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 #include <linux/user_namespace.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_dlm.h>
  50 #include <lustre_ver.h>
  51 #include "llite_internal.h"
  52
  53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  54                         struct lookup_intent *it);
  55
  56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  57 static int ll_test_inode(struct inode *inode, void *opaque)
  58 {
  59         struct ll_inode_info    *lli = ll_i2info(inode);
  60         struct lustre_md        *md = opaque;
  61
  62         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  63                 CERROR("MDS body missing FID\n");
  64                 return 0;
  65         }
  66
  67         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  68                 return 0;
  69
  70         return 1;
  71 }
  72
  73 static int ll_set_inode(struct inode *inode, void *opaque)
  74 {
  75         struct ll_inode_info *lli = ll_i2info(inode);
  76         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  77
  78         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  79                 CERROR("MDS body missing FID\n");
  80                 return -EINVAL;
  81         }
  82
  83         lli->lli_fid = body->mbo_fid1;
  84         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  85                 CERROR("Can not initialize inode "DFID" without object type: "
  86                        "valid = %#llx\n",
  87                        PFID(&lli->lli_fid), body->mbo_valid);
  88                 return -EINVAL;
  89         }
  90
  91         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  92         if (unlikely(inode->i_mode == 0)) {
  93                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  94                 return -EINVAL;
  95         }
  96
  97         ll_lli_init(lli);
  98
  99         return 0;
 100 }
 101
 102
 103 /**
 104  * Get an inode by inode number(@hash), which is already instantiated by
 105  * the intent lookup).
 106  */
 107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 108                       struct lustre_md *md)
 109 {
 110         struct inode    *inode;
 111         int             rc = 0;
 112
 113         ENTRY;
 114
 115         LASSERT(hash != 0);
 116         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 117         if (inode == NULL)
 118                 RETURN(ERR_PTR(-ENOMEM));
 119
 120         if (inode->i_state & I_NEW) {
 121                 rc = ll_read_inode2(inode, md);
 122                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 123                     ll_i2info(inode)->lli_clob == NULL)
 124                         rc = cl_file_inode_init(inode, md);
 125
 126                 if (rc != 0) {
 127                         /* Let's clear directory lsm here, otherwise
 128                          * make_bad_inode() will reset the inode mode
 129                          * to regular, then ll_clear_inode will not
 130                          * be able to clear lsm_md */
 131                         if (S_ISDIR(inode->i_mode))
 132                                 ll_dir_clear_lsm_md(inode);
 133                         make_bad_inode(inode);
 134                         unlock_new_inode(inode);
 135                         iput(inode);
 136                         inode = ERR_PTR(rc);
 137                 } else {
 138                         unlock_new_inode(inode);
 139                 }
 140         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 141                 rc = ll_update_inode(inode, md);
 142                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 143                        PFID(&md->body->mbo_fid1), inode, rc);
 144                 if (rc != 0) {
 145                         if (S_ISDIR(inode->i_mode))
 146                                 ll_dir_clear_lsm_md(inode);
 147                         iput(inode);
 148                         inode = ERR_PTR(rc);
 149                 }
 150         }
 151
 152         RETURN(inode);
 153 }
 154
 155 static void ll_invalidate_negative_children(struct inode *dir)
 156 {
 157         struct dentry *dentry, *tmp_subdir;
 158         DECLARE_LL_D_HLIST_NODE_PTR(p);
 159
 160         ll_lock_dcache(dir);
 161         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 162                 spin_lock(&dentry->d_lock);
 163                 if (!list_empty(&dentry->d_subdirs)) {
 164                         struct dentry *child;
 165
 166                         list_for_each_entry_safe(child, tmp_subdir,
 167                                                  &dentry->d_subdirs,
 168                                                  d_child) {
 169                                 if (child->d_inode == NULL)
 170                                         d_lustre_invalidate(child, 1);
 171                         }
 172                 }
 173                 spin_unlock(&dentry->d_lock);
 174         }
 175         ll_unlock_dcache(dir);
 176 }
 177
 178 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 179 {
 180         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 181 }
 182
 183 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 184                        void *data, int flag)
 185 {
 186         struct lustre_handle lockh;
 187         int rc;
 188         ENTRY;
 189
 190         switch (flag) {
 191         case LDLM_CB_BLOCKING:
 192                 ldlm_lock2handle(lock, &lockh);
 193                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 194                 if (rc < 0) {
 195                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 196                         RETURN(rc);
 197                 }
 198                 break;
 199         case LDLM_CB_CANCELING: {
 200                 struct inode *inode = ll_inode_from_resource_lock(lock);
 201                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 202
 203                 /* Inode is set to lock->l_resource->lr_lvb_inode
 204                  * for mdc - bug 24555 */
 205                 LASSERT(lock->l_ast_data == NULL);
 206
 207                 if (inode == NULL)
 208                         break;
 209
 210                 /* Invalidate all dentries associated with this inode */
 211                 LASSERT(ldlm_is_canceling(lock));
 212
 213                 if (!fid_res_name_eq(ll_inode2fid(inode),
 214                                      &lock->l_resource->lr_name)) {
 215                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 216                                    PFID(ll_inode2fid(inode)), inode);
 217                         LBUG();
 218                 }
 219
 220                 if (bits & MDS_INODELOCK_XATTR) {
 221                         if (S_ISDIR(inode->i_mode))
 222                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 223                         ll_xattr_cache_destroy(inode);
 224                         bits &= ~MDS_INODELOCK_XATTR;
 225                 }
 226
 227                 /* For OPEN locks we differentiate between lock modes
 228                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 229                 if (bits & MDS_INODELOCK_OPEN)
 230                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 231
 232                 if (bits & MDS_INODELOCK_OPEN) {
 233                         fmode_t fmode;
 234
 235                         switch (lock->l_req_mode) {
 236                         case LCK_CW:
