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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #include <linux/fs.h>
  34 #include <linux/sched.h>
  35 #include <linux/mm.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/security.h>
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include "../include/obd_support.h"
  44 #include "../include/lustre_fid.h"
  45 #include "../include/lustre_lite.h"
  46 #include "../include/lustre_dlm.h"
  47 #include "../include/lustre_ver.h"
  48 #include "llite_internal.h"
  49
  50 static int ll_create_it(struct inode *, struct dentry *,
  51                         int, struct lookup_intent *);
  52
  53 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
  54 static int ll_test_inode(struct inode *inode, void *opaque)
  55 {
  56         struct ll_inode_info *lli = ll_i2info(inode);
  57         struct lustre_md     *md = opaque;
  58
  59         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  60                 CERROR("MDS body missing FID\n");
  61                 return 0;
  62         }
  63
  64         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
  65                 return 0;
  66
  67         return 1;
  68 }
  69
  70 static int ll_set_inode(struct inode *inode, void *opaque)
  71 {
  72         struct ll_inode_info *lli = ll_i2info(inode);
  73         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  74
  75         if (unlikely(!(body->valid & OBD_MD_FLID))) {
  76                 CERROR("MDS body missing FID\n");
  77                 return -EINVAL;
  78         }
  79
  80         lli->lli_fid = body->fid1;
  81         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
  82                 CERROR("Can not initialize inode " DFID
  83                        " without object type: valid = %#llx\n",
  84                        PFID(&lli->lli_fid), body->valid);
  85                 return -EINVAL;
  86         }
  87
  88         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
  89         if (unlikely(inode->i_mode == 0)) {
  90                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  91                 return -EINVAL;
  92         }
  93
  94         ll_lli_init(lli);
  95
  96         return 0;
  97 }
  98
  99 /*
 100  * Get an inode by inode number (already instantiated by the intent lookup).
 101  * Returns inode or NULL
 102  */
 103 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 104                       struct lustre_md *md)
 105 {
 106         struct inode     *inode;
 107
 108         LASSERT(hash != 0);
 109         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 110
 111         if (inode) {
 112                 if (inode->i_state & I_NEW) {
 113                         int rc = 0;
 114
 115                         ll_read_inode2(inode, md);
 116                         if (S_ISREG(inode->i_mode) &&
 117                             !ll_i2info(inode)->lli_clob) {
 118                                 CDEBUG(D_INODE,
 119                                        "%s: apply lsm %p to inode " DFID ".\n",
 120                                        ll_get_fsname(sb, NULL, 0), md->lsm,
 121                                        PFID(ll_inode2fid(inode)));
 122                                 rc = cl_file_inode_init(inode, md);
 123                         }
 124                         if (rc != 0) {
 125                                 iget_failed(inode);
 126                                 inode = NULL;
 127                         } else {
 128                                 unlock_new_inode(inode);
 129                         }
 130                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 131                         ll_update_inode(inode, md);
 132                         CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
 133                                PFID(&md->body->fid1), inode);
 134                 }
 135         }
 136         return inode;
 137 }
 138
 139 static void ll_invalidate_negative_children(struct inode *dir)
 140 {
 141         struct dentry *dentry, *tmp_subdir;
 142
 143         spin_lock(&dir->i_lock);
 144         hlist_for_each_entry(dentry, &dir->i_dentry, d_u.d_alias) {
 145                 spin_lock(&dentry->d_lock);
 146                 if (!list_empty(&dentry->d_subdirs)) {
 147                         struct dentry *child;
 148
 149                         list_for_each_entry_safe(child, tmp_subdir,
 150                                                  &dentry->d_subdirs,
 151                                                  d_child) {
 152                                 if (d_really_is_negative(child))
 153                                         d_lustre_invalidate(child, 1);
 154                         }
 155                 }
 156                 spin_unlock(&dentry->d_lock);
 157         }
 158         spin_unlock(&dir->i_lock);
 159 }
 160
 161 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 162                        void *data, int flag)
 163 {
 164         struct lustre_handle lockh;
 165         int rc;
 166
 167         switch (flag) {
 168         case LDLM_CB_BLOCKING:
