kernel/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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include "../include/obd_support.h"
  48 #include "../include/lustre_fid.h"
  49 #include "../include/lustre_lite.h"
  50 #include "../include/lustre_dlm.h"
  51 #include "../include/lustre_ver.h"
  52 #include "llite_internal.h"
  53
  54 static int ll_create_it(struct inode *, struct dentry *,
  55                         int, struct lookup_intent *);
  56
  57 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
  58 static int ll_test_inode(struct inode *inode, void *opaque)
  59 {
  60         struct ll_inode_info *lli = ll_i2info(inode);
  61         struct lustre_md     *md = opaque;
  62
  63         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  64                 CERROR("MDS body missing FID\n");
  65                 return 0;
  66         }
  67
  68         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
  69                 return 0;
  70
  71         return 1;
  72 }
  73
  74 static int ll_set_inode(struct inode *inode, void *opaque)
  75 {
  76         struct ll_inode_info *lli = ll_i2info(inode);
  77         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  78
  79         if (unlikely(!(body->valid & OBD_MD_FLID))) {
  80                 CERROR("MDS body missing FID\n");
  81                 return -EINVAL;
  82         }
  83
  84         lli->lli_fid = body->fid1;
  85         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
  86                 CERROR("Can not initialize inode " DFID
  87                        " without object type: valid = %#llx\n",
  88                        PFID(&lli->lli_fid), body->valid);
  89                 return -EINVAL;
  90         }
  91
  92         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
  93         if (unlikely(inode->i_mode == 0)) {
  94                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  95                 return -EINVAL;
  96         }
  97
  98         ll_lli_init(lli);
  99
 100         return 0;
 101 }
 102
 103
 104 /*
 105  * Get an inode by inode number (already instantiated by the intent lookup).
 106  * Returns inode or NULL
 107  */
 108 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 109                       struct lustre_md *md)
 110 {
 111         struct inode     *inode;
 112
 113         LASSERT(hash != 0);
 114         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 115
 116         if (inode) {
 117                 if (inode->i_state & I_NEW) {
 118                         int rc = 0;
 119
 120                         ll_read_inode2(inode, md);
 121                         if (S_ISREG(inode->i_mode) &&
 122                             ll_i2info(inode)->lli_clob == NULL) {
 123                                 CDEBUG(D_INODE,
 124                                         "%s: apply lsm %p to inode "DFID".\n",
 125                                         ll_get_fsname(sb, NULL, 0), md->lsm,
 126                                         PFID(ll_inode2fid(inode)));
 127                                 rc = cl_file_inode_init(inode, md);
 128                         }
 129                         if (rc != 0) {
 130                                 make_bad_inode(inode);
 131                                 unlock_new_inode(inode);
 132                                 iput(inode);
 133                                 inode = ERR_PTR(rc);
 134                         } else
 135                                 unlock_new_inode(inode);
 136                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 137                         ll_update_inode(inode, md);
 138                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 139                        inode, PFID(&md->body->fid1));
 140         }
 141         return inode;
 142 }
 143
 144 static void ll_invalidate_negative_children(struct inode *dir)
 145 {
 146         struct dentry *dentry, *tmp_subdir;
 147         struct ll_d_hlist_node *p;
 148
 149         ll_lock_dcache(dir);
 150         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_u.d_alias) {
 151                 spin_lock(&dentry->d_lock);
 152                 if (!list_empty(&dentry->d_subdirs)) {
 153                         struct dentry *child;
 154
 155                         list_for_each_entry_safe(child, tmp_subdir,
 156                                                  &dentry->d_subdirs,
 157                                                  d_child) {
 158                                 if (d_really_is_negative(child))
 159                                         d_lustre_invalidate(child, 1);
 160                         }
 161                 }
 162                 spin_unlock(&dentry->d_lock);
 163         }
 164         ll_unlock_dcache(dir);
 165 }
 166
 167 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 168                        void *data, int flag)
 169 {
 170         struct lustre_handle lockh;
 171         int rc;
 172
 173         switch (flag) {
 174         case LDLM_CB_BLOCKING:
 175                 ldlm_lock2handle(lock, &lockh);
 176                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 177                 if (rc < 0) {
 178                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 179                         return rc;
 180                 }
 181                 break;
 182         case LDLM_CB_CANCELING: {
 183                 struct inode *inode = ll_inode_from_resource_lock(lock);
 184                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 185
 186                 /* Inode is set to lock->l_resource->lr_lvb_inode
 187                  * for mdc - bug 24555 */
 188                 LASSERT(lock->l_ast_data == NULL);