 237                                 fmode = FMODE_WRITE;
 238                                 break;
 239                         case LCK_PR:
 240                                 fmode = FMODE_EXEC;
 241                                 break;
 242                         case LCK_CR:
 243                                 fmode = FMODE_READ;
 244                                 break;
 245                         default:
 246                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 247                                 LBUG();
 248                         }
 249
 250                         ll_md_real_close(inode, fmode);
 251
 252                         bits &= ~MDS_INODELOCK_OPEN;
 253                 }
 254
 255                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 256                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 257                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 258
 259                 if (bits & MDS_INODELOCK_LAYOUT) {
 260                         struct cl_object_conf conf = {
 261                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 262                                 .coc_inode = inode,
 263                         };
 264
 265                         rc = ll_layout_conf(inode, &conf);
 266                         if (rc < 0)
 267                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 268                                        DFID": rc = %d\n",
 269                                        PFID(ll_inode2fid(inode)), rc);
 270                 }
 271
 272                 if (bits & MDS_INODELOCK_UPDATE) {
 273                         struct ll_inode_info *lli = ll_i2info(inode);
 274
 275                         spin_lock(&lli->lli_lock);
 276                         LTIME_S(inode->i_mtime) = 0;
 277                         LTIME_S(inode->i_atime) = 0;
 278                         LTIME_S(inode->i_ctime) = 0;
 279                         spin_unlock(&lli->lli_lock);
 280                 }
 281
 282                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 283                         struct ll_inode_info *lli = ll_i2info(inode);
 284
 285                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 286                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 287                                lli, PFID(&lli->lli_pfid));
 288                         truncate_inode_pages(inode->i_mapping, 0);
 289
 290                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 291                                 struct inode *master_inode = NULL;
 292                                 unsigned long hash;
 293
 294                                 /* This is slave inode, since all of the child
 295                                  * dentry is connected on the master inode, so
 296                                  * we have to invalidate the negative children
 297                                  * on master inode */
 298                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 299                                        PFID(ll_inode2fid(inode)),
 300                                        PFID(&lli->lli_pfid));
 301
 302                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 303                                         ll_need_32bit_api(ll_i2sbi(inode)));
 304
 305                                 /* Do not lookup the inode with ilookup5,
 306                                  * otherwise it will cause dead lock,
 307                                  *
 308                                  * 1. Client1 send chmod req to the MDT0, then
 309                                  * on MDT0, it enqueues master and all of its
 310                                  * slaves lock, (mdt_attr_set() ->
 311                                  * mdt_lock_slaves()), after gets master and
 312                                  * stripe0 lock, it will send the enqueue req
 313                                  * (for stripe1) to MDT1, then MDT1 finds the
 314                                  * lock has been granted to client2. Then MDT1
 315                                  * sends blocking ast to client2.
 316                                  *
 317                                  * 2. At the same time, client2 tries to unlink
 318                                  * the striped dir (rm -rf striped_dir), and
 319                                  * during lookup, it will hold the master inode
 320                                  * of the striped directory, whose inode state
 321                                  * is NEW, then tries to revalidate all of its
 322                                  * slaves, (ll_prep_inode()->ll_iget()->
 323                                  * ll_read_inode2()-> ll_update_inode().). And
 324                                  * it will be blocked on the server side because
 325                                  * of 1.
 326                                  *
 327                                  * 3. Then the client get the blocking_ast req,
 328                                  * cancel the lock, but being blocked if using
 329                                  * ->ilookup5()), because master inode state is
 330                                  *  NEW. */
 331                                 master_inode = ilookup5_nowait(inode->i_sb,
 332                                                     hash, ll_test_inode_by_fid,
 333                                                         (void *)&lli->lli_pfid);
 334                                 if (master_inode != NULL &&
 335                                         !IS_ERR(master_inode)) {
 336                                         ll_invalidate_negative_children(
 337                                                                 master_inode);
 338                                         iput(master_inode);
 339                                 }
 340                         } else {
 341                                 ll_invalidate_negative_children(inode);
 342                         }
 343                 }
 344
 345                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 346                     inode->i_sb->s_root != NULL &&
 347                     inode != inode->i_sb->s_root->d_inode)
 348                         ll_invalidate_aliases(inode);
 349
 350                 iput(inode);
 351                 break;
 352         }
 353         default:
 354                 LBUG();
 355         }
 356
 357         RETURN(0);
 358 }
 359
 360 __u32 ll_i2suppgid(struct inode *i)
 361 {
 362         if (in_group_p(i->i_gid))
 363                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 364         else
 365                 return (__u32) __kgid_val(INVALID_GID);
 366 }
 367
 368 /* Pack the required supplementary groups into the supplied groups array.
 369  * If we don't need to use the groups from the target inode(s) then we
 370  * instead pack one or more groups from the user's supplementary group
 371  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 372 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 373 {
 374         LASSERT(i1 != NULL);
 375         LASSERT(suppgids != NULL);
 376
 377         suppgids[0] = ll_i2suppgid(i1);
 378
 379         if (i2)
 380                 suppgids[1] = ll_i2suppgid(i2);
 381         else
 382                 suppgids[1] = -1;
 383 }
 384
 385 /*
 386  * try to reuse three types of dentry:
 387  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 388  *    by concurrent .revalidate).
