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, 2015, 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  * 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                                 iget_failed(inode);
 130                                 inode = ERR_PTR(rc);
 131                         } else
 132                                 unlock_new_inode(inode);
 133                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 134                         ll_update_inode(inode, md);
 135                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 136                        inode, PFID(&md->body->fid1));
 137         }
 138         return inode;
 139 }
 140
 141 static void ll_invalidate_negative_children(struct inode *dir)
 142 {
 143         struct dentry *dentry, *tmp_subdir;
 144
 145         ll_lock_dcache(dir);
 146         hlist_for_each_entry(dentry, &dir->i_dentry, d_u.d_alias) {
 147                 spin_lock(&dentry->d_lock);
 148                 if (!list_empty(&dentry->d_subdirs)) {
 149                         struct dentry *child;
 150
 151                         list_for_each_entry_safe(child, tmp_subdir,
 152                                                  &dentry->d_subdirs,
 153                                                  d_child) {
 154                                 if (d_really_is_negative(child))
 155                                         d_lustre_invalidate(child, 1);
 156                         }
 157                 }
 158                 spin_unlock(&dentry->d_lock);
 159         }
 160         ll_unlock_dcache(dir);
 161 }
 162
 163 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 164                        void *data, int flag)
 165 {
 166         struct lustre_handle lockh;
 167         int rc;
 168
 169         switch (flag) {
 170         case LDLM_CB_BLOCKING:
 171                 ldlm_lock2handle(lock, &lockh);
 172                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 173                 if (rc < 0) {
 174                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 175                         return rc;
 176                 }
 177                 break;
 178         case LDLM_CB_CANCELING: {
 179                 struct inode *inode = ll_inode_from_resource_lock(lock);
 180                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 181
 182                 /* Inode is set to lock->l_resource->lr_lvb_inode
 183                  * for mdc - bug 24555 */
 184                 LASSERT(lock->l_ast_data == NULL);
 185
 186                 if (inode == NULL)
 187                         break;
 188
 189                 /* Invalidate all dentries associated with this inode */
 190                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 191
 192                 if (!fid_res_name_eq(ll_inode2fid(inode),
 193                                      &lock->l_resource->lr_name)) {
 194                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 195                                    PFID(ll_inode2fid(inode)), inode);
 196                         LBUG();
 197                 }
 198
 199                 if (bits & MDS_INODELOCK_XATTR) {
 200                         ll_xattr_cache_destroy(inode);
 201                         bits &= ~MDS_INODELOCK_XATTR;
 202                 }
 203
 204                 /* For OPEN locks we differentiate between lock modes
 205                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 206                 if (bits & MDS_INODELOCK_OPEN)
 207                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 208
 209                 if (bits & MDS_INODELOCK_OPEN) {
 210                         fmode_t fmode;
 211
 212                         switch (lock->l_req_mode) {
 213                         case LCK_CW:
 214                                 fmode = FMODE_WRITE;
 215                                 break;
 216                         case LCK_PR:
 217                                 fmode = FMODE_EXEC;
 218                                 break;
 219                         case LCK_CR:
 220                                 fmode = FMODE_READ;
 221                                 break;
 222                         default:
 223                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 224                                 LBUG();
 225                         }
 226
 227                         ll_md_real_close(inode, fmode);
 228                 }
 229
 230                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 231                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 232                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 233
 234                 if (bits & MDS_INODELOCK_LAYOUT) {
 235                         struct cl_object_conf conf = {
 236                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 237                                 .coc_inode = inode,
 238                         };
 239
 240                         rc = ll_layout_conf(inode, &conf);
 241                         if (rc < 0)
 242                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 243                                        DFID": rc = %d\n",
 244                                        PFID(ll_inode2fid(inode)), rc);
 245                 }
 246
 247                 if (bits & MDS_INODELOCK_UPDATE) {
 248                         struct ll_inode_info *lli = ll_i2info(inode);
 249
 250                         spin_lock(&lli->lli_lock);
 251                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 252                         spin_unlock(&lli->lli_lock);
 253                 }
 254
 255                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 256                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 257                                inode->i_ino);
 258                         truncate_inode_pages(inode->i_mapping, 0);
 259                         ll_invalidate_negative_children(inode);
 260                 }
 261
 262                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 263                     inode->i_sb->s_root != NULL &&
 264                     !is_root_inode(inode))
 265                         ll_invalidate_aliases(inode);
 266
 267                 iput(inode);
 268                 break;
 269         }
 270         default:
 271                 LBUG();
 272         }
 273
 274         return 0;
 275 }
 276
 277 __u32 ll_i2suppgid(struct inode *i)
 278 {
 279         if (in_group_p(i->i_gid))
 280                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 281         else
 282                 return (__u32)(-1);
 283 }
 284
 285 /* Pack the required supplementary groups into the supplied groups array.
