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) {
 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                  */
 185                 LASSERT(!lock->l_ast_data);
 186
 187                 if (!inode)
 188                         break;
 189
 190                 /* Invalidate all dentries associated with this inode */
 191                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 192
 193                 if (!fid_res_name_eq(ll_inode2fid(inode),
 194                                      &lock->l_resource->lr_name)) {
 195                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 196                                    PFID(ll_inode2fid(inode)), inode);
 197                         LBUG();
 198                 }
 199
 200                 if (bits & MDS_INODELOCK_XATTR) {
 201                         ll_xattr_cache_destroy(inode);
 202                         bits &= ~MDS_INODELOCK_XATTR;
 203                 }
 204
 205                 /* For OPEN locks we differentiate between lock modes
 206                  * LCK_CR, LCK_CW, LCK_PR - bug 22891
 207                  */
 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 &&
 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  */
 292 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 293 {
 294         LASSERT(i1);
 295
 296         suppgids[0] = ll_i2suppgid(i1);
 297
 298         if (i2)
 299                 suppgids[1] = ll_i2suppgid(i2);
 300                 else
 301                         suppgids[1] = -1;
 302 }
 303
 304 /*
 305  * try to reuse three types of dentry:
 306  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 307  *    by concurrent .revalidate).
 308  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 309  *    be cleared by others calling d_lustre_revalidate).
 310  * 3. DISCONNECTED alias.
 311  */
 312 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 313 {
 314         struct dentry *alias, *discon_alias, *invalid_alias;
 315
 316         if (hlist_empty(&inode->i_dentry))
 317                 return NULL;
 318
 319         discon_alias = invalid_alias = NULL;
 320
 321         ll_lock_dcache(inode);
 322         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 323                 LASSERT(alias != dentry);
 324
 325                 spin_lock(&alias->d_lock);
 326                 if (alias->d_flags & DCACHE_DISCONNECTED)
 327                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 328                         discon_alias = alias;
 329                 else if (alias->d_parent == dentry->d_parent         &&
 330                          alias->d_name.hash == dentry->d_name.hash       &&
 331                          alias->d_name.len == dentry->d_name.len         &&
 332                          memcmp(alias->d_name.name, dentry->d_name.name,
 333                                 dentry->d_name.len) == 0)
 334                         invalid_alias = alias;
 335                 spin_unlock(&alias->d_lock);
 336
 337                 if (invalid_alias)
 338                         break;
 339         }
 340         alias = invalid_alias ?: discon_alias ?: NULL;
 341         if (alias) {
 342                 spin_lock(&alias->d_lock);
 343                 dget_dlock(alias);
 344                 spin_unlock(&alias->d_lock);
 345         }
 346         ll_unlock_dcache(inode);
 347
 348         return alias;
 349 }
 350
 351 /*
 352  * Similar to d_splice_alias(), but lustre treats invalid alias
 353  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 354  */
 355 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 356 {
 357         struct dentry *new;
 358         int rc;
 359
 360         if (inode) {
 361                 new = ll_find_alias(inode, de);
 362                 if (new) {
 363                         rc = ll_d_init(new);
 364                         if (rc < 0) {
 365                                 dput(new);
 366                                 return ERR_PTR(rc);
 367                         }
 368                         d_move(new, de);
 369                         iput(inode);
 370                         CDEBUG(D_DENTRY,
 371                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 372                               new, d_inode(new), d_count(new), new->d_flags);
 373                         return new;
 374                 }
 375         }
 376         rc = ll_d_init(de);
 377         if (rc < 0)
 378                 return ERR_PTR(rc);
 379         d_add(de, inode);
 380         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 381                de, d_inode(de), d_count(de), de->d_flags);
 382         return de;
 383 }
 384
 385 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 386                                struct lookup_intent *it,
 387                                struct inode *parent, struct dentry **de)
 388 {
 389         struct inode *inode = NULL;
 390         __u64 bits = 0;
 391         int rc = 0;
 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->d.lustre.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         /* Only hash *de if it is unhashed (new dentry).
 417          * Atoimc_open may passing hashed dentries for open.
 418          */
 419         if (d_unhashed(*de)) {
 420                 struct dentry *alias;
 421
 422                 alias = ll_splice_alias(inode, *de);
 423                 if (IS_ERR(alias)) {
 424                         rc = PTR_ERR(alias);
 425                         goto out;
 426                 }
 427                 *de = alias;
 428         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 429                    !it_disposition(it, DISP_OPEN_CREATE)) {
 430                 /* With DISP_OPEN_CREATE dentry will be
 431                  * instantiated in ll_create_it.