 169                 ldlm_lock2handle(lock, &lockh);
 170                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 171                 if (rc < 0) {
 172                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 173                         return rc;
 174                 }
 175                 break;
 176         case LDLM_CB_CANCELING: {
 177                 struct inode *inode = ll_inode_from_resource_lock(lock);
 178                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 179
 180                 /* Inode is set to lock->l_resource->lr_lvb_inode
 181                  * for mdc - bug 24555
 182                  */
 183                 LASSERT(!lock->l_ast_data);
 184
 185                 if (!inode)
 186                         break;
 187
 188                 /* Invalidate all dentries associated with this inode */
 189                 LASSERT(ldlm_is_canceling(lock));
 190
 191                 if (!fid_res_name_eq(ll_inode2fid(inode),
 192                                      &lock->l_resource->lr_name)) {
 193                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 194                                    PFID(ll_inode2fid(inode)), inode);
 195                         LBUG();
 196                 }
 197
 198                 if (bits & MDS_INODELOCK_XATTR) {
 199                         ll_xattr_cache_destroy(inode);
 200                         bits &= ~MDS_INODELOCK_XATTR;
 201                 }
 202
 203                 /* For OPEN locks we differentiate between lock modes
 204                  * LCK_CR, LCK_CW, LCK_PR - bug 22891
 205                  */
 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 "DFID"\n",
 257                                PFID(ll_inode2fid(inode)));
 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 &&
 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  */
 290 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 291 {
 292         LASSERT(i1);
 293
 294         suppgids[0] = ll_i2suppgid(i1);
 295
 296         if (i2)
 297                 suppgids[1] = ll_i2suppgid(i2);
 298                 else
 299                         suppgids[1] = -1;
 300 }
 301
 302 /*
 303  * try to reuse three types of dentry:
 304  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 305  *    by concurrent .revalidate).
 306  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 307  *    be cleared by others calling d_lustre_revalidate).
 308  * 3. DISCONNECTED alias.
 309  */
 310 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 311 {
 312         struct dentry *alias, *discon_alias, *invalid_alias;
 313
 314         if (hlist_empty(&inode->i_dentry))
 315                 return NULL;
 316
 317         discon_alias = NULL;
 318         invalid_alias = NULL;
 319
 320         spin_lock(&inode->i_lock);
 321         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 322                 LASSERT(alias != dentry);
 323
 324                 spin_lock(&alias->d_lock);
 325                 if (alias->d_flags & DCACHE_DISCONNECTED)
 326                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 327                         discon_alias = alias;
 328                 else if (alias->d_parent == dentry->d_parent         &&
 329                          alias->d_name.hash == dentry->d_name.hash       &&
 330                          alias->d_name.len == dentry->d_name.len         &&
 331                          memcmp(alias->d_name.name, dentry->d_name.name,
 332                                 dentry->d_name.len) == 0)
 333                         invalid_alias = alias;
 334                 spin_unlock(&alias->d_lock);
 335
 336                 if (invalid_alias)
 337                         break;
 338         }
 339         alias = invalid_alias ?: discon_alias ?: NULL;
 340         if (alias) {
 341                 spin_lock(&alias->d_lock);
 342                 dget_dlock(alias);
 343                 spin_unlock(&alias->d_lock);
 344         }
 345         spin_unlock(&inode->i_lock);
 346
 347         return alias;
 348 }
 349
 350 /*
 351  * Similar to d_splice_alias(), but lustre treats invalid alias
 352  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 353  */
 354 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 355 {
 356         struct dentry *new;
 357         int rc;
 358
 359         if (inode) {
 360                 new = ll_find_alias(inode, de);
 361                 if (new) {
 362                         rc = ll_d_init(new);
 363                         if (rc < 0) {
 364                                 dput(new);
 365                                 return ERR_PTR(rc);
 366                         }
 367                         d_move(new, de);
 368                         iput(inode);
 369                         CDEBUG(D_DENTRY,
 370                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 371                               new, d_inode(new), d_count(new), new->d_flags);
 372                         return new;
 373                 }
 374         }
 375         rc = ll_d_init(de);
 376         if (rc < 0)
 377                 return ERR_PTR(rc);
 378         d_add(de, inode);