 189
 190                 if (inode == NULL)
 191                         break;
 192
 193                 /* Invalidate all dentries associated with this inode */
 194                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 195
 196                 if (!fid_res_name_eq(ll_inode2fid(inode),
 197                                      &lock->l_resource->lr_name)) {
 198                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 199                                    PFID(ll_inode2fid(inode)), inode);
 200                         LBUG();
 201                 }
 202
 203                 if (bits & MDS_INODELOCK_XATTR) {
 204                         ll_xattr_cache_destroy(inode);
 205                         bits &= ~MDS_INODELOCK_XATTR;
 206                 }
 207
 208                 /* For OPEN locks we differentiate between lock modes
 209                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 210                 if (bits & MDS_INODELOCK_OPEN)
 211                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 212
 213                 if (bits & MDS_INODELOCK_OPEN) {
 214                         fmode_t fmode;
 215
 216                         switch (lock->l_req_mode) {
 217                         case LCK_CW:
 218                                 fmode = FMODE_WRITE;
 219                                 break;
 220                         case LCK_PR:
 221                                 fmode = FMODE_EXEC;
 222                                 break;
 223                         case LCK_CR:
 224                                 fmode = FMODE_READ;
 225                                 break;
 226                         default:
 227                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 228                                 LBUG();
 229                         }
 230
 231                         ll_md_real_close(inode, fmode);
 232                 }
 233
 234                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 235                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 236                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 237
 238                 if (bits & MDS_INODELOCK_LAYOUT) {
 239                         struct cl_object_conf conf = {
 240                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 241                                 .coc_inode = inode,
 242                         };
 243
 244                         rc = ll_layout_conf(inode, &conf);
 245                         if (rc < 0)
 246                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 247                                        DFID": rc = %d\n",
 248                                        PFID(ll_inode2fid(inode)), rc);
 249                 }
 250
 251                 if (bits & MDS_INODELOCK_UPDATE) {
 252                         struct ll_inode_info *lli = ll_i2info(inode);
 253
 254                         spin_lock(&lli->lli_lock);
 255                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 256                         spin_unlock(&lli->lli_lock);
 257                 }
 258
 259                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 260                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 261                                inode->i_ino);
 262                         truncate_inode_pages(inode->i_mapping, 0);
 263                         ll_invalidate_negative_children(inode);
 264                 }
 265
 266                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 267                     inode->i_sb->s_root != NULL &&
 268                     !is_root_inode(inode))
 269                         ll_invalidate_aliases(inode);
 270
 271                 iput(inode);
 272                 break;
 273         }
 274         default:
 275                 LBUG();
 276         }
 277
 278         return 0;
 279 }
 280
 281 __u32 ll_i2suppgid(struct inode *i)
 282 {
 283         if (in_group_p(i->i_gid))
 284                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 285         else
 286                 return (__u32)(-1);
 287 }
 288
 289 /* Pack the required supplementary groups into the supplied groups array.
 290  * If we don't need to use the groups from the target inode(s) then we
 291  * instead pack one or more groups from the user's supplementary group
 292  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 293 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 294 {
 295 #if 0
 296         int i;
 297 #endif
 298
 299         LASSERT(i1 != NULL);
 300         LASSERT(suppgids != NULL);
 301
 302         suppgids[0] = ll_i2suppgid(i1);
 303
 304         if (i2)
 305                 suppgids[1] = ll_i2suppgid(i2);
 306                 else
 307                         suppgids[1] = -1;
 308
 309 #if 0
 310         for (i = 0; i < current_ngroups; i++) {
 311                 if (suppgids[0] == -1) {
 312                         if (current_groups[i] != suppgids[1])
 313                                 suppgids[0] = current_groups[i];
 314                         continue;
 315                 }
 316                 if (suppgids[1] == -1) {
 317                         if (current_groups[i] != suppgids[0])
 318                                 suppgids[1] = current_groups[i];
 319                         continue;
 320                 }
 321                 break;
 322         }
 323 #endif
 324 }
 325
 326 /*
 327  * try to reuse three types of dentry:
 328  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 329  *    by concurrent .revalidate).
 330  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 331  *    be cleared by others calling d_lustre_revalidate).