 432                  */
 433                 LASSERT(!d_inode(*de));
 434                 d_instantiate(*de, inode);
 435         }
 436
 437         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 438                 /* we have lookup look - unhide dentry */
 439                 if (bits & MDS_INODELOCK_LOOKUP)
 440                         d_lustre_revalidate(*de);
 441         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 442                 /* If file created on server, don't depend on parent UPDATE
 443                  * lock to unhide it. It is left hidden and next lookup can
 444                  * find it in ll_splice_alias.
 445                  */
 446                 /* Check that parent has UPDATE lock. */
 447                 struct lookup_intent parent_it = {
 448                                         .it_op = IT_GETATTR,
 449                                         .d.lustre.it_lock_handle = 0 };
 450
 451                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 452                                        &ll_i2info(parent)->lli_fid, NULL)) {
 453                         d_lustre_revalidate(*de);
 454                         ll_intent_release(&parent_it);
 455                 }
 456         }
 457
 458 out:
 459         if (rc != 0 && it->it_op & IT_OPEN)
 460                 ll_open_cleanup((*de)->d_sb, request);
 461
 462         return rc;
 463 }
 464
 465 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 466                                    struct lookup_intent *it, int lookup_flags)
 467 {
 468         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 469         struct dentry *save = dentry, *retval;
 470         struct ptlrpc_request *req = NULL;
 471         struct inode *inode;
 472         struct md_op_data *op_data;
 473         __u32 opc;
 474         int rc;
 475
 476         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 477                 return ERR_PTR(-ENAMETOOLONG);
 478
 479         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 480                dentry, parent->i_ino,
 481                parent->i_generation, parent, LL_IT2STR(it));
 482
 483         if (d_mountpoint(dentry))
 484                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 485
 486         if (!it || it->it_op == IT_GETXATTR)
 487                 it = &lookup_it;
 488
 489         if (it->it_op == IT_GETATTR) {
 490                 rc = ll_statahead_enter(parent, &dentry, 0);
 491                 if (rc == 1) {
 492                         if (dentry == save)
 493                                 retval = NULL;
 494                         else
 495                                 retval = dentry;
 496                         goto out;
 497                 }
 498         }
 499
 500         if (it->it_op & IT_CREAT)
 501                 opc = LUSTRE_OPC_CREATE;
 502         else
 503                 opc = LUSTRE_OPC_ANY;
 504
 505         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 506                                      dentry->d_name.len, lookup_flags, opc,
 507                                      NULL);
 508         if (IS_ERR(op_data))
 509                 return (void *)op_data;
 510
 511         /* enforce umask if acl disabled or MDS doesn't support umask */
 512         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 513                 it->it_create_mode &= ~current_umask();
 514
 515         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 516                             lookup_flags, &req, ll_md_blocking_ast, 0);
 517         ll_finish_md_op_data(op_data);
 518         if (rc < 0) {
 519                 retval = ERR_PTR(rc);
 520                 goto out;
 521         }
 522
 523         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 524         if (rc != 0) {
 525                 ll_intent_release(it);
 526                 retval = ERR_PTR(rc);
 527                 goto out;
 528         }
 529
 530         inode = d_inode(dentry);
 531         if ((it->it_op & IT_OPEN) && inode &&
 532             !S_ISREG(inode->i_mode) &&
 533             !S_ISDIR(inode->i_mode)) {
 534                 ll_release_openhandle(inode, it);
 535         }
 536         ll_lookup_finish_locks(it, inode);
 537
 538         if (dentry == save)
 539                 retval = NULL;
 540         else
 541                 retval = dentry;
 542  out:
 543         if (req)
 544                 ptlrpc_req_finished(req);
 545         if (it->it_op == IT_GETATTR && (!retval || retval == dentry))
 546                 ll_statahead_mark(parent, dentry);
 547         return retval;
 548 }
 549
 550 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 551                                    unsigned int flags)
 552 {
 553         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 554         struct dentry *de;
 555
 556         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u\n",
 557                dentry, parent->i_ino,
 558                parent->i_generation, parent, flags);
 559
 560         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 561         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 562                 return NULL;
 563
 564         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 565                 itp = NULL;
 566         else
 567                 itp = &it;
 568         de = ll_lookup_it(parent, dentry, itp, 0);
 569
 570         if (itp)
 571                 ll_intent_release(itp);
 572
 573         return de;
 574 }
 575
 576 /*
 577  * For cached negative dentry and new dentry, handle lookup/create/open
 578  * together.