 286  * If we don't need to use the groups from the target inode(s) then we
 287  * instead pack one or more groups from the user's supplementary group
 288  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 289 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 290 {
 291 #if 0
 292         int i;
 293 #endif
 294
 295         LASSERT(i1 != NULL);
 296         LASSERT(suppgids != NULL);
 297
 298         suppgids[0] = ll_i2suppgid(i1);
 299
 300         if (i2)
 301                 suppgids[1] = ll_i2suppgid(i2);
 302                 else
 303                         suppgids[1] = -1;
 304
 305 #if 0
 306         for (i = 0; i < current_ngroups; i++) {
 307                 if (suppgids[0] == -1) {
 308                         if (current_groups[i] != suppgids[1])
 309                                 suppgids[0] = current_groups[i];
 310                         continue;
 311                 }
 312                 if (suppgids[1] == -1) {
 313                         if (current_groups[i] != suppgids[0])
 314                                 suppgids[1] = current_groups[i];
 315                         continue;
 316                 }
 317                 break;
 318         }
 319 #endif
 320 }
 321
 322 /*
 323  * try to reuse three types of dentry:
 324  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 325  *    by concurrent .revalidate).
 326  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 327  *    be cleared by others calling d_lustre_revalidate).
 328  * 3. DISCONNECTED alias.
 329  */
 330 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 331 {
 332         struct dentry *alias, *discon_alias, *invalid_alias;
 333
 334         if (hlist_empty(&inode->i_dentry))
 335                 return NULL;
 336
 337         discon_alias = invalid_alias = NULL;
 338
 339         ll_lock_dcache(inode);
 340         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 341                 LASSERT(alias != dentry);
 342
 343                 spin_lock(&alias->d_lock);
 344                 if (alias->d_flags & DCACHE_DISCONNECTED)
 345                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 346                         discon_alias = alias;
 347                 else if (alias->d_parent == dentry->d_parent         &&
 348                          alias->d_name.hash == dentry->d_name.hash       &&
 349                          alias->d_name.len == dentry->d_name.len         &&
 350                          memcmp(alias->d_name.name, dentry->d_name.name,
 351                                 dentry->d_name.len) == 0)
 352                         invalid_alias = alias;
 353                 spin_unlock(&alias->d_lock);
 354
 355                 if (invalid_alias)
 356                         break;
 357         }
 358         alias = invalid_alias ?: discon_alias ?: NULL;
 359         if (alias) {
 360                 spin_lock(&alias->d_lock);
 361                 dget_dlock(alias);
 362                 spin_unlock(&alias->d_lock);
 363         }
 364         ll_unlock_dcache(inode);
 365
 366         return alias;
 367 }
 368
 369 /*
 370  * Similar to d_splice_alias(), but lustre treats invalid alias
 371  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 372  */
 373 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 374 {
 375         struct dentry *new;
 376         int rc;
 377
 378         if (inode) {
 379                 new = ll_find_alias(inode, de);
 380                 if (new) {
 381                         rc = ll_d_init(new);
 382                         if (rc < 0) {
 383                                 dput(new);
 384                                 return ERR_PTR(rc);
 385                         }
 386                         d_move(new, de);
 387                         iput(inode);
 388                         CDEBUG(D_DENTRY,
 389                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 390                               new, d_inode(new), d_count(new), new->d_flags);
 391                         return new;
 392                 }
 393         }
 394         rc = ll_d_init(de);
 395         if (rc < 0)
 396                 return ERR_PTR(rc);
 397         d_add(de, inode);
 398         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 399                de, d_inode(de), d_count(de), de->d_flags);
 400         return de;
 401 }
 402
 403 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 404                                struct lookup_intent *it,
 405                                struct inode *parent, struct dentry **de)
 406 {
 407         struct inode *inode = NULL;
 408         __u64 bits = 0;
 409         int rc = 0;
 410
 411         /* NB 1 request reference will be taken away by ll_intent_lock()
 412          * when I return */
 413         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 414                it->d.lustre.it_disposition);
 415         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 416                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 417                 if (rc)
 418                         return rc;
 419
 420                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 421
 422                 /* We used to query real size from OSTs here, but actually
 423                    this is not needed. For stat() calls size would be updated
 424                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 425                    2.4 and
 426                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 427                    Everybody else who needs correct file size would call
 428                    ll_glimpse_size or some equivalent themselves anyway.