 379         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 380                de, d_inode(de), d_count(de), de->d_flags);
 381         return de;
 382 }
 383
 384 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 385                                struct lookup_intent *it,
 386                                struct inode *parent, struct dentry **de)
 387 {
 388         struct inode *inode = NULL;
 389         __u64 bits = 0;
 390         int rc = 0;
 391         struct dentry *alias;
 392
 393         /* NB 1 request reference will be taken away by ll_intent_lock()
 394          * when I return
 395          */
 396         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 397                it->it_disposition);
 398         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 399                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 400                 if (rc)
 401                         return rc;
 402
 403                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 404
 405                 /* We used to query real size from OSTs here, but actually
 406                  * this is not needed. For stat() calls size would be updated
 407                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 408                  * 2.4 and
 409                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 410                  * Everybody else who needs correct file size would call
 411                  * ll_glimpse_size or some equivalent themselves anyway.
 412                  * Also see bug 7198.
 413                  */
 414         }
 415
 416         alias = ll_splice_alias(inode, *de);
 417         if (IS_ERR(alias)) {
 418                 rc = PTR_ERR(alias);
 419                 goto out;
 420         }
 421         *de = alias;
 422
 423         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 424                 /* we have lookup look - unhide dentry */
 425                 if (bits & MDS_INODELOCK_LOOKUP)
 426                         d_lustre_revalidate(*de);
 427         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 428                 /* If file created on server, don't depend on parent UPDATE
 429                  * lock to unhide it. It is left hidden and next lookup can
 430                  * find it in ll_splice_alias.
 431                  */
 432                 /* Check that parent has UPDATE lock. */
 433                 struct lookup_intent parent_it = {
 434                                         .it_op = IT_GETATTR,
 435                                         .it_lock_handle = 0 };
 436
 437                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 438                                        &ll_i2info(parent)->lli_fid, NULL)) {
 439                         d_lustre_revalidate(*de);
 440                         ll_intent_release(&parent_it);
 441                 }
 442         }
 443
 444 out:
 445         if (rc != 0 && it->it_op & IT_OPEN)
 446                 ll_open_cleanup((*de)->d_sb, request);
 447
 448         return rc;
 449 }
 450
 451 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 452                                    struct lookup_intent *it, int lookup_flags)
 453 {
 454         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 455         struct dentry *save = dentry, *retval;
 456         struct ptlrpc_request *req = NULL;
 457         struct inode *inode;
 458         struct md_op_data *op_data;
 459         __u32 opc;
 460         int rc;
 461
 462         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 463                 return ERR_PTR(-ENAMETOOLONG);
 464
 465         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),intent=%s\n",
 466                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 467
 468         if (d_mountpoint(dentry))
 469                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 470
 471         if (!it || it->it_op == IT_GETXATTR)
 472                 it = &lookup_it;
 473
 474         if (it->it_op == IT_GETATTR) {
 475                 rc = ll_statahead_enter(parent, &dentry, 0);
 476                 if (rc == 1) {
 477                         if (dentry == save)
 478                                 retval = NULL;
 479                         else
 480                                 retval = dentry;
 481                         goto out;
 482                 }
 483         }
 484
 485         if (it->it_op & IT_CREAT)
 486                 opc = LUSTRE_OPC_CREATE;
 487         else
 488                 opc = LUSTRE_OPC_ANY;
 489
 490         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 491                                      dentry->d_name.len, lookup_flags, opc,
 492                                      NULL);
 493         if (IS_ERR(op_data))
 494                 return (void *)op_data;
 495
 496         /* enforce umask if acl disabled or MDS doesn't support umask */
 497         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 498                 it->it_create_mode &= ~current_umask();
 499
 500         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 501                             lookup_flags, &req, ll_md_blocking_ast, 0);