 579  */
 580 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 581                           struct file *file, unsigned open_flags,
 582                           umode_t mode, int *opened)
 583 {
 584         struct lookup_intent *it;
 585         struct dentry *de;
 586         long long lookup_flags = LOOKUP_OPEN;
 587         int rc = 0;
 588
 589         CDEBUG(D_VFSTRACE,
 590                "VFS Op:name=%pd,dir=%lu/%u(%p),file %p,open_flags %x,mode %x opened %d\n",
 591                dentry, dir->i_ino,
 592                dir->i_generation, dir, file, open_flags, mode, *opened);
 593
 594         it = kzalloc(sizeof(*it), GFP_NOFS);
 595         if (!it)
 596                 return -ENOMEM;
 597
 598         it->it_op = IT_OPEN;
 599         if (open_flags & O_CREAT) {
 600                 it->it_op |= IT_CREAT;
 601                 lookup_flags |= LOOKUP_CREATE;
 602         }
 603         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 604         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 605
 606         /* Dentry added to dcache tree in ll_lookup_it */
 607         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 608         if (IS_ERR(de))
 609                 rc = PTR_ERR(de);
 610         else if (de)
 611                 dentry = de;
 612
 613         if (!rc) {
 614                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 615                         /* Dentry instantiated in ll_create_it. */
 616                         rc = ll_create_it(dir, dentry, mode, it);
 617                         if (rc) {
 618                                 /* We dget in ll_splice_alias. */
 619                                 if (de)
 620                                         dput(de);
 621                                 goto out_release;
 622                         }
 623
 624                         *opened |= FILE_CREATED;
 625                 }
 626                 if (d_really_is_positive(dentry) && it_disposition(it, DISP_OPEN_OPEN)) {
 627                         /* Open dentry. */
 628                         if (S_ISFIFO(d_inode(dentry)->i_mode)) {
 629                                 /* We cannot call open here as it would
 630                                  * deadlock.
 631                                  */
 632                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 633                                         ptlrpc_req_finished(
 634                                                        (struct ptlrpc_request *)
 635                                                           it->d.lustre.it_data);
 636                                 rc = finish_no_open(file, de);
 637                         } else {
 638                                 file->private_data = it;
 639                                 rc = finish_open(file, dentry, NULL, opened);
 640                                 /* We dget in ll_splice_alias. finish_open takes
 641                                  * care of dget for fd open.
 642                                  */
 643                                 if (de)
 644                                         dput(de);
 645                         }
 646                 } else {
 647                         rc = finish_no_open(file, de);
 648                 }
 649         }
 650
 651 out_release:
 652         ll_intent_release(it);
 653         kfree(it);
 654
 655         return rc;
 656 }
 657
 658 /* We depend on "mode" being set with the proper file type/umask by now */
 659 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 660 {
 661         struct inode *inode = NULL;
 662         struct ptlrpc_request *request = NULL;
 663         struct ll_sb_info *sbi = ll_i2sbi(dir);
 664         int rc;
 665
 666         LASSERT(it && it->d.lustre.it_disposition);
 667
 668         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 669         request = it->d.lustre.it_data;
 670         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 671         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 672         if (rc) {
 673                 inode = ERR_PTR(rc);
 674                 goto out;
 675         }
 676
 677         LASSERT(hlist_empty(&inode->i_dentry));
 678
 679         /* We asked for a lock on the directory, but were granted a
 680          * lock on the inode.  Since we finally have an inode pointer,
 681          * stuff it in the lock.
 682          */
 683         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 684                inode, inode->i_ino, inode->i_generation);
 685         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 686  out:
 687         ptlrpc_req_finished(request);
 688         return inode;
 689 }
 690
 691 /*
 692  * By the time this is called, we already have created the directory cache
 693  * entry for the new file, but it is so far negative - it has no inode.
 694  *
 695  * We defer creating the OBD object(s) until open, to keep the intent and
 696  * non-intent code paths similar, and also because we do not have the MDS
 697  * inode number before calling ll_create_node() (which is needed for LOV),
 698  * so we would need to do yet another RPC to the MDS to store the LOV EA
 699  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 700  * lmm_size in datalen (the MDS still has code which will handle that).