 389  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 390  *    be cleared by others calling d_lustre_revalidate).
 391  * 3. DISCONNECTED alias.
 392  */
 393 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 394 {
 395         struct dentry *alias, *discon_alias, *invalid_alias;
 396         DECLARE_LL_D_HLIST_NODE_PTR(p);
 397
 398         if (ll_d_hlist_empty(&inode->i_dentry))
 399                 return NULL;
 400
 401         discon_alias = invalid_alias = NULL;
 402
 403         ll_lock_dcache(inode);
 404         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 405                 LASSERT(alias != dentry);
 406
 407                 spin_lock(&alias->d_lock);
 408                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 409                     S_ISDIR(inode->i_mode))
 410                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 411                         discon_alias = alias;
 412                 else if (alias->d_parent == dentry->d_parent             &&
 413                          alias->d_name.hash == dentry->d_name.hash       &&
 414                          alias->d_name.len == dentry->d_name.len         &&
 415                          memcmp(alias->d_name.name, dentry->d_name.name,
 416                                 dentry->d_name.len) == 0)
 417                         invalid_alias = alias;
 418                 spin_unlock(&alias->d_lock);
 419
 420                 if (invalid_alias)
 421                         break;
 422         }
 423         alias = invalid_alias ?: discon_alias ?: NULL;
 424         if (alias) {
 425                 spin_lock(&alias->d_lock);
 426                 dget_dlock(alias);
 427                 spin_unlock(&alias->d_lock);
 428         }
 429         ll_unlock_dcache(inode);
 430
 431         return alias;
 432 }
 433
 434 /*
 435  * Similar to d_splice_alias(), but lustre treats invalid alias
 436  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 437  */
 438 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 439 {
 440         struct dentry *new;
 441         int rc;
 442
 443         if (inode) {
 444                 new = ll_find_alias(inode, de);
 445                 if (new) {
 446                         rc = ll_d_init(new);
 447                         if (rc < 0) {
 448                                 dput(new);
 449                                 return ERR_PTR(rc);
 450                         }
 451                         d_move(new, de);
 452                         iput(inode);
 453                         CDEBUG(D_DENTRY,
 454                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 455                               new, new->d_inode, ll_d_count(new), new->d_flags);
 456                         return new;
 457                 }
 458         }
 459         rc = ll_d_init(de);
 460         if (rc < 0)
 461                 return ERR_PTR(rc);
 462         d_add(de, inode);
 463         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 464                de, de->d_inode, ll_d_count(de), de->d_flags);
 465         return de;
 466 }
 467
 468 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 469                                struct lookup_intent *it,
 470                                struct inode *parent, struct dentry **de)
 471 {
 472         struct inode             *inode = NULL;
 473         __u64                     bits = 0;
 474         int                       rc;
 475         struct dentry *alias;
 476         ENTRY;
 477
 478         /* NB 1 request reference will be taken away by ll_intent_lock()
 479          * when I return */
 480         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 481                it->it_disposition);
 482         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 483                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 484                 if (rc)
 485                         RETURN(rc);
 486
 487                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 488
 489                 /* We used to query real size from OSTs here, but actually
 490                    this is not needed. For stat() calls size would be updated
 491                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 492                    2.4 and
 493                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 494                    Everybody else who needs correct file size would call
 495                    ll_glimpse_size or some equivalent themselves anyway.
 496                    Also see bug 7198. */
 497         }
 498
 499         /* Only hash *de if it is unhashed (new dentry).
 500          * Atoimc_open may passin hashed dentries for open.
 501          */
 502         alias = ll_splice_alias(inode, *de);
 503         if (IS_ERR(alias))
 504                 GOTO(out, rc = PTR_ERR(alias));
 505
 506         *de = alias;
 507
 508         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 509                 /* we have lookup look - unhide dentry */
 510                 if (bits & MDS_INODELOCK_LOOKUP)
 511                         d_lustre_revalidate(*de);
 512         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 513                 /* If file created on server, don't depend on parent UPDATE
 514                  * lock to unhide it. It is left hidden and next lookup can
 515                  * find it in ll_splice_alias.