 502         ll_finish_md_op_data(op_data);
 503         if (rc < 0) {
 504                 retval = ERR_PTR(rc);
 505                 goto out;
 506         }
 507
 508         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 509         if (rc != 0) {
 510                 ll_intent_release(it);
 511                 retval = ERR_PTR(rc);
 512                 goto out;
 513         }
 514
 515         inode = d_inode(dentry);
 516         if ((it->it_op & IT_OPEN) && inode &&
 517             !S_ISREG(inode->i_mode) &&
 518             !S_ISDIR(inode->i_mode)) {
 519                 ll_release_openhandle(inode, it);
 520         }
 521         ll_lookup_finish_locks(it, inode);
 522
 523         if (dentry == save)
 524                 retval = NULL;
 525         else
 526                 retval = dentry;
 527  out:
 528         if (req)
 529                 ptlrpc_req_finished(req);
 530         if (it->it_op == IT_GETATTR && (!retval || retval == dentry))
 531                 ll_statahead_mark(parent, dentry);
 532         return retval;
 533 }
 534
 535 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 536                                    unsigned int flags)
 537 {
 538         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 539         struct dentry *de;
 540
 541         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),flags=%u\n",
 542                dentry, PFID(ll_inode2fid(parent)), parent, flags);
 543
 544         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 545         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 546                 return NULL;
 547
 548         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 549                 itp = NULL;
 550         else
 551                 itp = &it;
 552         de = ll_lookup_it(parent, dentry, itp, 0);
 553
 554         if (itp)
 555                 ll_intent_release(itp);
 556
 557         return de;
 558 }
 559
 560 /*
 561  * For cached negative dentry and new dentry, handle lookup/create/open
 562  * together.
 563  */
 564 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 565                           struct file *file, unsigned open_flags,
 566                           umode_t mode, int *opened)
 567 {
 568         struct lookup_intent *it;
 569         struct dentry *de;
 570         long long lookup_flags = LOOKUP_OPEN;
 571         int rc = 0;
 572
 573         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),file %p,open_flags %x,mode %x opened %d\n",
 574                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 575                *opened);
 576
 577         /* Only negative dentries enter here */
 578         LASSERT(!d_inode(dentry));
 579
 580         if (!d_in_lookup(dentry)) {
 581                 /* A valid negative dentry that just passed revalidation,
 582                  * there's little point to try and open it server-side,
 583                  * even though there's a minuscle chance it might succeed.
 584                  * Either way it's a valid race to just return -ENOENT here.
 585                  */
 586                 if (!(open_flags & O_CREAT))
 587                         return -ENOENT;
 588
 589                 /* Otherwise we just unhash it to be rehashed afresh via
 590                  * lookup if necessary
 591                  */
 592                 d_drop(dentry);
 593         }
 594
 595         it = kzalloc(sizeof(*it), GFP_NOFS);
 596         if (!it)
 597                 return -ENOMEM;
 598
 599         it->it_op = IT_OPEN;
 600         if (open_flags & O_CREAT) {
 601                 it->it_op |= IT_CREAT;
 602                 lookup_flags |= LOOKUP_CREATE;
 603         }
 604         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 605         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 606
 607         /* Dentry added to dcache tree in ll_lookup_it */
 608         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 609         if (IS_ERR(de))
 610                 rc = PTR_ERR(de);
 611         else if (de)
 612                 dentry = de;
 613
 614         if (!rc) {
 615                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 616                         /* Dentry instantiated in ll_create_it. */
 617                         rc = ll_create_it(dir, dentry, mode, it);
 618                         if (rc) {
 619                                 /* We dget in ll_splice_alias. */
 620                                 if (de)
 621                                         dput(de);
 622                                 goto out_release;
 623                         }
 624
 625                         *opened |= FILE_CREATED;
 626                 }
 627                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 628                         /* Open dentry. */
 629                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 630                                 /* We cannot call open here as it might
 631                                  * deadlock. This case is unreachable in
 632                                  * practice because of OBD_CONNECT_NODEVOH.