 701  *
 702  * If the create succeeds, we fill in the inode information
 703  * with d_instantiate().
 704  */
 705 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 706                         struct lookup_intent *it)
 707 {
 708         struct inode *inode;
 709         int rc = 0;
 710
 711         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),intent=%s\n",
 712                dentry, dir->i_ino,
 713                dir->i_generation, dir, LL_IT2STR(it));
 714
 715         rc = it_open_error(DISP_OPEN_CREATE, it);
 716         if (rc)
 717                 return rc;
 718
 719         inode = ll_create_node(dir, it);
 720         if (IS_ERR(inode))
 721                 return PTR_ERR(inode);
 722
 723         d_instantiate(dentry, inode);
 724         return 0;
 725 }
 726
 727 static void ll_update_times(struct ptlrpc_request *request,
 728                             struct inode *inode)
 729 {
 730         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 731                                                        &RMF_MDT_BODY);
 732
 733         LASSERT(body);
 734         if (body->valid & OBD_MD_FLMTIME &&
 735             body->mtime > LTIME_S(inode->i_mtime)) {
 736                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
 737                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 738                 LTIME_S(inode->i_mtime) = body->mtime;
 739         }
 740         if (body->valid & OBD_MD_FLCTIME &&
 741             body->ctime > LTIME_S(inode->i_ctime))
 742                 LTIME_S(inode->i_ctime) = body->ctime;
 743 }
 744
 745 static int ll_new_node(struct inode *dir, struct dentry *dentry,
 746                        const char *tgt, int mode, int rdev,
 747                        __u32 opc)
 748 {
 749         struct ptlrpc_request *request = NULL;
 750         struct md_op_data *op_data;
 751         struct inode *inode = NULL;
 752         struct ll_sb_info *sbi = ll_i2sbi(dir);
 753         int tgt_len = 0;
 754         int err;
 755
 756         if (unlikely(tgt))
 757                 tgt_len = strlen(tgt) + 1;
 758
 759         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 760                                      dentry->d_name.name,
 761                                      dentry->d_name.len,
 762                                      0, opc, NULL);
 763         if (IS_ERR(op_data)) {
 764                 err = PTR_ERR(op_data);
 765                 goto err_exit;
 766         }
 767
 768         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 769                         from_kuid(&init_user_ns, current_fsuid()),
 770                         from_kgid(&init_user_ns, current_fsgid()),
 771                         cfs_curproc_cap_pack(), rdev, &request);
 772         ll_finish_md_op_data(op_data);
 773         if (err)
 774                 goto err_exit;
 775
 776         ll_update_times(request, dir);
 777
 778         err = ll_prep_inode(&inode, request, dir->i_sb, NULL);
 779         if (err)
 780                 goto err_exit;
 781
 782         d_instantiate(dentry, inode);
 783 err_exit:
 784         ptlrpc_req_finished(request);
 785
 786         return err;
 787 }
 788
 789 static int ll_mknod(struct inode *dir, struct dentry *dchild,
 790                     umode_t mode, dev_t rdev)
 791 {
 792         int err;
 793
 794         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p) mode %o dev %x\n",
 795                dchild, dir->i_ino, dir->i_generation, dir,
 796                mode, old_encode_dev(rdev));
 797
 798         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 799                 mode &= ~current_umask();
 800
 801         switch (mode & S_IFMT) {
 802         case 0:
 803                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 804         case S_IFREG:
 805         case S_IFCHR:
 806         case S_IFBLK:
 807         case S_IFIFO:
 808         case S_IFSOCK:
 809                 err = ll_new_node(dir, dchild, NULL, mode,
 810                                   old_encode_dev(rdev),
 811                                   LUSTRE_OPC_MKNOD);
 812                 break;
 813         case S_IFDIR:
 814                 err = -EPERM;
 815                 break;
 816         default:
 817                 err = -EINVAL;
 818         }
 819
 820         if (!err)
 821                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 822
 823         return err;
 824 }
 825
 826 /*
 827  * Plain create. Intent create is handled in atomic_open.