 332  * 3. DISCONNECTED alias.
 333  */
 334 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 335 {
 336         struct dentry *alias, *discon_alias, *invalid_alias;
 337         struct ll_d_hlist_node *p;
 338
 339         if (ll_d_hlist_empty(&inode->i_dentry))
 340                 return NULL;
 341
 342         discon_alias = invalid_alias = NULL;
 343
 344         ll_lock_dcache(inode);
 345         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_u.d_alias) {
 346                 LASSERT(alias != dentry);
 347
 348                 spin_lock(&alias->d_lock);
 349                 if (alias->d_flags & DCACHE_DISCONNECTED)
 350                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 351                         discon_alias = alias;
 352                 else if (alias->d_parent == dentry->d_parent         &&
 353                          alias->d_name.hash == dentry->d_name.hash       &&
 354                          alias->d_name.len == dentry->d_name.len         &&
 355                          memcmp(alias->d_name.name, dentry->d_name.name,
 356                                 dentry->d_name.len) == 0)
 357                         invalid_alias = alias;
 358                 spin_unlock(&alias->d_lock);
 359
 360                 if (invalid_alias)
 361                         break;
 362         }
 363         alias = invalid_alias ?: discon_alias ?: NULL;
 364         if (alias) {
 365                 spin_lock(&alias->d_lock);
 366                 dget_dlock(alias);
 367                 spin_unlock(&alias->d_lock);
 368         }
 369         ll_unlock_dcache(inode);
 370
 371         return alias;
 372 }
 373
 374 /*
 375  * Similar to d_splice_alias(), but lustre treats invalid alias
 376  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 377  */
 378 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 379 {
 380         struct dentry *new;
 381         int rc;
 382
 383         if (inode) {
 384                 new = ll_find_alias(inode, de);
 385                 if (new) {
 386                         rc = ll_d_init(new);
 387                         if (rc < 0) {
 388                                 dput(new);
 389                                 return ERR_PTR(rc);
 390                         }
 391                         d_move(new, de);
 392                         iput(inode);
 393                         CDEBUG(D_DENTRY,
 394                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 395                               new, d_inode(new), d_count(new), new->d_flags);
 396                         return new;
 397                 }
 398         }
 399         rc = ll_d_init(de);
 400         if (rc < 0)
 401                 return ERR_PTR(rc);
 402         d_add(de, inode);
 403         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 404                de, d_inode(de), d_count(de), de->d_flags);
 405         return de;
 406 }
 407
 408 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 409                                struct lookup_intent *it,
 410                                struct inode *parent, struct dentry **de)
 411 {
 412         struct inode *inode = NULL;
 413         __u64 bits = 0;
 414         int rc;
 415
 416         /* NB 1 request reference will be taken away by ll_intent_lock()
 417          * when I return */
 418         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 419                it->d.lustre.it_disposition);
 420         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 421                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 422                 if (rc)
 423                         return rc;
 424
 425                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 426
 427                 /* We used to query real size from OSTs here, but actually
 428                    this is not needed. For stat() calls size would be updated
 429                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 430                    2.4 and
 431                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 432                    Everybody else who needs correct file size would call
 433                    ll_glimpse_size or some equivalent themselves anyway.
 434                    Also see bug 7198. */
 435         }
 436
 437         /* Only hash *de if it is unhashed (new dentry).
 438          * Atoimc_open may passing hashed dentries for open.
 439          */
 440         if (d_unhashed(*de)) {
 441                 struct dentry *alias;
 442
 443                 alias = ll_splice_alias(inode, *de);
 444                 if (IS_ERR(alias))
 445                         return PTR_ERR(alias);
 446                 *de = alias;
 447         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 448                    !it_disposition(it, DISP_OPEN_CREATE)) {
 449                 /* With DISP_OPEN_CREATE dentry will
 450                    instantiated in ll_create_it. */
 451                 LASSERT(d_inode(*de) == NULL);
 452                 d_instantiate(*de, inode);
 453         }
 454
 455         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 456                 /* we have lookup look - unhide dentry */
 457                 if (bits & MDS_INODELOCK_LOOKUP)
 458                         d_lustre_revalidate(*de);
 459         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 460                 /* If file created on server, don't depend on parent UPDATE
 461                  * lock to unhide it. It is left hidden and next lookup can
 462                  * find it in ll_splice_alias.