 633                                  */
 634                                 rc = finish_no_open(file, de);
 635                         } else {
 636                                 file->private_data = it;
 637                                 rc = finish_open(file, dentry, NULL, opened);
 638                                 /* We dget in ll_splice_alias. finish_open takes
 639                                  * care of dget for fd open.
 640                                  */
 641                                 if (de)
 642                                         dput(de);
 643                         }
 644                 } else {
 645                         rc = finish_no_open(file, de);
 646                 }
 647         }
 648
 649 out_release:
 650         ll_intent_release(it);
 651         kfree(it);
 652
 653         return rc;
 654 }
 655
 656 /* We depend on "mode" being set with the proper file type/umask by now */
 657 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 658 {
 659         struct inode *inode = NULL;
 660         struct ptlrpc_request *request = NULL;
 661         struct ll_sb_info *sbi = ll_i2sbi(dir);
 662         int rc;
 663
 664         LASSERT(it && it->it_disposition);
 665
 666         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 667         request = it->it_request;
 668         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 669         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 670         if (rc) {
 671                 inode = ERR_PTR(rc);
 672                 goto out;
 673         }
 674
 675         LASSERT(hlist_empty(&inode->i_dentry));
 676
 677         /* We asked for a lock on the directory, but were granted a
 678          * lock on the inode.  Since we finally have an inode pointer,
 679          * stuff it in the lock.
 680          */
 681         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 682                PFID(ll_inode2fid(dir)), inode);
 683         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 684  out:
 685         ptlrpc_req_finished(request);
 686         return inode;
 687 }
 688
 689 /*
 690  * By the time this is called, we already have created the directory cache
 691  * entry for the new file, but it is so far negative - it has no inode.
 692  *
 693  * We defer creating the OBD object(s) until open, to keep the intent and
 694  * non-intent code paths similar, and also because we do not have the MDS
 695  * inode number before calling ll_create_node() (which is needed for LOV),
 696  * so we would need to do yet another RPC to the MDS to store the LOV EA
 697  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 698  * lmm_size in datalen (the MDS still has code which will handle that).
 699  *
 700  * If the create succeeds, we fill in the inode information
 701  * with d_instantiate().
 702  */
 703 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 704                         struct lookup_intent *it)
 705 {
 706         struct inode *inode;
 707         int rc = 0;
 708
 709         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 710                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 711
 712         rc = it_open_error(DISP_OPEN_CREATE, it);
 713         if (rc)
 714                 return rc;
 715
 716         inode = ll_create_node(dir, it);
 717         if (IS_ERR(inode))
 718                 return PTR_ERR(inode);
 719
 720         d_instantiate(dentry, inode);
 721         return 0;
 722 }
 723
 724 static void ll_update_times(struct ptlrpc_request *request,
 725                             struct inode *inode)
 726 {
 727         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 728                                                        &RMF_MDT_BODY);
 729
 730         LASSERT(body);
 731         if (body->valid & OBD_MD_FLMTIME &&
 732             body->mtime > LTIME_S(inode->i_mtime)) {
 733                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu\n",
 734                        PFID(ll_inode2fid(inode)), LTIME_S(inode->i_mtime),
 735                        body->mtime);
 736                 LTIME_S(inode->i_mtime) = body->mtime;
 737         }
 738         if (body->valid & OBD_MD_FLCTIME &&
 739             body->ctime > LTIME_S(inode->i_ctime))
 740                 LTIME_S(inode->i_ctime) = body->ctime;
 741 }
 742