 828  */
 829 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 830                         umode_t mode, bool want_excl)
 831 {
 832         int rc;
 833
 834         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),flags=%u, excl=%d\n",
 835                dentry, dir->i_ino,
 836                dir->i_generation, dir, mode, want_excl);
 837
 838         rc = ll_mknod(dir, dentry, mode, 0);
 839
 840         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 841
 842         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
 843                dentry, d_unhashed(dentry));
 844
 845         return rc;
 846 }
 847
 848 static inline void ll_get_child_fid(struct dentry *child, struct lu_fid *fid)
 849 {
 850         if (d_really_is_positive(child))
 851                 *fid = *ll_inode2fid(d_inode(child));
 852 }
 853
 854 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 855 {
 856         struct mdt_body *body;
 857         struct lov_mds_md *eadata;
 858         struct lov_stripe_md *lsm = NULL;
 859         struct obd_trans_info oti = { 0 };
 860         struct obdo *oa;
 861         int rc;
 862
 863         /* req is swabbed so this is safe */
 864         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 865         if (!(body->valid & OBD_MD_FLEASIZE))
 866                 return 0;
 867
 868         if (body->eadatasize == 0) {
 869                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 870                 rc = -EPROTO;
 871                 goto out;
 872         }
 873
 874         /* The MDS sent back the EA because we unlinked the last reference
 875          * to this file. Use this EA to unlink the objects on the OST.
 876          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
 877          * check it is complete and sensible.
 878          */
 879         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
 880                                               body->eadatasize);
 881         LASSERT(eadata);
 882
 883         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
 884         if (rc < 0) {
 885                 CERROR("obd_unpackmd: %d\n", rc);
 886                 goto out;
 887         }
 888         LASSERT(rc >= sizeof(*lsm));
 889
 890         oa = kmem_cache_alloc(obdo_cachep, GFP_NOFS | __GFP_ZERO);
 891         if (!oa) {
 892                 rc = -ENOMEM;
 893                 goto out_free_memmd;
 894         }
 895
 896         oa->o_oi = lsm->lsm_oi;
 897         oa->o_mode = body->mode & S_IFMT;
 898         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 899
 900         if (body->valid & OBD_MD_FLCOOKIE) {
 901                 oa->o_valid |= OBD_MD_FLCOOKIE;
 902                 oti.oti_logcookies =
 903                         req_capsule_server_sized_get(&request->rq_pill,
 904                                                      &RMF_LOGCOOKIES,
 905                                                    sizeof(struct llog_cookie) *
 906                                                      lsm->lsm_stripe_count);
 907                 if (!oti.oti_logcookies) {
 908                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 909                         body->valid &= ~OBD_MD_FLCOOKIE;
 910                 }
 911         }
 912
 913         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
 914                          ll_i2mdexp(dir));
 915         if (rc)
 916                 CERROR("obd destroy objid "DOSTID" error %d\n",
 917                        POSTID(&lsm->lsm_oi), rc);
 918 out_free_memmd:
 919         obd_free_memmd(ll_i2dtexp(dir), &lsm);
 920         kmem_cache_free(obdo_cachep, oa);
 921 out:
 922         return rc;
 923 }
 924
 925 /* ll_unlink() doesn't update the inode with the new link count.
 926  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
 927  * is any lock existing. They will recycle dentries and inodes based upon locks
 928  * too. b=20433
 929  */
 930 static int ll_unlink(struct inode *dir, struct dentry *dentry)
 931 {
 932         struct ptlrpc_request *request = NULL;
 933         struct md_op_data *op_data;
 934         int rc;
 935
 936         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 937                dentry, dir->i_ino, dir->i_generation, dir);
 938
 939         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 940                                      dentry->d_name.name,
 941                                      dentry->d_name.len,
 942                                      0, LUSTRE_OPC_ANY, NULL);
 943         if (IS_ERR(op_data))
 944                 return PTR_ERR(op_data);
 945
 946         ll_get_child_fid(dentry, &op_data->op_fid3);
 947         op_data->op_fid2 = op_data->op_fid3;
 948         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 949         ll_finish_md_op_data(op_data);
 950         if (rc)
 951                 goto out;
 952
 953         ll_update_times(request, dir);
 954         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 955
 956         rc = ll_objects_destroy(request, dir);
 957  out:
 958         ptlrpc_req_finished(request);
 959         return rc;
 960 }
 961
 962 static int ll_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 963 {
 964         int err;
 965
