kernel/drivers/staging/lustre/lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include "../include/obd_support.h"
  48 #include "../include/lustre_fid.h"
  49 #include "../include/lustre_lite.h"
  50 #include "../include/lustre_dlm.h"
  51 #include "../include/lustre_ver.h"
  52 #include "llite_internal.h"
  53
  54 static int ll_create_it(struct inode *, struct dentry *,
  55                         int, struct lookup_intent *);
  56
  57 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
  58 static int ll_test_inode(struct inode *inode, void *opaque)
  59 {
  60         struct ll_inode_info *lli = ll_i2info(inode);
  61         struct lustre_md     *md = opaque;
  62
  63         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  64                 CERROR("MDS body missing FID\n");
  65                 return 0;
  66         }
  67
  68         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
  69                 return 0;
  70
  71         return 1;
  72 }
  73
  74 static int ll_set_inode(struct inode *inode, void *opaque)
  75 {
  76         struct ll_inode_info *lli = ll_i2info(inode);
  77         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  78
  79         if (unlikely(!(body->valid & OBD_MD_FLID))) {
  80                 CERROR("MDS body missing FID\n");
  81                 return -EINVAL;
  82         }
  83
  84         lli->lli_fid = body->fid1;
  85         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
  86                 CERROR("Can not initialize inode " DFID
  87                        " without object type: valid = %#llx\n",
  88                        PFID(&lli->lli_fid), body->valid);
  89                 return -EINVAL;
  90         }
  91
  92         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
  93         if (unlikely(inode->i_mode == 0)) {
  94                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  95                 return -EINVAL;
  96         }
  97
  98         ll_lli_init(lli);
  99
 100         return 0;
 101 }
 102
 103 /*
 104  * 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
 147         ll_lock_dcache(dir);
 148         hlist_for_each_entry(dentry, &dir->i_dentry, d_u.d_alias) {
 149                 spin_lock(&dentry->d_lock);
 150                 if (!list_empty(&dentry->d_subdirs)) {
 151                         struct dentry *child;
 152
 153                         list_for_each_entry_safe(child, tmp_subdir,
 154                                                  &dentry->d_subdirs,
 155                                                  d_child) {
 156                                 if (d_really_is_negative(child))
 157                                         d_lustre_invalidate(child, 1);
 158                         }
 159                 }
 160                 spin_unlock(&dentry->d_lock);
 161         }
 162         ll_unlock_dcache(dir);
 163 }
 164
 165 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 166                        void *data, int flag)
 167 {
 168         struct lustre_handle lockh;
 169         int rc;
 170
 171         switch (flag) {
 172         case LDLM_CB_BLOCKING:
 173                 ldlm_lock2handle(lock, &lockh);
 174                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 175                 if (rc < 0) {
 176                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 177                         return rc;
 178                 }
 179                 break;
 180         case LDLM_CB_CANCELING: {
 181                 struct inode *inode = ll_inode_from_resource_lock(lock);
 182                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 183
 184                 /* Inode is set to lock->l_resource->lr_lvb_inode
 185                  * for mdc - bug 24555 */
 186                 LASSERT(lock->l_ast_data == NULL);
 187
 188                 if (inode == NULL)
 189                         break;
 190
 191                 /* Invalidate all dentries associated with this inode */
 192                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 193
 194                 if (!fid_res_name_eq(ll_inode2fid(inode),
 195                                      &lock->l_resource->lr_name)) {
 196                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 197                                    PFID(ll_inode2fid(inode)), inode);
 198                         LBUG();
 199                 }
 200
 201                 if (bits & MDS_INODELOCK_XATTR) {
 202                         ll_xattr_cache_destroy(inode);
 203                         bits &= ~MDS_INODELOCK_XATTR;
 204                 }
 205
 206                 /* For OPEN locks we differentiate between lock modes
 207                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 208                 if (bits & MDS_INODELOCK_OPEN)
