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 " without object type: valid = %#llx\n",
  87                        PFID(&lli->lli_fid), body->valid);
  88                 return -EINVAL;
  89         }
  90
  91         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->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 (already instantiated by the intent lookup).
 105  * Returns inode or NULL
 106  */
 107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 108                       struct lustre_md *md)
 109 {
 110         struct inode     *inode;
 111
 112         LASSERT(hash != 0);
 113         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 114
 115         if (inode) {
 116                 if (inode->i_state & I_NEW) {
 117                         int rc = 0;
 118
 119                         ll_read_inode2(inode, md);
 120                         if (S_ISREG(inode->i_mode) &&
 121                             ll_i2info(inode)->lli_clob == NULL) {
 122                                 CDEBUG(D_INODE,
 123                                         "%s: apply lsm %p to inode "DFID".\n",
 124                                         ll_get_fsname(sb, NULL, 0), md->lsm,
 125                                         PFID(ll_inode2fid(inode)));
 126                                 rc = cl_file_inode_init(inode, md);
 127                         }
 128                         if (rc != 0) {
 129                                 make_bad_inode(inode);
 130                                 unlock_new_inode(inode);
 131                                 iput(inode);
 132                                 inode = ERR_PTR(rc);
 133                         } else
 134                                 unlock_new_inode(inode);
 135                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 136                         ll_update_inode(inode, md);
 137                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 138                        inode, PFID(&md->body->fid1));
 139         }
 140         return inode;
 141 }
 142
 143 static void ll_invalidate_negative_children(struct inode *dir)
 144 {
 145         struct dentry *dentry, *tmp_subdir;
 146         struct ll_d_hlist_node *p;
 147
 148         ll_lock_dcache(dir);
 149         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_u.d_alias) {
 150                 spin_lock(&dentry->d_lock);
 151                 if (!list_empty(&dentry->d_subdirs)) {
 152                         struct dentry *child;
 153
 154                         list_for_each_entry_safe(child, tmp_subdir,
 155                                                  &dentry->d_subdirs,
 156                                                  d_child) {
 157                                 if (d_really_is_negative(child))
 158                                         d_lustre_invalidate(child, 1);
 159                         }
 160                 }
 161                 spin_unlock(&dentry->d_lock);
 162         }
 163         ll_unlock_dcache(dir);
 164 }
 165
 166 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 167                        void *data, int flag)
 168 {
 169         struct lustre_handle lockh;
 170         int rc;
 171
 172         switch (flag) {
 173         case LDLM_CB_BLOCKING:
 174                 ldlm_lock2handle(lock, &lockh);
 175                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 176                 if (rc < 0) {
 177                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 178                         return rc;
 179                 }
 180                 break;
 181         case LDLM_CB_CANCELING: {
 182                 struct inode *inode = ll_inode_from_resource_lock(lock);
 183                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 184
 185                 /* Inode is set to lock->l_resource->lr_lvb_inode
 186                  * for mdc - bug 24555 */
 187                 LASSERT(lock->l_ast_data == NULL);
 188
 189                 if (inode == NULL)
 190                         break;
 191
 192                 /* Invalidate all dentries associated with this inode */
 193                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 194
 195                 if (!fid_res_name_eq(ll_inode2fid(inode),
 196                                      &lock->l_resource->lr_name)) {
 197                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 198                                    PFID(ll_inode2fid(inode)), inode);
 199                         LBUG();
 200                 }
 201
 202                 if (bits & MDS_INODELOCK_XATTR) {
 203                         ll_xattr_cache_destroy(inode);
 204                         bits &= ~MDS_INODELOCK_XATTR;
 205                 }
 206
 207                 /* For OPEN locks we differentiate between lock modes
