rmdir        = ll_rmdir,
1155         .symlink            = ll_symlink,
1156         .link          = ll_link,
1157         .rename      = ll_rename,
1158         .setattr            = ll_setattr,
1159         .getattr            = ll_getattr,
1160         .permission      = ll_inode_permission,
1161         .setxattr          = generic_setxattr,
1162         .getxattr          = generic_getxattr,
1163         .listxattr        = ll_listxattr,
1164         .removexattr    = generic_removexattr,
1165         .get_acl            = ll_get_acl,
1166 };
1167
1168 const struct inode_operations ll_special_inode_operations = {
1169         .setattr        = ll_setattr,
1170         .getattr        = ll_getattr,
1171         .permission     = ll_inode_permission,
1172         .setxattr       = generic_setxattr,
1173         .getxattr       = generic_getxattr,
1174         .listxattr      = ll_listxattr,
1175         .removexattr    = generic_removexattr,
1176         .get_acl            = ll_get_acl,
1177 };