 209                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 210
 211                 if (bits & MDS_INODELOCK_OPEN) {
 212                         fmode_t fmode;
 213
 214                         switch (lock->l_req_mode) {
 215                         case LCK_CW:
 216                                 fmode = FMODE_WRITE;
 217                                 break;
 218                         case LCK_PR:
 219                                 fmode = FMODE_EXEC;
 220                                 break;
 221                         case LCK_CR:
 222                                 fmode = FMODE_READ;
 223                                 break;
 224                         default:
 225                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 226                                 LBUG();
 227                         }
 228
 229                         ll_md_real_close(inode, fmode);
 230                 }
 231
 232                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 233                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 234                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 235
 236                 if (bits & MDS_INODELOCK_LAYOUT) {
 237                         struct cl_object_conf conf = {
 238                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 239                                 .coc_inode = inode,
 240                         };
 241
 242                         rc = ll_layout_conf(inode, &conf);
 243                         if (rc < 0)
 244                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 245                                        DFID": rc = %d\n",
 246                                        PFID(ll_inode2fid(inode)), rc);
 247                 }
 248
 249                 if (bits & MDS_INODELOCK_UPDATE) {
 250                         struct ll_inode_info *lli = ll_i2info(inode);
 251
 252                         spin_lock(&lli->lli_lock);
 253                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 254                         spin_unlock(&lli->lli_lock);
 255                 }
 256
 257                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 258                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 259                                inode->i_ino);
 260                         truncate_inode_pages(inode->i_mapping, 0);
 261                         ll_invalidate_negative_children(inode);
 262                 }
 263
 264                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 265                     inode->i_sb->s_root != NULL &&
 266                     !is_root_inode(inode))
 267                         ll_invalidate_aliases(inode);
 268
 269                 iput(inode);
 270                 break;
 271         }
 272         default:
 273                 LBUG();
 274         }
 275
 276         return 0;
 277 }
 278
 279 __u32 ll_i2suppgid(struct inode *i)
 280 {
 281         if (in_group_p(i->i_gid))
 282                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 283         else
 284                 return (__u32)(-1);
 285 }
 286
 287 /* Pack the required supplementary groups into the supplied groups array.
 288  * If we don't need to use the groups from the target inode(s) then we
 289  * instead pack one or more groups from the user's supplementary group
 290  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 291 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 292 {
 293 #if 0
 294         int i;
 295 #endif
 296
 297         LASSERT(i1 != NULL);
 298         LASSERT(suppgids != NULL);
 299
 300         suppgids[0] = ll_i2suppgid(i1);
 301
 302         if (i2)
 303                 suppgids[1] = ll_i2suppgid(i2);
 304                 else
 305                         suppgids[1] = -1;
 306
 307 #if 0
 308         for (i = 0; i < current_ngroups; i++) {
 309                 if (suppgids[0] == -1) {
 310                         if (current_groups[i] != suppgids[1])
 311                                 suppgids[0] = current_groups[i];
 312                         continue;
 313                 }
 314                 if (suppgids[1] == -1) {
 315                         if (current_groups[i] != suppgids[0])
 316                                 suppgids[1] = current_groups[i];
 317                         continue;
 318                 }
 319                 break;
 320         }
 321 #endif
 322 }
 323
 324 /*
 325  * try to reuse three types of dentry:
 326  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 327  *    by concurrent .revalidate).
 328  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 329  *    be cleared by others calling d_lustre_revalidate).
 330  * 3. DISCONNECTED alias.