 208                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 209                 if (bits & MDS_INODELOCK_OPEN)
 210                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 211
 212                 if (bits & MDS_INODELOCK_OPEN) {
 213                         fmode_t fmode;
 214
 215                         switch (lock->l_req_mode) {
 216                         case LCK_CW:
 217                                 fmode = FMODE_WRITE;
 218                                 break;
 219                         case LCK_PR:
 220                                 fmode = FMODE_EXEC;
 221                                 break;
 222                         case LCK_CR:
 223                                 fmode = FMODE_READ;
 224                                 break;
 225                         default:
 226                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 227                                 LBUG();
 228                         }
 229
 230                         ll_md_real_close(inode, fmode);
 231                 }
 232
 233                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 234                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 235                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 236
 237                 if (bits & MDS_INODELOCK_LAYOUT) {
 238                         struct cl_object_conf conf = {
 239                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 240                                 .coc_inode = inode,
 241                         };
 242
 243                         rc = ll_layout_conf(inode, &conf);
 244                         if (rc < 0)
 245                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 246                                        DFID": rc = %d\n",
 247                                        PFID(ll_inode2fid(inode)), rc);
 248                 }
 249
 250                 if (bits & MDS_INODELOCK_UPDATE) {
 251                         struct ll_inode_info *lli = ll_i2info(inode);
 252
 253                         spin_lock(&lli->lli_lock);
 254                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 255                         spin_unlock(&lli->lli_lock);
 256                 }
 257
 258                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 259                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 260                                inode->i_ino);
 261                         truncate_inode_pages(inode->i_mapping, 0);
 262                         ll_invalidate_negative_children(inode);
 263                 }
 264
 265                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 266                     inode->i_sb->s_root != NULL &&
 267                     !is_root_inode(inode))
 268                         ll_invalidate_aliases(inode);
 269
 270                 iput(inode);
 271                 break;
 272         }
 273         default:
 274                 LBUG();
 275         }
 276
 277         return 0;
 278 }
 279
 280 __u32 ll_i2suppgid(struct inode *i)
 281 {
 282         if (in_group_p(i->i_gid))
 283                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 284         else
 285                 return (__u32)(-1);
 286 }
 287
 288 /* Pack the required supplementary groups into the supplied groups array.
 289  * If we don't need to use the groups from the target inode(s) then we
 290  * instead pack one or more groups from the user's supplementary group
 291  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 292 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 293 {
 294 #if 0
 295         int i;
 296 #endif
 297
 298         LASSERT(i1 != NULL);
 299         LASSERT(suppgids != NULL);
 300
 301         suppgids[0] = ll_i2suppgid(i1);
 302
 303         if (i2)
 304                 suppgids[1] = ll_i2suppgid(i2);
 305                 else
 306                         suppgids[1] = -1;
 307
 308 #if 0
 309         for (i = 0; i < current_ngroups; i++) {
 310                 if (suppgids[0] == -1) {
 311                         if (current_groups[i] != suppgids[1])
 312                                 suppgids[0] = current_groups[i];
 313                         continue;
 314                 }
 315                 if (suppgids[1] == -1) {
 316                         if (current_groups[i] != suppgids[0])
 317                                 suppgids[1] = current_groups[i];
 318                         continue;
 319                 }
 320                 break;
 321         }
 322 #endif
 323 }
 324
 325 /*
 326  * try to reuse three types of dentry:
 327  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 328  *    by concurrent .revalidate).
 329  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 330  *    be cleared by others calling d_lustre_revalidate).
 331  * 3. DISCONNECTED alias.