 463                  */
 464                 /* Check that parent has UPDATE lock. */
 465                 struct lookup_intent parent_it = {
 466                                         .it_op = IT_GETATTR,
 467                                         .d.lustre.it_lock_handle = 0 };
 468
 469                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 470                                        &ll_i2info(parent)->lli_fid, NULL)) {
 471                         d_lustre_revalidate(*de);
 472                         ll_intent_release(&parent_it);
 473                 }
 474         }
 475
 476         return 0;
 477 }
 478
 479 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 480                                    struct lookup_intent *it, int lookup_flags)
 481 {
 482         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 483         struct dentry *save = dentry, *retval;
 484         struct ptlrpc_request *req = NULL;
 485         struct inode *inode;
 486         struct md_op_data *op_data;
 487         __u32 opc;
 488         int rc;
 489
 490         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 491                 return ERR_PTR(-ENAMETOOLONG);
 492
 493         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 494                dentry, parent->i_ino,
 495                parent->i_generation, parent, LL_IT2STR(it));
 496
 497         if (d_mountpoint(dentry))
 498                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 499
 500         if (it == NULL || it->it_op == IT_GETXATTR)
 501                 it = &lookup_it;
 502
 503         if (it->it_op == IT_GETATTR) {
 504                 rc = ll_statahead_enter(parent, &dentry, 0);
 505                 if (rc == 1) {
 506                         if (dentry == save)
 507                                 retval = NULL;
 508                         else
 509                                 retval = dentry;
 510                         goto out;
 511                 }
 512         }
 513
 514         if (it->it_op & IT_CREAT)
 515                 opc = LUSTRE_OPC_CREATE;
 516         else
 517                 opc = LUSTRE_OPC_ANY;
 518
 519         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 520                                      dentry->d_name.len, lookup_flags, opc,
 521                                      NULL);
 522         if (IS_ERR(op_data))
 523                 return (void *)op_data;
 524
 525         /* enforce umask if acl disabled or MDS doesn't support umask */
 526         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 527                 it->it_create_mode &= ~current_umask();
 528
 529         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 530                             lookup_flags, &req, ll_md_blocking_ast, 0);
 531         ll_finish_md_op_data(op_data);
 532         if (rc < 0) {
 533                 retval = ERR_PTR(rc);
 534                 goto out;
 535         }
 536
 537         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 538         if (rc != 0) {
 539                 ll_intent_release(it);
 540                 retval = ERR_PTR(rc);
 541                 goto out;
 542         }
 543
 544         inode = d_inode(dentry);
 545         if ((it->it_op & IT_OPEN) && inode &&
 546             !S_ISREG(inode->i_mode) &&
 547             !S_ISDIR(inode->i_mode)) {
 548                 ll_release_openhandle(inode, it);
 549         }
 550         ll_lookup_finish_locks(it, inode);
 551
 552         if (dentry == save)
 553                 retval = NULL;
 554         else
 555                 retval = dentry;
 556         goto out;
 557  out:
 558         if (req)
 559                 ptlrpc_req_finished(req);
 560         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 561                 ll_statahead_mark(parent, dentry);
 562         return retval;
 563 }
 564
 565 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 566                                    unsigned int flags)
 567 {
 568         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 569         struct dentry *de;
 570
 571         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u\n",
 572                dentry, parent->i_ino,
 573                parent->i_generation, parent, flags);
 574
 575         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 576         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 577                 return NULL;
 578
 579         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 580                 itp = NULL;
 581         else
 582                 itp = &it;
 583         de = ll_lookup_it(parent, dentry, itp, 0);
 584
 585         if (itp != NULL)
 586                 ll_intent_release(itp);
 587
 588         return de;
 589 }
 590
 591 /*
 592  * For cached negative dentry and new dentry, handle lookup/create/open
 593  * together.