 743 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 744                        const char *tgt, int mode, int rdev,
 745                        __u32 opc)
 746 {
 747         struct ptlrpc_request *request = NULL;
 748         struct md_op_data *op_data;
 749         struct inode *inode = NULL;
 750         struct ll_sb_info *sbi = ll_i2sbi(dir);
 751         int tgt_len = 0;
 752         int err;
 753
 754         if (unlikely(tgt))
 755                 tgt_len = strlen(tgt) + 1;
 756
 757         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 758                                      dentry->d_name.name,
 759                                      dentry->d_name.len,
 760                                      0, opc, NULL);
 761         if (IS_ERR(op_data)) {
 762                 err = PTR_ERR(op_data);
 763                 goto err_exit;
 764         }
 765
 766         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 767                         from_kuid(&init_user_ns, current_fsuid()),
 768                         from_kgid(&init_user_ns, current_fsgid()),
 769                         cfs_curproc_cap_pack(), rdev, &request);
 770         ll_finish_md_op_data(op_data);
 771         if (err)
 772                 goto err_exit;
 773
 774         ll_update_times(request, dir);
 775
 776         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 777         if (err)
 778                 goto err_exit;
 779
 780         d_instantiate(dentry, inode);
 781 err_exit:
 782         ptlrpc_req_finished(request);
 783
 784         return err;
 785 }
 786
 787 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 788                     umode_t mode, dev_t rdev)
 789 {
 790         int err;
 791
 792         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
 793                dchild, PFID(ll_inode2fid(dir)), dir, mode,
 794                old_encode_dev(rdev));
 795
 796         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 797                 mode &= ~current_umask();
 798
 799         switch (mode & S_IFMT) {
 800         case 0:
 801                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 802         case S_IFREG:
 803         case S_IFCHR:
 804         case S_IFBLK:
 805         case S_IFIFO:
 806         case S_IFSOCK:
 807                 err = ll_new_node(dir, dchild, NULL, mode,
 808                                   old_encode_dev(rdev),
 809                                   LUSTRE_OPC_MKNOD);
 810                 break;
 811         case S_IFDIR:
 812                 err = -EPERM;
 813                 break;
 814         default:
 815                 err = -EINVAL;
 816         }
 817
 818         if (!err)
 819                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 820
 821         return err;
 822 }
 823
 824 /*
 825  * Plain create. Intent create is handled in atomic_open.
 826  */
 827 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 828                         umode_t mode, bool want_excl)
 829 {
 830         int rc;
 831
 832         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
 833                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
 834
 835         rc = ll_mknod(dir, dentry, mode, 0);
 836
 837         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 838
 839         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 840                dentry, d_unhashed(dentry));
 841
 842         return rc;
 843 }
 844
 845 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 846 {
 847         struct mdt_body *body;
 848         struct lov_mds_md *eadata;
 849         struct lov_stripe_md *lsm = NULL;
 850         struct obd_trans_info oti = { 0 };
 851         struct obdo *oa;
 852         int rc;
 853
 854         /* req is swabbed so this is safe */
 855         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 856         if (!(body->valid & OBD_MD_FLEASIZE))
 857                 return 0;
 858
 859         if (body->eadatasize == 0) {
 860                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 861                 rc = -EPROTO;
 862                 goto out;
 863         }
 864
 865         /* The MDS sent back the EA because we unlinked the last reference
 866          * to this file. Use this EA to unlink the objects on the OST.
 867          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 868          * check it is complete and sensible.