 429                    Also see bug 7198. */
 430         }
 431
 432         /* Only hash *de if it is unhashed (new dentry).
 433          * Atoimc_open may passing hashed dentries for open.
 434          */
 435         if (d_unhashed(*de)) {
 436                 struct dentry *alias;
 437
 438                 alias = ll_splice_alias(inode, *de);
 439                 if (IS_ERR(alias)) {
 440                         rc = PTR_ERR(alias);
 441                         goto out;
 442                 }
 443                 *de = alias;
 444         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 445                    !it_disposition(it, DISP_OPEN_CREATE)) {
 446                 /* With DISP_OPEN_CREATE dentry will
 447                    instantiated in ll_create_it. */
 448                 LASSERT(d_inode(*de) == NULL);
 449                 d_instantiate(*de, inode);
 450         }
 451
 452         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 453                 /* we have lookup look - unhide dentry */
 454                 if (bits & MDS_INODELOCK_LOOKUP)
 455                         d_lustre_revalidate(*de);
 456         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 457                 /* If file created on server, don't depend on parent UPDATE
 458                  * lock to unhide it. It is left hidden and next lookup can
 459                  * find it in ll_splice_alias.
 460                  */
 461                 /* Check that parent has UPDATE lock. */
 462                 struct lookup_intent parent_it = {
 463                                         .it_op = IT_GETATTR,
 464                                         .d.lustre.it_lock_handle = 0 };
 465
 466                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 467                                        &ll_i2info(parent)->lli_fid, NULL)) {
 468                         d_lustre_revalidate(*de);
 469                         ll_intent_release(&parent_it);
 470                 }
 471         }
 472
 473 out:
 474         if (rc != 0 && it->it_op & IT_OPEN)
 475                 ll_open_cleanup((*de)->d_sb, request);
 476
 477         return rc;
 478 }
 479
 480 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 481                                    struct lookup_intent *it, int lookup_flags)
 482 {
 483         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 484         struct dentry *save = dentry, *retval;
 485         struct ptlrpc_request *req = NULL;
 486         struct inode *inode;
 487         struct md_op_data *op_data;
 488         __u32 opc;
 489         int rc;
 490
 491         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 492                 return ERR_PTR(-ENAMETOOLONG);
 493
 494         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 495                dentry, parent->i_ino,
 496                parent->i_generation, parent, LL_IT2STR(it));
 497
 498         if (d_mountpoint(dentry))
 499                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 500
 501         if (it == NULL || it->it_op == IT_GETXATTR)
 502                 it = &lookup_it;
 503
 504         if (it->it_op == IT_GETATTR) {
 505                 rc = ll_statahead_enter(parent, &dentry, 0);
 506                 if (rc == 1) {
 507                         if (dentry == save)
 508                                 retval = NULL;
 509                         else
 510                                 retval = dentry;
 511                         goto out;
 512                 }
 513         }
 514
 515         if (it->it_op & IT_CREAT)
 516                 opc = LUSTRE_OPC_CREATE;
 517         else
 518                 opc = LUSTRE_OPC_ANY;
 519
 520         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 521                                      dentry->d_name.len, lookup_flags, opc,
 522                                      NULL);
 523         if (IS_ERR(op_data))
 524                 return (void *)op_data;
 525
 526         /* enforce umask if acl disabled or MDS doesn't support umask */
 527         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 528                 it->it_create_mode &= ~current_umask();
 529
 530         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 531                             lookup_flags, &req, ll_md_blocking_ast, 0);
 532         ll_finish_md_op_data(op_data);
 533         if (rc < 0) {
 534                 retval = ERR_PTR(rc);