 516                  */
 517                 /* Check that parent has UPDATE lock. */
 518                 struct lookup_intent parent_it = {
 519                                         .it_op = IT_GETATTR,
 520                                         .it_lock_handle = 0 };
 521                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 522
 523                 /* If it is striped directory, get the real stripe parent */
 524                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 525                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 526                                                  ll_i2info(parent)->lli_lsm_md,
 527                                                  (*de)->d_name.name,
 528                                                  (*de)->d_name.len, &fid);
 529                         if (rc != 0)
 530                                 GOTO(out, rc);
 531                 }
 532
 533                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 534                                        NULL)) {
 535                         d_lustre_revalidate(*de);
 536                         ll_intent_release(&parent_it);
 537                 }
 538         }
 539
 540         GOTO(out, rc = 0);
 541
 542 out:
 543         if (rc != 0 && it->it_op & IT_OPEN)
 544                 ll_open_cleanup((*de)->d_sb, request);
 545
 546         return rc;
 547 }
 548
 549 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 550                                    struct lookup_intent *it)
 551 {
 552         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 553         struct dentry *save = dentry, *retval;
 554         struct ptlrpc_request *req = NULL;
 555         struct md_op_data *op_data = NULL;
 556         __u32 opc;
 557         int rc;
 558         ENTRY;
 559
 560         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 561                 RETURN(ERR_PTR(-ENAMETOOLONG));
 562
 563         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 564                dentry->d_name.len, dentry->d_name.name,
 565                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 566
 567         if (d_mountpoint(dentry))
 568                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 569
 570         if (it == NULL || it->it_op == IT_GETXATTR)
 571                 it = &lookup_it;
 572
 573         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 574                 rc = ll_statahead(parent, &dentry, 0);
 575                 if (rc == 1)
 576                         RETURN(dentry == save ? NULL : dentry);
 577         }
 578
 579         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 580             dentry->d_sb->s_flags & MS_RDONLY)
 581                 RETURN(ERR_PTR(-EROFS));
 582
 583         if (it->it_op & IT_CREAT)
 584                 opc = LUSTRE_OPC_CREATE;
 585         else
 586                 opc = LUSTRE_OPC_ANY;
 587
 588         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 589                                      dentry->d_name.len, 0, opc, NULL);
 590         if (IS_ERR(op_data))
 591                 GOTO(out, retval = ERR_CAST(op_data));
 592
 593         /* enforce umask if acl disabled or MDS doesn't support umask */
 594         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 595                 it->it_create_mode &= ~current_umask();
 596
 597         if (it->it_op & IT_CREAT &&
 598             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 599                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 600                                              &dentry->d_name,
 601                                              &op_data->op_file_secctx_name,
 602                                              &op_data->op_file_secctx,
 603                                              &op_data->op_file_secctx_size);
 604                 if (rc < 0)
 605                         GOTO(out, retval = ERR_PTR(rc));
 606         }
 607
 608         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 609                             &ll_md_blocking_ast, 0);
 610         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 611          * client does not know which suppgid should be sent to the MDS, or
 612          * some other(s) changed the target file's GID after this RPC sent
 613          * to the MDS with the suppgid as the original GID, then we should
 614          * try again with right suppgid. */
 615         if (rc == -EACCES && it->it_op & IT_OPEN &&
 616             it_disposition(it, DISP_OPEN_DENY)) {
 617                 struct mdt_body *body;
 618
 619                 LASSERT(req != NULL);
 620
 621                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 622                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 623                     op_data->op_suppgids[1] == body->mbo_gid ||
 624                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 625                         GOTO(out, retval = ERR_PTR(-EACCES));
 626
 627                 fid_zero(&op_data->op_fid2);
 628                 op_data->op_suppgids[1] = body->mbo_gid;
 629                 ptlrpc_req_finished(req);
 630                 req = NULL;
 631                 ll_intent_release(it);
 632                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 633                                     &ll_md_blocking_ast, 0);
 634         }
 635
 636         if (rc < 0)
 637                 GOTO(out, retval = ERR_PTR(rc));
 638
 639         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 640         if (rc != 0) {
 641                 ll_intent_release(it);
 642                 GOTO(out, retval = ERR_PTR(rc));
 643         }
 644
 645         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 646             !S_ISREG(dentry->d_inode->i_mode) &&
 647             !S_ISDIR(dentry->d_inode->i_mode)) {
 648                 ll_release_openhandle(dentry, it);
 649         }
 650         ll_lookup_finish_locks(it, dentry);
 651
 652         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 653
 654 out:
 655         if (op_data != NULL && !IS_ERR(op_data))
 656                 ll_finish_md_op_data(op_data);
 657
 658         ptlrpc_req_finished(req);
 659         return retval;
 660 }
 661
 662 #ifdef HAVE_IOP_ATOMIC_OPEN
 663 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 664                                    unsigned int flags)
 665 {
 666         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 667         struct dentry *de;
 668
 669         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 670                dentry->d_name.len, dentry->d_name.name,
 671                PFID(ll_inode2fid(parent)), parent, flags);
 672
 673         /*
 674          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 675          * but only if we have write permissions there, otherwise we need
 676          * to proceed with lookup. LU-4185
 677          */
 678         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 679             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 680                 return NULL;
 681
 682         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 683                 itp = NULL;
 684         else
 685                 itp = &it;
 686         de = ll_lookup_it(parent, dentry, itp);
 687
 688         if (itp != NULL)
 689                 ll_intent_release(itp);
 690
 691         return de;
 692 }
 693
 694 /*
 695  * For cached negative dentry and new dentry, handle lookup/create/open
 696  * together.
 697  */
 698 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 699                           struct file *file, unsigned open_flags,
 700                           umode_t mode, int *opened)
 701 {
 702         struct lookup_intent *it;
 703         struct dentry *de;
 704         long long lookup_flags = LOOKUP_OPEN;
 705         int rc = 0;
 706         ENTRY;
 707
 708         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 709                            "open_flags %x, mode %x opened %d\n",
 710                dentry->d_name.len, dentry->d_name.name,
 711                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 712
 713         /* Only negative dentries enter here */
 714         LASSERT(dentry->d_inode == NULL);
 715
 716         if (!d_unhashed(dentry)) {
 717                 /* A valid negative dentry that just passed revalidation,
 718                  * there's little point to try and open it server-side,
 719                  * even though there's a minuscule chance it might succeed.
 720                  * Either way it's a valid race to just return -ENOENT here.