 331  */
 332 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 333 {
 334         struct dentry *alias, *discon_alias, *invalid_alias;
 335
 336         if (hlist_empty(&inode->i_dentry))
 337                 return NULL;
 338
 339         discon_alias = invalid_alias = NULL;
 340
 341         ll_lock_dcache(inode);
 342         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 343                 LASSERT(alias != dentry);
 344
 345                 spin_lock(&alias->d_lock);
 346                 if (alias->d_flags & DCACHE_DISCONNECTED)
 347                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 348                         discon_alias = alias;
 349                 else if (alias->d_parent == dentry->d_parent         &&
 350                          alias->d_name.hash == dentry->d_name.hash       &&
 351                          alias->d_name.len == dentry->d_name.len         &&
 352                          memcmp(alias->d_name.name, dentry->d_name.name,
 353                                 dentry->d_name.len) == 0)
 354                         invalid_alias = alias;
 355                 spin_unlock(&alias->d_lock);
 356
 357                 if (invalid_alias)
 358                         break;
 359         }
 360         alias = invalid_alias ?: discon_alias ?: NULL;
 361         if (alias) {
 362                 spin_lock(&alias->d_lock);
 363                 dget_dlock(alias);
 364                 spin_unlock(&alias->d_lock);
 365         }
 366         ll_unlock_dcache(inode);
 367
 368         return alias;
 369 }
 370
 371 /*
 372  * Similar to d_splice_alias(), but lustre treats invalid alias
 373  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 374  */
 375 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 376 {
 377         struct dentry *new;
 378         int rc;
 379
 380         if (inode) {
 381                 new = ll_find_alias(inode, de);
 382                 if (new) {
 383                         rc = ll_d_init(new);
 384                         if (rc < 0) {
 385                                 dput(new);
 386                                 return ERR_PTR(rc);
 387                         }
 388                         d_move(new, de);
 389                         iput(inode);
 390                         CDEBUG(D_DENTRY,
 391                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 392                               new, d_inode(new), d_count(new), new->d_flags);
 393                         return new;
 394                 }
 395         }
 396         rc = ll_d_init(de);
 397         if (rc < 0)
 398                 return ERR_PTR(rc);
 399         d_add(de, inode);
 400         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 401                de, d_inode(de), d_count(de), de->d_flags);
 402         return de;
 403 }
 404
 405 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 406                                struct lookup_intent *it,
 407                                struct inode *parent, struct dentry **de)
 408 {
 409         struct inode *inode = NULL;
 410         __u64 bits = 0;
 411         int rc = 0;
 412
 413         /* NB 1 request reference will be taken away by ll_intent_lock()
 414          * when I return */
 415         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 416                it->d.lustre.it_disposition);
 417         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 418                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 419                 if (rc)
 420                         return rc;
 421
 422                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 423
 424                 /* We used to query real size from OSTs here, but actually
 425                    this is not needed. For stat() calls size would be updated
 426                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 427                    2.4 and
 428                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 429                    Everybody else who needs correct file size would call
 430                    ll_glimpse_size or some equivalent themselves anyway.
 431                    Also see bug 7198. */
 432         }
 433
 434         /* Only hash *de if it is unhashed (new dentry).
 435          * Atoimc_open may passing hashed dentries for open.
 436          */
 437         if (d_unhashed(*de)) {
 438                 struct dentry *alias;
 439
 440                 alias = ll_splice_alias(inode, *de);
 441                 if (IS_ERR(alias)) {
 442                         rc = PTR_ERR(alias);
 443                         goto out;
 444                 }
 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 out:
 476         if (rc != 0 && it->it_op & IT_OPEN)
 477                 ll_open_cleanup((*de)->d_sb, request);
 478
 479         return rc;
 480 }
 481
 482 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 483                                    struct lookup_intent *it, int lookup_flags)
 484 {
 485         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 486         struct dentry *save = dentry, *retval;
 487         struct ptlrpc_request *req = NULL;
 488         struct inode *inode;
 489         struct md_op_data *op_data;
 490         __u32 opc;
 491         int rc;
 492
 493         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 494                 return ERR_PTR(-ENAMETOOLONG);
 495
 496         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 497                dentry, parent->i_ino,
 498                parent->i_generation, parent, LL_IT2STR(it));
 499
 500         if (d_mountpoint(dentry))