 535                 goto out;
 536         }
 537
 538         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 539         if (rc != 0) {
 540                 ll_intent_release(it);
 541                 retval = ERR_PTR(rc);
 542                 goto out;
 543         }
 544
 545         inode = d_inode(dentry);
 546         if ((it->it_op & IT_OPEN) && inode &&
 547             !S_ISREG(inode->i_mode) &&
 548             !S_ISDIR(inode->i_mode)) {
 549                 ll_release_openhandle(inode, it);
 550         }
 551         ll_lookup_finish_locks(it, inode);
 552
 553         if (dentry == save)
 554                 retval = NULL;
 555         else
 556                 retval = dentry;
 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         kfree(it);
 669
 670         return rc;
 671 }
 672
 673 /* We depend on "mode" being set with the proper file type/umask by now */
 674 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 675 {
 676         struct inode *inode = NULL;
 677         struct ptlrpc_request *request = NULL;
 678         struct ll_sb_info *sbi = ll_i2sbi(dir);
 679         int rc;
 680
 681         LASSERT(it && it->d.lustre.it_disposition);
 682
 683         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 684         request = it->d.lustre.it_data;
 685         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 686         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 687         if (rc) {
 688                 inode = ERR_PTR(rc);
 689                 goto out;
 690         }
 691
 692         LASSERT(hlist_empty(&inode->i_dentry));
 693
 694         /* We asked for a lock on the directory, but were granted a
 695          * lock on the inode.  Since we finally have an inode pointer,
 696          * stuff it in the lock. */
 697         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 698                inode, inode->i_ino, inode->i_generation);
 699         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 700  out:
 701         ptlrpc_req_finished(request);
 702         return inode;
 703 }
 704
 705 /*
 706  * By the time this is called, we already have created the directory cache
 707  * entry for the new file, but it is so far negative - it has no inode.
 708  *
 709  * We defer creating the OBD object(s) until open, to keep the intent and
 710  * non-intent code paths similar, and also because we do not have the MDS
 711  * inode number before calling ll_create_node() (which is needed for LOV),
 712  * so we would need to do yet another RPC to the MDS to store the LOV EA
 713  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 714  * lmm_size in datalen (the MDS still has code which will handle that).
 715  *
 716  * If the create succeeds, we fill in the inode information
 717  * with d_instantiate().
 718  */
 719 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 720                         struct lookup_intent *it)
 721 {
 722         struct inode *inode;
 723         int rc = 0;
 724
 725         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 726                dentry, dir->i_ino,
 727                dir->i_generation, dir, LL_IT2STR(it));
 728
 729         rc = it_open_error(DISP_OPEN_CREATE, it);
 730         if (rc)
 731                 return rc;
 732
 733         inode = ll_create_node(dir, it);
 734         if (IS_ERR(inode))
 735                 return PTR_ERR(inode);
 736
 737         d_instantiate(dentry, inode);
 738         return 0;
 739 }
 740
 741 static void ll_update_times(struct ptlrpc_request *request,
 742                             struct inode *inode)
 743 {
 744         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 745                                                        &RMF_MDT_BODY);
 746
 747         LASSERT(body);
 748         if (body->valid & OBD_MD_FLMTIME &&
 749             body->mtime > LTIME_S(inode->i_mtime)) {
 750                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
 751                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 752                 LTIME_S(inode->i_mtime) = body->mtime;
 753         }
 754         if (body->valid & OBD_MD_FLCTIME &&
 755             body->ctime > LTIME_S(inode->i_ctime))
 756                 LTIME_S(inode->i_ctime) = body->ctime;
 757 }
 758
 759 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 760                        const char *tgt, int mode, int rdev,
 761                        __u32 opc)