 332  */
 333 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 334 {
 335         struct dentry *alias, *discon_alias, *invalid_alias;
 336         struct ll_d_hlist_node *p;
 337
 338         if (ll_d_hlist_empty(&inode->i_dentry))
 339                 return NULL;
 340
 341         discon_alias = invalid_alias = NULL;
 342
 343         ll_lock_dcache(inode);
 344         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_u.d_alias) {
 345                 LASSERT(alias != dentry);
 346
 347                 spin_lock(&alias->d_lock);
 348                 if (alias->d_flags & DCACHE_DISCONNECTED)
 349                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 350                         discon_alias = alias;
 351                 else if (alias->d_parent == dentry->d_parent         &&
 352                          alias->d_name.hash == dentry->d_name.hash       &&
 353                          alias->d_name.len == dentry->d_name.len         &&
 354                          memcmp(alias->d_name.name, dentry->d_name.name,
 355                                 dentry->d_name.len) == 0)
 356                         invalid_alias = alias;
 357                 spin_unlock(&alias->d_lock);
 358
 359                 if (invalid_alias)
 360                         break;
 361         }
 362         alias = invalid_alias ?: discon_alias ?: NULL;
 363         if (alias) {
 364                 spin_lock(&alias->d_lock);
 365                 dget_dlock(alias);
 366                 spin_unlock(&alias->d_lock);
 367         }
 368         ll_unlock_dcache(inode);
 369
 370         return alias;
 371 }
 372
 373 /*
 374  * Similar to d_splice_alias(), but lustre treats invalid alias
 375  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 376  */
 377 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 378 {
 379         struct dentry *new;
 380         int rc;
 381
 382         if (inode) {
 383                 new = ll_find_alias(inode, de);
 384                 if (new) {
 385                         rc = ll_d_init(new);
 386                         if (rc < 0) {
 387                                 dput(new);
 388                                 return ERR_PTR(rc);
 389                         }
 390                         d_move(new, de);
 391                         iput(inode);
 392                         CDEBUG(D_DENTRY,
 393                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 394                               new, d_inode(new), d_count(new), new->d_flags);
 395                         return new;
 396                 }
 397         }
 398         rc = ll_d_init(de);
 399         if (rc < 0)
 400                 return ERR_PTR(rc);
 401         d_add(de, inode);
 402         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 403                de, d_inode(de), d_count(de), de->d_flags);
 404         return de;
 405 }
 406
 407 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 408                                struct lookup_intent *it,
 409                                struct inode *parent, struct dentry **de)
 410 {
 411         struct inode *inode = NULL;
 412         __u64 bits = 0;
 413         int rc;
 414
 415         /* NB 1 request reference will be taken away by ll_intent_lock()
 416          * when I return */
 417         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 418                it->d.lustre.it_disposition);
 419         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 420                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 421                 if (rc)
 422                         return rc;
 423
 424                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 425
 426                 /* We used to query real size from OSTs here, but actually
 427                    this is not needed. For stat() calls size would be updated
 428                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 429                    2.4 and
 430                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 431                    Everybody else who needs correct file size would call
 432                    ll_glimpse_size or some equivalent themselves anyway.
 433                    Also see bug 7198. */
 434         }
 435
 436         /* Only hash *de if it is unhashed (new dentry).
 437          * Atoimc_open may passing hashed dentries for open.
 438          */
 439         if (d_unhashed(*de)) {
 440                 struct dentry *alias;
 441
 442                 alias = ll_splice_alias(inode, *de);
 443                 if (IS_ERR(alias))
 444                         return PTR_ERR(alias);
 445                 *de = alias;
 446         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 447                    !it_disposition(it, DISP_OPEN_CREATE)) {
 448                 /* With DISP_OPEN_CREATE dentry will
 449                    instantiated in ll_create_it. */
 450                 LASSERT(d_inode(*de) == NULL);
 451                 d_instantiate(*de, inode);
 452         }
 453
 454         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 455                 /* we have lookup look - unhide dentry */
 456                 if (bits & MDS_INODELOCK_LOOKUP)
 457                         d_lustre_revalidate(*de);
 458         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 459                 /* If file created on server, don't depend on parent UPDATE
 460                  * lock to unhide it. It is left hidden and next lookup can
 461                  * find it in ll_splice_alias.