 501                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 502
 503         if (it == NULL || it->it_op == IT_GETXATTR)
 504                 it = &lookup_it;
 505
 506         if (it->it_op == IT_GETATTR) {
 507                 rc = ll_statahead_enter(parent, &dentry, 0);
 508                 if (rc == 1) {
 509                         if (dentry == save)
 510                                 retval = NULL;
 511                         else
 512                                 retval = dentry;
 513                         goto out;
 514                 }
 515         }
 516
 517         if (it->it_op & IT_CREAT)
 518                 opc = LUSTRE_OPC_CREATE;
 519         else
 520                 opc = LUSTRE_OPC_ANY;
 521
 522         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 523                                      dentry->d_name.len, lookup_flags, opc,
 524                                      NULL);
 525         if (IS_ERR(op_data))
 526                 return (void *)op_data;
 527
 528         /* enforce umask if acl disabled or MDS doesn't support umask */
 529         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 530                 it->it_create_mode &= ~current_umask();
 531
 532         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 533                             lookup_flags, &req, ll_md_blocking_ast, 0);
 534         ll_finish_md_op_data(op_data);
 535         if (rc < 0) {
 536                 retval = ERR_PTR(rc);
 537                 goto out;
 538         }
 539
 540         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 541         if (rc != 0) {
 542                 ll_intent_release(it);
 543                 retval = ERR_PTR(rc);
 544                 goto out;
 545         }
 546
 547         inode = d_inode(dentry);
 548         if ((it->it_op & IT_OPEN) && inode &&
 549             !S_ISREG(inode->i_mode) &&
 550             !S_ISDIR(inode->i_mode)) {
 551                 ll_release_openhandle(inode, it);
 552         }
 553         ll_lookup_finish_locks(it, inode);
 554
 555         if (dentry == save)
 556                 retval = NULL;
 557         else
 558                 retval = dentry;
 559         goto out;
 560  out:
 561         if (req)
 562                 ptlrpc_req_finished(req);
 563         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 564                 ll_statahead_mark(parent, dentry);
 565         return retval;
 566 }
 567
 568 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 569                                    unsigned int flags)
 570 {
 571         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 572         struct dentry *de;
 573
 574         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u\n",
 575                dentry, parent->i_ino,
 576                parent->i_generation, parent, flags);
 577
 578         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 579         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 580                 return NULL;
 581
 582         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 583                 itp = NULL;
 584         else
 585                 itp = &it;
 586         de = ll_lookup_it(parent, dentry, itp, 0);
 587
 588         if (itp != NULL)
 589                 ll_intent_release(itp);
 590
 591         return de;
 592 }
 593
 594 /*
 595  * For cached negative dentry and new dentry, handle lookup/create/open
 596  * together.
 597  */
 598 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 599                           struct file *file, unsigned open_flags,
 600                           umode_t mode, int *opened)
 601 {
 602         struct lookup_intent *it;
 603         struct dentry *de;
 604         long long lookup_flags = LOOKUP_OPEN;
 605         int rc = 0;
 606
 607         CDEBUG(D_VFSTRACE,
 608                "VFS Op:name=%pd,dir=%lu/%u(%p),file %p,open_flags %x,mode %x opened %d\n",
 609                dentry, dir->i_ino,
 610                dir->i_generation, dir, file, open_flags, mode, *opened);
 611
 612         it = kzalloc(sizeof(*it), GFP_NOFS);
 613         if (!it)
 614                 return -ENOMEM;
 615
 616         it->it_op = IT_OPEN;
 617         if (open_flags & O_CREAT) {
 618                 it->it_op |= IT_CREAT;
 619                 lookup_flags |= LOOKUP_CREATE;
 620         }
 621         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 622         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 623
 624         /* Dentry added to dcache tree in ll_lookup_it */
 625         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 626         if (IS_ERR(de))
 627                 rc = PTR_ERR(de);
 628         else if (de != NULL)
 629                 dentry = de;
 630
 631         if (!rc) {
 632                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 633                         /* Dentry instantiated in ll_create_it. */
 634                         rc = ll_create_it(dir, dentry, mode, it);
 635                         if (rc) {
 636                                 /* We dget in ll_splice_alias. */
 637                                 if (de != NULL)
 638                                         dput(de);
 639                                 goto out_release;
 640                         }
 641
 642                         *opened |= FILE_CREATED;
 643                 }
 644                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 645                         /* Open dentry. */
 646                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 647                                 /* We cannot call open here as it would
 648                                  * deadlock.