 594  */
 595 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 596                           struct file *file, unsigned open_flags,
 597                           umode_t mode, int *opened)
 598 {
 599         struct lookup_intent *it;
 600         struct dentry *de;
 601         long long lookup_flags = LOOKUP_OPEN;
 602         int rc = 0;
 603
 604         CDEBUG(D_VFSTRACE,
 605                "VFS Op:name=%pd,dir=%lu/%u(%p),file %p,open_flags %x,mode %x opened %d\n",
 606                dentry, dir->i_ino,
 607                dir->i_generation, dir, file, open_flags, mode, *opened);
 608
 609         it = kzalloc(sizeof(*it), GFP_NOFS);
 610         if (!it)
 611                 return -ENOMEM;
 612
 613         it->it_op = IT_OPEN;
 614         if (open_flags & O_CREAT) {
 615                 it->it_op |= IT_CREAT;
 616                 lookup_flags |= LOOKUP_CREATE;
 617         }
 618         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 619         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 620
 621         /* Dentry added to dcache tree in ll_lookup_it */
 622         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 623         if (IS_ERR(de))
 624                 rc = PTR_ERR(de);
 625         else if (de != NULL)
 626                 dentry = de;
 627
 628         if (!rc) {
 629                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 630                         /* Dentry instantiated in ll_create_it. */
 631                         rc = ll_create_it(dir, dentry, mode, it);
 632                         if (rc) {
 633                                 /* We dget in ll_splice_alias. */
 634                                 if (de != NULL)
 635                                         dput(de);
 636                                 goto out_release;
 637                         }
 638
 639                         *opened |= FILE_CREATED;
 640                 }
 641                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 642                         /* Open dentry. */
 643                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 644                                 /* We cannot call open here as it would
 645                                  * deadlock.
 646                                  */
 647                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 648                                         ptlrpc_req_finished(
 649                                                        (struct ptlrpc_request *)
 650                                                           it->d.lustre.it_data);
 651                                 rc = finish_no_open(file, de);
 652                         } else {
 653                                 file->private_data = it;
 654                                 rc = finish_open(file, dentry, NULL, opened);
 655                                 /* We dget in ll_splice_alias. finish_open takes
 656                                  * care of dget for fd open.
 657                                  */
 658                                 if (de != NULL)
 659                                         dput(de);
 660                         }
 661                 } else {
 662                         rc = finish_no_open(file, de);
 663                 }
 664         }
 665
 666 out_release:
 667         ll_intent_release(it);
 668         OBD_FREE(it, sizeof(*it));
 669
 670         return rc;
 671 }
 672
 673
 674 /* We depend on "mode" being set with the proper file type/umask by now */
 675 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 676 {
 677         struct inode *inode = NULL;
 678         struct ptlrpc_request *request = NULL;
 679         struct ll_sb_info *sbi = ll_i2sbi(dir);
 680         int rc;
 681
 682         LASSERT(it && it->d.lustre.it_disposition);
 683
 684         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 685         request = it->d.lustre.it_data;
 686         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 687         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 688         if (rc) {
 689                 inode = ERR_PTR(rc);
 690                 goto out;
 691         }
 692
 693         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 694
 695         /* We asked for a lock on the directory, but were granted a
 696          * lock on the inode.  Since we finally have an inode pointer,
 697          * stuff it in the lock. */
 698         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 699                inode, inode->i_ino, inode->i_generation);
 700         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 701  out:
 702         ptlrpc_req_finished(request);
 703         return inode;
 704 }
 705
 706 /*
 707  * By the time this is called, we already have created the directory cache
 708  * entry for the new file, but it is so far negative - it has no inode.
 709  *
 710  * We defer creating the OBD object(s) until open, to keep the intent and
 711  * non-intent code paths similar, and also because we do not have the MDS
 712  * inode number before calling ll_create_node() (which is needed for LOV),
 713  * so we would need to do yet another RPC to the MDS to store the LOV EA
 714  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 715  * lmm_size in datalen (the MDS still has code which will handle that).
 716  *
 717  * If the create succeeds, we fill in the inode information
 718  * with d_instantiate().