 462                  */
 463                 /* Check that parent has UPDATE lock. */
 464                 struct lookup_intent parent_it = {
 465                                         .it_op = IT_GETATTR,
 466                                         .d.lustre.it_lock_handle = 0 };
 467
 468                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 469                                        &ll_i2info(parent)->lli_fid, NULL)) {
 470                         d_lustre_revalidate(*de);
 471                         ll_intent_release(&parent_it);
 472                 }
 473         }
 474
 475         return 0;
 476 }
 477
 478 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 479                                    struct lookup_intent *it, int lookup_flags)
 480 {
 481         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 482         struct dentry *save = dentry, *retval;
 483         struct ptlrpc_request *req = NULL;
 484         struct inode *inode;
 485         struct md_op_data *op_data;
 486         __u32 opc;
 487         int rc;
 488
 489         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 490                 return ERR_PTR(-ENAMETOOLONG);
 491
 492         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 493                dentry, parent->i_ino,
 494                parent->i_generation, parent, LL_IT2STR(it));
 495
 496         if (d_mountpoint(dentry))
 497                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 498
 499         if (it == NULL || it->it_op == IT_GETXATTR)
 500                 it = &lookup_it;
 501
 502         if (it->it_op == IT_GETATTR) {
 503                 rc = ll_statahead_enter(parent, &dentry, 0);
 504                 if (rc == 1) {
 505                         if (dentry == save)
 506                                 retval = NULL;
 507                         else
 508                                 retval = dentry;
 509                         goto out;
 510                 }
 511         }
 512
 513         if (it->it_op & IT_CREAT)
 514                 opc = LUSTRE_OPC_CREATE;
 515         else
 516                 opc = LUSTRE_OPC_ANY;
 517
 518         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 519                                      dentry->d_name.len, lookup_flags, opc,
 520                                      NULL);
 521         if (IS_ERR(op_data))
 522                 return (void *)op_data;
 523
 524         /* enforce umask if acl disabled or MDS doesn't support umask */
 525         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 526                 it->it_create_mode &= ~current_umask();
 527
 528         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 529                             lookup_flags, &req, ll_md_blocking_ast, 0);
 530         ll_finish_md_op_data(op_data);
 531         if (rc < 0) {
 532                 retval = ERR_PTR(rc);
 533                 goto out;
 534         }
 535
 536         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 537         if (rc != 0) {
 538                 ll_intent_release(it);
 539                 retval = ERR_PTR(rc);
 540                 goto out;
 541         }
 542
 543         inode = d_inode(dentry);
 544         if ((it->it_op & IT_OPEN) && inode &&
 545             !S_ISREG(inode->i_mode) &&
 546             !S_ISDIR(inode->i_mode)) {
 547                 ll_release_openhandle(inode, it);
 548         }
 549         ll_lookup_finish_locks(it, inode);
 550
 551         if (dentry == save)
 552                 retval = NULL;
 553         else
 554                 retval = dentry;
 555         goto out;
 556  out:
 557         if (req)
 558                 ptlrpc_req_finished(req);
 559         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 560                 ll_statahead_mark(parent, dentry);
 561         return retval;
 562 }
 563
 564 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 565                                    unsigned int flags)
 566 {
 567         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 568         struct dentry *de;
 569
 570         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u\n",
 571                dentry, parent->i_ino,
 572                parent->i_generation, parent, flags);
 573
 574         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 575         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 576                 return NULL;
 577
 578         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 579                 itp = NULL;
 580         else
 581                 itp = &it;
 582         de = ll_lookup_it(parent, dentry, itp, 0);
 583
 584         if (itp != NULL)
 585                 ll_intent_release(itp);
 586
 587         return de;
 588 }
 589
 590 /*
 591  * For cached negative dentry and new dentry, handle lookup/create/open
 592  * together.
 593  */
 594 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 595                           struct file *file, unsigned open_flags,
 596                           umode_t mode, int *opened)
 597 {
 598         struct lookup_intent *it;
 599         struct dentry *de;
 600         long long lookup_flags = LOOKUP_OPEN;
 601         int rc = 0;
 602
 603         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),file %p,open_flags %x,mode %x opened %d\n",
 604                dentry, dir->i_ino,
 605                dir->i_generation, dir, file, open_flags, mode, *opened);
 606
 607         it = kzalloc(sizeof(*it), GFP_NOFS);
 608         if (!it)
 609                 return -ENOMEM;
 610
 611         it->it_op = IT_OPEN;
 612         if (open_flags & O_CREAT) {
 613                 it->it_op |= IT_CREAT;
 614                 lookup_flags |= LOOKUP_CREATE;
 615         }
 616         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 617         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 618
 619         /* Dentry added to dcache tree in ll_lookup_it */
 620         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 621         if (IS_ERR(de))
 622                 rc = PTR_ERR(de);
 623         else if (de != NULL)
 624                 dentry = de;
 625
 626         if (!rc) {
 627                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 628                         /* Dentry instantiated in ll_create_it. */
 629                         rc = ll_create_it(dir, dentry, mode, it);
 630                         if (rc) {
 631                                 /* We dget in ll_splice_alias. */
 632                                 if (de != NULL)
 633                                         dput(de);
 634                                 goto out_release;
 635                         }
 636
 637                         *opened |= FILE_CREATED;
 638                 }
 639                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 640                         /* Open dentry. */
 641                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 642                                 /* We cannot call open here as it would
 643                                  * deadlock.