 762 {
 763         struct ptlrpc_request *request = NULL;
 764         struct md_op_data *op_data;
 765         struct inode *inode = NULL;
 766         struct ll_sb_info *sbi = ll_i2sbi(dir);
 767         int tgt_len = 0;
 768         int err;
 769
 770         if (unlikely(tgt != NULL))
 771                 tgt_len = strlen(tgt) + 1;
 772
 773         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 774                                      dentry->d_name.name,
 775                                      dentry->d_name.len,
 776                                      0, opc, NULL);
 777         if (IS_ERR(op_data)) {
 778                 err = PTR_ERR(op_data);
 779                 goto err_exit;
 780         }
 781
 782         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 783                         from_kuid(&init_user_ns, current_fsuid()),
 784                         from_kgid(&init_user_ns, current_fsgid()),
 785                         cfs_curproc_cap_pack(), rdev, &request);
 786         ll_finish_md_op_data(op_data);
 787         if (err)
 788                 goto err_exit;
 789
 790         ll_update_times(request, dir);
 791
 792         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 793         if (err)
 794                 goto err_exit;
 795
 796         d_instantiate(dentry, inode);
 797 err_exit:
 798         ptlrpc_req_finished(request);
 799
 800         return err;
 801 }
 802
 803 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 804                     umode_t mode, dev_t rdev)
 805 {
 806         int err;
 807
 808         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p) mode %o dev %x\n",
 809                dchild, dir->i_ino, dir->i_generation, dir,
 810                mode, old_encode_dev(rdev));
 811
 812         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 813                 mode &= ~current_umask();
 814
 815         switch (mode & S_IFMT) {
 816         case 0:
 817                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 818         case S_IFREG:
 819         case S_IFCHR:
 820         case S_IFBLK:
 821         case S_IFIFO:
 822         case S_IFSOCK:
 823                 err = ll_new_node(dir, dchild, NULL, mode,
 824                                   old_encode_dev(rdev),
 825                                   LUSTRE_OPC_MKNOD);
 826                 break;
 827         case S_IFDIR:
 828                 err = -EPERM;
 829                 break;
 830         default:
 831                 err = -EINVAL;
 832         }
 833
 834         if (!err)
 835                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 836
 837         return err;
 838 }
 839
 840 /*
 841  * Plain create. Intent create is handled in atomic_open.
 842  */
 843 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 844                         umode_t mode, bool want_excl)
 845 {
 846         int rc;
 847
 848         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u, excl=%d\n",
 849                dentry, dir->i_ino,
 850                dir->i_generation, dir, mode, want_excl);
 851
 852         rc = ll_mknod(dir, dentry, mode, 0);
 853
 854         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 855
 856         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 857                dentry, d_unhashed(dentry));
 858
 859         return rc;
 860 }
 861
 862 static inline void ll_get_child_fid(struct dentry *child, struct lu_fid *fid)
 863 {
 864         if (d_really_is_positive(child))
 865                 *fid = *ll_inode2fid(d_inode(child));
 866 }
 867
 868 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 869 {
 870         struct mdt_body *body;
 871         struct lov_mds_md *eadata;
 872         struct lov_stripe_md *lsm = NULL;
 873         struct obd_trans_info oti = { 0 };
 874         struct obdo *oa;
 875         int rc;
 876
 877         /* req is swabbed so this is safe */
 878         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 879         if (!(body->valid & OBD_MD_FLEASIZE))
 880                 return 0;
 881
 882         if (body->eadatasize == 0) {
 883                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 884                 rc = -EPROTO;
 885                 goto out;
 886         }
 887
 888         /* The MDS sent back the EA because we unlinked the last reference
 889          * to this file. Use this EA to unlink the objects on the OST.