 649                                  */
 650                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 651                                         ptlrpc_req_finished(
 652                                                        (struct ptlrpc_request *)
 653                                                           it->d.lustre.it_data);
 654                                 rc = finish_no_open(file, de);
 655                         } else {
 656                                 file->private_data = it;
 657                                 rc = finish_open(file, dentry, NULL, opened);
 658                                 /* We dget in ll_splice_alias. finish_open takes
 659                                  * care of dget for fd open.
 660                                  */
 661                                 if (de != NULL)
 662                                         dput(de);
 663                         }
 664                 } else {
 665                         rc = finish_no_open(file, de);
 666                 }
 667         }
 668
 669 out_release:
 670         ll_intent_release(it);
 671         kfree(it);
 672
 673         return rc;
 674 }
 675
 676 /* We depend on "mode" being set with the proper file type/umask by now */
 677 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 678 {
 679         struct inode *inode = NULL;
 680         struct ptlrpc_request *request = NULL;
 681         struct ll_sb_info *sbi = ll_i2sbi(dir);
 682         int rc;
 683
 684         LASSERT(it && it->d.lustre.it_disposition);
 685
 686         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 687         request = it->d.lustre.it_data;
 688         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 689         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 690         if (rc) {
 691                 inode = ERR_PTR(rc);
 692                 goto out;
 693         }
 694
 695         LASSERT(hlist_empty(&inode->i_dentry));
 696
 697         /* We asked for a lock on the directory, but were granted a
 698          * lock on the inode.  Since we finally have an inode pointer,
 699          * stuff it in the lock. */
 700         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 701                inode, inode->i_ino, inode->i_generation);
 702         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 703  out:
 704         ptlrpc_req_finished(request);
 705         return inode;
 706 }
 707
 708 /*
 709  * By the time this is called, we already have created the directory cache
 710  * entry for the new file, but it is so far negative - it has no inode.
 711  *
 712  * We defer creating the OBD object(s) until open, to keep the intent and
 713  * non-intent code paths similar, and also because we do not have the MDS
 714  * inode number before calling ll_create_node() (which is needed for LOV),
 715  * so we would need to do yet another RPC to the MDS to store the LOV EA
 716  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 717  * lmm_size in datalen (the MDS still has code which will handle that).
 718  *
 719  * If the create succeeds, we fill in the inode information
 720  * with d_instantiate().
 721  */
 722 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 723                         struct lookup_intent *it)
 724 {
 725         struct inode *inode;
 726         int rc = 0;
 727
 728         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 729                dentry, dir->i_ino,
 730                dir->i_generation, dir, LL_IT2STR(it));
 731
 732         rc = it_open_error(DISP_OPEN_CREATE, it);
 733         if (rc)
 734                 return rc;
 735
 736         inode = ll_create_node(dir, it);
 737         if (IS_ERR(inode))
 738                 return PTR_ERR(inode);
 739
 740         d_instantiate(dentry, inode);
 741         return 0;
 742 }
 743
 744 static void ll_update_times(struct ptlrpc_request *request,
 745                             struct inode *inode)
 746 {
 747         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 748                                                        &RMF_MDT_BODY);
 749
 750         LASSERT(body);
 751         if (body->valid & OBD_MD_FLMTIME &&
 752             body->mtime > LTIME_S(inode->i_mtime)) {
 753                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
 754                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 755                 LTIME_S(inode->i_mtime) = body->mtime;
 756         }
 757         if (body->valid & OBD_MD_FLCTIME &&
 758             body->ctime > LTIME_S(inode->i_ctime))
 759                 LTIME_S(inode->i_ctime) = body->ctime;