 719  */
 720 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 721                         struct lookup_intent *it)
 722 {
 723         struct inode *inode;
 724         int rc = 0;
 725
 726         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 727                dentry, dir->i_ino,
 728                dir->i_generation, dir, LL_IT2STR(it));
 729
 730         rc = it_open_error(DISP_OPEN_CREATE, it);
 731         if (rc)
 732                 return rc;
 733
 734         inode = ll_create_node(dir, it);
 735         if (IS_ERR(inode))
 736                 return PTR_ERR(inode);
 737
 738         d_instantiate(dentry, inode);
 739         return 0;
 740 }
 741
 742 static void ll_update_times(struct ptlrpc_request *request,
 743                             struct inode *inode)
 744 {
 745         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 746                                                        &RMF_MDT_BODY);
 747
 748         LASSERT(body);
 749         if (body->valid & OBD_MD_FLMTIME &&
 750             body->mtime > LTIME_S(inode->i_mtime)) {
 751                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
 752                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 753                 LTIME_S(inode->i_mtime) = body->mtime;
 754         }
 755         if (body->valid & OBD_MD_FLCTIME &&
 756             body->ctime > LTIME_S(inode->i_ctime))
 757                 LTIME_S(inode->i_ctime) = body->ctime;
 758 }
 759
 760 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 761                        const char *tgt, int mode, int rdev,
 762                        __u32 opc)
 763 {
 764         struct ptlrpc_request *request = NULL;
 765         struct md_op_data *op_data;
 766         struct inode *inode = NULL;
 767         struct ll_sb_info *sbi = ll_i2sbi(dir);
 768         int tgt_len = 0;
 769         int err;
 770
 771         if (unlikely(tgt != NULL))
 772                 tgt_len = strlen(tgt) + 1;
 773
 774         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 775                                      dentry->d_name.name,
 776                                      dentry->d_name.len,
 777                                      0, opc, NULL);
 778         if (IS_ERR(op_data)) {
 779                 err = PTR_ERR(op_data);
 780                 goto err_exit;
 781         }
 782
 783         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 784                         from_kuid(&init_user_ns, current_fsuid()),
 785                         from_kgid(&init_user_ns, current_fsgid()),
 786                         cfs_curproc_cap_pack(), rdev, &request);
 787         ll_finish_md_op_data(op_data);
 788         if (err)
 789                 goto err_exit;
 790
 791         ll_update_times(request, dir);
 792
 793         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 794         if (err)
 795                 goto err_exit;
 796
 797         d_instantiate(dentry, inode);
 798 err_exit:
 799         ptlrpc_req_finished(request);
 800
 801         return err;
 802 }
 803
 804 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 805                     umode_t mode, dev_t rdev)
 806 {
 807         int err;
 808
 809         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p) mode %o dev %x\n",
 810                dchild, dir->i_ino, dir->i_generation, dir,
 811                mode, old_encode_dev(rdev));
 812
 813         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 814                 mode &= ~current_umask();
 815
 816         switch (mode & S_IFMT) {
 817         case 0:
 818                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 819         case S_IFREG:
 820         case S_IFCHR:
 821         case S_IFBLK:
 822         case S_IFIFO:
 823         case S_IFSOCK:
 824                 err = ll_new_node(dir, dchild, NULL, mode,
 825                                   old_encode_dev(rdev),
 826                                   LUSTRE_OPC_MKNOD);
 827                 break;
 828         case S_IFDIR:
 829                 err = -EPERM;
 830                 break;
 831         default:
 832                 err = -EINVAL;
 833         }
 834
 835         if (!err)
 836                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 837
 838         return err;
 839 }
 840
 841 /*
 842  * Plain create. Intent create is handled in atomic_open.
 843  */
 844 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 845                         umode_t mode, bool want_excl)
 846 {
 847         int rc;
 848
 849         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u, excl=%d\n",
 850                dentry, dir->i_ino,
 851                dir->i_generation, dir, mode, want_excl);
 852
 853         rc = ll_mknod(dir, dentry, mode, 0);
 854
 855         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 856
 857         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 858                dentry, d_unhashed(dentry));
 859
 860         return rc;
 861 }
 862
 863 static inline void ll_get_child_fid(struct dentry *child, struct lu_fid *fid)
 864 {
 865         if (d_really_is_positive(child))
 866                 *fid = *ll_inode2fid(d_inode(child));
 867 }
 868
 869 /**
 870  * Remove dir entry
 871  **/
 872 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
 873 {
 874         struct ptlrpc_request *request = NULL;
 875         struct md_op_data *op_data;
 876         int rc;
 877
 878         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
 879                namelen, name, dir->i_ino, dir->i_generation, dir);
 880
 881         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
 882                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
 883         if (IS_ERR(op_data))
 884                 return PTR_ERR(op_data);
 885         op_data->op_cli_flags |= CLI_RM_ENTRY;
 886         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 887         ll_finish_md_op_data(op_data);
 888         if (rc == 0) {
 889                 ll_update_times(request, dir);
 890                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
 891         }
 892
 893         ptlrpc_req_finished(request);
 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         struct obd_capa *oc = NULL;
 905         int rc;
 906
 907         /* req is swabbed so this is safe */
 908         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 909         if (!(body->valid & OBD_MD_FLEASIZE))
 910                 return 0;
 911
 912         if (body->eadatasize == 0) {
 913                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 914                 rc = -EPROTO;
 915                 goto out;
 916         }
 917
 918         /* The MDS sent back the EA because we unlinked the last reference
 919          * to this file. Use this EA to unlink the objects on the OST.