 721                  */
 722                 if (!(open_flags & O_CREAT))
 723                         return -ENOENT;
 724
 725                 /* Otherwise we just unhash it to be rehashed afresh via
 726                  * lookup if necessary
 727                  */
 728                 d_drop(dentry);
 729         }
 730
 731         OBD_ALLOC(it, sizeof(*it));
 732         if (!it)
 733                 RETURN(-ENOMEM);
 734
 735         it->it_op = IT_OPEN;
 736         if (open_flags & O_CREAT) {
 737                 it->it_op |= IT_CREAT;
 738                 lookup_flags |= LOOKUP_CREATE;
 739         }
 740         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 741         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 742         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 743
 744         /* Dentry added to dcache tree in ll_lookup_it */
 745         de = ll_lookup_it(dir, dentry, it);
 746         if (IS_ERR(de))
 747                 rc = PTR_ERR(de);
 748         else if (de != NULL)
 749                 dentry = de;
 750
 751         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 752
 753         if (!rc) {
 754                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 755                         /* Dentry instantiated in ll_create_it. */
 756                         rc = ll_create_it(dir, dentry, it);
 757                         if (rc) {
 758                                 /* We dget in ll_splice_alias. */
 759                                 if (de != NULL)
 760                                         dput(de);
 761                                 goto out_release;
 762                         }
 763
 764                         *opened |= FILE_CREATED;
 765                 }
 766                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 767                         /* Open dentry. */
 768                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 769                                 /* We cannot call open here as it might
 770                                  * deadlock. This case is unreachable in
 771                                  * practice because of OBD_CONNECT_NODEVOH. */
 772                                 rc = finish_no_open(file, de);
 773                         } else {
 774                                 file->private_data = it;
 775                                 rc = finish_open(file, dentry, NULL, opened);
 776                                 /* We dget in ll_splice_alias. finish_open takes
 777                                  * care of dget for fd open.
 778                                  */
 779                                 if (de != NULL)
 780                                         dput(de);
 781                         }
 782                 } else {
 783                         rc = finish_no_open(file, de);
 784                 }
 785         }
 786
 787 out_release:
 788         ll_intent_release(it);
 789         OBD_FREE(it, sizeof(*it));
 790
 791         RETURN(rc);
 792 }
 793
 794 #else /* !HAVE_IOP_ATOMIC_OPEN */
 795 static struct lookup_intent *
 796 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 797 {
 798         struct lookup_intent *it;
 799
 800         OBD_ALLOC_PTR(it);
 801         if (!it)
 802                 return ERR_PTR(-ENOMEM);
 803
 804         if (lookup_flags & LOOKUP_OPEN) {
 805                 it->it_op = IT_OPEN;
 806                 if (lookup_flags & LOOKUP_CREATE)
 807                         it->it_op |= IT_CREAT;
 808                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 809                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 810                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 811         } else {
 812                 it->it_op = IT_GETATTR;
 813         }
 814
 815         return it;
 816 }
 817
 818 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 819                                    struct nameidata *nd)
 820 {
 821         struct dentry *de;
 822         ENTRY;
 823
 824         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 825                 struct lookup_intent *it;
 826
 827                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 828                         it = ll_d2d(dentry)->lld_it;
 829                         ll_d2d(dentry)->lld_it = NULL;
 830                 } else {
 831                         /*
 832                          * Optimize away (CREATE && !OPEN). Let .create handle
 833                          * the race. But only if we have write permissions
 834                          * there, otherwise we need to proceed with lookup.
 835                          * LU-4185
 836                          */
 837                         if ((nd->flags & LOOKUP_CREATE) &&
 838                             !(nd->flags & LOOKUP_OPEN) &&
 839                             (inode_permission(parent,
 840                                               MAY_WRITE | MAY_EXEC) == 0))
 841                                 RETURN(NULL);
 842
 843                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 844                         if (IS_ERR(it))
 845                                 RETURN((struct dentry *)it);
 846                 }
 847
 848                 de = ll_lookup_it(parent, dentry, it);
 849                 if (de)
 850                         dentry = de;
 851                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 852                         if (dentry->d_inode &&
 853                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 854                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 855                                         /* We cannot call open here as it might
 856                                          * deadlock. This case is unreachable in
 857                                          * practice because of
 858                                          * OBD_CONNECT_NODEVOH. */
 859                                 } else {
 860                                         struct file *filp;
 861
 862                                         nd->intent.open.file->private_data = it;
 863                                         filp = lookup_instantiate_filp(nd,
 864                                                                        dentry,
 865                                                                        NULL);
 866                                         if (IS_ERR(filp)) {
 867                                                 if (de)
 868                                                         dput(de);
 869                                                 de = (struct dentry *)filp;
 870                                         }
 871                                 }
 872                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 873                                 // XXX This can only reliably work on assumption
 874                                 // that there are NO hashed negative dentries.