 869          */
 870         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 871                                               body->eadatasize);
 872         LASSERT(eadata);
 873
 874         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 875         if (rc < 0) {
 876                 CERROR("obd_unpackmd: %d\n", rc);
 877                 goto out;
 878         }
 879         LASSERT(rc >= sizeof(*lsm));
 880
 881         oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
 882         if (!oa) {
 883                 rc = -ENOMEM;
 884                 goto out_free_memmd;
 885         }
 886
 887         oa->o_oi = lsm->lsm_oi;
 888         oa->o_mode = body->mode & S_IFMT;
 889         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 890
 891         if (body->valid & OBD_MD_FLCOOKIE) {
 892                 oa->o_valid |= OBD_MD_FLCOOKIE;
 893                 oti.oti_logcookies =
 894                         req_capsule_server_sized_get(&request->rq_pill,
 895                                                      &RMF_LOGCOOKIES,
 896                                                    sizeof(struct llog_cookie) *
 897                                                      lsm->lsm_stripe_count);
 898                 if (!oti.oti_logcookies) {
 899                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 900                         body->valid &= ~OBD_MD_FLCOOKIE;
 901                 }
 902         }
 903
 904         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 905                          ll_i2mdexp(dir));
 906         if (rc)
 907                 CERROR("obd destroy objid "DOSTID" error %d\n",
 908                        POSTID(&lsm->lsm_oi), rc);
 909 out_free_memmd:
 910         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 911         kmem_cache_free(obdo_cachep, oa);
 912 out:
 913         return rc;
 914 }
 915
 916 /* ll_unlink() doesn't update the inode with the new link count.
 917  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 918  * is any lock existing. They will recycle dentries and inodes based upon locks
 919  * too. b=20433
 920  */
 921 static int ll_unlink(struct inode *dir, struct dentry *dchild)
 922 {
 923         struct ptlrpc_request *request = NULL;
 924         struct md_op_data *op_data;
 925         int rc;
 926
 927         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 928                dchild, dir->i_ino, dir->i_generation, dir);
 929
 930         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 931                                      dchild->d_name.name,
 932                                      dchild->d_name.len,
 933                                      0, LUSTRE_OPC_ANY, NULL);
 934         if (IS_ERR(op_data))
 935                 return PTR_ERR(op_data);
 936
 937         if (dchild && dchild->d_inode)
 938                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
 939
 940         op_data->op_fid2 = op_data->op_fid3;
 941         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 942         ll_finish_md_op_data(op_data);
 943         if (rc)
 944                 goto out;
 945
 946         ll_update_times(request, dir);
 947         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 948
 949         rc = ll_objects_destroy(request, dir);
 950  out:
 951         ptlrpc_req_finished(request);
 952         return rc;
 953 }
 954
 955 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 956 {
 957         int err;
 958
 959         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir"DFID"(%p)\n",
 960                dentry, PFID(ll_inode2fid(dir)), dir);
 961
 962         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 963                 mode &= ~current_umask();
 964         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
 965         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
 966
 967         if (!err)
 968                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 969
 970         return err;
 971 }
 972
 973 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
 974 {
 975         struct ptlrpc_request *request = NULL;
 976         struct md_op_data *op_data;
 977         int rc;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
 980                dchild, PFID(ll_inode2fid(dir)), dir);
 981
 982         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 983                                      dchild->d_name.name,
 984                                      dchild->d_name.len,
 985                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
 986         if (IS_ERR(op_data))
 987                 return PTR_ERR(op_data);
 988
 989         if (dchild && dchild->d_inode)
 990                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
 991
 992         op_data->op_fid2 = op_data->op_fid3;
 993         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 994         ll_finish_md_op_data(op_data);
 995         if (rc == 0) {
 996                 ll_update_times(request, dir);
 997                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