 890          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 891          * check it is complete and sensible. */
 892         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 893                                               body->eadatasize);
 894         LASSERT(eadata != NULL);
 895
 896         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 897         if (rc < 0) {
 898                 CERROR("obd_unpackmd: %d\n", rc);
 899                 goto out;
 900         }
 901         LASSERT(rc >= sizeof(*lsm));
 902
 903         oa = kmem_cache_alloc(obdo_cachep, GFP_NOFS | __GFP_ZERO);
 904         if (oa == NULL) {
 905                 rc = -ENOMEM;
 906                 goto out_free_memmd;
 907         }
 908
 909         oa->o_oi = lsm->lsm_oi;
 910         oa->o_mode = body->mode & S_IFMT;
 911         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 912
 913         if (body->valid & OBD_MD_FLCOOKIE) {
 914                 oa->o_valid |= OBD_MD_FLCOOKIE;
 915                 oti.oti_logcookies =
 916                         req_capsule_server_sized_get(&request->rq_pill,
 917                                                      &RMF_LOGCOOKIES,
 918                                                    sizeof(struct llog_cookie) *
 919                                                      lsm->lsm_stripe_count);
 920                 if (oti.oti_logcookies == NULL) {
 921                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 922                         body->valid &= ~OBD_MD_FLCOOKIE;
 923                 }
 924         }
 925
 926         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 927                          ll_i2mdexp(dir));
 928         if (rc)
 929                 CERROR("obd destroy objid "DOSTID" error %d\n",
 930                        POSTID(&lsm->lsm_oi), rc);
 931 out_free_memmd:
 932         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 933         kmem_cache_free(obdo_cachep, oa);
 934 out:
 935         return rc;
 936 }
 937
 938 /* ll_unlink() doesn't update the inode with the new link count.
 939  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 940  * is any lock existing. They will recycle dentries and inodes based upon locks
 941  * too. b=20433 */
 942 static int ll_unlink(struct inode *dir, struct dentry *dentry)
 943 {
 944         struct ptlrpc_request *request = NULL;
 945         struct md_op_data *op_data;
 946         int rc;
 947
 948         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 949                dentry, dir->i_ino, dir->i_generation, dir);
 950
 951         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 952                                      dentry->d_name.name,
 953                                      dentry->d_name.len,
 954                                      0, LUSTRE_OPC_ANY, NULL);
 955         if (IS_ERR(op_data))
 956                 return PTR_ERR(op_data);
 957
 958         ll_get_child_fid(dentry, &op_data->op_fid3);
 959         op_data->op_fid2 = op_data->op_fid3;
 960         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 961         ll_finish_md_op_data(op_data);
 962         if (rc)
 963                 goto out;
 964
 965         ll_update_times(request, dir);
 966         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 967
 968         rc = ll_objects_destroy(request, dir);
 969  out:
 970         ptlrpc_req_finished(request);
 971         return rc;
 972 }
 973
 974 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 975 {
 976         int err;
 977
 978         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 979                dentry, dir->i_ino, dir->i_generation, dir);
 980
 981         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 982                 mode &= ~current_umask();
 983         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
 984         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
 985
 986         if (!err)
 987                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 988
 989         return err;
 990 }
 991
 992 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
 993 {
 994         struct ptlrpc_request *request = NULL;
 995         struct md_op_data *op_data;
 996         int rc;
 997
 998         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 999                dentry, dir->i_ino, dir->i_generation, dir);
1000
1001         op_data = ll_prep_md_op_data(NULL, dir, NULL,
1002                                      dentry->d_name.name,
1003                                      dentry->d_name.len,
1004                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1005         if (IS_ERR(op_data))
1006                 return PTR_ERR(op_data);
1007
1008         ll_get_child_fid(dentry, &op_data->op_fid3);
1009         op_data->op_fid2 = op_data->op_fid3;
1010         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1011         ll_finish_md_op_data(op_data);
1012         if (rc == 0) {
1013                 ll_update_times(request, dir);