 644                                  */
 645                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 646                                         ptlrpc_req_finished(
 647                                                        (struct ptlrpc_request *)
 648                                                           it->d.lustre.it_data);
 649                                 rc = finish_no_open(file, de);
 650                         } else {
 651                                 file->private_data = it;
 652                                 rc = finish_open(file, dentry, NULL, opened);
 653                                 /* We dget in ll_splice_alias. finish_open takes
 654                                  * care of dget for fd open.
 655                                  */
 656                                 if (de != NULL)
 657                                         dput(de);
 658                         }
 659                 } else {
 660                         rc = finish_no_open(file, de);
 661                 }
 662         }
 663
 664 out_release:
 665         ll_intent_release(it);
 666         OBD_FREE(it, sizeof(*it));
 667
 668         return rc;
 669 }
 670
 671
 672 /* We depend on "mode" being set with the proper file type/umask by now */
 673 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 674 {
 675         struct inode *inode = NULL;
 676         struct ptlrpc_request *request = NULL;
 677         struct ll_sb_info *sbi = ll_i2sbi(dir);
 678         int rc;
 679
 680         LASSERT(it && it->d.lustre.it_disposition);
 681
 682         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 683         request = it->d.lustre.it_data;
 684         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 685         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 686         if (rc) {
 687                 inode = ERR_PTR(rc);
 688                 goto out;
 689         }
 690
 691         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 692
 693         /* We asked for a lock on the directory, but were granted a
 694          * lock on the inode.  Since we finally have an inode pointer,
 695          * stuff it in the lock. */
 696         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 697                inode, inode->i_ino, inode->i_generation);
 698         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 699  out:
 700         ptlrpc_req_finished(request);
 701         return inode;
 702 }
 703
 704 /*
 705  * By the time this is called, we already have created the directory cache
 706  * entry for the new file, but it is so far negative - it has no inode.
 707  *
 708  * We defer creating the OBD object(s) until open, to keep the intent and
 709  * non-intent code paths similar, and also because we do not have the MDS
 710  * inode number before calling ll_create_node() (which is needed for LOV),
 711  * so we would need to do yet another RPC to the MDS to store the LOV EA
 712  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 713  * lmm_size in datalen (the MDS still has code which will handle that).
 714  *
 715  * If the create succeeds, we fill in the inode information
 716  * with d_instantiate().
 717  */
 718 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 719                         struct lookup_intent *it)
 720 {
 721         struct inode *inode;
 722         int rc = 0;
 723
 724         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 725                dentry, dir->i_ino,
 726                dir->i_generation, dir, LL_IT2STR(it));
 727
 728         rc = it_open_error(DISP_OPEN_CREATE, it);
 729         if (rc)
 730                 return rc;
 731
 732         inode = ll_create_node(dir, it);
 733         if (IS_ERR(inode))
 734                 return PTR_ERR(inode);
 735
 736         d_instantiate(dentry, inode);
 737         return 0;
 738 }
 739
 740 static void ll_update_times(struct ptlrpc_request *request,
 741                             struct inode *inode)
 742 {
 743         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 744                                                        &RMF_MDT_BODY);
 745
 746         LASSERT(body);
 747         if (body->valid & OBD_MD_FLMTIME &&
 748             body->mtime > LTIME_S(inode->i_mtime)) {
 749                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
 750                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 751                 LTIME_S(inode->i_mtime) = body->mtime;
 752         }
 753         if (body->valid & OBD_MD_FLCTIME &&
 754             body->ctime > LTIME_S(inode->i_ctime))
 755                 LTIME_S(inode->i_ctime) = body->ctime;
 756 }
 757
 758 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 759                        const char *tgt, int mode, int rdev,
 760                        __u32 opc)
 761 {
 762         struct ptlrpc_request *request = NULL;
 763         struct md_op_data *op_data;
 764         struct inode *inode = NULL;
 765         struct ll_sb_info *sbi = ll_i2sbi(dir);
 766         int tgt_len = 0;
 767         int err;
 768
 769         if (unlikely(tgt != NULL))