 760 }
 761
 762 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 763                        const char *tgt, int mode, int rdev,
 764                        __u32 opc)
 765 {
 766         struct ptlrpc_request *request = NULL;
 767         struct md_op_data *op_data;
 768         struct inode *inode = NULL;
 769         struct ll_sb_info *sbi = ll_i2sbi(dir);
 770         int tgt_len = 0;
 771         int err;
 772
 773         if (unlikely(tgt != NULL))
 774                 tgt_len = strlen(tgt) + 1;
 775
 776         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 777                                      dentry->d_name.name,
 778                                      dentry->d_name.len,
 779                                      0, opc, NULL);
 780         if (IS_ERR(op_data)) {
 781                 err = PTR_ERR(op_data);
 782                 goto err_exit;
 783         }
 784
 785         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 786                         from_kuid(&init_user_ns, current_fsuid()),
 787                         from_kgid(&init_user_ns, current_fsgid()),
 788                         cfs_curproc_cap_pack(), rdev, &request);
 789         ll_finish_md_op_data(op_data);
 790         if (err)
 791                 goto err_exit;
 792
 793         ll_update_times(request, dir);
 794
 795         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 796         if (err)
 797                 goto err_exit;
 798
 799         d_instantiate(dentry, inode);
 800 err_exit:
 801         ptlrpc_req_finished(request);
 802
 803         return err;
 804 }
 805
 806 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 807                     umode_t mode, dev_t rdev)
 808 {
 809         int err;
 810
 811         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p) mode %o dev %x\n",
 812                dchild, dir->i_ino, dir->i_generation, dir,
 813                mode, old_encode_dev(rdev));
 814
 815         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 816                 mode &= ~current_umask();
 817
 818         switch (mode & S_IFMT) {
 819         case 0:
 820                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 821         case S_IFREG:
 822         case S_IFCHR:
 823         case S_IFBLK:
 824         case S_IFIFO:
 825         case S_IFSOCK:
 826                 err = ll_new_node(dir, dchild, NULL, mode,
 827                                   old_encode_dev(rdev),
 828                                   LUSTRE_OPC_MKNOD);
 829                 break;
 830         case S_IFDIR:
 831                 err = -EPERM;
 832                 break;
 833         default:
 834                 err = -EINVAL;
 835         }
 836
 837         if (!err)
 838                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 839
 840         return err;
 841 }
 842
 843 /*
 844  * Plain create. Intent create is handled in atomic_open.
 845  */
 846 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 847                         umode_t mode, bool want_excl)
 848 {
 849         int rc;
 850
 851         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u, excl=%d\n",
 852                dentry, dir->i_ino,
 853                dir->i_generation, dir, mode, want_excl);
 854
 855         rc = ll_mknod(dir, dentry, mode, 0);
 856
 857         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 858
 859         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 860                dentry, d_unhashed(dentry));
 861
 862         return rc;
 863 }
 864
 865 static inline void ll_get_child_fid(struct dentry *child, struct lu_fid *fid)
 866 {
 867         if (d_really_is_positive(child))
 868                 *fid = *ll_inode2fid(d_inode(child));
 869 }
 870
 871 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 872 {
 873         struct mdt_body *body;
 874         struct lov_mds_md *eadata;
 875         struct lov_stripe_md *lsm = NULL;
 876         struct obd_trans_info oti = { 0 };
 877         struct obdo *oa;
 878         int rc;
 879
 880         /* req is swabbed so this is safe */
 881         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 882         if (!(body->valid & OBD_MD_FLEASIZE))
 883                 return 0;
 884
 885         if (body->eadatasize == 0) {
 886                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 887                 rc = -EPROTO;
 888                 goto out;
 889         }
 890
 891         /* The MDS sent back the EA because we unlinked the last reference
 892          * to this file. Use this EA to unlink the objects on the OST.