 920          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 921          * check it is complete and sensible. */
 922         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 923                                               body->eadatasize);
 924         LASSERT(eadata != NULL);
 925
 926         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 927         if (rc < 0) {
 928                 CERROR("obd_unpackmd: %d\n", rc);
 929                 goto out;
 930         }
 931         LASSERT(rc >= sizeof(*lsm));
 932
 933         OBDO_ALLOC(oa);
 934         if (oa == NULL) {
 935                 rc = -ENOMEM;
 936                 goto out_free_memmd;
 937         }
 938
 939         oa->o_oi = lsm->lsm_oi;
 940         oa->o_mode = body->mode & S_IFMT;
 941         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 942
 943         if (body->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 == NULL) {
 951                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 952                         body->valid &= ~OBD_MD_FLCOOKIE;
 953                 }
 954         }
 955
 956         if (body->valid & OBD_MD_FLOSSCAPA) {
 957                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
 958                 if (rc)
 959                         goto out_free_memmd;
 960         }
 961
 962         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 963                          ll_i2mdexp(dir), oc);
 964         capa_put(oc);
 965         if (rc)
 966                 CERROR("obd destroy objid "DOSTID" error %d\n",
 967                        POSTID(&lsm->lsm_oi), rc);
 968 out_free_memmd:
 969         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 970         OBDO_FREE(oa);
 971 out:
 972         return rc;
 973 }
 974
 975 /* ll_unlink() doesn't update the inode with the new link count.
 976  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 977  * is any lock existing. They will recycle dentries and inodes based upon locks
 978  * too. b=20433 */
 979 static int ll_unlink(struct inode *dir, struct dentry *dentry)
 980 {
 981         struct ptlrpc_request *request = NULL;
 982         struct md_op_data *op_data;
 983         int rc;
 984
 985         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 986                dentry, dir->i_ino, dir->i_generation, dir);
 987
 988         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 989                                      dentry->d_name.name,
 990                                      dentry->d_name.len,
 991                                      0, LUSTRE_OPC_ANY, NULL);
 992         if (IS_ERR(op_data))
 993                 return PTR_ERR(op_data);
 994
 995         ll_get_child_fid(dentry, &op_data->op_fid3);
 996         op_data->op_fid2 = op_data->op_fid3;
 997         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 998         ll_finish_md_op_data(op_data);
 999         if (rc)
1000                 goto out;
1001
1002         ll_update_times(request, dir);
1003         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1004
1005         rc = ll_objects_destroy(request, dir);
1006  out:
1007         ptlrpc_req_finished(request);
1008         return rc;
1009 }
1010
1011 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1012 {
1013         int err;
1014
1015         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
1016                dentry, dir->i_ino, dir->i_generation, dir);
1017
1018         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1019                 mode &= ~current_umask();
1020         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1021         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1022
1023         if (!err)
1024                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1025
1026         return err;
1027 }
1028
1029 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1030 {
1031         struct ptlrpc_request *request = NULL;
1032         struct md_op_data *op_data;
1033         int rc;
1034
1035         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
1036                dentry, dir->i_ino, dir->i_generation, dir);
1037
1038         op_data = ll_prep_md_op_data(NULL, dir, NULL,
1039                                      dentry->d_name.name,
1040                                      dentry->d_name.len,
1041                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1042         if (IS_ERR(op_data))
1043                 return PTR_ERR(op_data);
1044
1045         ll_get_child_fid(dentry, &op_data->op_fid3);
1046         op_data->op_fid2 = op_data->op_fid3;
1047         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1048         ll_finish_md_op_data(op_data);
1049         if (rc == 0) {
1050                 ll_update_times(request, dir);
1051                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1052         }
1053
1054         ptlrpc_req_finished(request);
1055         return rc;