 770                 tgt_len = strlen(tgt) + 1;
 771
 772         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 773                                      dentry->d_name.name,
 774                                      dentry->d_name.len,
 775                                      0, opc, NULL);
 776         if (IS_ERR(op_data)) {
 777                 err = PTR_ERR(op_data);
 778                 goto err_exit;
 779         }
 780
 781         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 782                         from_kuid(&init_user_ns, current_fsuid()),
 783                         from_kgid(&init_user_ns, current_fsgid()),
 784                         cfs_curproc_cap_pack(), rdev, &request);
 785         ll_finish_md_op_data(op_data);
 786         if (err)
 787                 goto err_exit;
 788
 789         ll_update_times(request, dir);
 790
 791         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 792         if (err)
 793                 goto err_exit;
 794
 795         d_instantiate(dentry, inode);
 796 err_exit:
 797         ptlrpc_req_finished(request);
 798
 799         return err;
 800 }
 801
 802 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 803                     umode_t mode, dev_t rdev)
 804 {
 805         int err;
 806
 807         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p) mode %o dev %x\n",
 808                dchild, dir->i_ino, dir->i_generation, dir,
 809                mode, old_encode_dev(rdev));
 810
 811         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 812                 mode &= ~current_umask();
 813
 814         switch (mode & S_IFMT) {
 815         case 0:
 816                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 817         case S_IFREG:
 818         case S_IFCHR:
 819         case S_IFBLK:
 820         case S_IFIFO:
 821         case S_IFSOCK:
 822                 err = ll_new_node(dir, dchild, NULL, mode,
 823                                   old_encode_dev(rdev),
 824                                   LUSTRE_OPC_MKNOD);
 825                 break;
 826         case S_IFDIR:
 827                 err = -EPERM;
 828                 break;
 829         default:
 830                 err = -EINVAL;
 831         }
 832
 833         if (!err)
 834                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 835
 836         return err;
 837 }
 838
 839 /*
 840  * Plain create. Intent create is handled in atomic_open.
 841  */
 842 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 843                         umode_t mode, bool want_excl)
 844 {
 845         int rc;
 846
 847         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u, excl=%d\n",
 848                dentry, dir->i_ino,
 849                dir->i_generation, dir, mode, want_excl);
 850
 851         rc = ll_mknod(dir, dentry, mode, 0);
 852
 853         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 854
 855         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 856                dentry, d_unhashed(dentry));
 857
 858         return rc;
 859 }
 860
 861 static inline void ll_get_child_fid(struct dentry *child, struct lu_fid *fid)
 862 {
 863         if (d_really_is_positive(child))
 864                 *fid = *ll_inode2fid(d_inode(child));
 865 }
 866
 867 /**
 868  * Remove dir entry
 869  **/
 870 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
 871 {
 872         struct ptlrpc_request *request = NULL;
 873         struct md_op_data *op_data;
 874         int rc;
 875
 876         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
 877                namelen, name, dir->i_ino, dir->i_generation, dir);
 878
 879         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
 880                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
 881         if (IS_ERR(op_data))
 882                 return PTR_ERR(op_data);
 883         op_data->op_cli_flags |= CLI_RM_ENTRY;
 884         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 885         ll_finish_md_op_data(op_data);
 886         if (rc == 0) {
 887                 ll_update_times(request, dir);
 888                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
 889         }
 890
 891         ptlrpc_req_finished(request);
 892         return rc;
 893 }
 894
 895 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 896 {
 897         struct mdt_body *body;
 898         struct lov_mds_md *eadata;
 899         struct lov_stripe_md *lsm = NULL;
 900         struct obd_trans_info oti = { 0 };
 901         struct obdo *oa;
 902         struct obd_capa *oc = NULL;
 903         int rc;
 904
 905         /* req is swabbed so this is safe */
 906         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 907         if (!(body->valid & OBD_MD_FLEASIZE))
 908                 return 0;
 909
 910         if (body->eadatasize == 0) {
 911                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 912                 rc = -EPROTO;
 913                 goto out;
 914         }
 915
 916         /* The MDS sent back the EA because we unlinked the last reference
 917          * to this file. Use this EA to unlink the objects on the OST.