 893          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 894          * check it is complete and sensible. */
 895         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 896                                               body->eadatasize);
 897         LASSERT(eadata != NULL);
 898
 899         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 900         if (rc < 0) {
 901                 CERROR("obd_unpackmd: %d\n", rc);
 902                 goto out;
 903         }
 904         LASSERT(rc >= sizeof(*lsm));
 905
 906         oa = kmem_cache_alloc(obdo_cachep, GFP_NOFS | __GFP_ZERO);
 907         if (oa == NULL) {
 908                 rc = -ENOMEM;
 909                 goto out_free_memmd;
 910         }
 911
 912         oa->o_oi = lsm->lsm_oi;
 913         oa->o_mode = body->mode & S_IFMT;
 914         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 915
 916         if (body->valid & OBD_MD_FLCOOKIE) {
 917                 oa->o_valid |= OBD_MD_FLCOOKIE;
 918                 oti.oti_logcookies =
 919                         req_capsule_server_sized_get(&request->rq_pill,
 920                                                      &RMF_LOGCOOKIES,
 921                                                    sizeof(struct llog_cookie) *
 922                                                      lsm->lsm_stripe_count);
 923                 if (oti.oti_logcookies == NULL) {
 924                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 925                         body->valid &= ~OBD_MD_FLCOOKIE;
 926                 }
 927         }
 928
 929         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 930                          ll_i2mdexp(dir));
 931         if (rc)
 932                 CERROR("obd destroy objid "DOSTID" error %d\n",
 933                        POSTID(&lsm->lsm_oi), rc);
 934 out_free_memmd:
 935         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 936         kmem_cache_free(obdo_cachep, oa);
 937 out:
 938         return rc;
 939 }
 940
 941 /* ll_unlink() doesn't update the inode with the new link count.
 942  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 943  * is any lock existing. They will recycle dentries and inodes based upon locks
 944  * too. b=20433 */
 945 static int ll_unlink(struct inode *dir, struct dentry *dentry)
 946 {
 947         struct ptlrpc_request *request = NULL;
 948         struct md_op_data *op_data;
 949         int rc;
 950
 951         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 952                dentry, dir->i_ino, dir->i_generation, dir);
 953
 954         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 955                                      dentry->d_name.name,
 956                                      dentry->d_name.len,
 957                                      0, LUSTRE_OPC_ANY, NULL);
 958         if (IS_ERR(op_data))
 959                 return PTR_ERR(op_data);
 960
 961         ll_get_child_fid(dentry, &op_data->op_fid3);
 962         op_data->op_fid2 = op_data->op_fid3;
 963         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 964         ll_finish_md_op_data(op_data);
 965         if (rc)
 966                 goto out;
 967
 968         ll_update_times(request, dir);
 969         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 970
 971         rc = ll_objects_destroy(request, dir);
 972  out:
 973         ptlrpc_req_finished(request);
 974         return rc;
 975 }
 976
 977 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 978 {
 979         int err;
 980
 981         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 982                dentry, dir->i_ino, dir->i_generation, dir);
 983
 984         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 985                 mode &= ~current_umask();
 986         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
 987         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
 988
 989         if (!err)
 990                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 991
 992         return err;
 993 }
 994
 995 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
 996 {
 997         struct ptlrpc_request *request = NULL;
 998         struct md_op_data *op_data;
 999         int rc;
1000
1001         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
1002                dentry, dir->i_ino, dir->i_generation, dir);
1003
1004         op_data = ll_prep_md_op_data(NULL, dir, NULL,
1005                                      dentry->d_name.name,
1006                                      dentry->d_name.len,
1007                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1008         if (IS_ERR(op_data))
1009                 return PTR_ERR(op_data);
1010
1011         ll_get_child_fid(dentry, &op_data->op_fid3);
1012         op_data->op_fid2 = op_data->op_fid3;
1013         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1014         ll_finish_md_op_data(op_data);
1015         if (rc == 0) {
1016                 ll_update_times(request, dir);