1056 }
1057
1058 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1059                       const char *oldname)
1060 {
1061         int err;
1062
1063         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),target=%.*s\n",
1064                dentry, dir->i_ino, dir->i_generation,
1065                dir, 3000, oldname);
1066
1067         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1068                         0, LUSTRE_OPC_SYMLINK);
1069
1070         if (!err)
1071                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1072
1073         return err;
1074 }
1075
1076 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1077                    struct dentry *new_dentry)
1078 {
1079         struct inode *src = d_inode(old_dentry);
1080         struct ll_sb_info *sbi = ll_i2sbi(dir);
1081         struct ptlrpc_request *request = NULL;
1082         struct md_op_data *op_data;
1083         int err;
1084
1085         CDEBUG(D_VFSTRACE,
1086                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%pd\n",
1087                src->i_ino, src->i_generation, src, dir->i_ino,
1088                dir->i_generation, dir, new_dentry);
1089
1090         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1091                                      new_dentry->d_name.len,
1092                                      0, LUSTRE_OPC_ANY, NULL);
1093         if (IS_ERR(op_data))
1094                 return PTR_ERR(op_data);
1095
1096         err = md_link(sbi->ll_md_exp, op_data, &request);
1097         ll_finish_md_op_data(op_data);
1098         if (err)
1099                 goto out;
1100
1101         ll_update_times(request, dir);
1102         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1103 out:
1104         ptlrpc_req_finished(request);
1105         return err;
1106 }
1107
1108 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1109                      struct inode *new_dir, struct dentry *new_dentry)
1110 {
1111         struct ptlrpc_request *request = NULL;
1112         struct ll_sb_info *sbi = ll_i2sbi(old_dir);
1113         struct md_op_data *op_data;
1114         int err;
1115
1116         CDEBUG(D_VFSTRACE,
1117                "VFS Op:oldname=%pd,src_dir=%lu/%u(%p),newname=%pd,tgt_dir=%lu/%u(%p)\n",
1118                old_dentry, old_dir->i_ino, old_dir->i_generation, old_dir,
1119                new_dentry, new_dir->i_ino, new_dir->i_generation, new_dir);
1120
1121         op_data = ll_prep_md_op_data(NULL, old_dir, new_dir, NULL, 0, 0,
1122                                      LUSTRE_OPC_ANY, NULL);
1123         if (IS_ERR(op_data))
1124                 return PTR_ERR(op_data);
1125
1126         ll_get_child_fid(old_dentry, &op_data->op_fid3);
1127         ll_get_child_fid(new_dentry, &op_data->op_fid4);
1128         err = md_rename(sbi->ll_md_exp, op_data,
1129                         old_dentry->d_name.name,
1130                         old_dentry->d_name.len,
1131                         new_dentry->d_name.name,
1132                         new_dentry->d_name.len, &request);
1133         ll_finish_md_op_data(op_data);
1134         if (!err) {
1135                 ll_update_times(request, old_dir);
1136                 ll_update_times(request, new_dir);
1137                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1138                 err = ll_objects_destroy(request, old_dir);
1139         }
1140
1141         ptlrpc_req_finished(request);
1142         if (!err)
1143                 d_move(old_dentry, new_dentry);
1144         return err;
1145 }
1146
1147 const struct inode_operations ll_dir_inode_operations = {
1148         .mknod        = ll_mknod,
1149         .atomic_open        = ll_atomic_open,
1150         .lookup      = ll_lookup_nd,
1151         .create      = ll_create_nd,
1152         /* We need all these non-raw things for NFSD, to not patch it. */
1153         .unlink      = ll_unlink,
1154         .mkdir        = ll_mkdir,
1155         .rmdir        = ll_rmdir,
1156         .symlink            = ll_symlink,
1157         .link          = ll_link,
1158         .rename      = ll_rename,
1159         .setattr            = ll_setattr,
1160         .getattr            = ll_getattr,
1161         .permission      = ll_inode_permission,
1162         .setxattr          = ll_setxattr,
1163         .getxattr          = ll_getxattr,
1164         .listxattr        = ll_listxattr,
1165         .removexattr    = ll_removexattr,
1166         .get_acl            = ll_get_acl,
1167 };
1168
1169 const struct inode_operations ll_special_inode_operations = {
1170         .setattr        = ll_setattr,
1171         .getattr        = ll_getattr,
1172         .permission     = ll_inode_permission,
1173         .setxattr       = ll_setxattr,
1174         .getxattr       = ll_getxattr,
1175         .listxattr      = ll_listxattr,
1176         .removexattr    = ll_removexattr,
1177         .get_acl            = ll_get_acl,
1178 };