1014                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1015         }
1016
1017         ptlrpc_req_finished(request);
1018         return rc;
1019 }
1020
1021 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1022                       const char *oldname)
1023 {
1024         int err;
1025
1026         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),target=%.*s\n",
1027                dentry, dir->i_ino, dir->i_generation,
1028                dir, 3000, oldname);
1029
1030         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1031                         0, LUSTRE_OPC_SYMLINK);
1032
1033         if (!err)
1034                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1035
1036         return err;
1037 }
1038
1039 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1040                    struct dentry *new_dentry)
1041 {
1042         struct inode *src = d_inode(old_dentry);
1043         struct ll_sb_info *sbi = ll_i2sbi(dir);
1044         struct ptlrpc_request *request = NULL;
1045         struct md_op_data *op_data;
1046         int err;
1047
1048         CDEBUG(D_VFSTRACE,
1049                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%pd\n",
1050                src->i_ino, src->i_generation, src, dir->i_ino,
1051                dir->i_generation, dir, new_dentry);
1052
1053         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1054                                      new_dentry->d_name.len,
1055                                      0, LUSTRE_OPC_ANY, NULL);
1056         if (IS_ERR(op_data))
1057                 return PTR_ERR(op_data);
1058
1059         err = md_link(sbi->ll_md_exp, op_data, &request);
1060         ll_finish_md_op_data(op_data);
1061         if (err)
1062                 goto out;
1063
1064         ll_update_times(request, dir);
1065         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1066 out:
1067         ptlrpc_req_finished(request);
1068         return err;
1069 }
1070
1071 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1072                      struct inode *new_dir, struct dentry *new_dentry)
1073 {
1074         struct ptlrpc_request *request = NULL;
1075         struct ll_sb_info *sbi = ll_i2sbi(old_dir);
1076         struct md_op_data *op_data;
1077         int err;
1078
1079         CDEBUG(D_VFSTRACE,
1080                "VFS Op:oldname=%pd,src_dir=%lu/%u(%p),newname=%pd,tgt_dir=%lu/%u(%p)\n",
1081                old_dentry, old_dir->i_ino, old_dir->i_generation, old_dir,
1082                new_dentry, new_dir->i_ino, new_dir->i_generation, new_dir);
1083
1084         op_data = ll_prep_md_op_data(NULL, old_dir, new_dir, NULL, 0, 0,
1085                                      LUSTRE_OPC_ANY, NULL);
1086         if (IS_ERR(op_data))
1087                 return PTR_ERR(op_data);
1088
1089         ll_get_child_fid(old_dentry, &op_data->op_fid3);
1090         ll_get_child_fid(new_dentry, &op_data->op_fid4);
1091         err = md_rename(sbi->ll_md_exp, op_data,
1092                         old_dentry->d_name.name,
1093                         old_dentry->d_name.len,
1094                         new_dentry->d_name.name,
1095                         new_dentry->d_name.len, &request);
1096         ll_finish_md_op_data(op_data);
1097         if (!err) {
1098                 ll_update_times(request, old_dir);
1099                 ll_update_times(request, new_dir);
1100                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1101                 err = ll_objects_destroy(request, old_dir);
1102         }
1103
1104         ptlrpc_req_finished(request);
1105         if (!err)
1106                 d_move(old_dentry, new_dentry);
1107         return err;
1108 }
1109
1110 const struct inode_operations ll_dir_inode_operations = {
1111         .mknod        = ll_mknod,
1112         .atomic_open        = ll_atomic_open,
1113         .lookup      = ll_lookup_nd,
1114         .create      = ll_create_nd,
1115         /* We need all these non-raw things for NFSD, to not patch it. */
1116         .unlink      = ll_unlink,
1117         .mkdir        = ll_mkdir,
1118         .rmdir        = ll_rmdir,
1119         .symlink            = ll_symlink,
1120         .link          = ll_link,
1121         .rename      = ll_rename,
1122         .setattr            = ll_setattr,
1123         .getattr            = ll_getattr,
1124         .permission      = ll_inode_permission,
1125         .setxattr          = ll_setxattr,
1126         .getxattr          = ll_getxattr,
1127         .listxattr        = ll_listxattr,
1128         .removexattr    = ll_removexattr,
1129         .get_acl            = ll_get_acl,
1130 };
1131
1132 const struct inode_operations ll_special_inode_operations = {
1133         .setattr        = ll_setattr,
1134         .getattr        = ll_getattr,
1135         .permission     = ll_inode_permission,
1136         .setxattr       = ll_setxattr,
1137         .getxattr       = ll_getxattr,
1138         .listxattr      = ll_listxattr,
1139         .removexattr    = ll_removexattr,
1140         .get_acl            = ll_get_acl,
1141 };