1017                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1018         }
1019
1020         ptlrpc_req_finished(request);
1021         return rc;
1022 }
1023
1024 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1025                       const char *oldname)
1026 {
1027         int err;
1028
1029         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),target=%.*s\n",
1030                dentry, dir->i_ino, dir->i_generation,
1031                dir, 3000, oldname);
1032
1033         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1034                         0, LUSTRE_OPC_SYMLINK);
1035
1036         if (!err)
1037                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1038
1039         return err;
1040 }
1041
1042 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1043                    struct dentry *new_dentry)
1044 {
1045         struct inode *src = d_inode(old_dentry);
1046         struct ll_sb_info *sbi = ll_i2sbi(dir);
1047         struct ptlrpc_request *request = NULL;
1048         struct md_op_data *op_data;
1049         int err;
1050
1051         CDEBUG(D_VFSTRACE,
1052                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%pd\n",
1053                src->i_ino, src->i_generation, src, dir->i_ino,
1054                dir->i_generation, dir, new_dentry);
1055
1056         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1057                                      new_dentry->d_name.len,
1058                                      0, LUSTRE_OPC_ANY, NULL);
1059         if (IS_ERR(op_data))
1060                 return PTR_ERR(op_data);
1061
1062         err = md_link(sbi->ll_md_exp, op_data, &request);
1063         ll_finish_md_op_data(op_data);
1064         if (err)
1065                 goto out;
1066
1067         ll_update_times(request, dir);
1068         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1069 out:
1070         ptlrpc_req_finished(request);
1071         return err;
1072 }
1073
1074 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1075                      struct inode *new_dir, struct dentry *new_dentry)
1076 {
1077         struct ptlrpc_request *request = NULL;
1078         struct ll_sb_info *sbi = ll_i2sbi(old_dir);
1079         struct md_op_data *op_data;
1080         int err;
1081
1082         CDEBUG(D_VFSTRACE,
1083                "VFS Op:oldname=%pd,src_dir=%lu/%u(%p),newname=%pd,tgt_dir=%lu/%u(%p)\n",
1084                old_dentry, old_dir->i_ino, old_dir->i_generation, old_dir,
1085                new_dentry, new_dir->i_ino, new_dir->i_generation, new_dir);
1086
1087         op_data = ll_prep_md_op_data(NULL, old_dir, new_dir, NULL, 0, 0,
1088                                      LUSTRE_OPC_ANY, NULL);
1089         if (IS_ERR(op_data))
1090                 return PTR_ERR(op_data);
1091
1092         ll_get_child_fid(old_dentry, &op_data->op_fid3);
1093         ll_get_child_fid(new_dentry, &op_data->op_fid4);
1094         err = md_rename(sbi->ll_md_exp, op_data,
1095                         old_dentry->d_name.name,
1096                         old_dentry->d_name.len,
1097                         new_dentry->d_name.name,
1098                         new_dentry->d_name.len, &request);
1099         ll_finish_md_op_data(op_data);
1100         if (!err) {
1101                 ll_update_times(request, old_dir);
1102                 ll_update_times(request, new_dir);
1103                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1104                 err = ll_objects_destroy(request, old_dir);
1105         }
1106
1107         ptlrpc_req_finished(request);
1108         if (!err)
1109                 d_move(old_dentry, new_dentry);
1110         return err;
1111 }
1112
1113 const struct inode_operations ll_dir_inode_operations = {
1114         .mknod        = ll_mknod,
1115         .atomic_open        = ll_atomic_open,
1116         .lookup      = ll_lookup_nd,
1117         .create      = ll_create_nd,
1118         /* We need all these non-raw things for NFSD, to not patch it. */
1119         .unlink      = ll_unlink,
1120         .mkdir        = ll_mkdir,
1121         .rmdir        = ll_rmdir,
1122         .symlink            = ll_symlink,
1123         .link          = ll_link,
1124         .rename      = ll_rename,
1125         .setattr            = ll_setattr,
1126         .getattr            = ll_getattr,
1127         .permission      = ll_inode_permission,
1128         .setxattr          = ll_setxattr,
1129         .getxattr          = ll_getxattr,
1130         .listxattr        = ll_listxattr,
1131         .removexattr    = ll_removexattr,
1132         .get_acl            = ll_get_acl,
1133 };
1134
1135 const struct inode_operations ll_special_inode_operations = {
1136         .setattr        = ll_setattr,
1137         .getattr        = ll_getattr,
1138         .permission     = ll_inode_permission,
1139         .setxattr       = ll_setxattr,
1140         .getxattr       = ll_getxattr,
1141         .listxattr      = ll_listxattr,
1142         .removexattr    = ll_removexattr,
1143         .get_acl            = ll_get_acl,
1144 };