 966         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 967                dentry, dir->i_ino, dir->i_generation, dir);
 968
 969         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 970                 mode &= ~current_umask();
 971         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
 972         err = ll_new_node(dir, dentry, NULL, mode, 0, LUSTRE_OPC_MKDIR);
 973
 974         if (!err)
 975                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 976
 977         return err;
 978 }
 979
 980 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
 981 {
 982         struct ptlrpc_request *request = NULL;
 983         struct md_op_data *op_data;
 984         int rc;
 985
 986         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
 987                dentry, dir->i_ino, dir->i_generation, dir);
 988
 989         op_data = ll_prep_md_op_data(NULL, dir, NULL,
 990                                      dentry->d_name.name,
 991                                      dentry->d_name.len,
 992                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
 993         if (IS_ERR(op_data))
 994                 return PTR_ERR(op_data);
 995
 996         ll_get_child_fid(dentry, &op_data->op_fid3);
 997         op_data->op_fid2 = op_data->op_fid3;
 998         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
 999         ll_finish_md_op_data(op_data);
1000         if (rc == 0) {
1001                 ll_update_times(request, dir);
1002                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1003         }
1004
1005         ptlrpc_req_finished(request);
1006         return rc;
1007 }
1008
1009 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1010                       const char *oldname)
1011 {
1012         int err;
1013
1014         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd,dir=%lu/%u(%p),target=%.*s\n",
1015                dentry, dir->i_ino, dir->i_generation,
1016                dir, 3000, oldname);
1017
1018         err = ll_new_node(dir, dentry, oldname, S_IFLNK | S_IRWXUGO,
1019                         0, LUSTRE_OPC_SYMLINK);
1020
1021         if (!err)
1022                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1023
1024         return err;
1025 }
1026
1027 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1028                    struct dentry *new_dentry)
1029 {
1030         struct inode *src = d_inode(old_dentry);
1031         struct ll_sb_info *sbi = ll_i2sbi(dir);
1032         struct ptlrpc_request *request = NULL;
1033         struct md_op_data *op_data;
1034         int err;
1035
1036         CDEBUG(D_VFSTRACE,
1037                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%pd\n",
1038                src->i_ino, src->i_generation, src, dir->i_ino,
1039                dir->i_generation, dir, new_dentry);
1040
1041         op_data = ll_prep_md_op_data(NULL, src, dir, new_dentry->d_name.name,
1042                                      new_dentry->d_name.len,
1043                                      0, LUSTRE_OPC_ANY, NULL);
1044         if (IS_ERR(op_data))
1045                 return PTR_ERR(op_data);
1046
1047         err = md_link(sbi->ll_md_exp, op_data, &request);
1048         ll_finish_md_op_data(op_data);
1049         if (err)
1050                 goto out;
1051
1052         ll_update_times(request, dir);
1053         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1054 out:
1055         ptlrpc_req_finished(request);
1056         return err;
1057 }
1058
1059 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1060                      struct inode *new_dir, struct dentry *new_dentry)
1061 {
1062         struct ptlrpc_request *request = NULL;
1063         struct ll_sb_info *sbi = ll_i2sbi(old_dir);
1064         struct md_op_data *op_data;
1065         int err;
1066
1067         CDEBUG(D_VFSTRACE,
1068                "VFS Op:oldname=%pd,src_dir=%lu/%u(%p),newname=%pd,tgt_dir=%lu/%u(%p)\n",
1069                old_dentry, old_dir->i_ino, old_dir->i_generation, old_dir,
1070                new_dentry, new_dir->i_ino, new_dir->i_generation, new_dir);
1071
1072         op_data = ll_prep_md_op_data(NULL, old_dir, new_dir, NULL, 0, 0,
1073                                      LUSTRE_OPC_ANY, NULL);
1074         if (IS_ERR(op_data))
1075                 return PTR_ERR(op_data);
1076
1077         ll_get_child_fid(old_dentry, &op_data->op_fid3);
1078         ll_get_child_fid(new_dentry, &op_data->op_fid4);
1079         err = md_rename(sbi->ll_md_exp, op_data,
1080                         old_dentry->d_name.name,
1081                         old_dentry->d_name.len,
1082                         new_dentry->d_name.name,
1083                         new_dentry->d_name.len, &request);
1084         ll_finish_md_op_data(op_data);
1085         if (!err) {
1086                 ll_update_times(request, old_dir);
1087                 ll_update_times(request, new_dir);
1088                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1089                 err = ll_objects_destroy(request, old_dir);
1090         }
1091
1092         ptlrpc_req_finished(request);
1093         if (!err)
1094                 d_move(old_dentry, new_dentry);
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 };