 918          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 919          * check it is complete and sensible. */
 920         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 921                                               body->eadatasize);
 922         LASSERT(eadata != NULL);
 923
 924         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 925         if (rc < 0) {
 926                 CERROR("obd_unpackmd: %d\n", rc);
 927                 goto out;
 928         }
 929         LASSERT(rc >= sizeof(*lsm));
 930
 931         OBDO_ALLOC(oa);
 932         if (oa == NULL) {
 933                 rc = -ENOMEM;
 934                 goto out_free_memmd;
 935         }
 936
 937         oa->o_oi = lsm->lsm_oi;
 938         oa->o_mode = body->mode & S_IFMT;
 939         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 940
 941         if (body->valid & OBD_MD_FLCOOKIE) {
 942                 oa->o_valid |= OBD_MD_FLCOOKIE;
 943                 oti.oti_logcookies =
 944                         req_capsule_server_sized_get(&request->rq_pill,
 945                                                      &RMF_LOGCOOKIES,
 946                                                    sizeof(struct llog_cookie) *
 947                                                      lsm->lsm_stripe_count);
 948                 if (oti.oti_logcookies == NULL) {
 949                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 950                         body->valid &= ~OBD_MD_FLCOOKIE;
 951                 }
 952         }
 953
 954         if (body->valid & OBD_MD_FLOSSCAPA) {
 955                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
 956                 if (rc)
 957                         goto out_free_memmd;
 958         }
 959
 960         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 961                          ll_i2mdexp(dir), oc);
 962         capa_put(oc);
 963         if (rc)
 964                 CERROR("obd destroy objid "DOSTID" error %d\n",
 965                        POSTID(&lsm->lsm_oi), rc);
 966 out_free_memmd:
 967         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 968         OBDO_FREE(oa);
 969 out:
 970         return rc;
 971 }
 972
 973 /* ll_unlink() doesn't update the inode with the new link count.
 974  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 975  * is any lock existing. They will recycle dentries and inodes based upon locks
 976  * too. b=20433 */
 977 static int ll_unlink(struct inode * dir, struct dentry *dentry)
 978 {
 979         struct ptlrpc_request *request = NULL;
 980         struct md_op_data *op_data;
 981         int rc;
 982         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 983                dentry, dir->i_ino, dir->i_generation, dir);
 984
 985         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 986                                      dentry->d_name.name,
 987                                      dentry->d_name.len,
 988                                      0, LUSTRE_OPC_ANY, NULL);
 989         if (IS_ERR(op_data))
 990                 return PTR_ERR(op_data);
 991
 992         ll_get_child_fid(dentry, &op_data->op_fid3);
 993         op_data->op_fid2 = op_data->op_fid3;
 994         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 995         ll_finish_md_op_data(op_data);
 996         if (rc)
 997                 goto out;
 998
 999         ll_update_times(request, dir);
1000         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1001
1002         rc = ll_objects_destroy(request, dir);
1003  out:
1004         ptlrpc_req_finished(request);
1005         return rc;
1006 }
1007
1008 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1009 {
1010         int err;
1011
1012         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
1013                dentry, dir->i_ino, dir->i_generation, dir);
1014
1015         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1016                 mode &= ~current_umask();
1017         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1018         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1019
1020         if (!err)
1021                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1022
1023         return err;
1024 }
1025
1026 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1027 {
1028         struct ptlrpc_request *request = NULL;
1029         struct md_op_data *op_data;
1030         int rc;
1031
1032         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
1033                dentry, dir->i_ino, dir->i_generation, dir);
1034
1035         op_data = ll_prep_md_op_data(NULL, dir, NULL,
1036                                      dentry->d_name.name,
1037                                      dentry->d_name.len,
1038                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1039         if (IS_ERR(op_data))
1040                 return PTR_ERR(op_data);
1041
1042         ll_get_child_fid(dentry, &op_data->op_fid3);
1043         op_data->op_fid2 = op_data->op_fid3;
1044         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1045         ll_finish_md_op_data(op_data);
1046         if (rc == 0) {