 998         }
 999
1000         ptlrpc_req_finished(request);
1001         return rc;
1002 }
1003
1004 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1005                       const char *oldname)
1006 {
1007         int err;
1008
1009         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p),target=%.*s\n",
1010                dentry, PFID(ll_inode2fid(dir)), dir, 3000, oldname);
1011
1012         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1013                           0, LUSTRE_OPC_SYMLINK);
1014
1015         if (!err)
1016                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1017
1018         return err;
1019 }
1020
1021 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1022                    struct dentry *new_dentry)
1023 {
1024         struct inode *src = d_inode(old_dentry);
1025         struct ll_sb_info *sbi = ll_i2sbi(dir);
1026         struct ptlrpc_request *request = NULL;
1027         struct md_op_data *op_data;
1028         int err;
1029
1030         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1031                PFID(ll_inode2fid(src)), src, PFID(ll_inode2fid(dir)), dir,
1032                new_dentry);
1033
1034         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1035                                      new_dentry->d_name.len,
1036                                      0, LUSTRE_OPC_ANY, NULL);
1037         if (IS_ERR(op_data))
1038                 return PTR_ERR(op_data);
1039
1040         err = md_link(sbi->ll_md_exp, op_data, &request);
1041         ll_finish_md_op_data(op_data);
1042         if (err)
1043                 goto out;
1044
1045         ll_update_times(request, dir);
1046         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1047 out:
1048         ptlrpc_req_finished(request);
1049         return err;
1050 }
1051
1052 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1053                      struct inode *tgt, struct dentry *tgt_dchild,
1054                      unsigned int flags)
1055 {
1056         struct ptlrpc_request *request = NULL;
1057         struct ll_sb_info *sbi = ll_i2sbi(src);
1058         struct md_op_data *op_data;
1059         int err;
1060
1061         if (flags)
1062                 return -EINVAL;
1063
1064         CDEBUG(D_VFSTRACE,
1065                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1066                src_dchild, PFID(ll_inode2fid(src)), src,
1067                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1068
1069         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1070                                      LUSTRE_OPC_ANY, NULL);
1071         if (IS_ERR(op_data))
1072                 return PTR_ERR(op_data);
1073
1074         if (src_dchild && src_dchild->d_inode)
1075                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1076         if (tgt_dchild && tgt_dchild->d_inode)
1077                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1078
1079         err = md_rename(sbi->ll_md_exp, op_data,
1080                         src_dchild->d_name.name,
1081                         src_dchild->d_name.len,
1082                         tgt_dchild->d_name.name,
1083                         tgt_dchild->d_name.len, &request);
1084         ll_finish_md_op_data(op_data);
1085         if (!err) {
1086                 ll_update_times(request, src);
1087                 ll_update_times(request, tgt);
1088                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1089                 err = ll_objects_destroy(request, src);
1090         }
1091
1092         ptlrpc_req_finished(request);
1093         if (!err)
1094                 d_move(src_dchild, tgt_dchild);
1095         return err;
1096 }
1097
1098 const struct inode_operations ll_dir_inode_operations = {
1099         .mknod        = ll_mknod,
1100         .atomic_open        = ll_atomic_open,
1101         .lookup      = ll_lookup_nd,
1102         .create      = ll_create_nd,
1103         /* We need all these non-raw things for NFSD, to not patch it. */
1104         .unlink      = ll_unlink,
1105         .mkdir        = ll_mkdir,
1106         .rmdir        = ll_rmdir,
1107         .symlink            = ll_symlink,
1108         .link          = ll_link,
1109         .rename         = ll_rename,
1110         .setattr            = ll_setattr,
1111         .getattr            = ll_getattr,
1112         .permission      = ll_inode_permission,
1113         .setxattr          = ll_setxattr,
1114         .getxattr          = ll_getxattr,
1115         .listxattr        = ll_listxattr,
1116         .removexattr    = ll_removexattr,
1117         .get_acl            = ll_get_acl,
1118 };
1119
1120 const struct inode_operations ll_special_inode_operations = {
1121         .setattr        = ll_setattr,
1122         .getattr        = ll_getattr,
1123         .permission     = ll_inode_permission,
1124         .setxattr       = ll_setxattr,
1125         .getxattr       = ll_getxattr,
1126         .listxattr      = ll_listxattr,
1127         .removexattr    = ll_removexattr,
1128         .get_acl            = ll_get_acl,
1129 };