 875                                 ll_d2d(dentry)->lld_it = it;
 876                                 it = NULL; /* Will be freed in ll_create_nd */
 877                                 /* We absolutely depend on ll_create_nd to be
 878                                  * called to not leak this intent and possible
 879                                  * data attached to it */
 880                         }
 881                 }
 882
 883                 if (it) {
 884                         ll_intent_release(it);
 885                         OBD_FREE(it, sizeof(*it));
 886                 }
 887         } else {
 888                 de = ll_lookup_it(parent, dentry, NULL);
 889         }
 890
 891         RETURN(de);
 892 }
 893 #endif /* HAVE_IOP_ATOMIC_OPEN */
 894
 895 /* We depend on "mode" being set with the proper file type/umask by now */
 896 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 897 {
 898         struct inode *inode = NULL;
 899         struct ptlrpc_request *request = NULL;
 900         struct ll_sb_info *sbi = ll_i2sbi(dir);
 901         int rc;
 902         ENTRY;
 903
 904         LASSERT(it && it->it_disposition);
 905
 906         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 907         request = it->it_request;
 908         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 909         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 910         if (rc)
 911                 GOTO(out, inode = ERR_PTR(rc));
 912
 913         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 914
 915         /* We asked for a lock on the directory, but were granted a
 916          * lock on the inode.  Since we finally have an inode pointer,
 917          * stuff it in the lock. */
 918         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 919                PFID(ll_inode2fid(inode)), inode);
 920         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 921         EXIT;
 922  out:
 923         ptlrpc_req_finished(request);
 924         return inode;
 925 }
 926
 927 /*
 928  * By the time this is called, we already have created the directory cache
 929  * entry for the new file, but it is so far negative - it has no inode.
 930  *
 931  * We defer creating the OBD object(s) until open, to keep the intent and
 932  * non-intent code paths similar, and also because we do not have the MDS
 933  * inode number before calling ll_create_node() (which is needed for LOV),
 934  * so we would need to do yet another RPC to the MDS to store the LOV EA
 935  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 936  * lmm_size in datalen (the MDS still has code which will handle that).
 937  *
 938  * If the create succeeds, we fill in the inode information
 939  * with d_instantiate().
 940  */
 941 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 942                         struct lookup_intent *it)
 943 {
 944         struct inode *inode;
 945         int rc = 0;
 946         ENTRY;
 947
 948         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 949                dentry->d_name.len, dentry->d_name.name,
 950                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 951
 952         rc = it_open_error(DISP_OPEN_CREATE, it);
 953         if (rc)
 954                 RETURN(rc);
 955
 956         inode = ll_create_node(dir, it);
 957         if (IS_ERR(inode))
 958                 RETURN(PTR_ERR(inode));
 959
 960         d_instantiate(dentry, inode);
 961
 962         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
 963                 rc = ll_inode_init_security(dentry, inode, dir);
 964                 if (rc)
 965                         RETURN(rc);
 966         }
 967
 968         RETURN(0);
 969 }
 970
 971 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 972 {
 973         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 974                                                        &RMF_MDT_BODY);
 975
 976         LASSERT(body);
 977         if (body->mbo_valid & OBD_MD_FLMTIME &&
 978             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 979                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
 980                        "\n", PFID(ll_inode2fid(inode)),
 981                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 982                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 983         }
 984
 985         if (body->mbo_valid & OBD_MD_FLCTIME &&
 986             body->mbo_ctime > LTIME_S(inode->i_ctime))
 987                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 988 }
 989
 990 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 991                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 992 {
 993         struct qstr *name = &dchild->d_name;
 994         struct ptlrpc_request *request = NULL;
 995         struct md_op_data *op_data;
 996         struct inode *inode = NULL;
 997         struct ll_sb_info *sbi = ll_i2sbi(dir);
 998         int tgt_len = 0;
 999         int err;
1000
1001         ENTRY;
1002         if (unlikely(tgt != NULL))
1003                 tgt_len = strlen(tgt) + 1;
1004
1005 again:
1006         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1007                                      name->len, 0, opc, NULL);
1008         if (IS_ERR(op_data))
1009                 GOTO(err_exit, err = PTR_ERR(op_data));
1010
1011         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1012                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1013                                               &op_data->op_file_secctx_name,
1014                                               &op_data->op_file_secctx,
1015                                               &op_data->op_file_secctx_size);
1016                 if (err < 0)
1017                         GOTO(err_exit, err);
1018         }
1019
1020         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1021                         from_kuid(&init_user_ns, current_fsuid()),
1022                         from_kgid(&init_user_ns, current_fsgid()),
1023                         cfs_curproc_cap_pack(), rdev, &request);
1024         ll_finish_md_op_data(op_data);
1025         op_data = NULL;
1026         if (err < 0 && err != -EREMOTE)
1027                 GOTO(err_exit, err);
1028
1029         /* If the client doesn't know where to create a subdirectory (or
1030          * in case of a race that sends the RPC to the wrong MDS), the
1031          * MDS will return -EREMOTE and the client will fetch the layout
1032          * of the directory, then create the directory on the right MDT. */
1033         if (unlikely(err == -EREMOTE)) {
1034                 struct ll_inode_info    *lli = ll_i2info(dir);
1035                 struct lmv_user_md      *lum;
1036                 int                     lumsize;
1037                 int                     err2;
1038
1039                 ptlrpc_req_finished(request);
1040                 request = NULL;
1041
1042                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1043                                         OBD_MD_DEFAULT_MEA);
1044                 if (err2 == 0) {
1045                         /* Update stripe_offset and retry */
1046                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1047                 } else if (err2 == -ENODATA &&
1048                            lli->lli_def_stripe_offset != -1) {
1049                         /* If there are no default stripe EA on the MDT, but the
1050                          * client has default stripe, then it probably means
1051                          * default stripe EA has just been deleted. */
1052                         lli->lli_def_stripe_offset = -1;
1053                 } else {
1054                         GOTO(err_exit, err);
1055                 }
1056
1057                 ptlrpc_req_finished(request);
1058                 request = NULL;
1059                 goto again;
1060         }
1061
1062         ll_update_times(request, dir);
1063
1064         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1065
1066         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1067         if (err)
1068                 GOTO(err_exit, err);
1069
1070         d_instantiate(dchild, inode);
1071
1072         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1073                 err = ll_inode_init_security(dchild, inode, dir);
1074                 if (err)
1075                         GOTO(err_exit, err);
1076         }
1077
1078         EXIT;
1079 err_exit:
1080         if (request != NULL)
1081                 ptlrpc_req_finished(request);
1082
1083         if (!IS_ERR_OR_NULL(op_data))
1084                 ll_finish_md_op_data(op_data);
1085
1086         return err;
1087 }
1088
1089 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1090                     dev_t rdev)
1091 {
1092         struct qstr *name = &dchild->d_name;
1093         int err;
1094         ENTRY;
1095
1096         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1097                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1098                mode, rdev);
1099
1100         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1101                 mode &= ~current_umask();
1102
1103         switch (mode & S_IFMT) {
1104         case 0:
1105                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1106         case S_IFREG:
1107         case S_IFCHR:
1108         case S_IFBLK:
1109         case S_IFIFO:
1110         case S_IFSOCK:
1111                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1112                                   LUSTRE_OPC_MKNOD);
1113                 break;
1114         case S_IFDIR:
1115                 err = -EPERM;
1116                 break;
1117         default:
1118                 err = -EINVAL;
1119         }
1120
1121         if (!err)
1122                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1123
1124         RETURN(err);
1125 }
1126
1127 #ifdef HAVE_IOP_ATOMIC_OPEN
1128 /*
1129  * Plain create. Intent create is handled in atomic_open.