1047                 ll_update_times(request, dir);
1048                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1049         }
1050
1051         ptlrpc_req_finished(request);
1052         return rc;
1053 }
1054
1055 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1056                       const char *oldname)
1057 {
1058         int err;
1059
1060         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),target=%.*s\n",
1061                dentry, dir->i_ino, dir->i_generation,
1062                dir, 3000, oldname);
1063
1064         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1065                         0, LUSTRE_OPC_SYMLINK);
1066
1067         if (!err)
1068                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1069
1070         return err;
1071 }
1072
1073 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1074                    struct dentry *new_dentry)
1075 {
1076         struct inode *src = d_inode(old_dentry);
1077         struct ll_sb_info *sbi = ll_i2sbi(dir);
1078         struct ptlrpc_request *request = NULL;
1079         struct md_op_data *op_data;
1080         int err;
1081
1082         CDEBUG(D_VFSTRACE,
1083                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%pd\n",
1084                src->i_ino, src->i_generation, src, dir->i_ino,
1085                dir->i_generation, dir, new_dentry);
1086
1087         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1088                                      new_dentry->d_name.len,
1089                                      0, LUSTRE_OPC_ANY, NULL);
1090         if (IS_ERR(op_data))
1091                 return PTR_ERR(op_data);
1092
1093         err = md_link(sbi->ll_md_exp, op_data, &request);
1094         ll_finish_md_op_data(op_data);
1095         if (err)
1096                 goto out;
1097
1098         ll_update_times(request, dir);
1099         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1100 out:
1101         ptlrpc_req_finished(request);
1102         return err;
1103 }
1104
1105 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1106                      struct inode *new_dir, struct dentry *new_dentry)
1107 {
1108         struct ptlrpc_request *request = NULL;
1109         struct ll_sb_info *sbi = ll_i2sbi(old_dir);
1110         struct md_op_data *op_data;
1111         int err;
1112
1113         CDEBUG(D_VFSTRACE,
1114                "VFS Op:oldname=%pd,src_dir=%lu/%u(%p),newname=%pd,"
1115                "tgt_dir=%lu/%u(%p)\n", old_dentry,
1116                old_dir->i_ino, old_dir->i_generation, old_dir, new_dentry,
1117                new_dir->i_ino, new_dir->i_generation, new_dir);
1118
1119         op_data = ll_prep_md_op_data(NULL, old_dir, new_dir, NULL, 0, 0,
1120                                      LUSTRE_OPC_ANY, NULL);
1121         if (IS_ERR(op_data))
1122                 return PTR_ERR(op_data);
1123
1124         ll_get_child_fid(old_dentry, &op_data->op_fid3);
1125         ll_get_child_fid(new_dentry, &op_data->op_fid4);
1126         err = md_rename(sbi->ll_md_exp, op_data,
1127                         old_dentry->d_name.name,
1128                         old_dentry->d_name.len,
1129                         new_dentry->d_name.name,
1130                         new_dentry->d_name.len, &request);
1131         ll_finish_md_op_data(op_data);
1132         if (!err) {
1133                 ll_update_times(request, old_dir);
1134                 ll_update_times(request, new_dir);
1135                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1136                 err = ll_objects_destroy(request, old_dir);
1137         }
1138
1139         ptlrpc_req_finished(request);
1140         if (!err)
1141                 d_move(old_dentry, new_dentry);
1142         return err;
1143 }
1144
1145 const struct inode_operations ll_dir_inode_operations = {
1146         .mknod        = ll_mknod,
1147         .atomic_open        = ll_atomic_open,
1148         .lookup      = ll_lookup_nd,
1149         .create      = ll_create_nd,
1150         /* We need all these non-raw things for NFSD, to not patch it. */
1151         .unlink      = ll_unlink,
1152         .mkdir        = ll_mkdir,
1153         .rmdir        = ll_rmdir,
1154         .symlink            = ll_symlink,
1155         .link          = ll_link,
1156         .rename      = ll_rename,
1157         .setattr            = ll_setattr,
1158         .getattr            = ll_getattr,
1159         .permission      = ll_inode_permission,
1160         .setxattr          = ll_setxattr,
1161         .getxattr          = ll_getxattr,
1162         .listxattr        = ll_listxattr,
1163         .removexattr    = ll_removexattr,
1164         .get_acl            = ll_get_acl,
1165 };
1166
1167 const struct inode_operations ll_special_inode_operations = {
1168         .setattr        = ll_setattr,
1169         .getattr        = ll_getattr,
1170         .permission     = ll_inode_permission,
1171         .setxattr       = ll_setxattr,
1172         .getxattr       = ll_getxattr,
1173         .listxattr      = ll_listxattr,
1174         .removexattr    = ll_removexattr,
1175         .get_acl            = ll_get_acl,
1176 };