1130  */
1131 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1132                         umode_t mode, bool want_excl)
1133 {
1134         int rc;
1135
1136         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1137
1138         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1139                            "flags=%u, excl=%d\n", dentry->d_name.len,
1140                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1141                dir, mode, want_excl);
1142
1143         rc = ll_mknod(dir, dentry, mode, 0);
1144
1145         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1146
1147         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1148                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1149
1150         return rc;
1151 }
1152 #else /* !HAVE_IOP_ATOMIC_OPEN */
1153 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1154                         ll_umode_t mode, struct nameidata *nd)
1155 {
1156         struct ll_dentry_data *lld = ll_d2d(dentry);
1157         struct lookup_intent *it = NULL;
1158         int rc;
1159
1160         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1161
1162         if (lld != NULL)
1163                 it = lld->lld_it;
1164
1165         if (!it)
1166                 return ll_mknod(dir, dentry, mode, 0);
1167
1168         lld->lld_it = NULL;
1169
1170         /* Was there an error? Propagate it! */
1171         if (it->it_status) {
1172                 rc = it->it_status;
1173                 goto out;
1174         }
1175
1176         rc = ll_create_it(dir, dentry, it);
1177         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1178                 struct file *filp;
1179
1180                 nd->intent.open.file->private_data = it;
1181                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1182                 if (IS_ERR(filp))
1183                         rc = PTR_ERR(filp);
1184         }
1185
1186 out:
1187         ll_intent_release(it);
1188         OBD_FREE(it, sizeof(*it));
1189
1190         if (!rc)
1191                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1192
1193         return rc;
1194 }
1195 #endif /* HAVE_IOP_ATOMIC_OPEN */
1196
1197 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1198                       const char *oldpath)
1199 {
1200         struct qstr *name = &dchild->d_name;
1201         int err;
1202         ENTRY;
1203
1204         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1205                name->len, name->name, PFID(ll_inode2fid(dir)),
1206                dir, 3000, oldpath);
1207
1208         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1209                           LUSTRE_OPC_SYMLINK);
1210
1211         if (!err)
1212                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1213
1214         RETURN(err);
1215 }
1216
1217 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1218                    struct dentry *new_dentry)
1219 {
1220         struct inode *src = old_dentry->d_inode;
1221         struct qstr *name = &new_dentry->d_name;
1222         struct ll_sb_info *sbi = ll_i2sbi(dir);
1223         struct ptlrpc_request *request = NULL;
1224         struct md_op_data *op_data;
1225         int err;
1226
1227         ENTRY;
1228         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1229                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1230                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1231
1232         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1233                                      0, LUSTRE_OPC_ANY, NULL);
1234         if (IS_ERR(op_data))
1235                 RETURN(PTR_ERR(op_data));
1236
1237         err = md_link(sbi->ll_md_exp, op_data, &request);
1238         ll_finish_md_op_data(op_data);
1239         if (err)
1240                 GOTO(out, err);
1241
1242         ll_update_times(request, dir);
1243         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1244         EXIT;
1245 out:
1246         ptlrpc_req_finished(request);
1247         RETURN(err);
1248 }
1249
1250 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1251 {
1252         struct qstr *name = &dchild->d_name;
1253         int err;
1254         ENTRY;
1255
1256         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1257                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1258
1259         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1260                 mode &= ~current_umask();
1261
1262         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1263
1264         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1265         if (err == 0)
1266                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1267
1268         RETURN(err);
1269 }
1270
1271 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1272 {
1273         struct qstr *name = &dchild->d_name;
1274         struct ptlrpc_request *request = NULL;
1275         struct md_op_data *op_data;
1276         int rc;
1277         ENTRY;
1278
1279         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1280                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1281
1282         if (unlikely(d_mountpoint(dchild)))
1283                 RETURN(-EBUSY);
1284
1285         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1286                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1287         if (IS_ERR(op_data))
1288                 RETURN(PTR_ERR(op_data));
1289
1290         if (dchild->d_inode != NULL)
1291                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1292
1293         op_data->op_fid2 = op_data->op_fid3;
1294         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1295         ll_finish_md_op_data(op_data);
1296         if (rc == 0) {
1297                 ll_update_times(request, dir);
1298                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1299         }
1300
1301         ptlrpc_req_finished(request);
1302         RETURN(rc);
1303 }
1304
1305 /**
1306  * Remove dir entry
1307  **/
1308 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1309 {
1310         struct ptlrpc_request *request = NULL;
1311         struct md_op_data *op_data;
1312         int rc;
1313         ENTRY;
1314
1315         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1316                namelen, name, PFID(ll_inode2fid(dir)), dir);
1317
1318         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1319                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1320         if (IS_ERR(op_data))
1321                 RETURN(PTR_ERR(op_data));
1322         op_data->op_cli_flags |= CLI_RM_ENTRY;
1323         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1324         ll_finish_md_op_data(op_data);
1325         if (rc == 0) {
1326                 ll_update_times(request, dir);
1327                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1328         }
1329
1330         ptlrpc_req_finished(request);
1331         RETURN(rc);
1332 }
1333
1334 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1335 {
1336         struct qstr *name = &dchild->d_name;
1337         struct ptlrpc_request *request = NULL;
1338         struct md_op_data *op_data;
1339         int rc;
1340         ENTRY;
1341         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1342                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1343
1344         /*
1345          * XXX: unlink bind mountpoint maybe call to here,
1346          * just check it as vfs_unlink does.
1347          */
1348         if (unlikely(d_mountpoint(dchild)))
1349                 RETURN(-EBUSY);
1350
1351         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1352                                      LUSTRE_OPC_ANY, NULL);
1353         if (IS_ERR(op_data))
1354                 RETURN(PTR_ERR(op_data));
1355
1356         if (dchild->d_inode != NULL)
1357                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1358
1359         op_data->op_fid2 = op_data->op_fid3;
1360         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1361         ll_finish_md_op_data(op_data);
1362         if (rc)
1363                 GOTO(out, rc);
1364
1365         ll_update_times(request, dir);
1366         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1367
1368  out:
1369         ptlrpc_req_finished(request);
1370         RETURN(rc);
1371 }
1372
1373 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1374                      struct inode *tgt, struct dentry *tgt_dchild)
1375 {
1376         struct qstr *src_name = &src_dchild->d_name;
1377         struct qstr *tgt_name = &tgt_dchild->d_name;
1378         struct ptlrpc_request *request = NULL;
1379         struct ll_sb_info *sbi = ll_i2sbi(src);
1380         struct md_op_data *op_data;
1381         int err;
1382         ENTRY;
1383         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1384                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1385                src_name->len, src_name->name,
1386                PFID(ll_inode2fid(src)), src, tgt_name->len,
1387                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1388
1389         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1390                 RETURN(-EBUSY);
1391
1392         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1393                                      LUSTRE_OPC_ANY, NULL);
1394         if (IS_ERR(op_data))
1395                 RETURN(PTR_ERR(op_data));
1396
1397         if (src_dchild->d_inode != NULL)
1398                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1399
1400         if (tgt_dchild->d_inode != NULL)
1401                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1402
1403         err = md_rename(sbi->ll_md_exp, op_data,
1404                         src_name->name, src_name->len,
1405                         tgt_name->name, tgt_name->len, &request);
1406         ll_finish_md_op_data(op_data);
1407         if (!err) {
1408                 ll_update_times(request, src);
1409                 ll_update_times(request, tgt);
1410                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1411         }
1412
1413         ptlrpc_req_finished(request);
1414
1415         if (err == 0)
1416                 d_move(src_dchild, tgt_dchild);
1417
1418         RETURN(err);
1419 }
1420
1421 const struct inode_operations ll_dir_inode_operations = {
1422         .mknod              = ll_mknod,
1423 #ifdef HAVE_IOP_ATOMIC_OPEN
1424         .atomic_open        = ll_atomic_open,
1425 #endif
1426         .lookup             = ll_lookup_nd,
1427         .create             = ll_create_nd,
1428         /* We need all these non-raw things for NFSD, to not patch it. */
1429         .unlink             = ll_unlink,
1430         .mkdir              = ll_mkdir,
1431         .rmdir              = ll_rmdir,
1432         .symlink            = ll_symlink,
1433         .link               = ll_link,
1434         .rename             = ll_rename,
1435         .setattr            = ll_setattr,
1436         .getattr            = ll_getattr,
1437         .permission         = ll_inode_permission,
1438         .setxattr           = ll_setxattr,
1439         .getxattr           = ll_getxattr,
1440         .listxattr          = ll_listxattr,
1441         .removexattr        = ll_removexattr,
1442 #ifdef HAVE_IOP_GET_ACL
1443         .get_acl            = ll_get_acl,
1444 #endif
1445 };
1446
1447 const struct inode_operations ll_special_inode_operations = {
1448         .setattr        = ll_setattr,
1449         .getattr        = ll_getattr,
1450         .permission     = ll_inode_permission,
1451         .setxattr       = ll_setxattr,
1452         .getxattr       = ll_getxattr,
1453         .listxattr      = ll_listxattr,
1454         .removexattr    = ll_removexattr,
1455 #ifdef HAVE_IOP_GET_ACL
1456         .get_acl            = ll_get_acl,
1457 #endif
1458 };